Stella Mook T3-108

8/10/2019 Stella Mook T3-108

http://slidepdf.com/reader/full/stella-mook-t3-108 1/248

Prognostic factors in

breast cancer

One fits all?

Stella Mook

8/10/2019 Stella Mook T3-108


Prognostic Factors in Breast Cancer

One fits all?

Stella Mook

8/10/2019 Stella Mook T3-108


Cover: Jantien Mook – www.jantienmook.nl

Layout: Gildeprint Drukkerijen – www.gildeprint.nl

Printed by: Gildeprint Drukkerijen – www.gildeprint.nl

ISBN: 978-94-6108-151-3

Online: http:// dare.uva.nl/document

The work described in this thesis was performed at the Netherlands Cancer Institute-Antoni

van Leeuwenhoek Hospital, Amsterdam, the Netherlands.

Financial support provided by:

Netherlands Cancer Institute, Academic Medical Center, Agendia BV, AstraZeneca,

Boehringer Ingelheim, GlaxoSmithKline, Hoofdredactie OncoMotief.nl - EURIN BV, Novartis,

Roche, Sanofi-Aventis.

© 2011 Stella Mook, Amsterdam, the Netherlands

8/10/2019 Stella Mook T3-108


Prognostic Factors in Breast Cancer

One fits all?

ACADEMISCH PROEFSCHRIFT

ter verkrijging van de graad van doctor

aan de Universiteit van Amsterdam

op gezag van de Rector Magnificus

prof. dr. D.C. van den Boom

ten overstaan van een door het college voor promoties

ingestelde commissie,in het openbaar te verdedigen in de Agnietenkapel

op donderdag 21 april 2011, te 12:00 uur

door

Stella Mook

geboren te Hoorn

8/10/2019 Stella Mook T3-108


Promotiecommissie

Promotores: Prof. dr. E.J.Th. Rutgers

Prof. dr. L.J. Van ‘t Veer

Overige Leden: Prof. dr. R. Bernards

Prof. dr. J.W. Coeberg

Prof. dr. C.C.E. Koning

Prof. dr. J.W.R. Nortier

Prof. dr. S. Rodenhuis

Prof. dr. M.J. van de Vijver

Dr. J.H.G. Klinkenbijl

Faculteit der Geneeskunde

8/10/2019 Stella Mook T3-108


Voor Marcel & mijn ouders

8/10/2019 Stella Mook T3-108


Table of contents

Chapter 1 Introduction and outline 11

Chapter 2 Individualization of therapy using Mammaprint: 23

from development to the MINDACT Trial.

Cancer Genomics Proteomics2007; 4: 147-155.

Chapter 3 Daily clinical practice of fresh tumour tissue freezing and 41

gene expression profiling; logistics pilot study preceding

the MINDACT trial.

Eur J Cancer 2009; 45: 1201-1208.

Chapter 4 The 70-gene prognosis signature predicts early metastasis in 57

breast cancer patients between 55 and 70 years of age.

Ann Oncol 2010; 21: 717-722.

Chapter 5 The 70-gene prognosis-signature predicts disease outcome 73

in breast cancer patients with 1-3 positive lymph nodes in an

independent validation study.

Breast Cancer Res Treat 2009; 116: 295-302.

Chapter 6 Metastatic potential of T1 breast cancer can be predicted by 97

the 70-gene MammaPrint signature.

Ann Surg Oncol 2010; 17: 1406-1413.

Chapter 7 The predictive value of the 70-gene signature for adjuvant 117

chemotherapy in early breast cancer


Chapter 8 Calibration and discriminatory accuracy of prognosis calculation 133

for breast cancer with the online Adjuvant! program:

a hospital-based retrospective cohort study

Lancet Oncol 2009; 10: 1070-1076.

Chapter 9 Independent prognostic value of screen detection in 157

invasive breast cancer

JNCI accepted for publication

8/10/2019 Stella Mook T3-108


Chapter 10 General discussion and future prospects 197

Chapter 11 Summary 215

Chapter 12 Nederlandse samenvatting 223

List of publications 233

Dankwoord 237

Curriculum vitae 245

Appendix Gene signature evaluation as a prognostic tool: 249

challenges in the design of the MINDACT trial.

Nat Clin Pract Oncol 2006; 3: 540-551.

8/10/2019 Stella Mook T3-108


Chapter 1

Introduction and outline

8/10/2019 Stella Mook T3-108


Chapter 1

12

8/10/2019 Stella Mook T3-108



13

1Introduction

Breast cancer

Breast cancer is the most frequently diagnosed malignancy in women worldwide. In the

Netherlands in 2008, 13,005 women were diagnosed with invasive breast cancer and 3,327

patients died of the disease.1 Although there is an increase in breast cancer incidence,

breast cancer mortality is decreasing in the last decennia. 2-4 This decrease in mortality is

mainly caused by both the introduction of breast cancer screening and the improvement

and more extensive use of adjuvant systemic therapy.2,3,5-9 Currently, approximately 2/3

of the patients who are diagnosed with breast cancer do not have nodal involvement at

diagnosis and about 2/3 of the patients are 55 years of age or older at diagnosis.10

Treatment of breast cancer

The treatment of early stage breast cancer consists of two aspects. The first is loco-regional

control, which is primarily achieved by surgery with or without radiotherapy. The second part

of breast cancer treatment focuses on preventing the development of distant metastases.

Distant metastases account for the majority of breast cancer deaths and are thought to

develop from undetectable micrometastases or circulating tumor cells that are already

present at time of diagnosis. Adjuvant systemic therapy (i.e. chemotherapy, hormonal

therapy and/or targeted therapy) can help eradicate micrometastases and circulatingtumor cells, thereby preventing distant metastases to occur and thus improving survival.

The incurable nature of metastatic breast cancer emphasizes the importance of selecting

patients for adjuvant systemic therapy who are at risk of developing distant metastases. In

patients with lymph node-negative disease, adjuvant chemotherapy improves survival on

average by 25%.11 On the other hand, especially chemotherapy can cause a wide range of

acute and long-term side effects.12

Adjuvant systemic therapy

Since the introduction in the early 1980s, there is a steady increase in the use of adjuvant

systemic therapy (AST) in the Netherlands.9 This increase is supported by data from the

Early Breast Cancer Trialists’ Collaborative Group (EBCTCG) overviews showing a significant

benefit of adjuvant systemic therapy for disease-free and overall survival. 11,13-15 In the

1990s, adjuvant systemic therapy was recommended mainly for patients with lymph

node-positive breast cancer. In 2000, the National Breast Cancer Consultation Netherlands

(NABON) developed the first national guideline for adjuvant systemic therapy.16 Tamoxifen

was recommended for lymph node-positive, estrogen receptor (ER)-positive tumors

in postmenopausal patients. For lymph node-positive premenopausal patients and

8/10/2019 Stella Mook T3-108


Chapter 1

14

for lymph node-positive postmenopausal patients with an estrogen receptor negative

tumor chemotherapy was recommended. In addition, it was recommended to consider

adjuvant systemic therapy for a subgroup of patients with lymph node-negative tumors

depending on tumor size and tumor grade. The use of AST for Dutch patients with early

stage breast cancer increased significantly over time, from 37% in the period 1990–1997, to

53% in 2002–2006.17,18 Currently, Dutch breast cancer patients are treated according to the

NABON and Dutch Institute for Healthcare Improvement (CBO) guidelines and adjuvant

systemic therapy is recommended for > 80% of all patients.19 As in the Netherlands, the

administration of AST increased substantially in the US, were the use of chemotherapy or

hormonal therapy tripled from 1987 to 2000 in women with node-negative disease.20 Only

1 in 5 women with node negative disease did not receive any form of adjuvant systemic

therapy in the US in the year 2000.20

Who to treat; prognostic factors

Patients who are at high risk of developing distant metastases are candidates for AST.

Prognostic factors help identify patients who are at high risk of distant metastases in

the absence of AST.21 An ideal prognostic factor tells us exactly ‘who to treat’, by reliably

distinguishing patients who are at high risk of developing distant metastases from those

who are at low risk. Nowadays, the selection of patients who are at high risk of recurrence

is based on clinical and pathological prognostic factors, such as age, menopausal status,

co-morbidity, tumor size, tumor grade, lymph node status and hormonal receptor status.22 These clinicopathological criteria are often combined into guidelines or models such as the

St. Gallen recommendations, the Nottingham Prognostic Index, the Dutch CBO guideline or

the Adjuvant! tool.19,23-25 However, tumors with the same clinicopathological characteristics

can have strikingly different outcomes. Consequently, AST recommendation according

to these guidelines is far from accurate. Although 60-70% of patients with lymph node-

negative breast cancer are likely to be cured by surgery and radiotherapy alone, the majority

of patients is currently treated with chemotherapy, hormonal therapy and/or targeted

therapy (Figure 1).11 As a result, a substantial proportion of patients will unnecessarily receive

AST and will be needlessly exposed to its toxicity. This overtreatment is due to the lack ofaccurate identification of patients with a low risk of developing distant metastases, who are

unlikely to benefit from adjuvant systemic therapy. Apparently, better prognostic factors

are urgently needed. Although, a number of single parameter prognostic biomarkers have

been studied, few have achieved the level of supporting evidence required for routine

clinical use.26 One of the more recently developed techniques that provides us with

promising new prognostic tools is the microarray gene expression technique.

8/10/2019 Stella Mook T3-108



15

1

Figure 1. Survival of early stage breast cancer patients after loco-regional treatment.

Gene expression profiling

The introduction of the new high-throughput microarray technology at the beginning of

this century, has introduced a new era of multi-parameter prognostic tests and causeda revolution in medicine, particularly in the oncology field. 27 In contrast to the single-

parameter biomarker, microarray analyses can measure the expression of thousands of

genes in the tumor simultaneously.28-30 The expression level of all genes together gives

insight in tumor biology and in this way provides the possibility to subdivide breast cancer

based on its biology. Since tumor behavior and clinical outcome depend largely on tumor

biology, gene expression profiles are anticipated to refine the prognostication of breast

cancer.

The first published molecular classification of breast cancer using microarray technology

displayed the molecular heterogeneity of the disease. Unsupervised analyses of microarraygene expression data of breast cancer patients have resulted in the identification of 4

molecular subtypes, according to gene expression profile: Luminal A, Luminal B, Basal-like

and ERBB2 breast cancers.31 Those gene expression profiles reflect biological diversity and

were shown to be associated with disease outcome as well.31,32 Many subsequent studies

have discovered several other prognostic gene expression profiles.31,33-40 Remarkably,

although the prognostic performance of these signatures in terms of individual patient

classification was similar, overlap in terms of gene identity was limited.41 However, it was

shown that these signatures reflect overlapping common biological processes and cellular

phenotypes that drive breast cancer prognosis.42,43

~30% die of breast canceradjuvant therapy can be beneficial

~70% survive breast canceradjuvant therapy is not beneficial

8/10/2019 Stella Mook T3-108


Chapter 1

16

In addition to unsupervised analyses, supervised analyses can be used to develop a

gene expression signature that can predict clinical outcome. In contrast to unsupervised

analyses that classify tumors based on the similarity of gene expression, supervised

analyses compares gene-expression data from patients with known clinical outcomes

(e.g. absence or presence of distant metastases) to identify genes that are associated

with prognosis. Such classification method was used to identify the 70-gene prognosis

signature (MammaPrint™).38 The 70-gene signature has been identified using frozen tumor

samples from 78 patients who were diagnosed at the Netherlands Cancer Institute-Antoni

van Leeuwenhoek hospital (NKI-AVL) with lymph node-negative breast cancer and who

were up to 55 years of age at diagnosis. Among these 78 patients, 44 remained free of

distant metastases for at least 5 years (defined as the good prognosis group), whereas 34

patients developed distant metastases within 5 years of diagnosis (poor prognosis group).

The signature consists of the top 70 genes that were differentially expressed between

the two prognosis groups and most accurately classified tumors in the good- or poor

prognosis group. The signature was validated in a consecutive second patient series from

the NKI-AVL, consisting of 151 lymph node-negative and 144 lymph node-positive patients

up to 53 years at diagnosis, and in a third independent patient series of 302 lymph node-

negative breast cancer patients from 5 European hospitals, who were up to 60 years of

age at diagnosis.44,45 Subsequently, the prognostic value of 70-gene signature has been

confirmed by others.46-48

In 2004, another prognostic test has been developed. The OncotypeDX™ is a RT-PCR based

assay performed on paraffin-embedded tumor samples that classifies tumors based on theexpression of 16 genes into a low Recurrence Score (RS), an intermediate RS or a high RS.36

A community-based validation study demonstrated that the RS could be used to predict

the outcome of node-negative patients receiving tamoxifen alone.49 Retrospective analysis

of the node-negative NSABP B20 and node-positive SWOG 8814 trial showed similar

prognostic value for the RS in patients treated with the combination of tamoxifen and

chemotherapy.50,51

Validation studies to assess the reliability and reproducibility are of utmost importance to

determine a signature’s clinical utility. Furthermore, practical issues of the implementation

of gene expression microarrays need to be addressed and quality of performance andstandardized procedures for a diagnostic test should be monitored by International

Organization for Standardization (ISO) or Clinical Laboratory Improvement Amendments

(CLIA) certification and should preferably fall under the regulatory oversight such as the

US Food and Drug Administration (FDA).52 Successful implementation of a gene expression

profile requires, in addition to thorough validation studies, the collection of good quality,

fresh frozen tumor tissue and close collaboration between different departments in the

hospital.

8/10/2019 Stella Mook T3-108



17

1Rationale and outline of this thesis

The overall aim of this thesis is to evaluate the accuracy and clinical utility of a relatively new

prognostic microarray test, the 70-gene signature, in several breast cancer subpopulations.

In addition, we evaluated the accuracy of the extensively used prognostic tool Adjuvant!,

which is based on clinicopathological characteristics. Finally, we evaluated whether the

method of detection of a tumor (i.e. screen-detected or symptomatic) affect prognosis and

should be taken into account to improve patient selection for AST.

The first part of this thesis focuses on the applicability of the 70-gene signature

(MammaPrint™) and the potential improvement of patient selection for adjuvant systemic

therapy by using this microarray test.

In chapter 2 the development of the 70-gene signature, its initial retrospective validation

studies and logistical feasibility studies are described. In addition, the currently conducted

prospective randomized clinical trial, the so-called MINDACT study (Microarray In Node-

negative and 1-3 positive lymph node disease may Avoid ChemoTherapy) which will

compare the prognostic value of the 70-gene signature with that of currently available

prognostic clinicopathological variables, is discussed. More detailed information about the

design of the MINDACT trial is provided in Appendix 1.

Chapter 3 presents the results of a European pilot study preceding the MINDACT trial to

test the feasibility and to optimize the logistics for the collection of good-quality fresh

frozen tumor tissue in order to perform the 70-gene signature. The 70-gene signature has been developed and so far mostly validated in premenopausal

patients with lymph node-negative breast cancer. However, the majority of breast

carcinomas is diagnosed in postmenopausal women. Therefore, we evaluated the accuracy

of the 70-gene signature in postmenopausal patients, which is described in chapter

4. Although lymph node metastases are a strong indicator of a poor prognosis, still

approximately 30-40% of patients with 1-3 positive lymph nodes at diagnosis will remain

free of distant metastases without adjuvant systemic therapy. Currently, there are no

biomarkers available to select these low risk lymph node-positive patients. In chapter 5 we

evaluated the ability of the 70-gene signature to identify patients with 1-3 positive lymphnodes who are at low risk of recurrence in an independent, retrospective validation study.

In addition to lymph node status, tumor size is known to be a powerful prognostic factor,

with small tumor size being thought to indicate a low risk of recurrence. Nevertheless, small

tumors still can metastasize, which leaves us with the question of the necessity of adjuvant

systemic therapy in patients with pT1 (≤20mm) tumors. In Chapter 6 the prognostic value

and clinical utility of the 70-gene signature in a pooled retrospective series of patients with

pT1 (≤20mm) breast carcinomas are discussed. Adjuvant treatment allocation based on

the 70-gene signature seems to be justified when the low risk of recurrence in the good

prognosis group is sufficiently low to withhold chemotherapy and the expected benefit

8/10/2019 Stella Mook T3-108


Chapter 1

18

from adjuvant chemotherapy is limited. In addition, administration of chemotherapy in

patients classified as high risk is legitimate when the benefit of treatment in these patients

is substantial. In chapter 7 we assessed this predictive value of the 70-gene signature in a

pooled analysis.

The 70-gene signature is currently studied in the prospective MINDACT (Microarray In Node-

negative and 1-3 positive lymph node disease may Avoid ChemoTherapy) trial, which will

evaluate whether patients who are considered high risk according to the currently available

prognostic tool Adjuvant! but classified as low risk by the 70-gene signature can be safely

spared chemotherapy. Adjuvant! combines clinicopathological characteristics, such as

patient age, co-morbidity, tumor size, lymph node involvement, histological grade and

estrogen receptor status, to forecast the overall and breast cancer-specific mortality and to

predict the benefit of additional chemotherapy and/or endocrine therapy. The Adjuvant!

model is based on information from breast cancer patients in the United States who were

diagnosed between 1988 and 1992 and recorded in the Surveillance, Epidemiology and

End Results (SEER) registry.24 In 2005, the model was retrospectively validated in breast

cancer patients from British Columbia.53 Since the European breast cancer populations

may differ from those in the US and Canada, the question remains whether outcome

predictions of the Adjuvant! model are applicable to the European population. Therefore,

we conducted a retrospective validation study to test the accuracy of Adjuvant! in a Dutch

breast cancer cohort of 5,830 patients, which is described in chapter 8. The aim of this

study was to assess both the ability of Adjuvant! to predict outcomes in (sub)groups ofDutch breast cancer patients (calibration) and its ability to distinguish individuals who will

experience different outcomes (discriminatory accuracy).

Awaiting the incorporation of gene expression profiles in prognostic tools, models such as

Adjuvant! are still suboptimal. Incorporation of other prognostic markers may also improve

these tools. It has been shown that mammographic screening detects breast cancer at an

earlier stage.54-56 Therefore, we investigated whether method of detection has additional

prognostic value that could improve the estimation of disease outcome, assuming that

screen-detected carcinomas are of a different tumor biology. This question is addressed inchapter 9, where we studied the accuracy of Adjuvant! in patients with a screen-detected

carcinoma as well as assessed the independent prognostic value of screen-detection in a

retrospective patient cohort.

This thesis ends with concluding remarks and future prospects in chapter 10 and a

summary of the results presented in chapter 11.

8/10/2019 Stella Mook T3-108



19

1Reference List

1. http://www.ikcnet.nl. Last accessed January 12, 2011.

2. Jatoi I and Miller AB. Why is breast-cancer mortality declining? Lancet Oncol 2003; 4: 251-254.

3. Smigal C, Jemal A, Ward E, et al. Trends in breast cancer by race and ethnicity: update 2006. CA Cancer J Clin

2006; 56: 168-183.

4. http://www.cancer.org. Cancer Facts & Figures 2010. Last accessed January 15, 2011.

5. Berry DA and Ravdin PM. Breast cancer trends: a marriage between clinical trial evidence and

epidemiology. J Natl Cancer Inst 2007; 99: 1139-1141.

6. Chu KC, Tarone RE, Kessler LG, et al. Recent trends in U.S. breast cancer incidence, survival, and mortality

rates. J Natl Cancer Inst 1996; 88: 1571-1579.

7. Garne JP, Aspegren K, Balldin G, Ranstam J. Increasing incidence of and declining mortality from breast

carcinoma. Trends in Malmo, Sweden, 1961-1992. Cancer 1997; 79: 69-74.

8. Otten JD, Broeders MJ, Fracheboud J, et al . Impressive time-related influence of the Dutch screening

programme on breast cancer incidence and mortality, 1975-2006. Int J Cancer 2008; 123: 1929-1934.

9. Vervoort MM, Draisma G, Fracheboud J, van de Poll-Franse LV, de Koning HJ. Trends in the usage of

adjuvant systemic therapy for breast cancer in the Netherlands and its effect on mortality. Br J Cancer

2004; 91: 242-247.

10. Ries LAG, Melbert D, Krapcho M et al. SEER Cancer Statistics Review, 1975-2005. Bethesda, MD: National

Cancer Institute 2008; http://seer.cancer.gov/csr/1975_2005/, based on November 2007 SEER data

submission, posted to the SEER web site. Last accessed 28 November, 2008.

11. Early Breast Cancer Trialists’ Collaborative Group. Effects of chemotherapy and hormonal therapy for

early breast cancer on recurrence and 15-year survival: an overview of the randomised trials. Lancet

2005; 365: 1687-1717.

12. Eifel P, Axelson JA, Costa J, et al . National Institutes of Health consensus development conference

statement: Adjuvant therapy for breast cancer, November 1-3, 2000. J Natl Cancer Inst 2001; 93: 979-989.

13. Clarke M, Coates AS, Darby SC, et al . Adjuvant chemotherapy in oestrogen-receptor-poor breast cancer:

patient-level meta-analysis of randomised trials. Lancet 2008; 371: 29-40.

14. Early Breast Cancer Trialists’ Collaborative Group. Tamoxifen for early breast cancer: an overview of the

randomised trials. Lancet 1998; 351: 1451-1467.

15. Early Breast Cancer Trialists’ Collaborative Group. Polychemotherapy for early breast cancer: anoverview of the randomised trials. Lancet 1998; 352: 930-942.

16. Bontenbal M, Nortier JW, Beex LV, et al . [Adjuvant systemic therapy for patients with resectable breast

cancer: guideline from the Dutch National Breast Cancer Platform and the Dutch Society for Medical

Oncology]. Ned Tijdschr Geneeskd 2000; 144: 984-989.

17. Van Steenbergen LN, Van de Poll-Franse LV, Wouters MW, et al . Variation in management of early breast

cancer in the Netherlands, 2003-2006. Eur J Surg Oncol 2010; 36 Suppl 1: S36-S43.

18. Sukel MP, Van de Poll-Franse LV, Nieuwenhuijzen GA, et al . Substantial increase in the use of adjuvant

systemic treatment for early stage breast cancer reflects changes in guidelines in the period 1990-2006

in the southeastern Netherlands. Eur J Cancer 2008; 44: 1846-1854.

8/10/2019 Stella Mook T3-108


Chapter 1

20

19. CBO Richtlijn Mammacarcinoom 2008. http://www.cbo.nl/product/richtlijnen/folder20021023121843/rl_mamma_08.pdf

20. Harlan LC, Clegg LX, Abrams J, Stevens JL, Ballard-Barbash R. Community-based use of chemotherapy

and hormonal therapy for early-stage breast cancer: 1987-2000. J Clin Oncol 2006; 24: 872-877.

21. Weigelt B, Peterse JL, van ‘t Veer LJ. Breast cancer metastasis: markers and models. Nat Rev Cancer 2005; 5:

591-602.

22. Harris JR. Staging of Breast Cancer. 2009; Fourth edition: 489-500.

23. Goldhirsch A, Ingle JN, Gelber RD, et al . Thresholds for therapies: highlights of the St. Gallen International

Expert Consensus on the primary therapy of early breast cancer 2009. Ann Oncol 2009; 20: 1319–1329.

24. Ravdin PM, Siminoff LA, Davis GJ, et al. Computer program to assist in making decisions about adjuvant

therapy for women with early breast cancer. J Clin Oncol 2001; 19: 980-991.

25. Todd JH, Dowle C, Williams MR, et al. Confirmation of a prognostic index in primary breast cancer. Br J

Cancer 1987; 56: 489-492.

26. Harris L, Fritsche H, Mennel R, et al. American Society of Clinical Oncology 2007 update of

recommendations for the use of tumor markers in breast cancer. J Clin Oncol 2007; 25: 5287-5312.

27. Schena M. Microarray Analysis. New York, Wiley-Liss. First edition 2002.

28. Brown PO and Botstein D. Exploring the new world of the genome with DNA microarrays. Nat Genet 1999;

21: 33-37.

29. DeRisi J, Penland L, Brown PO, et al. Use of a cDNA microarray to analyse gene expression patterns in

human cancer. Nat Genet 1996; 14: 457-460.

30. Schena M, Shalon D, Davis RW, Brown PO. Quantitative monitoring of gene expression patterns with a

complementary DNA microarray. Science 1995; 270: 467-470.

31. Perou CM, Sorlie T, Eisen MB, et al. Molecular portraits of human breast tumours. Nature 2000; 406: 747-

752.

32. Sorlie T, Perou CM, Tibshirani R, et al . Gene expression patterns of breast carcinomas distinguish tumor

subclasses with clinical implications. Proc Natl Acad Sci USA 2001; 98: 10869-10874.

33. Chang HY, Nuyten DSA, Sneddon JB, et al . Robustness, scalability, and integration of a wound-response

gene expression signature in predicting breast cancer survival. Proc Natl Acad Sci USA 2005; 102: 3738-3743.

34. Huang E, Cheng SH, Dressman H, et al . Gene expression predictors of breast cancer outcomes. Lancet

2003; 361: 1590-1596..

35. Ma XJ, Hilsenbeck SG, Wang W, et al. The HOXB13:IL17BR expression index is a prognostic factor in early-

stage breast cancer. J Clin Oncol 2006; 24: 4611-4619.36. Paik S, Shak S, Tang G, et al. A multigene assay to predict recurrence of tamoxifen-treated, node-negative

breast cancer. N Engl J Med 2004; 351: 2817-2826.

37. Sotiriou C, Neo SY, McShane LM, et al . Breast cancer classification and prognosis based on gene

expression profiles from a population-based study. Proc Natl Acad Sci USA 2003; 100: 10393-10398.

38. Van ‘t Veer LJ, Dai H, Van de Vijver MJ, et al . Gene expression profiling predicts clinical outcome of breast

cancer. Nature 2002; 415: 530-536.

39. Wang Y, Klijn JG, Zhang Y, et al. Gene-expression profiles to predict distant metastasis of lymph-node-

negative primary breast cancer. Lancet 2005; 365: 671-679.

8/10/2019 Stella Mook T3-108



21

140. West M, Blanchette C, Dressman H, et al . Predicting the clinical status of human breast cancer by using

gene expression profiles. Proc Natl Acad Sci USA 2001; 98: 11462-11467.

41. Fan C, Oh DS, Wessels L, et al. Concordance among gene-expression-based predictors for breast cancer.

N Engl J Med 2006; 355: 560-569.

42. Loi S, Haibe-Kains B, Desmedt C, et al. Definition of clinically distinct molecular subtypes in estrogen

receptor-positive breast carcinomas through genomic grade. J Clin Oncol 2007; 25: 1239-1246.

43. Yu JX, Sieuwerts AM, Zhang Y, et al. Pathway analysis of gene signatures predicting metastasis of node-

negative primary breast cancer. BMC Cancer 2007; 7: 182.

44. Buyse M, Loi S, Van ‘t Veer L, et al . Validation and clinical utility of a 70-gene prognostic signature for

women with node-negative breast cancer. J Natl Cancer Inst 2006; 98: 1183-1192.

45. Van de Vijver MJ, He YD, Van ‘t Veer LJ, et al . A gene-expression signature as a predictor of survival in


46. Bueno de Mesquita JM, Linn SC, Keijzer R et al. Validation of 70-gene prognosis signature in node-

negative breast cancer. Breast Cancer Res Treat 2009; 117: 483–495.

47. Ishitobi M, Goranova TE, Komoike Y, et al. Clinical utility of the 70-gene MammaPrint profile in a Japanese

population. Jpn J Clin Oncol 2010; 40: 508-512.

48. Wittner BS, Sgroi DC, Ryan PD, et al. Analysis of the MammaPrint breast cancer assay in a predominantly

postmenopausal cohort. Clin Cancer Res 2008; 14: 2988-2993.

49. Habel LA, Shak S, Jacobs MK, et al. A population-based study of tumor gene expression and risk of breast

cancer death among lymph node-negative patients. Breast Cancer Res 2006; 8: R25.

50. Paik S, Tang G, Shak S, et al. Gene expression and benefit of chemotherapy in women with node-

negative, estrogen receptor-positive breast cancer. J Clin Oncol 2006; 24: 3726-3734.

51. Albain KS, Barlow WE, Shak S, et al. Prognostic and predictive value of the 21-gene recurrence score

assay in postmenopausal women with node-positive, oestrogen-receptor-positive breast cancer on

chemotherapy: a retrospective analysis of a randomised trial. Lancet Oncol 2010; 11: 55-65.

52. Couzin J. Diagnostics: Amid debate, gene-based cancer test approved. Science 2007; 315: 924.

53. Olivotto IA, Bajdik CD, Ravdin PM, et al. Population-based validation of the prognostic model ADJUVANT!

for early breast cancer. J Clin Oncol 2005; 23: 2716-2725.

54. Chu KC, Smart CR, Tarone RE. Analysis of breast cancer mortality and stage distribution by age for the

Health Insurance Plan clinical trial. J Natl Cancer Inst 1988; 80: 1125-1132.

55. Connor RJ, Chu KC, Smart CR. Stage-shift cancer screening model. J Clin Epidemiol 1989; 42: 1083-1095.56. Fracheboud J, Otto SJ, van Dijck JA, et al . Decreased rates of advanced breast cancer due to

mammography screening in The Netherlands. Br J Cancer 2004; 91: 861-867.

8/10/2019 Stella Mook T3-108


Chapter 2

Individualization of therapy using

Mammaprint: from development to the

MINDACT Trial

Stella Mook

Laura J. Van ‘t Veer

Emiel J. Th. Rutgers

Martine J. Piccart-Gebhart

Fatima Cardoso

Cancer Genomics Proteomics 2007; 4: 147-155.

8/10/2019 Stella Mook T3-108


Chapter 2

24

Abstract

To date, most treatment decisions for adjuvant chemotherapy in breast cancer are based

on conventional clinicopathological criteria. Since breast cancer tumors with similar

clinicopathological characteristics can have strikingly different outcomes, the current

selection for adjuvant chemotherapy is far from accurate. Using high-throughput microarray

analysis, a 70-gene signature was identified which can accurately select early stage breast

cancer patients who are highly likely to develop distant metastases, and therefore, may

benefit the most from adjuvant chemotherapy. This review describes the development of

the 70-gene profile (MammaPrint™), its retrospective validation and feasibility studies, and

its prospective validation in the large adjuvant MINDACT (Microarray In Node-negative

Disease may Avoid ChemoTherapy) clinical trial.

8/10/2019 Stella Mook T3-108


Development of the MammaPrint

25

2

Introduction

The outcome of patients with breast cancer has improved in the last 20 years, due to

both early detection and the amelioration of adjuvant systemic treatment. The majority

of early stage breast cancer patients receive adjuvant systemic treatment, which may

include chemotherapy, hormonal therapy, immunotherapy or a combination. Nowadays,

patients who should receive chemotherapy are selected by using consensus guidelines like

the St. Gallen, or the National Comprehensive Cancer Network (NCCN) guidelines which

are based on the assessment of clinicopathological criteria such as age, tumor size and

grade, hormonal receptor status and axillary lymph node involvement.1-3 However, breast

cancer tumors with similar clinicopathological characteristics can have strikingly different

outcomes, reflecting the heterogeneity of the disease. Consequently, the current adjuvant

treatment decision-making process for breast cancer patients is far from accurate. The

majority of early stage breast cancer patients, particularly those with lymph node-negative

disease (60-70%), has a fairly good 10-year overall survival with locoregional treatment

alone, with only 30-40% developing distant metastases.4 Notwithstanding these facts,

most lymph node-negative breast cancer patients are offered chemotherapy, according to

the currently used guidelines, causing an important proportion of overtreatment.1-3 This is

justified largely by our inability to clearly identify those patients who will not relapse and

hence do not need adjuvant chemotherapy. Since metastatic breast cancer is an incurable

disease, the only chance for cure is in the adjuvant setting. However, overtreatment not

only unnecessarily exposes women to potential toxicity and side-effects of this treatment,but also increases the economic burden of breast cancer on society. It is thus quite clear

that robust and reliable prognostic markers to accurately select patients not requiring

aggressive adjuvant therapy are urgently needed.

With the introduction of new high-throughput methods, such as gene expression

microarray technologies, the expression level of tens of thousands of genes can be measured

simultaneously. Using microarray techniques, several studies have recently classified

breast tumors according to their gene expression profile and identified prognostic and

predictive classifiers.5-14 Although these studies appear to be very promising, microarray

analysis has some potential pitfalls. For example, the analysis of the large amount of dataobtained through this technology can cause process errors and overfitting. Furthermore,

retrospective studies using frozen tissue processed and stored many years ago could result

in different levels of gene expression due to differences in tissue handling and pertain to

patient populations which may be different from those diagnosed today. Taking all this into

account, validation studies, particularly prospective ones, are indispensable in assessing

the reliability and reproducibility of the results and in identifying the true benefit of a

classifier for clinical practice.

Here we provide an overview of the development of the 70-gene profile (MammaPrint™)

from discovery to application in clinical trials, including retrospective validation and

8/10/2019 Stella Mook T3-108


Chapter 2

26

feasibility studies, and its prospective validation in the large MINDACT (Microarray In Node-

negative Disease may Avoid ChemoTherapy) clinical trial.

The Development of the 70-gene Signature

By using gene expression profiling, Van ’t Veer and colleagues developed a 70-gene

classifier that accurately distinguished breast cancer patients who were likely to remain free

of distant metastases (good profile) from breast cancer patients at high risk of developing

distant metastases (poor profile).8 To develop this 70-gene profile, 78 tumors from women

with lymph node-negative breast cancer were studied. Patients were under 55 years of

age at diagnosis, had a primary invasive breast carcinoma less than 5 cm in diameter,

no previous malignancies and were treated at The Netherlands Cancer Institute (NKI).

All patients were treated by modified radical mastectomy or breast conserving therapy.

Five out of 78 patients received adjuvant systemic treatment, consisting of chemotherapy

(n = 3) or hormonal therapy (n = 2); all 5 patients developed distant metastases within

5 years of diagnosis. Forty-four patients remained free of distant metastases for at least

5 years (good-prognosis group), whereas the remaining 34 patients did develop distant

metastases within 5 years of diagnosis (poor-prognosis group). The mean follow-up of the

good prognosis group was 8.7 years, the mean time to distant metastases was 2.5 years.

From all 78 frozen tumor samples, the percentage of tumor cells was determined in a

hematoxylin and eosin stained section, before and after cutting sections for RNA isolation.

Only tumor samples with at least 50% tumor cells were eligible. RNA was isolated andlabeled with a fluorescent dye. An equal amount of RNA from all tumors was pooled

and provided reference RNA. Both tumor RNA and reference RNA were hybridized on an

oligonucleotide microarray platform containing approximately 25,000 genes, synthesized

by inkjet technology (produced by Agilent).15

In a first step, using a statistical analysis method called ‘supervised classification’, the

expression of 231 genes appeared to be significantly correlated with disease outcome

(distant metastases within 5 years). These 231 genes were ranked, based on their correlation

coefficient with disease outcome; the top 70 of these were shown to most accurately

classify tumors in either the good- or the poor-prognosis category.All 78 tumors were ranked according to their correlation with the average expression of the

70 genes of the patients who did not develop a distant metastasis (good-outcome patients).

Where the sensitivity was optimized by setting a threshold resulting in a misclassification

of less than 10% of patients with a poor disease outcome. Consequently, 3 out of the 34

patients with a poor disease outcome would erroneously be withheld chemotherapy

based on this new tool (9% misclassification).

This supervised classification strategy resulted in the 70-gene dichotomous risk classifier,

using the 78 tumors as a training set. To initially validate the 70-gene profile, an additional

set of 7 tumors from patients with a good clinical outcome (free from distant metastases for

8/10/2019 Stella Mook T3-108



27

2

at least 5 years after diagnosis) and 12 tumors from patients with a poor clinical outcome

(distant metastases within 5 years of diagnosis) were analyzed. The 70-gene profile

accurately predicted disease outcome in 17 out of the 19 patients, thereby confirming the

initial performance of the prognostic classifier.

Although the first results were very promising, one major comment on the development

of the 70-gene profile was the small sample size of both the training and the test sets.

Supervised analysis of a relatively small sample size, in combination with the enormous

number of parameters (genes) can result in what is called ‘overfitting’.16 Since the classifier

is developed and optimized to classify the tumors in the training set accurately, the model

will fit this training set but could predict disease outcome imprecisely in an independent

sample set. Therefore, well-designed validation studies were necessary to confirm these

earlier findings.

First Retrospective Validation Series Confirms the Prognostic Value of the 70-gene

Signature

The first validation of the 70-gene profile was performed by Van de Vijver and colleagues,

on a consecutive series of 295 breast cancer tumors; 144 tumors from lymph node-positive

and 151 tumors from lymph node-negative breast cancer patients.7 Sixty-one lymph node-

negative patients were also part of the previous series used to develop the prognostic

profile. All patients were diagnosed between 1984 and 1995 at the NKI and under the

age of 53 at diagnosis. Tumors were primary invasive breast carcinomas less than 5 cm,treated with locoregional therapy alone (56%), or in combination with adjuvant systemic

treatment (44%) consisting of chemotherapy alone (31%), hormonal therapy alone (7%),

or a combination (7%). The median follow-up was 7.8 years for the 207 patients without

metastasis as first event and the median time to distant metastases was 2.7 years. The

median follow-up among all 295 patients was 6.7 years.

For all 234 samples that were not part of the previous study, the correlation coefficient of

the average level of expression of the 70 genes with the previously established good profile

was calculated. Tumors with a correlation coefficient above the previously determined

threshold (above 0.4) were assigned to the good-profile group. For the 61 patients whowere included in the previous study, a threshold of 0.55 was used to correct for overfitting.

The profile accurately distinguished a good-prognosis group (of 115 tumors) with a 10-year

overall survival of 95% (±2.6%) from a poor-prognosis group (of 180 tumors) with a 10-

year overall survival of 55% (±4.4%). The 70-gene profile was associated with established

prognostic factors such as age, tumor grade and estrogen receptor (ER) status. Remarkably,

the 70-gene profile did not seem to depend on the lymph node status, since the 144

tumors from lymph node-positive patients were equally distributed over the poor- and

good-prognosis groups. In the multivariate analysis of the risk of distant metastases as first

event, the poor-prognosis signature, large tumor size, presence of vascular invasion and no

8/10/2019 Stella Mook T3-108


Chapter 2

28

chemotherapy treatment were the only significant independent factors for the prediction

of the likelihood of developing distant metastases. With an overall hazard ratio of 4.6, the

70-gene profile was by far the most powerful predictor of distant metastases (95% CI 2.3-

9.2; p < 0.001).

To assess the value of this new prognosticator in a clinical context, the 70-gene profile

was compared to the St. Gallen and NIH criteria used at that time. 17,18 The 70-gene profile

assigned 40% of the patients in the good-prognosis or low-risk group, compared with only

15% according to the St. Gallen consensus guidelines and 7% according to the NIH criteria.

Furthermore, patients identified as being at low risk (good prognosis) by the 70-gene profile

were more likely to remain free of distant metastases, compared with patients classified as

being at low risk according to the St. Gallen or the NIH criteria. On the other hand, patients

identified as being at high risk (poor prognosis) by the 70-gene profile had a higher risk of

developing distant metastases than the high-risk patients classified by the St. Gallen or NIH

criteria. The misclassification of patients using the clinicopathological criteria is even more

clearly perceptible when the high-risk group, according to the NIH (140 out of 151 lymph

node-negative breast cancer patients), is subdivided using the 70-gene classifier. This NIH

high-risk group includes 53 out of 140 patients with a good 70-gene prognosis and indeed

a good clinical outcome, indicating a better prediction of disease outcome when using the

70-gene profile.

In this validation series, the 70-gene profile had a high negative predictive value in all

subgroups; 97% for the new lymph node-negative patients; 96% for the lymph node-

positive patients; and 96% for all new patients, respectively. Due to the setting of thethreshold in the previous study, the profile was built to have a minimum of misclassified

patients with a poor disease outcome. Consequently, the positive predictive value was

only 38% for all new patients. Although this would still lead to overtreatment, the absolute

number of patients unnecessarily exposed to chemotherapy would still be reduced by

25-30%, compared to treatment selection based on the clinicopathological criteria, since

the total proportion of poor-prognosis patients identified by the 70-gene profile is much

smaller than the proportion of high-risk patients according to the St. Gallen or NIH criteria.

Moreover, the overall selection of patients who should receive chemotherapy and patients

who can safely be spared this treatment seems to be far more accurate.An important criticism of this first validation was that the series included 61 patients from

the study on which the classifier was established. Although this validation already showed

a significant prognostic value in patients that were not included in the previous study when

analyzed separately (OR 15.3; 95% CI 1.8-127; p = 0.003), this was further substantiated in

a recently published independent validation study performed by the TRANSBIG research

consortium.19 This independent validation also addressed the question whether the 70-

gene profile, which was developed and so far validated on a selected group of patients

(young patients with stage I or II tumors, from a single institution), could be applied to a

larger proportion of breast cancer patients.

8/10/2019 Stella Mook T3-108



29

2

Independent Multi-center Validation Established the Prognostic Value of the 70-gene

Profile

In the study from the TRANSBIG consortium, recently published in the JNCI, the 70-gene

profile was independently validated in 302 patients from 5 different European hospitals.19

Patients were up to 61 years at diagnosis, diagnosed before 1999, with a lymph node-

negative T1 or T2 breast carcinoma and had not received any adjuvant systemic therapy.

The median follow-up was 13.6 years.

The frozen tumor samples were sent to Agendia, a spin-off company of the NKI in

Amsterdam, where RNA was isolated and the microarray analysis was performed. The

samples were hybridized on the MammaPrint™, which is a custom-made microarray slide,

assessing the mRNA expression of the previously identified 70 genes in triplicate. A tumor

was classified as high risk if the correlation coefficient for the average expression of the

70-gene profile was under 0.4. Importantly, researchers at Agendia were blinded to the

clinical data while performing the genomic test. Clinical data from these patients were

collected, audited by two independent auditors and sent to an independent statistical

partner in Brussels. The researchers collecting the clinical data were blinded for the

genomic test results. Furthermore, a central pathology review was performed in Milan to

decrease the potential heterogeneity of results from different laboratories (ER status and

grade were centrally assessed in 80% of samples). Only the independent statistical office

had simultaneous access to both clinical and genomic data and performed the correlation

analysis. This independent validation confirmed that the 70-gene profile is a strong prognostic

factor for overall survival and time to distant metastases, with hazard ratios of 2.79 (95%

CI 1.60-4.87) and 2.32 (95% CI 1.35-4.0), respectively. The prognostic value of the 70-gene

profile remained statistically significant after adjustment for other risk classifications, using

clinicopathological criteria with known prognostic value, such as the St. Gallen consensus

guidelines, the Nottingham Prognostic Index and the prognostic evaluation tool Adjuvant!

Online. This last tool is a software program (www.adjuvantonline.com) which can calculate

a 10-year survival probability based on the patient’s age, co-morbidities, tumor size, grade

and ER status.20

The prognostic model is constructed using the risk estimates based on theobserved overall survival from thousands of breast cancer patients, recorded in the SEER

database, and was recently validated on more than 4000 breast cancer patients from British

Columbia.21 To distinguish a low-risk group from a high-risk group using Adjuvant!, the

TRANSBIG consortium decided the following: a low-risk group would be defined as patients

with a 10-year breast cancer survival of at least 88% for estrogen receptor (ER)-positive

patients and at least 92% for ER-negative patients. The rationale for these 2 different cutoffs

is the assumption that ER-positive patients would now all receive hormonal treatment

(with an estimated average absolute 10-year survival benefit of 4%) and patients in this

validation series were all untreated.19

8/10/2019 Stella Mook T3-108


Chapter 2

30

After adjustment for the clinical risk groups defined by Adjuvant!, the hazard ratios for

overall survival and time to distant metastases given by the 70-gene profile were 2.13

(95% CI 1.19-3.82) and 2.63 (95% CI 1.45-4.79), respectively. Moreover, patients in the

good-prognosis group according to the 70-gene profile had a 10-year survival rate of

88% and 89%, respectively, for low and high clinical risk as classified by Adjuvant!. On the

other hand, patients in the poor-prognosis group defined by the 70-gene profile had a

10-year survival rate of 69%, for both low and high clinical risk defined by Adjuvant!. These

findings suggest that the 70-gene profile predicts disease outcome independently of the

clinicopathological criteria.

The median follow-up time in the original series was less than half that of this validation

series (6.7 years versus 13.6 years, respectively). Therefore, the 70-gene profile hazard ratios

were also calculated with arbitrary censoring of all observations at different time points.

A strong time dependency of the 70-gene profile was observed, with adjusted HR of

4.68 and 16.99 at 5 years, and 3.5 and 3.46 at 10 years for time to distant metastases and

overall survival, respectively, suggesting a better prediction of early distant metastases (i.e.

occurring during the first five years) by the 70-gene profile. The different duration of follow-

up could be a plausible explanation for the discrepancy in hazard ratios between the first

validation series and this independent validation series.

The results of this independent validation strengthen the previous findings that the 70-

gene profile is a strong independent prognostic marker in early stage breast cancer, also in

patients up to the age of 61. The substantiation of the prognostic value in this independent

validation study was a prerequisite for the initiation of a large prospective validation study,the MINDACT trial.

In the meantime, two other prognostic gene expression signatures were developed, using

the Affymetrix microarray platform: the 76-gene Veridex/Rotterdam signature22 and the

Genomic Grading Index.23

To decide which signature would be the best tool to move forward with in the large,

prospective MINDACT trial, the TRANSBIG consortium performed the retrospective

validation of these two signatures using the same methodology and the same patient

population as described for the 70-gene profile. The results have shown that the

three signatures performed in quite a similar way, all being superior to the classicalclinicopathological evaluation and all possess a strong time dependency (are better

predictors of early relapse).22 Since no significant differences were seen in the performances

of the three signatures and the 70-gene profile test is robust, with a good interlaboratory

reproducibility, and available for patient diagnostic testing, even as an FDA approved test,24

the TRANSBIG consortium has decided to move forward with this tool in the MINDACT trial

(Table 1).

8/10/2019 Stella Mook T3-108



31

2

Table 1. Summary of development and validation of the 70-gene profile.

Nature paper8 NEJM paper7 TRANSBIG paper19

Purpose Development of breast

cancer prognosis 70-gene

profile

Validation of the 70-gene

profile in consecutive series

of breast cancer patients

Independent European

validation of the 70-gene

profile

Patient & tumor

characteristics

n = 78,

< 55 years,

pT1-2,

node-negative,

50% ER-positive

n = 295,

< 53 years,

pT1-2,

node-negative/node-

positive,

77% ER-positive

n = 302,

< 61 years,

pT1-2,

node-negative,

70% ER-positive

Year of diagnosis 1983-1996 1984-1995 < 1999

Adjuvantsystemic

treatment

Chemotherapy 4%hormonal therapy 3%

Chemotherapy 31%hormonal therapy 7%

both 7%

No adjuvant systemictreatment

Follow-up 8.7 years (mean) in the

good- prognosis group

6.7 years (median) 13.6 years (median)

Good profile 35 (45%) 115 (39%) 111 (37%)

Comments Multivariate OR of 18

(95% CI: 3.3-94; p = 1.4 *

10-4) for distant metastases

< 5 years

DMFS by 70-gene profile

at 10 years: poor-prognosis

profile 55% (±4.4), good-

prognosis profile 95%

(±2.6).

Multivariate HR for distant

metastases as first event 4.6

(95% CI: 2.3-9.2; p < 0.001)

(poor versus good profile)

DMFS by 70-gene profile

at 10 years: poor-prognosis

profile 69%, good-

prognosis profile 88%.

Univariate HR for overall

survival 2.79

(95% CI: 1.60-4.87; p < 0.001)

Strong time dependency

ER, estrogen receptor; OR, odds ratio; HR, hazard ratio; DMFS, distant metastases-free survival; CI, confidence

interval.

8/10/2019 Stella Mook T3-108


Chapter 2

32

Implementation of 70-gene Profile in Daily Clinical Practice Requires Adjustments to

Standard Procedures

In addition to thorough validation studies, the implementation of a new test in daily clinical

practice should also be feasible, before it can be applied in a prospective trial. One major

obstacle for the implementation of microarray techniques such as the 70-gene profile is

the requirement for good quality RNA. Since RNA is very unstable, the tumor tissue must

be preserved either by snap freezing or in a special preservation fluid based on RNAlater®

(Qiagen), rather than in paraffin. The logistics for the collection of fresh frozen tissue is

complex and varies from hospital to hospital. Therefore, performing microarray analysis,

especially on a real-time basis, can cause some logistical problems such as insufficient

freezing procedures, or transport-related issues. Thus, close collaboration between

pathologists, surgeons and oncologists is of paramount importance.

To investigate whether MammaPrint™ could be implemented in daily clinical practice,

the Netherlands Cancer Institute (with financial support from the Dutch Health Care

Insurance Board) performed a multi-center feasibility study: the RASTER study (Figure 1).

The first aim of this study was to assess the feasibility of collecting good quality tissue

from several community hospitals in the Netherlands to be used to perform the 70-gene

profile analysis.25 In this RASTER study, the so-called Constructive Technology Assessment

(CTA) was used as a tool to facilitate the introduction of the 70-gene profile in daily

clinical practice by evaluating aspects of the dynamics of its implementation, such as

communication, logistics, juridical-ethical matters and cost effectiveness.26 The results ofthese evaluations will be used in decision-making concerning the large-scale application

of the 70-gene profile in daily clinical practice and related guidelines. Other aims of the

RASTER study were to determine the proportion of good- and poor-profile patients and

to establish the concordance between the risk assessment defined by the 70-gene profile

and the one defined by the Dutch breast cancer guidelines, which are based on common

clinicopathological criteria.25

Lymph node-negative breast cancer patients under the age of 61, with a T1 or T2 tumor,

were eligible. A tumor sample from the excised specimen was obtained from all patients,

using a 6 mm biopsy punch, and placed in the commercially available preservationfluid at room temperature. Subsequently, the sample was sent by conventional mail to

the NKI, were it was frozen in liquid nitrogen and stored at -80°C until processing. The

MammaPrint™ was performed at Agendia and patients were classified into good- and poor-

prognosis groups. Preliminary results show the feasibility of collecting good quality tissue

for microarray analysis from several community hospitals. A minority of the samples were

lost due to processing errors, such as initial storage in formalin resulting in an insufficient

RNA quality.25,27 The study ended in December 2006 and the final results are expected in

due course.

8/10/2019 Stella Mook T3-108



33

2

Figure 1. RASTER study design.

From Sauter G and Simon R. Predictive Molecular Pathology. N Engl J Med 347(25): 1995-1996, 2002.Copyright © 2002 Massachusetts Medical Society. All rights reserved. Adapted with permission,

2007.

The Prospective Validation of the 70-gene Profile in a Large Randomized Clinical

Trial: the MINDACT Study

The 70-gene profile has been extensively validated in a retrospective series. Furthermore,

the logistics concerning fresh frozen tissue collection were tested and adjusted wherenecessary. The final step before the implementation of the 70-gene profile in clinical

practice is its prospective validation in the MINDACT trial. This trial is a multicentric large

prospective randomized study, comparing the 70-gene profile with currently used tools for

selecting lymph node-negative breast cancer patients for adjuvant systemic treatment. The

primary aim of the study is to show that patients defined as at low risk using the 70-gene

profile but who are at high risk according to the current clinicopathological criteria can be

safely spared chemotherapy, without deterioration of the clinical outcome. The study will

enroll 6,000 node-negative breast cancer patients who will have their risk assessed by both

the 70-gene profile and the currently used clinicopathological criteria through an updated

Good signature

Low risk

Poor signature

High risk DNA Microarray

70 genes

Dutch Health Care Insurance Board (CVZ)/ NKI

‘Raster trial’

Tumor RNA

Unfixed sample of

tumor tissue

Surgical removal oftumor tissue

Labeled tumorcDNA or cRNA

Labeled control

cDNA or cRNA

8/10/2019 Stella Mook T3-108


Chapter 2

34

version of Adjuvant! Online. We estimate that about 55% will be classified as at high risk by

both methods and these patients will be offered adjuvant chemotherapy; about 10% will be

classified as at low risk by both methods and will not be offered adjuvant chemotherapy. The

patients with a discordant risk assessment (approx. 35%), i.e. a high genomic risk (according

to the 70-gene profile) and a low clinical risk (according to Adjuvant! software) or vice versa

(low genomic risk and high clinical risk) will be randomized for the treatment decision tool.

In other words, 50% of those patients will receive adjuvant treatment according to their

genomic risk and 50% will be treated according to their clinical risk (Figure 2).

Other objectives of the study are related to the type of adjuvant systemic treatment. A

second randomization will compare an anthracycline-based regimen to a docetaxel-

capecitabine regimen, which might be associated with increased efficacy and reduced

long-term side-effects, particularly cardiotoxicity and leukemia. A third randomization,

which will be offered to all postmenopausal hormone receptor-positive patients, will

compare 2 years of tamoxifen followed by 5 years of letrozole to 7 years of letrozole

upfront (Figure 2). Women aged 18-70 years with an operable invasive breast carcinoma and

a negative sentinel node or axillary clearance are eligible. An overview of the MINDACT

trial was recently published in Nature Clinical Practice of Oncology, explaining in detail the

rationale behind the study design.28

An additional and important effort has been made to perform not only the analysis

of the 70-gene profile, but whole genome arrays for all 6,000 eligible patients. This will

potentially allow for the discovery of new genomic profiles with prognostic or predictive

value and eventually new drug targets. Fresh frozen tissue, paraffin-embedded tissue andblood samples from all 6,000 patients will be stored in an independent biobank repository,

representing an invaluable resource for future research.

The Collection of Fresh Frozen Tissue for the MINDACT Trial

Although microarray experiments are becoming more and more standardized, operator

and technical variability are well known to influence the measurement of gene expression

levels. For all samples in the MINDACT trial, RNA isolation, quality controls and microarray

analysis will be performed at Agendia, Amsterdam, to avoid the bias of potentialinterlaboratory reproducibility. Consequently, tumor samples from all over the world will

be shipped to the Netherlands within a fixed timeframe. Additionally, since one of the aims

of the MINDACT trial is the establishment of a biological material bank for future research,

also in the field of proteomics, frozen tumor samples and blood samples will be collected

from all patients. Preservation of the tumor samples in RNAlater® might influence several

processes in the tissue, such as the level of proteins, therefore, temporary preservation

in RNAlater®, as tested in the RASTER study, is not suitable and material frozen in liquid

nitrogen will be mandatory for the MINDACT trial.

8/10/2019 Stella Mook T3-108



35

2

To test the logistics of collection, freezing and shipment of the samples, the authors

performed a logistics pilot study in 6 European hospitals. For this feasibility study, patients

with early stage breast cancer under the age of 71 were eligible. All patients signed an

informed consent to donate a piece of tumor tissue for research. The pathologist obtained

a representative tumor sample within one hour of surgery, using a 6mm biopsy punch,

according to a standardized procedure. The tumor samples were snap frozen in liquid

nitrogen and stored at -80°C until shipment. All samples were shipped on dry ice by a

specifically contracted courier specialized in transportation of frozen material at -80 °C. At

Agendia, the percentage of tumor cells in the samples was determined as described by Van

‘t Veer and colleagues.8 When the sample was representative of the tumor ( i.e. tumor cell

≥ 50%), RNA was isolated and, after measurement of its quality and quantity, the 70-gene

profile was performed. The primary endpoint of this logistics pilot study was the success

rate of hybridization. Preliminary results show that, in general, it is feasible to collect and

ship good quality fresh frozen tumor samples from several locations throughout Europe to

Amsterdam. The procedures (of tissue collection, freezing and transportation) tested in this

study formed the basis of the standard operating procedures written for the MINDACT trial.

The final results of this pilot study will be published in 2007.

Future Prospects

The MINDACT study will determine the clinical relevance of the 70-gene profile and its

prognostic value compared to the currently available prognostic clinicopathologicalcriteria. Moreover, as tumor and blood material and whole genome microarray data will

be collected from all patients, a valuable bank of material will be established, providing an

opportunity for the identification of predictive gene expression profiles and potential drug

targets. Nowadays, the choice among treatment options is based upon patient and tumor

characteristics, such as age and estrogen receptor status, but overall it is extrapolated from

the percentage of risk reduction measured in a large population to the individual patient.

In the future, we might be able to identify the genomic fingerprint of each individual tumor,

telling us not only if a given patient needs adjuvant systemic treatment, but also which

treatment will have the best response and which treatment should be avoided because ofpotentially serious side-effects.

8/10/2019 Stella Mook T3-108


Chapter 2

36

Figure 2. MINDACT study design.Abbreviations: R-T = Treatment decision randomization; CT = Chemotherapy

Acknowledgements

The studies mentioned in this article were supported by the European Commission

Framework Programme VI, the Center of Biomedical Genetics, the Dutch Health Care

Insurance Board, the Dutch National Genomic Initiative - Cancer Genomics Program, the

Breast Cancer Research Foundation and the European Organisation for Research and Treatment of Cancer (EORTC) - Breast Cancer Group. S. Mook was partially supported by the

traineeship program of TRANSBIG network. The authors thank the numerous individuals

who have contributed to the studies mentioned in this review, especially those from

the TRANSBIG consortium and the EORTC, and to all the patients who have and still are

participating in these studies.

Conflicts of Interest

Dr. L.J. Van ’t Veer is a named inventor on a patent application for MammaPrint™ and reports

holding equity in Agendia BV.

8/10/2019 Stella Mook T3-108



37

2

References

1. Goldhirsch A, Glick JH, Gelber RD, et al . Meeting highlights: International expert consensus on the

primary therapy of early breast cancer 2005. Ann Oncol 2005; 16: 1569-1583.

2. Carlson RW and McCormick B. Update: NCCN breast cancer clinical practice guidelines. J Natl Compr Canc

Netw 2005; 3 Suppl 1: S7-11.

3. NCCN Clinical Practice Guidelines in Oncology. Breast Cancer V.I.2007. www.nccn.org 2007.



2005; 365: 1687-1717.








cancer. Nature 2002; 415: 530-536.

9. Sotiriou C, Neo SY, McShane LM, et al . Breast cancer classification and prognosis based on gene

expression profiles from a population-based study. Proc Natl Acad Sci USA 2003; 100: 10393-10398.



11. Paik S, Shak S, Tang G, et al . A multigene assay to predict recurrence of tamoxifen-treated, node-negative


12. Jansen MP, Foekens JA, Van Staveren IL, et al . Molecular classification of tamoxifen-resistant breast

carcinomas by gene expression profiling. J Clin Oncol 2005; 23: 732-740.


2003; 361: 1590-1596.

14. Chang JC, Wooten EC, Tsimelzon A, et al . Gene expression profiling for the prediction of therapeutic

response to docetaxel in patients with breast cancer. Lancet 2003; 362: 362-369.

15. Hughes TR, Mao M, Jones AR, et al . Expression profiling using microarrays fabricated by an ink-jetoligonucleotide synthesizer. Nat Biotech 2001; 19: 342-347.

16. Simon R, Radmacher MD, Dobbin K, McShane LM. Pitfalls in the use of DNA microarray data for

diagnostic and prognostic classification. J Natl Cancer Inst 2003; 95: 14-18.

17. Eifel P, Axelson JA, Costa J, et al . National Institutes of Health Consensus development conference


18. Goldhirsch A, Glick JH, Gelber RD, Coates AS, Senn HJ. Meeting highlights: International consensus

panel on the treatment of primary breast cancer. J Clin Oncol 2001; 19: 3817-3827.



8/10/2019 Stella Mook T3-108


Chapter 2

38

20. Ravdin PM, Siminoff LA, Davis GJ, et al . Computer program to assist in making decisions about adjuvant


21. Olivotto IA, Bajdik CD, Ravdin PM, et al . Population-based validation of the prognostic model ADJUVANT!


22. Desmedt C, Piette F, Loi S, et al . Strong time dependence of the 76-gene prognostic signature for node-

negative breast cancer patients in the TRANSBIG multicenter independent validation series. Clin Cancer

Res 2007; 13: 3207-3214.

23. Sotiriou C, Wirapati P, Loi S, et al . Gene expression profiling in breast cancer: understanding the molecular

basis of histologic grade to improve prognosis. J Natl Cancer Inst 2006; 98: 262-272.

24. Couzin J. Diagnostics: Amid debate, gene-based cancer test approved. Science 2007; 315: 924.

25. Bueno de Mesquita JM, Van De Vijver MJ, Peterse JL, et al . Feasibility of gene expression profiling in

community hospitals; preliminary results of a pilot study in N0 breast cancer patients (abstract 309).

Breast Cancer Res Treat 2005 2005; 94 (Suppl 1): A 309.

26. Douma KFL, Karsenberg K, Bueno de Mesquita JM, Hummel JM, Van Harten WH. Methodology of

constructive technology assessment in health care. Int J Technol Assess in Health Care 2007; 23: 162-168.

27. Van de Vijver M. Gene-expression profiling and the future of adjuvant therapy. Oncologist 2005; 10: 30-34.

28. Bogaerts J, Cardoso F, Buyse M, et al . Gene signature evaluation as a prognostic tool: challenges in the

design of the MINDACT trial. Nat Clin Pract Oncol 2006; 3: 540-551.

8/10/2019 Stella Mook T3-108


Chapter 3

Daily clinical practice of fresh tumor tissue

freezing and gene expression profiling;

logistics pilot study preceding

the MINDACT trial

Stella Mook

Hervé Bonnefoi

Giancarlo Pruneri

Denis Larsimont

Janusz Jaskiewicz

Maria D Sabadell

Gaëten MacGrogan

Laura J. Van ’t VeerFatima Cardoso

Emiel J.Th. Rutgers

Eur J Cancer 2009; 45: 1201-1208.

8/10/2019 Stella Mook T3-108


Chapter 3

42

Abstract

Purpose

The 70-gene prognosis-signature is a prognostic tool for early breast cancer analysis.

In addition to scientific evidence, implementation of the signature in clinical trials and daily

practice requires logistical feasibility. The aim of our study was to test logistics for gene

expression profiling on fresh frozen tumor tissue in the preparation for the prospective,

multinational Microarray In Node-negative Disease may Avoid ChemoTherapy (MINDACT)

trial.

Methods

Sixty-four patients were included in six European hospitals. Fresh frozen tumor samples

were shipped on dry ice to Agendia B.V., where RNA was isolated and subsequently

hybridized on the 70-gene prognosis-signature (MammaPrint™).

Results

Tumor samples were obtained in 60 of 64 patients. Among the 60 samples, 11 contained

insufficient tumor cells (< 50%) and three contained insufficient RNA quality. All 46 samples

eligible for genomic profiling were successfully hybridized, and the results were reportedon average within 4-5 d.

Conclusion

Gene expression profiling on fresh frozen tissue is feasible in daily clinical practice.

8/10/2019 Stella Mook T3-108


Logistics of gene expression profiling

43

3

Introduction

In the past 20 years, important advances have been made in the knowledge of the biology

of breast cancer. Using new high-throughput techniques, such as microarray-based

gene expression profiling, both prognostic and predictive profiles were established, and

breast cancer was re-classified based on molecular characteristics.1-16 One of these gene

expression classifiers is the 70-gene prognosis-signature (MammaPrint™).2,3 This 70-gene

dichotomous classifier can accurately distinguish breast tumors with a high metastatic

capacity from tumors with a low risk of developing distant metastases, by measuring the

expression level of 70 genes in tumor tissue. Several retrospective validation studies have

confirmed its prognostic value.3,17-20

Implementation of gene expression profiles requires logistical feasibility, in addition to

scientific evidence provided by validation studies. An essential part of this logistics is the

procurement of fresh frozen tissue as source of high-quality RNA. Traditional fixation of

fresh tissue in formaldehyde results in degradation of RNA and cross-linking, which makes

it unsuitable for comprehensive microarray analysis.21 Moreover, RNA becomes heavily

fragmented during storage of paraffin-embedded tissue.22 In addition, slow freezing of

samples promotes the formation of ice crystals, which can also provoke RNA damage.23

Consequently, collection of snap-frozen tissue or fresh tissue preserved in RNA preservation

fluid, such as RNARetain™ (Asuragen Inc., TX, USA),24 is at present mandatory to obtain high-

quality RNA and successful gene expression profiling. Recently, Bueno-de-Mesquita and

colleagues described the successful implementation of RNARetain™ tissue preservationand centralized MammaPrint™ testing for 16 community hospitals in the Netherlands for

the prospective RASTER trial.24 Evaluation of logistics of frozen tissue collection, centralized

microarray testing and swift reporting of results in the preparation of a multinational

multicentre clinical trial is described here.

In 2007, the MINDACT trial (Microarray In Node-negative Disease may Avoid ChemoTherapy;

EORTC 10041/ BIG 3-04) started to prospectively evaluate the 70-gene prognosis-signature

as a risk assessment and decision-making tool.25-27 This trial will enroll 6000 breast cancer

patients throughout Europe, who will have their risk of disease recurrence assessed by

both traditional clinicopathological criteria and the 70-gene prognosis-signature. Sincedecision-making for adjuvant treatment is based on both the risk assessments, the 70-

gene prognosis-signature test result has to be available within a fixed timeframe suitable

for daily clinical practice. Moreover, to avoid interlaboratory variability, which may

artificially influence gene expression levels, all samples are obtained and frozen at local

sites and shipped to Agendia B.V., Amsterdam, for RNA isolation and microarray analysis.

Consequently, participation in the MINDACT trial demands personnel at local hospitals who

will collect and freeze tumor samples. These local procedures, frozen sample shipment and

sample analysis within a fixed timeframe entail complex logistics that requires a thorough

organization.

8/10/2019 Stella Mook T3-108


Chapter 3

44

In the preparation for the MINDACT trial, we conducted a pilot study to test the logistics

for gene expression profiling in a multicentre and multinational setting. The first aim of

this pilot study was to test and if necessary to improve the logistics to collect good-quality

fresh frozen tissue at individual hospitals for microarray testing. The second aim was to

determine the proportion of samples that was hybridized successfully. The last aim was to

define Standard Operating Procedures (SOPs) for the tissue logistics in the MINDACT trial.

Together with the Dutch RASTER trial,24 this pilot study provided crucial information for the

feasibility of the MINDACT trial.

Patients and Methods

This logistics pilot study was coordinated by the Netherlands Cancer Institute (NKI) and was

conducted in six European hospitals. The study was approved by the institutional ethical

review board of each participating hospital, and all patients gave their written informed

consent before surgery, for the donation of a piece of tumor tissue to test the logistics for

genomic profiling.

Patients

Women under the age of 71 years at diagnosis with a unifocal, unilateral pT1-pT2, invasive

breast carcinoma and a clinically negative axillary lymph node status were eligible forinclusion. Patients with carcinoma in situ were eligible, provided that invasive cancer was

present. Patients who received neoadjuvant therapy were not included. Each hospital

included at least eight patients.

On-site training

Before the start of the study, the study coordinator organized on-site instruction

meetings in each participating hospital. These instruction meetings were attended by

a multidisciplinary team, i.e. breast surgeons, medical oncologists, pathologists, datamanagers and research nurses. During this instruction visit, the logistical scheme was

discussed and incorporated in the local standard procedures. Additionally, all study-

specific standard procedures were explained in detail in a manual of operations and were

summarized on provided pocket summaries, to support standardized procedures for tissue

collection, freezing and shipment.

8/10/2019 Stella Mook T3-108



45

3

Figure 1A. Biopsy puncher for standardized tumor sampling.

B. Tumor specimen after sampling, using the 6 mm biopsy puncher (by courtesy of J.F. Egger).

C. Hematoxylin and eosin (H&E) stained section of the tumor specimen (shown in B). After

sampling, intact morphology is shown, and appropriate grading and staging of the tumor are

allowed (by courtesy of J.F. Egger).

A

B

C

8/10/2019 Stella Mook T3-108


Chapter 3

46

Pre-assembled kits

The study coordinator provided hospitals with pre-assembled sample kits for each patient,

consisting of all case report forms (CRFs), a 6 mm biopsy puncher (Figure 1A) and a printed

sticker sheet with a unique identification number (Sample ID), suitable for freezing in liquid

nitrogen and prolonged storage at -80°C.

Tumor sampling

After surgical resection, tumor specimens were immediately transported from the

operating room to the pathology department in a tumor container without fixatives, e.g.

formalin. To ensure standardized tumor sampling, 6 mm biopsy punchers were provided

(Figure 1A). The pathologist obtained a tumor sample within 1 h of surgery, using this biopsy

puncher. Samples were placed in an Eppendorf tube, labeled with a sample ID sticker.

For tumors smaller than 1 cm (at macroscopic examination), a 3 mm biopsy puncher was

allowed to obtain a sample. Obviously, standard diagnostic pathology examination had

priority over the procurement of a research sample, i.e. the pathologist only obtained a

tumor sample for gene expression profiling when he/she judged that there was a sufficient

amount of tumor tissue.

Snap-freezing and storage

Eppendorf tubes containing tumor samples were snap-frozen by submerging the tubes

in liquid nitrogen for at least 1 min. After snap-freezing of the sample, the total time from

transportation of tissue to the pathology department till freezing of the samples was

recorded. Samples were stored in a -80°C freezer until shipment.

Shipment

Frozen samples were shipped on dry ice by a contracted courier, specialized in

transportation of frozen material. Samples were shipped as ‘Biological Substance CategoryB UN 3373’ (Exempt Human Specimen) in the applicable packaging material provided by

the courier, i.e. an inner sealed plastic bag with absorbent material, an outer packaging

and a polystyrene box with dry ice. Packaging and shipment complied with International

Air Transport Association (IATA) criteria (http://www.iata.org). Samples were shipped as a

batch of three samples or once in every 3 weeks, to reduce costs. Samples were shipped

and delivered at Agendia B.V., Amsterdam, within 1 working day after collection at the local

hospitals. Samples collected on Friday were delivered on Monday. The amount of dry ice

was maintained during the weekend, to prevent thawing of the samples.

8/10/2019 Stella Mook T3-108



47

3

Microarray analyses

Upon receipt of the samples at Agendia B.V., outer and inner packaging and Eppendorf

tubes were checked for damage and for the presence of an appropriate sticker with unique

sample ID. Samples were processed for microarray analysis, and the number of days required

to generate a 70-gene prognosis-signature result was registered. Frozen sections were cut

and stained with hematoxylin and eosin (H&E), before and after cutting the sections for

RNA isolation, to confirm the presence of tumor and to determine tumor cell percentage.

If the mean tumor cell percentage was < 50%, again two frozen sections were cut and

stained with H&E before and after cutting the sections for RNA isolation. Samples with less

than 50% tumor cells determined in duplicate were excluded from further analysis. RNA

isolation, amplification and labeling were performed at Agendia Laboratories, as described

previously.2,28 RNA quality was assessed using the Agilent bioanalyzer. Samples without

sample ID stickers or samples with damaged packaging material, less than 50% tumor cells

or insufficient RNA quality (RIN < 7) were excluded from further processing.

A total of 200 ng of tumor RNA was co-hybridized with a standard reference to a custom-

designed microarray (MammaPrint™), including eight identical subarrays, each containing

oligonucleotide probes for the 70 genes in triplicate.28 The standard reference sample

consisted of pooled RNA of 105 primary breast tumors selected from patients of the

retrospective validation series.3 For this feasibility study, results were only presented

as successful hybridization or exclusion, hence no good- or poor prognosis-signature

was reported. Consequently, all patients included in this feasibility study were treatedaccording to the standard national guidelines. The above-mentioned standard procedures

for the collection of good-quality fresh frozen tumor tissue for gene expression profiling

are shown in Figure 3, left panel.

Results

Between November 2005 and November 2006, 68 patients were enrolled in six hospitals

throughout Europe. Among the 68 patients, four were excluded (one patient had withdrawninformed consent, one patient was aged > 71 years and two had no detectable malignancy).

All 64 eligible patients underwent the surgery. Baseline characteristics are shown in Table 1.

The pathologist was able to obtain a tumor sample in 60 patients (94%): 55 samples were

obtained using a 6 mm biopsy puncher and five samples were obtained using a 3 mm

biopsy puncher. Among the 60 tumor samples, 14 samples were inadequate (11 samples

contained less than 50% tumor cells and three samples had insufficient RNA quality),

whereas all 46 adequate samples were successfully hybridized on the MammaPrint™. None

of the samples were lost due to processing errors, such as initial storage in formalin. In 4

of the 64 eligible patients no tumor sample was obtained; three patients had tumors that

8/10/2019 Stella Mook T3-108


Chapter 3

48

were too small to obtain a tumor sample and in one case the pathologist forgot to take a

tumor sample. A summary is given in Figure 2.

Remarkably, all 3 mm samples were representative and hybridized successfully. There was

no significant differences in tumor size between the samples that were inadequate because

of insufficient tumor cells and samples that were hybridized successfully (mean diameter

21 mm versus 19 mm; p = 0.7) (Table 1). The median time to freeze a tumor sample was 20 min

(range 5-235 min). For three samples that had poor RNA quality, the freezing time was <

20 min. The median time to generate and report a 70-gene prognosis-signature result from

the time of arrival at Agendia laboratories was 4 working days (range 3-14; mean 5.2).

Table 1. Baseline characteristics

Successful hybridization No hybridization P

N % N %

Age ns

≤ 50 years 16 35 4 29

51 - 60 years 14 30 3 21

61 - 70 years 16 35 7 50

Tumor size ns

pT1 (≤ 20mm) 29 66 7 54

pT2 (> 20mm) 15 34 6 46

Histology

Invasive ductal 38 83 8 57

Others 6 13 5 36

Missing 2 4 1 7

Grade ns

Grade 1 7 15 2 14

Grade 2 22 48 7 50

Grade 3 14 30 3 22

Missing 3 7 2 14

Estrogen receptor status ns

Positive 36 78 9 64

Negative 6 13 2 14

Missing 4 9 3 22

Lymph node status ns

Positive 13 28 3 22

Negative 31 68 10 71

Missing 2 4 1 7

Total 46 100 14 100

Missing value were not used for calculation of p-values.

8/10/2019 Stella Mook T3-108



49

3

F i g u r e

2 .

F l o w d i a g r a m o

f p a t i e n

t s e n r o l l e d a n d t u m o r s a m p l e s .

I n c l u s i o n

N = 6 8

E l i g i b l e

N = 6 4

S u r g e r y

N = 6 4

N o s a m p l e

n = 4 ( 4 / 6 4 = 6 % )

P r o c u r e m e n t o f

s a m p l e

n = 6 0 ( 6 0 / 6 4 = 9 4 % )

S h i p m e n t

n = 6 0 ( 6 0 / 6 4 = 9 4 % )

6 m m B i o p s y p u n c h e r

n = 5 5 ( 5 5 / 6 4 = 8 6 % )

3 m m B i o p s y p u n c h e r

n = 5 ( 5 / 6 4 = 8 % )

E x c l u d e d : < 5 0 %

t u m o r c e l l s

n = 1 1 ( 1 1 / 6 4 = 1 7 % )

E x c l u d e d : I n s u f f i c i e n t

R N A q u a l i t y

n = 3 ( 3 / 6 4 = 5 % )

R N A o b t a i n e d

n = 4 1 ( 4 1 / 6 4 = 6 4 % )

R N A o b t a i n e d

n = 5 ( 5 / 6 4 = 8 % )

H y b r i d i z a t i o n

s u c c e s s f u l

n =

4 6 ( 4 6 / 4 6 = 1 0 0 % )

T u m o r t o o s m a l l t o

o b t a i n s a m p l e

n = 3 ( 3 / 6 4 = 5 % )

F o r g o t t e n t o t a k e

s a m p l e

n = 1 ( 1 / 6 4 = 2 % )

E x c l u s i o n ( n = 4 )

−

N o m a l i g n a n c y d e t e c t e d ( n = 2 )

−

W i t h d r a w n i n f o r m e d c o n s e n

t ( n = 1 )

−

A g e d > 7 1 y e a r s ( n = 1 )

8/10/2019 Stella Mook T3-108


Chapter 3

50

Figure 3. Flow chart of standardized procedures for obtaining good-quality fresh frozen tumor

samples for microarray analyses (left panel) and adaptations to local standard procedures (right

panel).

Procedures Adaptations

Signed informed consent

Surgery

Transfer of tumor specimen in DRY tumorcontainer from OR to pathology department

Procurement of tumor sample within 1 hour

of surgery, following standard procedures

Tumor sampling using 6 mm biopsy puncher

Tumor sample in Eppendorf tube, labeled

with ID sticker

Submerge Eppendorf tube with tumor

sample in liquid nitrogen for at least 1

minute

Storage of tumor sample in -80°C freezer

Inform OR personnel that patient

participates in study

Tumor container WITHOUT fixative (e.g.formaline), labeled with study number

Time limit for tissue processing

− Selection of representative, non-sclerotic,

non-necrotic tumor area for sampling

− Availability biopsy puncher

Availability of stickers suitable for freezing in

liquid nitrogen and storage at -80°C

Availability of liquid nitrogen and personal

protective equipment at pathology

department

Availability of -80°C freezer

8/10/2019 Stella Mook T3-108



51

3

Discussion

Our study showed that collection and shipment of fresh frozen tumor tissue for gene

expression profiling are feasible in a multicentre and multinational practice setting, with

a success rate of 72% (46 out of 64). Provided that the pathologist was able to obtain a

tumor sample, the success rate increased to 77% (46 out of 60). When RNA was obtained, all

samples were successfully hybridized and a gene expression signature result was obtained

in 100% (46 out of 46). The main reason for sample failure was a non-representative tumor

sample; 18% (11 out of 60) of the samples contained < 50% tumor cells. This proportion of

non-representative samples is in agreement with the proportion reported by a previous

feasibility study.24

The pathologist obtains a sample after macroscopic evaluation of the tumor specimen

(Figure 1B). As shown in Figure 1C , tumor sampling does not alter morphology and allows

appropriate grading and staging of the tumor. The best area for sampling is the periphery

of the tumor, given that the central part is often sclerotic or necrotic and lack tumor

cells. However, sampling in the periphery of the tumor could increase the amount of

surrounding stroma in the sample. The balance between a sufficient amount of tumor cells

and a limited amount of stromal tissue can be improved in part by training and repetition.

Additionally, recent research has shown that samples containing > 30% tumor cells are

suitable for reliable 70-gene prognosis-signature read-out (Amendment 1, MINDACT trial;

EORTC 10041/ BIG 3-04). As a consequence, the cut-off for tumor cell percentage in the

MINDACT trial has been lowered to 30%, and hence sample inclusion will be increased. Inour study, inclusion of samples containing 30-50% tumor cells would have resulted in five

additional hybridizations (51 out of 60 = 85% success rate).

Although H&E stained sections of the material sampled for profiling were used to determine

if the sample contained a certain amount of malignant tissue, a tumor in itself can be very

heterogeneous.29 To test if the biopsy sample was also representative for the tumor in its

entirety, we compared the final pathology report with the genomic test result. In this pilot

study, the profile was associated with grade and estrogen receptor status (ER) ( p < 0.001),

which is in good agreement with previous validation studies, that have shown a strong

association between the profile and grade, ER status and disease outcome.3,17-20

Although gene expression profiling is becoming more and more standardized, operator

and technical variability are well known to influence the measurement of gene expression

levels.30-32 To avoid potential interlaboratory bias, all samples in the MINDACT will be

shipped on dry ice to Agendia, Amsterdam, where quality controls, RNA isolation and gene

expression analysis will be performed. Consequently, frozen tumor samples have to be

shipped from all over the world to Amsterdam within a fixed timeframe. In this pilot study,

samples were shipped once in every 3 weeks or as a batch, therefore time from sample

arrival at Agendia till reporting of the genomic test was measured, instead of the interval

between surgery and genomic test result. In our study, all tumor samples were delivered

8/10/2019 Stella Mook T3-108


Chapter 3

52

within one working day. Furthermore, a 70-gene prognosis-signature result was available

after a median of 4 d, thereby showing the feasibility of implementation of this signature

in clinical trials and daily practice, with regard to the needed timeframe for clinical decision

making.

Recently, Bueno de Mesquita and colleagues have shown that the collection of fresh tumor

tissue is feasible in community hospitals in the Netherlands. 24 In contrast to our study,

tumor samples were placed in a commercially available preservation fluid (RNARetain™) at

room temperature, and were sent by conventional mail to the Netherlands Cancer Institute,

where samples were subsequently frozen in liquid nitrogen. Although preservation of

tumor samples in RNARetain™ does not influence gene expression measurements,33,34 it is

unclear whether it might influence levels of proteins. Since one of the aims of MINDACT is

the establishment of a biological materials bank for future research, including proteomics,

temporarily preservation of tissue in RNARetain™ as done by Bueno de Mesquita and

colleagues is not suitable, and tumor tissue immediately frozen in liquid nitrogen was

chosen. The complex logistics involved in the collection and shipment of fresh frozen

tissue demands a thorough and detailed organization with adjustments to local standard

procedures. The major adjustments are shown in Figure 3, right panel. These adjustments

formed the basis of the standard operating procedures (SOPs) written for MINDACT.35

In conclusion, through detailed standard operating procedures, provision of necessary

devices and close collaboration between surgeons, medical oncologists, pathologists,

research nurses, data-managers and scientists, successful implementation of the logistics

for gene expression profiling on fresh frozen tissue is feasible.

Conflicts of interest statement

Laura J. Van ’t Veer is a named inventor on a patent application for MammaPrint™ and

reports holding equity in Agendia B.V.

Acknowledgements

This study was financially supported by a grant of the EORTC Breast Cancer Group. Weare indebted to the women who participated in this study. We thank J.F. Egger (Geneva

University Hospitals, Geneva, Switzerland); M.J. Piccart (Institute Jules Bordet, Brussels,

Belgium); G. Viale, A. Goldhirsh, M. Colleoni (European Institute of Oncology and University

of Milan, Milan, Italy); J. Jassem, K. Jaskiewicz (Medical University of Gdansk, Gdansk,

Poland); J. Baselga, F. Rojo (Vall d’Hebron University Hospital, Barcelona, Spain); L. Mauriac

(Institute Bergonié, Bordeaux, France) and all other doctors, nurses, and data managers for

their generous participation. We are indebted to Jillian Harrison, EORTC and Annuska Glas

from Agendia B.V. for their expert advice and to Marjanka Schmidt for critically reading the

manuscript.

8/10/2019 Stella Mook T3-108



53

3

References

1. Perou CM, Sorlie T, Eisen MB, et al . Molecular portraits of human breast tumours. Nature 2000; 406: 747-

752.


cancer. Nature 2002; 415: 530-536.



4. Chang JC, Wooten EC, Tsimelzon A, et al . Gene expression profiling for the prediction of therapeutic

response to docetaxel in patients with breast cancer. Lancet 2003; 362: 362-369.



6. Chang HY, Nuyten DSA, Sneddon JB, et al . Robustness, scalability, and integration of a wound-response


7. Gianni L, Zambetti M, Clark K, et al . Gene expression profiles in paraffin-embedded core biopsy tissue

predict response to chemotherapy in women with locally advanced breast cancer. J Clin Oncol 2005; 23:

7265-7277.

8. Hannemann J, Oosterkamp HM, Bosch CA, et al . Changes in gene expression associated with response

to neoadjuvant chemotherapy in breast cancer. J Clin Oncol 2005; 23: 3331-3342.

9. Iwao-Koizumi K, Matoba R, Ueno N, et al . Prediction of docetaxel response in human breast cancer by

gene expression profiling. J Clin Oncol 2005; 23: 422-431.

10. Rouzier R, Perou CM, Symmans WF, et al . Breast cancer molecular subtypes respond differently to

preoperative chemotherapy. Clin Cancer Res 2005; 11: 5678-5685.

11. Wang Y, Klijn JG, Zhang Y, et al . Gene-expression profiles to predict distant metastasis of lymph-node-


12. Hess KR, Anderson K, Symmans WF, et al . Pharmacogenomic predictor of sensitivity to preoperative

chemotherapy with paclitaxel and fluorouracil, doxorubicin, and cyclophosphamide in breast cancer.

J Clin Oncol 2006; 24: 4236-4244.

13. Potti A, Dressman HK, Bild A, et al . Genomic signatures to guide the use of chemotherapeutics. Nat Med

2006; 12: 1294-1300.

14. Sotiriou C, Wirapati P, Loi S, et al . Gene expression profiling in breast cancer: understanding the molecularbasis of histologic grade to improve prognosis. J Natl Cancer Inst 2006; 98: 262-272.

15. Thuerigen O, Schneeweiss A, Toedt G, et al . Gene expression signature predicting pathologic complete

response with gemcitabine, epirubicin, and docetaxel in primary breast cancer. J Clin Oncol 2006; 24: 1839-

1845.

16. Bonnefoi H, Potti A, Delorenzi M, et al . Validation of gene signatures that predict the response of breast

cancer to neoadjuvant chemotherapy: a substudy of the EORTC 10994/BIG 00-01 clinical trial. Lancet Oncol

2007; 8: 1071-1078.



8/10/2019 Stella Mook T3-108


Chapter 3

54



19. Mook S, Schmidt MK, Viale G, et al . The 70-gene prognosis-signature predicts disease outcome in breast

cancer patients with 1-3 positive lymph nodes in an independent validation study. Breast Cancer Res Treat

2009; 116: 295-302.

20. Wittner BS, Sgroi DC, Ryan PD, et al . Analysis of the MammaPrint breast cancer assay in a predominantly


21. Masuda N, Ohnishi T, Kawamoto S, Monden M, Okubo K. Analysis of chemical modification of RNA from

formalin-fixed samples and optimization of molecular biology applications for such samples. Nucleic Acids

Res 1999; 27: 4436-4443.

22. Cronin M, Pho M, Dutta D, et al . Measurement of gene expression in archival paraffin-embedded tissues:

development and performance of a 92-gene reverse transcriptase-polymerase chain reaction assay.

Am J Pathol 2004; 164: 35-42.

23. Vonsattel JP, Aizawa H, Ge P, et al . An improved approach to prepare human brains for research. J

Neuropathol Exp Neurol 1995; 54: 42-56.

24. Bueno de Mesquita JM, Van Harten WH, Retel VP, et al . Use of 70-gene signature to predict prognosis of

patients with node-negative breast cancer: a prospective community-based feasibility study (RASTER).

Lancet Oncol 2007; 8: 1079-1087.



26. Mook S, Van’t Veer LJ, Rutgers EJ, Piccart-Gebhart MJ, Cardoso F. Individualization of therapy using

Mammaprint: from development to the MINDACT Trial. Cancer Genomics Proteomics 2007; 4: 147-155.

27. Cardoso F, Van’t Veer L, Rutgers E, et al . Clinical application of the 70-gene profile: the MINDACT trial. J

Clin Oncol 2008; 26: 729-735.

28. Glas AM, Floore A, Delahaye LJ, et al . Converting a breast cancer microarray signature into a high-

throughput diagnostic test. BMC Genomics 2006; 7: 278-287.

29. Camp RL, Charette LA, Rimm DL. Validation of tissue microarray technology in breast carcinoma. Lab

Invest 2000; 80: 1943-1949.

30. Bammler T, Beyer RP, Bhattacharya S, et al . Standardizing global gene expression analysis between

laboratories and across platforms. Nat Methods 2005; 2: 351-356.

31. Dobbin KK, Beer DG, Meyerson M, et al . Interlaboratory comparability study of cancer gene expressionanalysis using oligonucleotide microarrays. Clin Cancer Res 2005; 11: 565-572.

32. Irizarry RA, Warren D, Spencer F, et al . Multiple-laboratory comparison of microarray platforms. Nat Methods

2005; 2: 345-350.

33. Florell SR, Coffin CM, Holden JA, et al . Preservation of RNA for functional genomic studies: a

multidisciplinary tumor bank protocol. Mod Pathol 2001; 14: 116-128.

34. Mutter GL, Zahrieh D, Liu C, et al . Comparison of frozen and RNALater solid tissue storage methods for

use in RNA expression microarrays. BMC Genomics 2004; 5: 88-94.

35. Leyland-Jones BR, Ambrosone CB, Bartlett J, et al . Recommendations for collection and handling of

specimens from group breast cancer clinical trials. J Clin Oncol 2008; 26: 5638-5644.

8/10/2019 Stella Mook T3-108


Chapter 4

The 70-gene prognosis signature predicts

early metastasis in breast cancer patients

between 55 and 70 years of age

Stella Mook

Marjanka K. Schmidt

Britta Weigelt

Bas Kreike

Inge Eekhout

Marc J. Van de Vijver

Annuska M. Glas

Arno FlooreEmiel J.Th. Rutgers

Laura J. Van ’t Veer

Ann Oncol 2010; 21: 717-722.

8/10/2019 Stella Mook T3-108


Chapter 4

58

Abstract

Background

The majority of breast cancer patients are postmenopausal women who are increasingly

being offered adjuvant chemotherapy. Since the beneficial effect of chemotherapy in

postmenopausal patients predominantly occurs in the first 5 years after diagnosis, a

prognostic marker for early events can be of use for adjuvant treatment decision-making.

The aim of this study was to evaluate the prognostic value of the 70-gene prognosis-

signature for early events in postmenopausal patients.

Methods

Frozen tumor samples from 148 patients aged 55-70 years were selected (T1-2, N0) and

classified by the 70-gene prognosis signature (MammaPrint™) into good or poor prognosis.

Eighteen percent received hormonal therapy.

Results

Breast cancer-specific survival (BCSS) at 5 years was 99% for the good-prognosis signature

versus 80% for the poor-prognosis signature group ( p = 0.036). The 70-gene prognosis

signature was a significant and independent predictor of BCCS during the first 5 years offollow-up with an adjusted hazard ratio of 14.4 (95% confidence interval 1.7-122.2; p = 0.01)

at 5 years.

Conclusion

The 70-gene prognosis signature can accurately select postmenopausal patients at low

risk of breast cancer-related death within 5 years of diagnosis and can be of clinical use in

selecting postmenopausal women for adjuvant chemotherapy.

8/10/2019 Stella Mook T3-108


The 70-gene signature in postmenopausal patients

59

4

Introduction

Approximately two-thirds of the newly diagnosed breast cancer patients are ≥ 55

years of age.1 These postmenopausal women are increasingly being offered adjuvant

chemotherapy despite the more favorable biological characteristics of their tumors and

their known favorable breast cancer-specific outcome in general.2-5 The Early Breast Cancer

Trialists’ Collaborative Group meta-analysis has shown that the benefit from chemotherapy

is influenced by age, with less benefit in older patients. Moreover, those data have shown

that the time course of chemotherapy efficacy differs between pre- and postmenopausal

patients; the benefit of chemotherapy in postmenopausal breast cancer patients occurs

predominantly in the first 5 years, while in premenopausal patients, the benefit sustains

throughout the first 10 years.6 Therefore, a prognostic marker that can accurately identify

postmenopausal patients who are at low risk of developing an early breast cancer-

related event can be of clinical use for selecting postmenopausal patients for adjuvant

chemotherapy. One of the prognostic markers in the field of breast cancer is the 70-gene

prognosis signature (MammaPrint™), which can accurately identify patients who have a

good prognosis and therefore might be safely spared chemotherapy.7-11 This signature has

been developed in a predefined subset of patients, i.e. women under the age of 55 years at

diagnosis with stage I or II, node-negative breast cancer. Therefore, the aim of our study was

to evaluate the prognostic value of the 70-gene prognosis signature in postmenopausal

women with node-negative breast cancer. Specifically, we investigated whether the

signature could select postmenopausal patients who are at low risk of developing anearly breast cancer-related event and thus can be safely spared chemotherapy, without

jeopardizing disease outcome.

Methods

Patient selection

A consecutive series of patients treated at the Netherlands Cancer Institute-Antoni vanLeeuwenhoek Hospital (NKI-AVL) from 1984 to 1996 were selected according to the

following criteria: female, unilateral T1 or T2 primary invasive breast carcinoma, negative

nodal status, aged between 55 and 71 years at diagnosis, no adjuvant chemotherapy and

fresh frozen tumor material available in the comprehensive NKI-AVL tissue bank. Patients

without complete axillary staging, patients with prior malignancies (except for non-

melanoma skin cancer and dysplasia of the cervix), bilateral synchronous breast tumors or

patients treated with neoadjuvant therapy were not included.

All patients (n = 148) had been treated by modified radical mastectomy or breast

-conserving surgery, including axillary lymph node dissection, followed by radiotherapy

8/10/2019 Stella Mook T3-108


Chapter 4

60

if indicated. Twenty-seven patients (18%) received endocrine therapy, which consisted of

tamoxifen for a median duration of 2.0 years (range 0.06-7.0 years). Patients were treated

according to consensus guidelines, taking into account patients’ will and consent. The

study was approved by the ethical review board of the NKI-AVL.

Tumor samples, RNA extraction and gene expression analysis

Frozen tumor samples were evaluated for MammaPrint™ (FDA 510(K) cleared) at Agendia’s

laboratories (ISO17025 certified and CLIA accredited; Amsterdam, the Netherlands)

blinded to clinical data, as previously described.7,12 Briefly, frozen sections were stained with

hematoxylin and eosin; only samples that contained at least 30% tumor cells were used

for RNA isolation. Labeled complementary RNAs were hybridized together with a standard

breast cancer reference pool to the custom-designed MammaPrint™ microarray. 12 Tumors

were classified according to their cosine correlation coefficient with the MammaPrint™

template. Tumors with a correlation coefficient above the threshold were classified as

good prognosis signature, whereas all other tumors were classified as poor prognosis

signature.7,12

Clinicopathological and follow-up data

Clinical data were retrieved from medical patient records, blinded to the 70-gene prognosis

signature. Follow-up was completed until October 2007. End points considered were timefrom surgery to distant metastasis as first event [distant metastasis-free survival (DMFS)]

and breast cancer-specific survival (BCSS), defined as the time from surgery to breast cancer-

related death. For the analysis of DMFS, we considered distant metastasis as first event as

failure; patients were censored on the date of local or regional recurrence, development

of a second primary, including contralateral breast cancer, death from any cause or date

of last follow-up visit. Tumor grade was defined according to Bloom-Richardson. Estrogen

receptor (ER) expression was estimated using ER messenger RNA levels as determined by

the microarray.8

Clinical risk was evaluated using Adjuvant! software version 8.0 (available at www.adjuvantonline.com). Adjuvant! calculates 10-year survival probability based on patient’s

age, co-morbidities (set to ‘average for age’), tumor size, tumor grade, ER-status and

number of positive axillary lymph nodes.13,14 Patients were classified as having low clinical

risk when the predicted 10-year BCSS was > 88% for ER-positive tumors and > 92% for ER-

negative tumors, respectively.9

8/10/2019 Stella Mook T3-108



61

4

Statistical analyses

Analyses were carried out using SPSS version 15.0 (SPSS Inc., Chicago, IL). Cox proportional

hazards regression analyses were used to calculate hazard ratios (HRs) and their 95%

confidence intervals (CIs). HRs for the risk groups as defined by the 70-gene signature were

estimated with stratification for clinical risk as defined by Adjuvant! (adjusted HRs).

Additionally, HRs for the risk groups as defined by Adjuvant! were estimated with

stratification for genomic risk as defined by the 70-gene prognosis signature. The impact

of duration of follow-up on HRs was analyzed by censoring observations at increasing time

points.

Results

The 70-gene prognosis signature (MammaPrint™) risk classification was assessed in tumor

tissues of a consecutive series of 148 postmenopausal patients with early-stage, lymph

node-negative, invasive breast cancer. Tumor samples of 173 patients fulfilled the selection

criteria, of which 25 contained insufficient tumor cells (n = 22) or had insufficient RNA

quality (n = 3). All 148 samples eligible for genomic profiling were successfully hybridized.

There was no difference in tumor or patient characteristics between the 25 samples that

could not be hybridized and the 148 analyzed samples with regard to age, tumor size,

histology, overall survival and BCSS (data not shown). The median duration of follow-upwas 12.5 years (range 0.4-20.2) for the 114 patients who did not die of breast cancer and

7.2 years (range 0.8-17.7) for the 34 patients who died of breast cancer. During follow-up,

83 patients had at least one event, among which were 42 distant metastases including 36

distant metastases as first event and 57 deaths of which 34 were breast cancer-related.

Twelve of the 34 breast cancer-related deaths occurred within 5 years after diagnosis.

Classification by 70-gene prognosis signature and disease outcome

The 70-gene prognosis signature classified 91 (61%) patients as good prognosis, whereas57 (39%) patients were classified as poor prognosis. A good prognosis signature was

associated with smaller, well-differentiated and ER -positive tumors (Table 1).

Patients classified as good prognosis by the signature had a 5-year DMFS probability of

93% [standard error (SE) 3%], compared with 72% (SE 6%) in the poor-prognosis signature

group (Figure 1A). DMFS at 5 years was significantly worse in the poor-prognosis signature

group, with a univariate HR of 4.6 (95%CI 1.8-12.0; p = 0.001). Over the entire follow-up

period, the HR for DMFS was 1.8 (95% CI 0.9-3.5; p = 0.07).

8/10/2019 Stella Mook T3-108


Chapter 4

62

Table 1. Baseline characteristics and association between clinicopathological characteristics

and the 70-gene prognosis profile

70-gene prognosis signature

Good prognosis

signature (n=91)

Poor prognosis

signature (n=57)

P value

Characteristics No. % No. %

Age (years) 0.25

< 60 26 28.6 24 42.1

60 - 64 34 37.4 15 26.3

65 - 71 31 34.1 18 31.6

Surgery 0.32

Breast conserving surgery 46 50.5 24 42.1

Mastectomy 45 49.5 33 57.9

Tumor size 0.007

pT1 (≤ 20 mm) 59 64.8 24 42.1

pT2 (> 20-50 mm) 32 35.2 33 57.9

Histological tumor type 0.02

Invasive ductal carcinoma 67 73.6 52 91.2

Invasive lobular carcinoma 17 18.7 0 0

Mixed IDC ILC 4 4.4 3 5.3

Other 3 3.3 2 3.5

Histological grade < 0.001

Grade I 52 57.1 3 5.3

Grade II 28 30.8 15 26.3

Grade III 11 12.1 39 68.4

Estrogen-receptor status < 0.001

Negative 3 3.3 29 50.9

Positive 88 96.7 28 49.1

Adjuvant endocrine therapy 0.86

No 74 81.3 47 82.5

Yes 17 18.7 10 17.5

IDC, invasive ductal carcinoma; ILC, invasive lobular carcinoma.

The 5-year BCSS probability was 99% (SE 1%) for the good-prognosis signature group and

80% (SE 5%) for the poor-prognosis signature group (Figure 1B). In addition, the difference

in BCSS between the poor–prognosis signature group and the good-prognosis signature

group over the entire follow-up period was significant with a univariate HR of 2.0 (95% CI

1.0-4.0; p = 0.04). This difference was most pronounced at 5 years, with a univariate HR of

19.1 (95%CI 2.5-148; p = 0.005) (Figure 2A).

8/10/2019 Stella Mook T3-108



63

4

Figure 1.

Kaplan-Meier curves by 70-gene prognosis signature.

A. Time to distant metastases as first event.

B. Breast cancer-specific survival

Time (years)

121086420

D i s t a n t m e t a s t a s i s - f r e e s u r v i v a l

1.0

0.8

0.6

0.4

0.2

0.0

91

57

86

2344 37 34 31

77 67 51 43 28

17 Poor signature

Good signature

Log rank p = 0.07

72 ± 6%

93 ± 3%

80 ± 5%

67 ± 7%

Time (years)

121086420

B r e a s t c a n c e r - s p e c i fi c s

u r v i v a l

1.0

0.8

0.6

0.4

0.2

0.0

91

57

89

2852 46 42 36

86 80 68 61 48

23 Poor signature

Good signature

80 ± 5%

69 ± 6%

99 ± 1%

90 ± 4%

Log rank p = 0.036

A

B

8/10/2019 Stella Mook T3-108


Chapter 4

64

Clinical risk assessment and discordance with 70-gene prognosis signature

Using the predefined cut-off, Adjuvant! classified 74 patients (50%) as clinical low risk and

74 patients (50%) as clinical high risk. Concordance was observed between Adjuvant! and

the 70-gene prognosis signature for 62 (42%) low-risk/good prognosis patients, whereas

45 patients (30%) were classified as high-risk/poor prognosis according to both risk

assessments. The clinical risk assessment was discordant with the genomic prognosis for

41 patients (28%); 12 (8%) were classified as clinical low risk and poor prognosis signature

and 29 (20%) were classified as clinical high risk though good prognosis signature.

Prediction of early breast cancer-specific death: time dependency

Since the benefit of chemotherapy in postmenopausal patients is predominantly seen in

the first 5 years and given the long follow-up time in this study compared with the original

validation study (11.6 versus 6.7 years, respectively), we calculated unadjusted HRs for the

signature and clinical risk assessment with censoring of all observations at increasing time

points (Figure 2, panels A and B). Remarkably, the 70-gene prognosis signature was a strong

predictor for early breast cancer-specific death (BCSD) with the strongest prognostic

value at 5 years as shown by the highest HR (HR 19.1; 95% CI 2.5-148; p = 0.005), whereas

the clinical risk classification predicted disease outcome more evenly with a tendency to

predict better after 5 years, with the strongest prognostic capacity at 10 years (HR 6.2;

95% CI 2.1-18.0; p = 0.001). To further evaluate the clinical utility of the 70-gene prognosissignature, we adjusted its performance for the clinical risk assessment, which showed that

the signature is a powerful predictor of early BCSD independent of the clinical risk, with an

adjusted HR at 5 years of 14.4 (95% CI 1.7-122; p = 0.01) (Figure 2, panel C ). The reverse analysis,

i.e. HRs for the clinical risk classification adjusted for the gene signature, showed that the

clinical risk classification is a prognostic factor of BCSD after 10 years independent of the

signature, with an adjusted HR of 4.4 (95% CI 1.4-13.6; p = 0.01) for BCSS at 10 years (Figure

2, panel D).

8/10/2019 Stella Mook T3-108



65

4

6,6

13,5 15,4 19,112,4

3,92,6

2,0 2,0

0,1

1,0

10,0

100,0

2 3 4 5 7 10 12,5 15 None

A. 70-gene prognosis-signature

4,1

8,3

4,2 5,3 4,8

6,24,6 4,3 3,6

0,1

1,0

10,0

100,0

B. Adjuvant! online

2,0

3,5

1,6 1,9 2,0

4,4 3,8 4,13,2

0,1

1,0

10,0

100,0

D. Adjuvant! adjusted for signature

4,98,4

12,6 14,49,4

2,21,6 1,2 1,3

0,1

1,0

10,0

100,0

C. Signature adjusted for Adjuvant!

15% 26% 29% 35%

47%

74%85%

94% 100%

2 3 4 5 7 10 12,5 15 None

Cumulativeproportion

of events

Censoring time (years)

Figure 2.

Hazard ratios (HRs) for breast cancer-specific death at increasing censoring times.

A). Univariate HRs for poor-prognosis signature versus good-prognosis signature groups.

B). Univariate HRs for clinical high-risk versus clinical low-risk groups as calculated by Adjuvant!.

C). HRs for 70-gene prognosis signature adjusted for clinical risk. D). HRs clinical risk assessment

by Adjuvant! adjusted for prognosis signature.

8/10/2019 Stella Mook T3-108


Chapter 4

66

Discussion

The present validation study shows that the 70-gene prognosis signature, which was

developed in premenopausal patients with early -stage breast cancer, is also a prognostic

factor in postmenopausal women, with especially strong prognostic capacity in the first

5 years after diagnosis. Since the beneficial effect of chemotherapy in postmenopausal

women mainly occurs in the first 5 years after diagnosis,6,15 accurate identification of

early events by the 70-gene prognosis signature can be of great value in selecting

postmenopausal patients for adjuvant chemotherapy.

In a previous validation study, Buyse et al .9 showed a strong time dependency of the signature.

Since we were especially interested in predicting early events which might be prevented by

chemotherapy, we also investigated the effect of duration of follow-up on the prognostic

value of the signature. The prognostic value of the signature was most pronounced within

the first 5 years of diagnosis, even after adjustment for clinicopathological risk classification

by Adjuvant! (Figure 2, panel C ). These results demonstrate the additional value of the 70-gene

prognosis signature over and above the clinical risk assessment in predicting early BCSD.

First, the signature enlarged the group of low-risk/ good prognosis patients as compared

with the clinical risk classification (from 50% to 61%). Secondly, despite this increase in

the low -risk group, the signature accurately classified 11 of 12 (92%) patients who died of

breast cancer within 5 years of diagnosis as poor prognosis (Figure 3), compared with 10 of 12

(83%) correctly classified by Adjuvant!.

Although our study confirmed that the signature can correctly predict early BCSD, lateBCSD was less accurately predicted by the signature (Figure 3), resulting in misclassification

of 15 BCSD after 5 years of diagnosis (compared with seven misclassified by Adjuvant!).

Remarkably, all 15 patients who were classified as good signature but died of breast cancer

after 5 years had ER-positive tumors, and only one patient received endocrine therapy.

Consequently, endocrine treatment could potentially have prevented at least part of these

late BCSDs (31% reduction of annual breast cancer death rate by adjuvant tamoxifen). 6

In our series, the 27 patients who received hormonal therapy were equally distributed

between the good- and the poor-prognosis signature groups. Moreover, the median

duration of hormonal therapy was only 2 years (according to Dutch treatment guidelinesin the years of diagnosis concerned), instead of the current standard treatment of at least

5 years. Separate analyses of the 121 hormonal therapy-naive patients showed that the

70-gene prognosis signature was also a predictor for early BCSD in untreated patients

between 55 and 71 years of age (adjusted HR at 5 years 10.8; 95% CI 1.2-94.7; p = 0.03).

The strong time dependency of the signature can be explained by the fact that the

signature was built to identify patients with distant metastases within 5 years. Moreover,

it supports the hypothesis that different biological mechanisms are involved in early and

late disease recurrences.16,17 ER-negative, high-grade tumors are more likely to metastasize

during the early years after diagnosis, whereas ER-positive, low-grade tumors more often

cause late recurrences.18

8/10/2019 Stella Mook T3-108



67

4

Figure 3.

Breast cancer-specific deaths by 70-gene prognosis signature. Each circle represents a patient

who died of breast cancer.

In our study, 61% of the patients were classified as good prognosis by the signature. In

previous validation studies, the 70-gene signature consistently classified ~40% to 50% of

the (predominantly premenopausal) patients as good prognosis.8-11 The increase in patients

classified as good prognosis in our series could be the reflection of the intrinsic low-risk

nature of breast cancer and mammographic screening in postmenopausal women.2-4

Recently, an independent validation study of the 70-gene prognosis signature in

predominantly postmenopausal women was published.19

In contrast to our study, themajority of their study population was classified as high risk (73% versus 39% in our series). This

discrepancy in risk classification can be attributed to differences in baseline characteristics,

i.e. more poorly differentiated tumors in their series. On the other hand, disease outcome

in our series is worse compared with the outcome in the series of Wittner et al .,19 which

can be caused by the difference in proportion of patients who received adjuvant systemic

therapy (18% hormonal therapy in our series versus 45% chemo- and/or hormonal therapy

in Wittner’s series).

Several other prognostic profiles have been studied, among which are the 76-gene profile

and the 21-recurrence score.20,21 Both profiles have been developed and so far validated in

0 5 10 15

Breast cancer-specific deaths

Poor signature

Good signature

Time (years)

8/10/2019 Stella Mook T3-108


Chapter 4

68

a mixed population of pre- and postmenopausal women.22,23 To our knowledge, this is the

first study that evaluates the prognostic value of a prognostic signature in an exclusively

postmenopausal patient series. Recently, Anders et al . 5 showed a significant difference in

gene expression patterns between tumors from pre- and postmenopausal breast cancer

patients. However, our study indicates that disease outcome in pre- and postmenopausal

patients can be determined by common denominators, which are captured by the 70-gene

signature.

In conclusion, our study indicates that application of the 70-gene prognosis signature in

breast cancer patients between 55 and 71 years of age could result in a more accurate

allocation of adjuvant systemic therapy. A poor prognosis signature would imply

chemotherapy treatment to prevent early breast cancer deaths, and patients with ER-

positive tumors should receive endocrine therapy to prevent late events. Furthermore,

given the results from the ATAC (Arimidex, Tamoxifen, Alone or in Combination) and

(Breast International Group) BIG 1-98 trial indicating that aromatase inhibitors (AIs) are

more effective in preventing early recurrences compared with tamoxifen, patients with

ER-positive tumors classified as poor prognosis by the signature might be candidates for

up-front AI treatment.24-26 This last question will also be addressed in the endocrine therapy

randomization of the MINDACT (Microarray for Node Negative and 1 to 3 Positive Node

Disease may Avoid Chemotherapy) trial; patients with hormone receptor-positive tumors

(good and poor prognosis signature) will be randomized between 2 years of tamoxifen

followed by 5 years of letrozole versus 7 years of letrozole up front, therefore endocrine

responsiveness can be related to the 70-gene prognosis signature.27,28

Funding

European Commission Framework Program VI-TRANSBIG (LSHC-CT-2004-503426); Dutch

National Genomics Initiative-Cancer Genomics Center (CGC 2008-2012); Agendia BV.

Acknowledgements

We are indebted to Mahasti Saghatchian and Marleen Kok for helpful discussions andcritically reading the manuscript.

Disclosures

LJvV and MJvdV are named inventor on a MammaPrint™ patent. LJvV reports holding

equity in Agendia BV. AF and AMG are employees of Agendia BV.

8/10/2019 Stella Mook T3-108



69

4

References

1. Ries LAG, Melbert D, Krapcho M et al. SEER Cancer Statistics Review, 1975-2005. Bethesda, MD: National

Cancer Institute 2008; http://seer.cancer.gov/csr/1975_2005/, based on November 2007 SEER data

submission, posted to the SEER web site. Last accessed 28 November, 2008.

2. Anderson WF, Chatterjee N, Ershler WB, Brawley OW. Estrogen receptor breast cancer phenotypes in

the Surveillance, Epidemiology, and End Results database. Breast Cancer Res Treat 2002; 76: 27-36.

3. Diab SG, Elledge RM, Clark GM. Tumor characteristics and clinical outcome of elderly women with

breast cancer. J Natl Cancer Inst 2000; 92: 550-556.

4. Fisher B, Wickerham DL, Brown A, Redmond CK. Breast cancer estrogen and progesterone receptor

values: their distribution, degree of concordance, and relation to number of positive axillary nodes. J

Clin Oncol 1983; 1: 349-358.

5. Anders C, Hsu D, Broadwater G, et al . Young age at diagnosis correlates with worse prognosis and defines

a subset of breast cancers with shared patterns of gene expression. J Clin Oncol 2008; 26: 3324-3330.



2005; 365: 1687-1717.


cancer. Nature 2002; 415: 530-536.







2009; 116: 295-302.





13. Ravdin PM, Siminoff LA, Davis GJ, et al . Computer program to assist in making decisions about adjuvanttherapy for women with early breast cancer. J Clin Oncol 2001; 19: 980-991.

14. Olivotto IA, Bajdik CD, Ravdin PM, et al . Population-based validation of the prognostic model ADJUVANT!


15. Demicheli R, Miceli R, Moliterni A, et al . Breast cancer recurrence dynamics following adjuvant CMF is

consistent with tumor dormancy and mastectomy-driven acceleration of the metastatic process. Ann

Oncol 2005; 16: 1449-1457.

16. Klein CA, Blankenstein TJ, Schmidt-Kittler O, et al . Genetic heterogeneity of single disseminated tumour

cells in minimal residual cancer. Lancet 2002; 360: 683-689.

8/10/2019 Stella Mook T3-108


Chapter 4

70

17. Schmidt-Kittler O, Ragg T, Daskalakis A, et al . From latent disseminated cells to overt metastasis: genetic

analysis of systemic breast cancer progression. Proc Natl Acad Sci USA 2003; 100: 7737-7742.

18. Saphner T, Tormey DC, Gray R. Annual hazard rates of recurrence for breast cancer after primary

therapy. J Clin Oncol 1996; 14: 2738-2746.





21. Wang Y, Klijn JG, Zhang Y, et al . Gene-expression profiles to predict distant metastasis of lymph-node-


22. Esteva FJ, Sahin AA, Cristofanilli M, et al . Prognostic role of a multigene reverse transcriptase-PCR assay

in patients with node-negative breast cancer not receiving adjuvant systemic therapy. Clin Cancer Res

2005; 11: 3315-3319.

23. Foekens JA, Atkins D, Zhang Y, et al . Multicenter validation of a gene expression-based prognostic

signature in lymph node-negative primary breast cancer. J Clin Oncol 2006; 24: 1665-1671.

24. Coates AS, Keshaviah A, Thurlimann B, et al . Five years of letrozole compared with tamoxifen as initial

adjuvant therapy for postmenopausal women with endocrine-responsive early breast cancer: update

of study BIG 1-98. J Clin Oncol 2007; 25: 486-492.

25. Mauriac L, Keshaviah A, Debled M, et al . Predictors of early relapse in postmenopausal women with

hormone receptor-positive breast cancer in the BIG 1-98 trial. Ann Oncol 2007; 18: 859-867.

26. Forbes JF, Cuzick J, Buzdar A, et al . Effect of anastrozole and tamoxifen as adjuvant treatment for early-

stage breast cancer: 100-month analysis of the ATAC trial. Lancet Oncol 2008; 9: 45-53.



28. Mook S, Van’t Veer LJ, Rutgers EJ, Piccart-Gebhart MJ, Cardoso F. Individualization of therapy using

Mammaprint: from development to the MINDACT Trial. Cancer Genomics Proteomics 2007; 4: 147-155.

8/10/2019 Stella Mook T3-108


Chapter 5

The 70-gene prognosis-signature predicts

disease outcome in breast cancer patients

with 1-3 positive lymph nodes in an

independent validation study

Stella Mook

Marjanka K. Schmidt

Giuseppe Viale

Giancarlo Pruneri

Inge Eekhout

Arno Floore

Annuska M. Glas

Jan BogaertsFatima Cardoso

Martine Piccart-Gebhart

Emiel J.Th. Rutgers


On behalf of the TRANSBIG consortium

Breast Cancer Res Treat 2009; 116: 295-302

8/10/2019 Stella Mook T3-108


Chapter 5

74

Abstract

Purpose

The 70-gene prognosis signature has shown to be a valid prognostic tool in node-negative

breast cancer. Although axillary lymph node status is considered to be one of the most

important prognostic factors, still 25–30% of node-positive breast cancer patients will

remain free of distant metastases, even without adjuvant systemic therapy. We therefore

investigated whether the 70-gene prognosis signature can accurately identify patients

with 1-3 positive lymph nodes who have an excellent disease outcome.

Methods

Frozen tumor samples from 241 patients with operable T1-3 breast cancer, and 1-3 positive

axillary lymph nodes, with a median follow-up of 7.8 years, were selected from 2 institutes.

Using a customized microarray, tumor samples were analyzed for the 70-gene tumor

expression signature. In addition, we reanalyzed part of a previously described cohort (n =

106) with extended follow-up.

Results

The 10-year distant metastasis-free (DMFS) and breast cancer specific survival (BCSS)probabilities were 91% (SE 4%) and 96% (SE 2%), respectively for the good prognosis

signature group (99 patients), and 76% (SE 4%) and 76% (SE 4%), respectively for the poor

prognosis signature group (142 patients). The 70-gene signature was significantly superior

to the traditional prognostic factors in predicting BCSS with a multivariate hazard ratio (HR)

of 7.17 (95% CI 1.81 to 28.43; p = 0.005).

Conclusions

The 70-gene prognosis signature outperforms traditional prognostic factors in predictingdisease outcome in patients with 1-3 positive nodes. Moreover, the signature can accurately

identify patients with an excellent disease outcome in node-positive breast cancer, who

may be safely spared adjuvant chemotherapy.

8/10/2019 Stella Mook T3-108


The 70-gene signature in pN1 breast cancer

75

5

Introduction

Axillary lymph node status is historically one of the most important prognostic factors

in breast cancer, with deterioration in disease outcome as the number of positive nodes

increases.1-3 Consequently, patients with axillary lymph node metastases are considered as

having a poor prognosis and hence are most likely to benefit from adjuvant chemotherapy,

with an absolute benefit of 6-15% at 5 years.4 However, up to 25-30% of node-positive

patients will remain free of distant metastases even without adjuvant systemic therapy.4,5

Thus, adjuvant treatment decision-making based on nodal status is only moderately

accurate and results in overtreatment, with unnecessary exposure to treatment toxicity.

Identifying robust and reliable prognostic factors that can select those node-positive

patients who do not require adjuvant chemotherapy is essential to reduce overtreatment.

One of the new prognostic markers which has been validated for lymph node-negative

breast cancer is the 70-gene prognosis signature (MammaPrint™).6-8 The original

retrospective validation study demonstrated that the signature was also a significant

prognostic factor in 144 node-positive patients.8 The aim of this study is to further

substantiate the prognostic value of the 70-gene signature in patients with 1-3 positive

nodes in a new independent dataset, and to assess its relation to standard prognostic

markers. Specifically, we investigated whether the 70-gene signature can select patients

with 1-3 positive nodes with an excellent survival, who might be safely spared adjuvant

chemotherapy.

Methods

Patients

Patients were selected from the Netherlands Cancer Institute-Antoni van Leeuwenhoek

hospital (NKI-AVL), Amsterdam, The Netherlands (n = 213, consecutive series) and the

European Institute of Oncology (EIO), Milan, Italy (n = 79, consecutive series), according

to the following criteria: unilateral T1, T2 or operable T3 invasive breast carcinoma, withmetastases in 1-3 axillary lymph nodes; frozen tumor tissue available; no prior malignancies,

no bilateral synchronous breast tumors, and no neoadjuvant therapy. Micrometastases

(tumor deposits > 0.2 and ≤ 2.0 mm) were considered as positive lymph nodes. Patients

were diagnosed between 1994 and 2001 and were under the age of 71 years at diagnosis.

Patients were treated with mastectomy or breast-conserving surgery, including dissection

of the axillary lymph nodes (ALND), followed by radiotherapy and adjuvant systemic

therapy if indicated. Adjuvant systemic therapy was administered according to national

guidelines, taking into account patients’ preferences and consent (Table 1). The proportion

of adjuvant systemic therapy in our study was similar to all patients at NKI who fulfilled

8/10/2019 Stella Mook T3-108


Chapter 5

76

the above mentioned selection criteria except for the availability of frozen tumor tissue in

the same time period (data not shown). The study received approval of the medical ethical

committee of NKI-AVL.

To allow more extensive analyses, follow-up data of all patients with 1-3 positive nodes

from the previously described series by Van de Vijver,8 were updated, blinded to the 70-

gene prognosis signature.9

Tumor samples, RNA extraction and gene expression analysis

Frozen tumor samples were processed in Agendia’s laboratories (Amsterdam, the

Netherlands), for RNA isolation, amplification and labeling as previously described.7,10

Samples were available for RNA isolation if they contained at least 30% tumor cells on

hematoxylin/eosin stained sections. Of the 292 samples processed, 10 were rejected on

the basis of RNA quality and 41 because of insufficient tumor cells. The 51 rejected samples

were obtained from slightly smaller tumors than the 241 samples that were hybridized

(mean tumor size 19 mm versus 23 mm; p = 0.04). However, there were no differences in age,

tumor grade, ER status, systemic treatment and proportion alive after 10 years.

To assess the mRNA expression level of the 70 genes, RNA was hybridized to a custom-

designed array (MammaPrint™) blinded to clinical data, at Agendia’s ISO17025-certified and

CLIA accredited laboratories. Tumors were classified as 70-gene good or poor prognosis

signature as described previously.6-8,10

Clinicopathological data

Clinical data were retrieved from medical records, blinded to the 70-gene prognosis

signature. Endpoints considered were time from surgery to distant metastasis as first

event (DMFS), and breast cancer specific survival (BCSS), defined as time from surgery to

breast cancer-related death. For the analysis of distant metastasis-free survival (DMFS)

we considered distant metastases as first event as failure; patients were censored on date

of local or regional recurrence, development of second primary including contralateral

breast cancer, death from any cause or date of last follow-up visit. Tumor grading wasdefined according to the Bloom-Richardson method. Estrogen receptor (ER) status and

progesterone receptor (PR) status were determined by immunohistochemistry and

interpreted positive if more than 10% of the cells stained. For patients treated at NKI-AVL,

HER2/NEU immunohistochemistry status was retrieved from the original pathology report.

For patients treated at EIO, HER2/NEU status was determined by immunohistochemistry; in

case of 2+ scores FISH analyses were used to determine amplification (ratio ≥ 2.2).

8/10/2019 Stella Mook T3-108



77

5

Clinical risk assessment by Adjuvant!

To assess the 70-gene prognosis signature in a clinical context, it was compared with

the clinicopathological risk as predicted by Adjuvant! The Adjuvant! Software version 8.0

(available at www.adjuvantonline.com) calculated 10-year survival probability based on

patient’s age, co-morbidities, tumor size, tumor grade, ER-status and number of positive

axillary lymph nodes.11,12 Patients were considered as having low clinical risk when the 10-

year BCSS as predicted by Adjuvant! was more than 88% for ER-positive tumors, and more

than 92% for ER-negative tumors, respectively.6


Analyses were performed using SPSS version 15.0 (SPSS Inc, Chicago, IL) and EPICURE

(Epicure release 2.0.Seattle: HiroSoft International Corporation, 1996). Kaplan-Meier

survival plots and log-rank tests were used to assess the difference in DMFS and BCSS of the

predicted good and poor prognosis groups. Cox proportional hazards regression analyses

were used to calculate uni- and multivariate hazard ratios (HR) and their 95% confidence

intervals (95% CI). In multivariate Cox regression analyses traditional clinicopathological

variables were used. An interaction term of gene signature and chemotherapy, within a

multivariate Cox regression model was tested for significance by the likelihood ratio test.

P -values are two-sided.

Results

The 70-gene prognosis signature (MammaPrint™) was assessed in tumor tissue of an

independent series of 241 invasive breast cancer patients with 1-3 positive lymph nodes.

Among the 241 patients, 99 (41%) were classified as good prognosis signature, whereas 142

(59%) patients were classified as poor prognosis signature. Patients with a poor prognosis

signature were more frequently diagnosed at EIO, and had more often received adjuvant

chemotherapy and less often received endocrine therapy. Moreover, tumors classified aspoor prognosis signature were larger and more often poorly differentiated, ER- and PR

negative, and HER2/NEU receptor positive (Table 1).

After a median follow-up of 7.8 years (range, 0.01-12.3) 66 patients had at least one event,

including 13 local recurrences, 9 regional recurrences, 6 contralateral breast cancers, 9

second primary cancers, 43 distant metastases, including 35 distant metastases as first

event, and 39 deaths of which 33 breast cancer-related deaths.

8/10/2019 Stella Mook T3-108


Chapter 5

78

T

a b l e 1 .

A s s o c i a t i o n b e t w e e n c l i n

i c o p a t h o l o g i c a l c h a r a c t e r i s t i c s a

n d t h e 7 0 - g e n e p r o g n o s i s s i g n a

t u r e f o r t h e n e w v a l i d a t i o n s e r i e s ( n =

2 4 1 ) .

7 0 - g e n e p r o g n o s i s s i g n a t u r e

G o o d p r o g n o s i s s i g n

a t u r e ( n = 9 9 )

P o o r p r o g n o

s i s s i g n a t u r e ( n = 1 4 2 )

P

v a l u e *

C h a r a c t e r i s t i c s

N o .

%

N o .

%

H o s p i t a l

< 0

. 0 0 1

N K I - A V L

8 4

8 4

. 8

9 0

6 3

. 4

E I O

1 5

1 5

. 2

5 2

3 6

. 6

A g e ( y e a r s )

0 . 1

8

< 4 0

6

6 . 1

1 7

1 2

. 0

4 0 -

4 9

4 1

4 1

. 4

6 1

4 3

. 0

5 0 -

5 9

3 9

3 9

. 4

4 7

3 3

. 0

6 0 -

7 0

1 3

1 3

. 1

1 7

1 2

. 0

S u r g e r y

0 . 1

7

B C T

5 4

5 4

. 5

9 0

6 3

. 4

M a s t e c t o m y

4 5

4 5

. 5

5 2

3 6

. 6

A x i l l a r y p r o c e d u r e

0 . 4

2

A L N D

6 2

6 2

. 6

9 6

6 7

. 6

S L N P & A L N D

3 7

3 7

. 4

4 6

3 2

. 4

N o d a l s t a t u s

0 . 9

3

1 p o s i t i v e n o d e

4 9

4 9

. 5

7 4

5 2

. 1

2 p o s i t i v e n o d e s

3 5

3 5

. 4

4 2

2 9

. 6

3 p o s i t i v e n o d e s

1 5

1 5

. 1

2 6

1 8

. 3

T u m o r s i z e ( p T N M )

0 . 0

1

p T 1 ( ≤ 2 0 m m )

5 8

5 8

. 6

5 9

4 1

. 5

p T 2 ( > 2 0 - 5

0 m m )

4 0

4 0

. 4

8 1

5 7

. 1

p T 3 ( > 5 0 m m )

1

1 . 0

2

1 . 4

H i s t o l o g i c a l t u m o r t y p e

< 0

. 0 0 1

D u c t a l

7 2

7 2

. 8

1 3 2

9 3

. 0

L o b u l a r

1 2

1 2

. 1

3

2 . 1

M i x e d

1 4

1 4

. 1

3

2 . 1

O t h e r

1

1 . 0

4

2 . 8

H i s t o l o g i c a l g r a d e

< 0

. 0 0 1

G o o d

4 5

4 6

. 4

1 2

8 . 5

M o d e r a t e

4 6

4 7

. 4

5 3

3 7

. 3

P o o r

6

6 . 2

7 7

5 4

. 2

U n k n o w n

2

0

8/10/2019 Stella Mook T3-108



79

5

7 0 - g e n e p r o g n o s i s s i g n a t u r e

G o o d p r o g n o s i s s i g n

a t u r e ( n = 9 9 )

P o o r p r o g n o s i s s i g n a t u r e ( n = 1 4 2 )

P

v a l u e *

C h a r a c t e r i s t i c s

N o .

%

N o .

%

E s t r o g e n - r e c e p t o r s t a t u s

< 0

. 0 0 1

N e g a t i v e

4

4 . 0

4 6

3 2

. 4

P o s i t i v e

9 5

9 6

. 0

9 6

6 7

. 6

P r o g e s t e r o n e - r e c e p t o r s t a t u s

< 0

. 0 0 1

N e g a t i v e

1 6

1 6

. 5

7 2

5 0

. 7

P o s i t i v e

8 1

8 3

. 5

7 0

4 9

. 3

U n k n o w n

2

0

H E R 2 / N E U r e c e p t o r s t a t u s

< 0

. 0 0 1

N e g a t i v e

9 5

9 7

. 9

1 0 3

7 4

. 6

P o s i t i v e

2

2 . 1

3 5

2 5

. 4

U n k n o w n

2

4

A d j u v a n t s y s t e m i c t r e a t m e n t

0 . 4

1

N o n e

7

7 . 3

3

2 . 3

C h e m o t h e r a p y o n l y

1 0

1 0

. 4

4 3

3 2

. 3

E n d o c r i n e t h e r a p y o n l y

5 0

5 2

. 1

4 1

3 0

. 8

B o t h

2 9

3 0

. 2

4 6

3 4

. 6

U n k n o w n

3

9

A d j u v a n t c h e m o t h e r a p y

< 0

. 0 0 1

N o

5 7

5 9

. 4

4 4

3 3

. 1

Y e s

3 9

4 0

. 6

8 9

6 6

. 9

U n k n o w n

3

9

A d j u v a n t e n d o c r i n e t h e r a p y

0 . 0

0 5

N o

1 7

1 7

. 7

4 6

3 4

. 6

Y e s

7 9

8 2

. 3

8 7

6 5

. 4

U n k n o w n

3

9

A

b b r e v i a t i o n s : N K I - A V L

, N e t h e r l a n d s

C a n c e r I n s t i t u t e - A n t o n i v a n L e e u w e n h o e k h o s p i t a l ; E I O

, E u r o p e a n I n s t i t u

t e o f O n c o l o g y ; B C T

, b r e a s t - c o n s e r v

i n g t h e r a p y ;

A

L N D

, a x i l l a r y l y m p h n o d e d i s s e c t i o n

; S L N P

, s e n t i n e l l y m p h n o d e p r o c e d u

r e .

*

M i s s i n g v a l u e s w e r e n o t u s e d f o r c a l c u l a t i o n o f p

– v a l u e s .

8/10/2019 Stella Mook T3-108


Chapter 5

80

Figure 1. Kaplan-Meier curves by 70-gene prognosis signature among the 241 patients.

A) Breast cancer specific survival.

B) Distant metastasis-free survival (distant metastasis as first event).

Time (years)

1086420

B r e a s t c a n c e r - s p e c i fi c s u r v i v a

l

1.0

0.8

0.6

0.4

0.2

0.0

99

142

96

10137 127 105 58

95 83 53 15

Poor prognosis-signature

Good prognosis-signatureNumbers at

risk

99%

96%

Log-rank p < 0.001

88%

76%

Time (years)

1086420

D i s t a n t m e t a s t a s e s - f r e e s u r v i v a l

1.0

0.8

0.6

0.4

0.2

0.0

99

142

96

8119 105 84 47

93 77 49 12



risk

Log-rank p = 0.001

98%

80%

76%

91%

A

B

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108



83

5

factor with an HR of 0.31 (95% CI 0.12-0.80; p = 0.02); the 70-gene signature and number of

positive nodes (3 versus 1) tended to be prognostic factors with HRs of 2.99 (95% CI 0.996-

8.99; p = 0.051) and 2.29 (95% CI 0.99-5.29; p = 0.053), respectively.

Adjuvant! classified 32 patients (13%) as clinical low risk and 209 patients (87%) as clinical

high risk, using the pre-defined cut-off (See methods). The clinical risk assessment was

discordant with the genomic risk by the 70-gene prognosis signature for 77 patients (32%);

5 were classified as clinical low risk and poor prognosis signature; 72 were classified as

clinical high risk and good prognosis signature. Remarkably, in the 27 patients defined as

both 70-gene good prognosis and clinical low risk none of the patients developed distant

metastases nor died (Figure 2). Moreover, when the clinical high risk group (n = 209) was

stratified by signature risk, the 10-year BCSS probability was 94% (SE 3%) for the good

prognosis signature group and 76% (SE 4%) for the poor prognosis signature group,

respectively [HR of 4.12 (95% CI 1.45-11.76; p = 0.008)]. This shows the additional value of

the 70-gene prognosis signature up to and above the Adjuvant! risk assessment.

Interestingly, the 70-gene signature was also predictive for BCSS in the 101 chemotherapy

naïve patients (HR 7.33; 95% CI 1.61-33.49; p = 0.01), 128 chemotherapy-treated patients (HR

4.70; 95% CI 1.09-20.17; p = 0.04) ( Supplementary Figure 3), 63 endocrine therapy naïve patients

(HR ∞ (infinity); 95% CI 2.97-∞; p = 0.001), and 166 endocrine therapy-treated patients (HR

3.63; 95% CI 1.21-10.94; p = 0.02). Moreover, the 70-gene signature accurately predicted

BCSS in the 191 patients with ER-positive tumors (HR 9.75; 95% CI 2.26-42.01; p = 0.002). The

group of 50 ER-negative patients of whom 4 were classified as good prognosis signature,

and the 10 adjuvant untreated patients, were too small to analyze separately.Among the 241 patients, 29 had solely micrometastatic axillary lymph node involvement

(22 patients in 1 node, 6 in 2 nodes, and 1 in 3 nodes, respectively) and 18 patients had

micrometastatic involvement in addition to macrometastases. The 70-gene signature

maintained its prognostic value when nodes with micrometastases were excluded

(multivariate HR for BCSS 6.68; 95% CI 1.65-27.08; p = 0.008).

The previously described validation of the 70-gene signature by Van de Vijver et al ., included

144 node-positive patients with no restriction to number of positive nodes. 8 To be able to

do more extensive analyses we selected all patients with only 1-3 positive nodes from this

series (n = 106).8

Follow-up was updated from a median of 7.4 years to 10.3 years (range, 1.6to 21.2 years).9 This patient series was significantly different from our here described new

series, with regard to age (median age 45 versus 50 years, respectively; p < 0.001), axillary

procedure (all ALND), adjuvant systemic therapy and survival probabilities ( Supplementary

Tables 3 and 4). Most differences can be attributed to the fact that these patients were selected

to be younger than 53 years and were diagnosed at earlier calendar years (before 1995)

when sentinel lymph node procedure was not available, and adjuvant systemic treatment

guidelines were not as comprehensive as today. The 10-year BCSS probability was 98% (SE

2%) for the good prognosis profile (43 patients), and 64% (SE 6%) for the poor prognosis

profile group (63 patients), respectively. In this series a poor prognosis signature was also

8/10/2019 Stella Mook T3-108


Chapter 5

84

associated with shorter BCSS, with a univariate HR of 6.60 (95% CI 1.97-22.10; p = 0.002) and

a multivariate HR (adjusted for the same variables as listed in table 2) of 3.63 (95% CI 0.88-

14.96; p = 0.07).

Figure 2. Kaplan-Meier curves by 70-gene prognosis signature and clinical risk groups among

the 241 patients. A) Breast cancer-specific survival. B) Distant metastasis-free survival (distant

metastasis as first event).

Time (years)

1086420

1.0

0.8

0.6

0.4

0.2

0.0

B r e a s t

c a n c e r - s p e c i fi c s u r v i v a l

Patients Events Risk g roup

5

72

27

137

1

4

0

28

Prognosis-signature good, clinical low risk

Prognosis-signature poor, clinical high risk

Prognosis-signature poor, clinical low risk

Prognosis-signature good, clinical high risk

Time (years)

1086420

1.0

0.8

0.6

0.4

0.2

0.0


Patients Events Risk g roup

5

72

27

137

1

6

0

28

Prognosis-signature good, clinical low risk

Prognosis-signature poor, clinical high risk

Prognosis-signature poor, clinical low risk

Prognosis-signature good, clinical high risk

A

B

8/10/2019 Stella Mook T3-108



85

5

Discussion

The present study demonstrates that molecular diagnostics can identify a group of low

risk patients within node-positive breast cancer patients who are traditionally viewed as

high risk for recurrence by conventional histopathological evaluation. As such, this study

underscores the added value of molecular diagnostics and more specifically of the 70-gene

prognosis signature in the tailoring of treatment for the individual patient.

The 70-gene prognosis signature, which was developed using tumors of lymph node-

negative patients, first demonstrated its prognostic power in node-positive breast cancer

in the paper by Van de Vijver et al ..8 In this study, patients with one up to any number of

positive nodes were included. Nevertheless, our present results are in good agreement

with this previous publication: the HR for DMFS of 4.13 (95% CI 1.72-9.96; p = 0.002) in our

series is similar to the prognostic value of the signature in the 151 node-positive patients

from the Van de Vijver study (HR for DMFS 4.5; 95% CI 2.0-10.2; p < 0.001).8

In our new independent validation series both the 70-gene prognosis signature and

traditional clinicopathological factors were predictive for BCSS. However, the multivariate

analyses clearly demonstrate that the 70-gene signature remained the most powerful

predictor for BCSS, even after adjustment for the clinicopathological factors, showing the

added value of the signature.

The signature performed as a significant prognostic factor for DMFS (DM as first event)

in the univariate analysis and retained this capacity at borderline significance when

adjusted for clinicopathological variables. For DMFS with distant metastasis as any eventthe signature remained a strong independent predictor (HR 3.83; 95% CI 1.40-10.47; p =

0.009). In addition, in a pooled multivariate analysis of our new independent series and

the 106 patients from the Van de Vijver study with extended follow-up, the HR for DMFS

(as first event) for the signature remained consistent at 2.79 (95% CI 1.29-6.02; p = 0.009)

( Supplementary Table 5A).

As a consequence of adjuvant treatment guidelines, a substantial proportion of patients

in this validation series (128 of 241 patients) received adjuvant chemotherapy, with or

without hormonal therapy. Patients classified as poor prognosis by the 70-gene signature

more often received adjuvant chemotherapy (67% versus 41%, respectively; p < 0.001). Tumor characteristics in the poor signature group, i.e. more ER-negative and poorly

differentiated, are generally believed to be associated with a higher likelihood of response

to chemotherapy.4 Moreover, Albain et al. recently presented data on the 21-gene recurrence

score (RS) in lymph node-positive patients, showing that node-positive patients classified

as high RS have more benefit from chemotherapy in addition to tamoxifen.13 The larger

efficacy of chemotherapy in combination with the larger proportion of chemotherapy-

treated patients in the poor prognosis signature group would imply that the prognostic

value of the 70-gene signature would potentially be higher in an untreated group. To

further investigate this, we performed subgroup analyses in the chemotherapy-treated

8/10/2019 Stella Mook T3-108


Chapter 5

86

and untreated group, and confirmed similar prognostic power in each subgroup (HRs 4.85

and 5.99, respectively). To determine potential heterogeneity of the prognostic value of

the signature among the chemotherapy-treated and untreated group, we also performed

a multivariate analysis including an interaction variable between the signature and

chemotherapy. In this multivariate analysis of our series and the 106 patients from the

Van de Vijver study combined, the 70-gene prognosis signature maintained its prognostic

value for BCSS (HR 5.50; 95% CI 1. 47-20.62; p = 0.01), while the interaction term did not

reach significance ( p = 0.95), showing no signal of potential difference in prognostic value

in the two groups ( Supplementary Table 5B).

The clinical utility of the 70-gene signature depends on its potential value in addition to

traditional prognostic factors. Therefore, we compared the signature to clinicopathological

risk assessment, by Adjuvant!.11,12 As anticipated, Adjuvant! classified the majority of these

node-positive patients as high clinical risk (87%). Interestingly, the 70-gene prognosis

signature classified 72 (34%) clinical high risk patients as good prognosis and indeed the

disease outcome in this discordant group (clinical high risk, good prognosis signature)

was remarkably good, with a 10-year BCSS of 94%, indicating that the use of this signature

could result in a substantial reduction of patients who would be recommended for

chemotherapy, without jeopardizing outcome.

Although several prognostic markers have been studied in breast cancer, the majority of

these markers have not been studied in node-positive breast cancer,14,15 or lack prognostic

value in node-positive disease.16 Some previously identified markers do have prognostic

value in node-positive breast cancer, however, since they do not identify a substantialgroup of patients with an excellent disease outcome, the clinical relevance as prognostic

marker for this node-positive patients’ group seems to be limited.13,17,18 The only other

signature that could identify a low risk group with a sufficiently good outcome within node-

positive patients was the wound signature.9 Since this wound signature is not available as a

diagnostic test, its value for clinical practice seems to be limited at this moment.

The strong prognostic power of the signature with respect to distant metastases

(hematogenous spread), regardless of nodal involvement, suggests that the molecular

mechanism of hematogenous metastases leading to distant metastases is different from

that of lymphogenic metastases leading to regional metastases.19

As stated by Fisher‘lymph node metastases seem to be only ‘‘indicators’’ and not ‘‘instigators’’ of metastatic

disease’.20 With the strong prognostic information provided by the 70-gene signature,

axillary staging might become less important for guiding adjuvant treatment. Since the

signature accurately classifies as many as 41% of patients with 1-3 positive nodes as good

prognosis, application of the 70-gene prognosis signature could result in a safe reduction

of chemotherapy treatment in up to 41% of these patients. The distant relapse rate of 3% at

10 years in chemotherapy-untreated patients who were classified as good prognosis by the

70-gene signature (data not shown), further substantiate that withholding chemotherapy

in this group seems justified, and implies a major change in the treatment of node-positive

breast cancer.

8/10/2019 Stella Mook T3-108



87

5

This independent retrospective validation study provides additional strong evidence that

the 70-gene signature is a powerful predictor of disease outcome in patients with 1-3

positive nodes, both in chemotherapy-treated and untreated patients. Based on the results

of this study the inclusion criteria of the MINDACT trial (EORTC 10041 BIG 3-04), which is

currently prospectively validating the 70-gene signature in node-negative patients, will be

enlarged to include patients with 1-3 positive nodes.21 Furthermore, our validation study

shows that the signature adds independent prognostic information to that provided by

traditional clinicopathological factors and can accurately identify patients with node-

positive breast cancer and an excellent disease outcome, which would allow a more

tailored approach for adjuvant systemic therapy in this patient group.

Acknowledgements

The authors would like to thank Hugo Horlings for providing immunohistochemistry

data and Dimitry Nuyten for updating the clinical data for the Van de Vijver series and

Michael Hauptman for helping with part of the statistical analyses. We are indebted to

Sjoerd Rodenhuis, Rene Bernards, Marleen Kok and Philippe Bedard for critically reading

the manuscript. This study was supported by the European Commission Framework

Programme VI-TRANSBIG, the Dutch National Genomics Initiative-Cancer Genomics Center,

and an unrestricted research grant from Agendia B.V.


Laura J Van ‘t Veer is a named inventor on a patent application for MammaPrint™ and

reports holding equity in Agendia B.V. Arno Floore and Annuska M Glas are employees of

Agendia B.V..

8/10/2019 Stella Mook T3-108


Chapter 5

88

References

1. Carter CL, Allen C, Henson DE. Relation of tumor size, lymph node status, and survival in 24,740 breast

cancer cases. Cancer 1989; 63: 181-187.

2. Page DL. Prognosis and breast cancer. Recognition of lethal and favorable prognostic types. Am J Surg

Pathol 1991; 15: 334-349.

3. Rosen PP, Groshen S, Saigo PE, Kinne DW, Hellman S. Pathological prognostic factors in stage I (T1N0M0)

and stage II (T1N1M0) breast carcinoma: a study of 644 patients with median follow-up of 18 years.

J Clin Oncol 1989; 7: 1239-1251.



2005; 365: 1687-1717.

5. Joensuu H, Pylkkanen L, Toikkanen S. Long-term survival in node-positive breast cancer treated by

locoregional therapy alone. Br J Cancer 1998; 78: 795-799.

6. Buyse M, Loi S, Van ‘t Veer L, et al. Validation and clinical utility of a 70-gene prognostic signature for


7. Van ‘t Veer LJ, Dai H, Van de Vijver MJ, et al. Gene expression profiling predicts clinical outcome of breast

cancer. Nature 2002; 415: 530-536.

8. Van de Vijver MJ, He YD, Van ‘t Veer LJ, et al. A gene-expression signature as a predictor of survival in


9. Chang HY, Nuyten DSA, Sneddon JB, et al. Robustness, scalability, and integration of a wound-response


10. Glas AM, Floore A, Delahaye LJ, et al. Converting a breast cancer microarray signature into a high-


11. Olivotto IA, Bajdik CD, Ravdin PM, et al. Population-based validation of the prognostic model

ADJUVANT! for early breast cancer. J Clin Oncol 2005; 23: 2716-2725.



13. Albain K, Barlow W, Shak S., et al. Prognostic and predictive value of the 21-gene recurrence score assay

in postmenopausal, node-positive, ER-positive breast cancer (S8814, INT0100). San Antonio Breast

Cancer Symposium 2007 (abstract no 10).14. Desmedt C, Piette F, Loi S, et al. Strong time dependence of the 76-gene prognostic signature for node-

negative breast cancer patients in the TRANSBIG multicenter independent validation series. Clin Cancer

Res 2007; 13: 3207-3214.

15. Foekens JA, Atkins D, Zhang Y, et al. Multicenter validation of a gene expression-based prognostic

signature in lymph node-negative primary breast cancer. J Clin Oncol 2006; 24: 1665-1671.

16. Ma XJ, Hilsenbeck SG, Wang W, et al. The HOXB13:IL17BR expression index is a prognostic factor in

early-stage breast cancer. J Clin Oncol 2006; 24: 4611-4619.

8/10/2019 Stella Mook T3-108



89

5

17. Look MP, van Putten WL, Duffy MJ, et al. Pooled analysis of prognostic impact of urokinase-type

plasminogen activator and its inhibitor PAI-1 in 8377 breast cancer patients. J Natl Cancer Inst 2002; 94:

116-128.

18. Sorlie T, Tibshirani R, Parker J, et al. Repeated observation of breast tumor subtypes in independent

gene expression data sets. Proc Natl Acad Sci USA 2003; 100: 8418-8423.

19. Weigelt B, Wessels LF, Bosma AJ, et al. No common denominator for breast cancer lymph node

metastasis. Br J Cancer 2005; 93: 924-932.

20. Fisher B. The evolution of paradigms for the management of breast cancer: a personal perspective.

Cancer Res 1992; 52: 2371-2383.

21. Bogaerts J, Cardoso F, Buyse M, et al. Gene signature evaluation as a prognostic tool: challenges in the


8/10/2019 Stella Mook T3-108


Chapter 5

90

Supplements Chapter 5

Supplementary table 3. Clinical and pathological characteristics of new validation series (NKI-

EIO; n=241) and of the 106 patients selected from the Van de Vijver study (Ref. Van de Vijver et

al, NEJM 2002).

NKI-EIO

n = 241

Van de Vijver

n = 106 P value*


Age (years) < 0.001

Mean (SD) 50 (7) 45 (5)

<40 23 9.5 16 15.0

40 - 49 102 42.4 68 64.2

50 - 59 86 35.7 22 20.860 - 70 30 12.4 0 .0

Surgery 0.48

BCT 144 59.8 59 55.7

Mastectomy 97 40.2 47 44.3

Axillary procedure < 0.001

ALND 158 65.6 106 100.0

SLNP & ALND 83 34.4 0 .0

Nodal status 0.78

1 positive node 123 51.0 58 54.7

2 positive nodes 77 32.0 27 25.5


Tumor size (pTNM) 0.23

pT1 (≤ 20mm) 117 48.5 65 61.3

pT2 (> 20 - 50mm) 121 50.3 41 38.7

pT3 (> 50mm) 3 1.2 0 .0

Histological tumor type NA

Ductal 204 84.6 0 .0

Lobular 15 6.2 0 .0

Mixed 17 7.1 0 .0

Other 5 2.1 0 .0

Unknown 0 106 100.0Histological grade 0.07

Good 57 23.8 35 33.0

Moderate 99 41.5 41 38.7

Poor 83 34.7 30 28.3

Unknown 2 0

Estrogen-receptor status 0.56

Negative 50 20.7 18 18.0

Positive 191 79.3 82 82.0

Unknown 0 6

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108


Chapter 5

92

Supplementary table 4. Association between clinicopathological characteristics and the 70-

gene prognosis-signature for 106 patients selected from the Van de Vijver study (Ref. Van de

Vijver NEJM 2002).

70-gene prognosis-signature

Good prognosis signature

(n=43)

Poor prognosis signature

(n=63) P value*


Age (years) 0.89

<40 4 9.3 12 19.1

40 - 49 32 74.4 36 57.1

50 - 55 7 16.3 15 23.8

Surgery 0.98

BCT 24 55.8 35 55.6Mastectomy 19 44.2 28 44.4

Axillary procedure NA

ALND 43 100.0 63 100.0

SLNP & ALND 0 .0 0 .0

Nodal status 0.82

1 positive node 24 55.8 34 54.0



Tumor size (pTNM) 0.14

pT1 (≤ 20mm) 30 69.8 35 55.6

pT2 (> 20 - 50mm) 13 30.2 28 44.4

Histological tumor type NA

Unknown 43 100.0 63 100.0

Histological grade < 0.001

Good 24 55.8 11 17.5

Moderate 18 41.9 23 36.5

Poor 1 2.3 29 46.0

Estrogen-receptor status 0.03

Negative 3 7.5 15 25.0

Positive 37 92.5 45 75.0

Unknown 3 3

Progesterone-receptor status < 0.001

Negative 4 10.0 28 46.7

Positive 36 90.0 32 53.3

Unknown 3 3

HER2/NEU receptor status 0.27

Negative 34 89.5 49 80.3

Positive 4 10.5 12 19.7

Unknown 5 2

8/10/2019 Stella Mook T3-108



93

5

70-gene prognosis-signature

Good prognosis signature(n=43)

Poor prognosis signature(n=63)

P value*

Adjuvant systemic treatment 0.28

None 6 14.0 13 20.6

Chemotherapy only 26 60.4 37 58.8

Endocrine therapy only 4 9.3 9 14.3

Both 7 16.3 4 6.3

Adjuvant chemotherapy 0.20

No 10 23.3 22 34.9

Yes 33 76.7 41 65.1

Adjuvant endocrine therapy 0.55

No 32 74.4 50 79.4

Yes 11 25.6 13 20.6

Abbreviations: BCT, breast-conserving therapy; ALND, axillary lymph node dissection; SLNP, sentinel lymph

node procedure.

* Missing values were not used for calculation of p- values.

Supplementary table 4. Continued

8/10/2019 Stella Mook T3-108


Chapter 5

94

Supplementary table 5. Multivariate Cox-regression analyses for new validation series and 106

patients of Van de Vijver study combined; n=320.

a. Multivariate analysis for DM as first event n=320*

Variable P Value Hazard Ratio (95% CI)

Age (years) 0.11 0.97 (0.93 - 1.01)

No. of positive nodes 0.05

2 versus 1 0.48 0.77 (0.38 – 1.58)

3 versus 1 0.05 1.88 (1.01 – 3.49)

Tumor size (> 20 mm versus ≤ 20 mm) 0.16 1.52 (0.85 – 2.73)

Histological grade 0.14

Moderate versus good 0.83 1.10 (0.46 – 2.62)

Poor versus good 0.13 2.08 (0.81 – 5.35)Estrogen-receptor status 0.60 1.19 (0.62 – 2.30)

HER2/NEU receptor status 0.98 0.99 (0.51 – 1.92)

Surgery (mastectomy versus BCT) 0.33 1.31 (0.76 – 2.26)

Chemotherapy 0.03 0.50 (0.27 – 0.94)

Endocrine therapy 0.001 0.36 (0.19 – 0.67)

Prognosis-signature (poor versus good) 0.009 2.79 (1.29 – 6.02)

b. Multivariate analysis for BCSS including an interaction term n=320*

Variable P Value Hazard Ratio (95% CI)

Age (years) 0.35 0.98 (0.94 – 1.02)

No. of positive nodes 0.002

2 versus 1 0.28 0.65 (0.29 – 1.42)

3 versus 1 0.006 2.34 (1.28 – 4.27)

Tumor size (> 20 mm versus ≤ 20 mm) 0.90 1.04 (0.57 – 1.91)

Histological grade 0.14

Moderate versus good 0.77 0.87 (0.34 – 2.23)

Poor versus good 0.27 1.73 (0.66 – 4.60)

Estrogen-receptor status 0.79 0.91 (0.48 – 1.74)

HER2/NEU receptor status 0.90 1.04 (0.54 – 2.02)

Surgery (mastectomy versus BCT) 0.29 1.35 (0.77 – 2.38)

Chemotherapy 0.41 0.50 (0.10 – 2.63)

Endocrine therapy 0.01 0.42 (0.22 – 0.81)

Prognosis-signature (poor versus good) 0.01 5.50 (1.47 – 20.62)

Prognosis-signature * chemotherapy 0.95 0.95 (0.17 – 5.39)

Abbreviations: CI, confidence interval; BCT, breast-conserving therapy.

* Multivariate model included 320 patients due to missing values in 27 patients.

8/10/2019 Stella Mook T3-108



95

5

Supplementary figure 3. Kaplan-Meier curves by 70-gene prognosis-signature for breast

cancer specific survival among the 241 patients.

A) Chemotherapy naïve patients (n=101)

B) Chemotherapy-treated patients (n=128).

Time (years)

1086420

B r e a s t c a n c e r - s p e c i fi c s u r v i v a l

1.0

0.8

0.6

0.4

0.2

0.0

57

44

55

343 41 32 19

54 49 31 6



risk

Log-rank p = 0.003

Time (years)

1086420

B r e a s t c a n c e r - s p e c i fi c s

u r v i v a l

1.0

0.8

0.6

0.4

0.2

0.0

39

89

38

686 78 65 31

38 31 19 8



risk

Log-rank p = 0.02

A

B

8/10/2019 Stella Mook T3-108


Chapter 6

Metastatic potential of T1 breast cancer

can be predicted by the 70-gene

MammaPrint signature

Stella Mook*

Michael Knauer*

Jolien M. Bueno de Mesquita

Valesca P. Retel

Jelle Wesseling

Sabine C. Linn


Emiel J.Th. Rutgers* Contributed equally

Ann Surg Oncol 2010; 17:1406-1413

8/10/2019 Stella Mook T3-108


Chapter 6

98

Abstract

Background

Mammographic screening and increased awareness has led to an increase in the detection

of T1 breast tumors that are generally estimated as having low risk of recurrence after

locoregional treatment. However, even small tumors can metastasize, which leaves us

with the question for the necessity of adjuvant treatment. Therefore, additional prognostic

markers are needed to tailor adjuvant systemic treatment for these relatively low-risk

patients. The aim of our study was to evaluate the accuracy of the 70-gene MammaPrint™

signature in T1 breast cancer.

Materials and Methods

We selected 964 patients from previously reported studies with pT1 tumors (≤ 2 cm).

Frozen tumor samples were hybridized on the 70-gene signature array at the time of the

initial study and classified as having good prognosis or poor prognosis.

Results

The median follow-up was 7.1 years (range 0.2–25.2). The 10-year distant metastasis-free

(DMFS) and breast cancer specific survival (BCSS) probabilities were 87% (SE 2%) and91% (SE 2%), respectively, for the good prognosis-signature group (n = 525), and 72% (SE

3%) and 72% (SE 3%), respectively, for the poor prognosis signature group (n = 439). The

signature was an independent prognostic factor for BCSS at 10 years (multivariate hazard

ratio [HR] 3.25 [95% confidence interval, CI, 1.92–5.51; p < 0.001]). Moreover, the 70-gene

MammaPrint™ signature predicted DMFS at 10 years for 139 patients with pT1ab cancers

(HR 3.45 [95% CI 1.04-11.50, P = 0.04]).

Conclusions

The 70-gene MammaPrint™ signature is an independent prognostic factor in patients with

pT1 tumors and can help to individualize adjuvant treatment recommendation in this

increasing breast cancer population.

8/10/2019 Stella Mook T3-108


The 70-gene signature in pT1 breast cancer

99

6

Introduction

Primary tumor size, in addition to axillary lymph node status, is considered to be one of

the most important prognostic factors in breast cancer, with small tumor size being an

indicator of good prognosis.1-5 However, even small tumors can metastasize, suggesting

that the ability to metastasize is an early and inherent genetic property. 6,7 Adjuvant

treatment decisions based on tumor size alone are only moderately accurate and could

result in undertreatment of T1ab and overtreatment of T1c tumors. The need for adjuvant

systemic therapy after locoregional therapy for patients with small tumors is unresolved.8,9

Currently used treatment guidelines give different recommendations for pT1ab and pT1c

tumors and often the advice ‘consider chemotherapy’ is given, without providing specific

advice for the use of prognostic factors.10-12

With the widespread introduction of breast cancer screening programs and increased

awareness, the proportion of patients presenting with small tumors is ever increasing;

therefore, robust and reliable prognostic factors that can identify patients who are at high

risk of developing distant metastases despite their small tumor are needed.13-15 In previous

validation studies, the 70-gene MammaPrint™ signature accurately distinguished patients

with a good prognosis from those with a poor prognosis in both node-negative and node-

positive breast cancer.16-21 The aim of our study was to evaluate the prognostic value of the

70-gene signature in small pT1 tumors. In addition, we investigated whether the 70-gene

signature could provide clinical utility; that is, if it was able to identify a subgroup of patients

with pT1ab tumors with a poor prognosis as indication for chemotherapy and a subgroupof patients with a pT1c tumor and a good prognosis as indication for no adjuvant treatment

or endocrine therapy only. We merged databases from previous studies to overcome the

underrepresentation of pT1 tumors.16-18,20-23

Methods

Patients

For this study we selected patients with pT1 tumors from previously reported studies.16-18,20-23

Selection criteria for the initially reported studies are depicted in Supplementary Figure 1. For 2

series (i.e., 295-series and RASTER-series) the follow-up was updated since initial publication

(median updated follow-up 10.3 and 2.4 years, respectively).16,22 Each of the series were

consecutive selections from the comprehensive institutional tissue banks. Patients in series

1, 4, 5, and 6 received adjuvant systemic therapy according to national guidelines applicable

at that time.16,18,20,21,24Patients from series 2 and 7 were selected based on adjuvant systemic

therapy received, that is, no adjuvant systemic therapy for patients in series 2 and adjuvant

tamoxifen monotherapy for patients in series 7.17,23 Patients from series 3 (prospective

8/10/2019 Stella Mook T3-108


Chapter 6

100

RASTER trial) were treated according to the national Dutch guideline and the result of the

70-gene signature.22,24 There were 15 patients, all classified as poor prognosis-signature

(1.6%), who received adjuvant trastuzumab. All individual studies were approved by the

ethical committee of the respective hospitals.

70-gene MammaPrint™ signature

Frozen tumor samples were processed at Agendia’s laboratories (Amsterdam, the

Netherlands), for RNA isolation, amplification, and labeling as previously described.7,25

Samples were eligible for RNA isolation if they contained at least 30–50% tumor cells on

hematoxylin/eosin stained sections. To assess the mRNA expression level of the 70 genes,

RNA was hybridized to a custom-designed array (MammaPrint™), blinded to clinical data,

at Agendia’s ISO17025-certified and CLIA accredited laboratories. Tumors were classified

as 70-gene good or poor prognosis signature at time of the initial series as described

previously.16-18,20-23 On average, the 70-gene signature could be performed in 81% of the

patients, which is in accordance with our previously published feasibility study.26 When

sufficient RNA could be extracted, the success rate of hybridization was 100%. For detailed

information about dropout of patients because of tumor cell content or RNA quality, we

refer to the initial publications.

Endpoints

Endpoints considered were time from surgery to distant metastasis (DMFS), and breast

cancer specific survival (BCSS), defined as time from surgery to breast cancer-related death.

For the analysis of distant metastasis-free survival (DMFS) we considered distant metastases

as failure; patients were censored on date of death from causes other than breast cancer or

date of last follow-up visit. For the analysis of BCSS, patients were censored on date of last

follow-up or date of death from causes other than breast cancer. Clinicopathological data

were collected as previously reported, and databases were pooled for our current study (M.

Knauer, unpublished).


Associations between 70-gene signature and classical clinicopathological factors were

studied using chi-square and Mann–Whitney tests. Kaplan–Meier survival analyses and

log-rank tests were used to assess the difference in distant metastasis-free survival (DMFS)

and breast cancer-specific survival (BCSS) of the predicted good- and poor prognosis

groups by the signature. Cox proportional hazard analyses were used to calculate

univariate and multivariate hazard ratios (HR) and their 95% confidence intervals (95% CI).

Multivariate Cox proportional hazard analyses included the 70-gene signature and known

8/10/2019 Stella Mook T3-108



101

6

clinicopathological prognostic factors. We missed information about grade (n = 10), nodal

status (n = 8), ER status (n = 1), and adjuvant systemic therapy (n = 4) for a small proportion

of the patients. These patients were excluded for the multivariate Cox proportional hazard

analyses. HRs for DMFS and BCSS at 10 years were calculated with right-censoring follow-

up > 10 years, because timing of collection of follow-up data differed for the 7 series.

Hazard ratios with their 95% CI for the 7 patient series were displayed on forest plots and

tested for heterogeneity using a chi-square test with 6 degrees of freedom. All p-values are

two-sided. Analyses were performed using SPSS version 15.0 (SPSS Inc, Chicago, IL) and

Revman 5 (Review Manager) (www.cc-ims.net/revman).

Results

A total of 964 patients with pT1 tumors were selected from the 7 studies ( Supplementary Figure 1).16-18,20-23 Among the 964 patients, 139 patients (14%) had a pT1ab tumor, 825 patients (86%)

had a pT1c tumor, 693 patients (72%) had node-negative breast cancer, and 263 patients

(27%) had node-positive breast cancer. During follow-up (median 7.1 years; range 0.2-25.2

years) 154 patients developed distant metastases and 155 patients died, of whom 130 of

breast cancer.

The signature classified 525 tumors (54%) as good prognosis and 439 (46%) tumors as

poor prognosis. A poor prognosis signature was associated with younger age at diagnosis,

invasive ductal carcinoma, poorly differentiated, ER negative and HER2 positive tumors.In addition, patients with a 70-gene poor prognosis tumor more often received adjuvant

systemic therapy (Table 1).

DMFS and BCSS were significantly better in the good prognosis group ( Figure 1A and B). The

probability of remaining free of distant metastases at 5 and 10 years were 95% (SE 1%)

and 87% (SE 2%), respectively for the good prognosis-signature group, and 80% (SE 2%)

and 72% (SE 3%), respectively for the poor prognosis-signature group (Figure 1A). A poor

prognosis-signature was associated with worse DMFS at 10 years, with a univariate hazard

ratio (HR) of 2.70 (95% CI 1.88-3.88; p < 0.001). The 5- and 10-year BCSS probabilities were

99% (SE 1%) and 91% (SE 2%), respectively, for the good prognosis group and 88% (SE 2%)and 72% (SE 3%), respectively, for the poor prognosis-signature group, with a univariate HR

of 4.22 (95% CI 2.70-6.60; p < 0.001) at 10 years (Figure 1B). Forest plots of univariate hazard

ratios for the signature in each individual series showed no significant heterogeneity for

the prognostic value of the signature for both DMFS (chi square = 6.18; p = 0.4) and BCSS

(chi square = 5.99; p = 0.4) ( Supplementary Figure 2).

8/10/2019 Stella Mook T3-108


Chapter 6

102

Table 1. Association between clinicopathological characteristics and the 70-gene signature.

Good prognosis profile Poor prognosis profile P -value*

Surgery 0.09BCT 402 76.6% 315 71.8%

Mastectomy 123 23.4% 124 28.2%

Age <0.001

≤ 50 yrs 282 53.7% 290 66.1%

> 50 yrs 243 46.3% 149 33.9%

Histology <0.001

IDC 435 82.9% 399 90.9%

ILC 54 10.3% 17 3.9%

Others 36 6.8% 23 5.2%

Tumor size 0.13pT1a/b 84 16.0% 55 12.5%

pT1c 441 84.0% 384 87.5%

Nodal status 0.69

Node negative 380 73.1% 313 71.8%

Node positive 140 26.9% 123 28.2%

Unknown 5 3

Grade <0.001

Grade 1 224 43.0% 56 12.9%

Grade 2 248 47.6% 164 37.9%

Grade 3 49 9.4% 213 19.2%

Unknown 4 6

Estrogen-receptor status <0.001

Positive 513 97.7% 295 67.4%

Negative 12 2.3% 143 32.6%

Unknown 0 1

HER2 status <0.001

Negative 402 95.7% 253 77.6%

Positive 18 4.3% 73 22.4%

Unknown 105 113

Adjuvant systemic therapy <0.001

None 357 68.3% 195 44.6%

HT only 113 21.6% 79 18.1%

CT only 22 4.2% 76 17.4%

HT & CT 31 5.9% 87 19.9%

Unknown 2 2

Total 525 100.0% 439 100.0%

* Missing data were not used for calculation of p-values

BCT, breast-conserving therapy; IDC, invasive ductal carcinoma; ILC, invasive lobular carcinoma; ER, estrogen

receptor; HT, hormonal therapy; CT, chemotherapy.

8/10/2019 Stella Mook T3-108



103

6

Figure 1. Kaplan-Meier curves and univariate hazard ratio (HR) for distant metastasis-free

survival (DMFS) and breast cancer-specific survival (BCSS) by 70-gene prognosis-signature for

964 patients with pT1 breast tumors (A and B), for 139 patients with pT1ab tumors (C and D), and

for 825 patients with pT1C tumors (E and F).

Time (years)

1086420

D i s t a n t m e t a s t a s e s - f r e e s u

r v i v a l

1.0

0.8

0.6

0.4

0.2

0.0

525

439

456

117384 261 221 172

336 290 234 169



risk

Log-rank p < 0.001

HR at 10 yrs: 2.70 (95% CI 1.88-3.88); p < 0.001

95%

80%

87%

72%

1086420


1.0

0.8

0.6

0.4

0.2

0.0

525

439

460

124400 284 239 183

344 300 252 179



risk

Log-rank p < 0.001

99%

88%

72%

91%

HR at 10 yrs: 4.22 (95% CI 2.70-6.60); p < 0.001

Time (years)

Time (years)

1086420


1.0

0.8

0.6

0.4

0.2

0.0

84

55

67

1049 28 22 16

49 37 33 25


Good prognosis-signatureNumbers atrisk

Log-rank p = 0.016

98%

86%

90%

76%

HR at 10 yrs: 3.45 (95% CI 1.04-11.50); p = 0.04

1086420


1.0

0.8

0.6

0.4

0.2

0.0

84

55

69

1051 28 24 17

50 39 34 25


Good prognosis-signatureNumbers atrisk

Time (years)

Log-rank p = 0.06

100%

90%

88%

73%

HR at 10 yrs: 3.12 (95% CI 0.91-10.67); p = 0.07

Time (years)

1086420


1.0

0.8

0.6

0.4

0.2

0.0

441

384

388

107335 234 199 156

287 253 201 144



risk

Log-rank p < 0.001

95%

80%

86%

72%

HR at 10 yrs: 2.61 (95% CI 1.78-3.82); p < 0.001

1086420


1.0

0.8

0.6

0.4

0.2

0.0

441

384

392

114350 256 215 168

294 261 218 154



risk

Time (years)

Log-rank p < 0.001

99%

88%

92%

72%

HR at 10 yrs: 4.42 (95% CI 2.73-7.17); p < 0.001

A

C

E

B

D

F

8/10/2019 Stella Mook T3-108


Chapter 6

104

Univariate analysis showed that besides the 70-gene signature, age, histology, tumor

grade, ER status, HER2 status, and type of surgery were significant predictors for DMFS at 10

years. The 70-gene signature, age, tumor grade, ER status, and HER2 status were significant

predictors for BCSS at 10 years ( Supplementary table 1). In a multivariate model, the 70-gene

signature was the strongest predictor for DMFS, with an adjusted HR of 2.43 (95% CI 1.56-

3.77; p < 0.001). In addition to the signature, nodal status and adjuvant systemic therapy

were independent significant predictors for DMFS. For BCSS, again the signature, nodal

status, and adjuvant chemotherapy were independent prognostic factors with adjusted

HRs of 3.25 (95% CI 1.92-5.51; p < 0.001), 1.70 (95% CI 1.12-2.57; p = 0.01) and 0.41 (95% CI

0.22-0.75; p = 0.004), respectively (Table 2).

Table 2. Multivariate Cox proportional hazard analyses for distant metastasis-free survival and

breast cancer-specific survival at 10 years.

VariableDistant metastases

Breast cancer-specific

survival

HR CI P-value HR CI P-value

MammaPrint (poor versus good signature) 2.43 1.56-3.77 <0.001 3.25 1.92-5.51 < 0.001

Age (years) 0.99 0.97-1.01 0.31 0.98 0.96-1.01 0.16

Histology

ILC (versus IDC) 1.41 0.70-2.83 0.33 1.70 0.81-3.57 0.16

Other (versus IDC) 0.39 0.12-1.26 0.12 0.46 0.14-1.48 0.19

Tumor size (11-20 mm versus ≤10 mm) 1.07 0.59-1.97 0.82 0.88 0.46-1.67 0.69

Nodal status (negative versus positive) 1.61 1.13-2.29 0.01 1.70 1.12-2.57 0.01

Grade

Grade 2 (versus grade 1) 1.28 0.77-2.11 0.34 1.31 0.72-2.39 0.38


ER status (negative versus positive) 0.92 0.58-1.46 0.72 1.36 0.85-2.17 0.20

HER2/NEU status

Positive (versus negative) 1.19 0.69-2.04 0.53 1.48 0.84-2.61 0.18

Unknown (versus negative) 0.75 0.48-1.19 0.22 0.94 0.57-1.55 0.80

Surgery (mastectomy versus BCT) 1.38 0.95-1.99 0.09 1.23 0.82-1.85 0.32

Hormonal therapy (versus no hormonal therapy) 0.55 0.34-0.89 0.02 0.61 0.34-1.07 0.09

Chemotherapy (versus no chemotherapy) 0.50 0.29-0.84 0.01 0.41 0.22-0.75 0.004

Multivariate models included 941 patients due to missing values for nodal status, grade, and/or ER status

in 23 patients.

IDC, invasive ductal carcinoma; ILC, invasive lobular carcinoma; ER, estrogen receptor; BCT, breast-

conserving therapy; HR, hazard ratio; CI 95%, confidence interval.

8/10/2019 Stella Mook T3-108



105

6

Besides our initial selection of patients with tumors ≤ 20 mm, we divided our study cohort

into patients with pT1ab tumors (≤ 10 mm) (n = 139) and patients with pT1c tumors (11-20

mm) (n = 825). Of the patients with a pT1ab tumor, 40% were classified as having a 70-gene

poor prognosis tumor. The DMFS in these 55 patients was significantly worse compared

with the DMFS in patients with a good prognosis signature tumor during the entire follow-

up (log rank p = 0.016) and at 10 years (HR 3.45 [95% CI 1.04-11.50; p = 0.044]). The same

trend was seen for BCSS at 10 years, albeit borderline significant (Figure 1, panel C and D). The

number of events was too small to calculate adjusted HRs in patients with pT1ab tumors.

In patients with a pT1c tumor the 70-gene signature was a prognostic factor for both DMFS

and BCSS at 10 years (HR 2.61 [95% CI 1.78-3.82; p < 0.001] and 4.42 [95% CI 2.73-7.17; p <

0.001], respectively) (Figure 1, panel E and F ).

Among the 964 patients in our study cohort, 552 patients (57%) received no adjuvant

systemic therapy, 408 patients (42%) received endocrine- and/or chemotherapy and for

4 patients (1%) adjuvant systemic therapy was unknown. The 70-gene signature retained

its prognostic value in adjuvantly untreated patients, with adjusted HRs of 2.54 (95% CI

1.49-4.34; p = 0.001) and 3.47 (95% CI 1.83-6.60; p < 0.001) for DMFS and BCSS, respectively

( Supplementary Figure 3). In addition, the 70-gene MammaPrint™ signature was an independent

prognostic factor in 788 patients with ER positive tumors for both DMFS and BCSS, with

adjusted HRs of 2.51 (95% CI 1.60-3.95; p < 0.001) and 3.43 (95% CI 1.98-5.95; p < 0.001),

respectively.

Discussion

Our study showed that the 70-gene MammaPrint™ signature that has been validated as

an independent prognostic factor in node-negative and node-positive breast cancer is

also an independent prognostic factor in patients with small breast tumors. The signature

accurately distinguished patients with a good outcome from those with a poor outcome in

our study cohort of patients with pT1 tumors. Interestingly, our results showed that even

a considerable proportion of small tumors have a substantial metastatic capacity, which

can be identified by the 70-gene signature (28% distant relapse rate at 10 years in tumorsclassified as poor prognosis by the signature). Therefore, the 70-gene signature can be of

use in daily clinical practice to optimize and individualize treatment decision-making in

this growing breast cancer population of patients with pT1 tumors.

Historical tumor banks contain particularly large tumors, which is in contrast to the actual

increase in proportion of small tumors diagnosed, due to mammographic screening.

As a consequence, most studies of potential prognostic markers to date included fewer

than 10% of tumors smaller than 10 mm.7,27-29 To overcome this potential problem of

underrepresentation of small tumors in repositories we selected patients with pT1 tumors

from previous studies. One of the potential limitations of our pooled database is therefore

8/10/2019 Stella Mook T3-108


Chapter 6

106

the heterogeneity of our study population, especially with regard to years of diagnosis and

adjuvant systemic therapy. However, there was no evidence of significant heterogeneity

among the various series that provided pT1 cases for our study, for both DMFS and BCSS

( Supplementary Figure 2). Another potential bias with regard to adjuvant systemic therapy in

our study cohort is the fact that patients from the more recent prospective RASTER trial

were partially treated based on the outcome of the 70-gene signature.22 However, when

we excluded these patients from our series (n = 301) the 70-gene prognosis signature

retained its independent prognostic value with adjusted HRs for DMFS and BCSS of 2.32

(95% CI 1.48-3.63; p < 0.001) and 3.09 (95% CI 1.82-5.24; p < 0.001), respectively. Patients

with pT1 tumors selected from series 1 included 31 patients whose data were used in the

development of the 70-gene signature, thereby potentially causing an overestimation of

the prognostic value of the signature.7 Excluding these patients from our analyses resulted

in similar adjusted HR for both DMFS and BCSS at 10 years (2.24 [1.41-3.54; p = 0.001] and

3.10 [1.79-5.37; p < 0.001] respectively).

Our study cohort contained 139 patients with pT1ab tumors, and in this small group of

patients with pT1ab tumors 15 distant metastases (DM) and 11 breast cancer-specific

deaths (BCSD) occurred. The 70-gene signature was able to distinguish patients with pT1ab

tumors who developed DM from those who did not (log rank p = 0.016 and HR 3.45, p

= 0.044). For BCSS the same trend was observed, but did not, however, reach statistical

significance, which is most likely because of the low number of events in combination

with the relatively small patient population with pT1ab tumors. The results of our study

suggest that the 70-gene signature can select patients with pT1ab tumors with a higherrisk of developing DM (24% at 10 years), who thus might be candidates for adjuvant

systemic therapy. In addition, the signature can identify patients with a pT1c tumor with

a relatively low risk of developing DM (14% at 10 years), who might be sufficiently treated

with endocrine therapy, as the large majority (98%) is ER positive. Since adjuvant systemic

treatment recommendation for patients with small tumors is a matter of debate, our results

provide evidence that selecting patients with pT1 tumors using the 70-gene signature

could be relevant for adjuvant systemic therapy recommendation. For patients with pT1ab

tumors the data suggest the same, though we will have to await results of further studies.

With the current development of RNA extraction from FNA and core biopsies for microarraygene expression analyses, gene expression profiles will become available to a larger extent

for patients with very small tumors.30

Hanrahan and colleagues showed that, in addition to a relatively wide range of observed

relapse-free survival rates in patients with pT1abN0 tumors, histological grade, ER negative

tumor, younger age at diagnosis (<50 years), lymphovascular invasion (LVI), high Ki-67,

HER2/NEU positivity, and larger tumor size were associated with poor outcome. 8,9 In our

database of patients with pT1 tumors, we confirmed the univariate prognostic value of age,

tumor grade, ER and HER2/NEU ( Supplementary Table 1). Among the 954 patients with known

tumor grade, 692 patients had a grade 1 or 2 tumor and 262 patients had a grade 3 tumor.

8/10/2019 Stella Mook T3-108



107

6

Patients with a grade 1–2 tumor had a significantly lower 10 -year distant relapse rate

compared with those with a grade 3 tumor (16 and 29%, respectively), with a univariate

HR of 2.24 (95% CI 1.54-3.08; p < 0.001) ( Supplementary Figure 4). However, the event rate in

patients considered to have a ‘good prognosis’ based on grade ( i.e., with grade 1-2 tumors)

was relatively large. Specifically, classification based on grade (i.e., grade 1-2 considered low

risk) resulted in misclassification of 43 additional DM and 41 additional BCSD, compared

with 4 additional misclassified DM and 3 BCSD when classified by the 70-gene signature.

Classification based on ER status, would result in a large ‘good-prognosis’ group (i.e., 84%

is ER positive), with a relatively high event rate in this group (19% DM at 10 years) (data

not shown). Therefore, ER status alone would not be useful to select patients with pT1

tumors for adjuvant systemic chemotherapy. These results show that while the proportion

of patients classified as having a ‘good prognosis’ by both grade and ER status increased

compared with the good prognosis group by the 70-gene signature, the prediction of

outcome becomes less accurate and an increased proportion of events were missed.

Moreover, results of multivariate analyses showed that the prognostic information that is

captured by ER and grade is not independent of other factors. In fact, only the 70-gene

signature, nodal status, and adjuvant systemic therapy were independent prognostic

factors for DMFS and BCSS in this study cohort.

Tumors identified as 70-gene signature low risk and grade 1/2 showed a considerable

proportion of events (51 DM and 32 BCSD, respectively). However, these misclassified

events occurred significantly later compared with the accurately classified events, 7.4 years

(SE 0.4) versus 3.1 years (SE 0.2), respectively, for DM ( p < 0.001) and 9.3 yrs (SE 0.5) versus 4.6 years (SE 0.3), respectively, for BCSD ( p < 0.001). Previous studies have already shown

a time dependency for the prognostic value of the 70-gene signature, and our results

support once more the hypothesis of a different biological mechanism for early and late

relapses.17,21 Moreover, our results show the still unmet need for markers to predict late

events. The accuracy of the 70-gene MammaPrint™ signature in predicting early events

coincides with the effect of chemotherapy, as that is known to be most beneficial in the

first 5-7 years after diagnosis and would thus potentially prevent the occurrence of early

metastasis in the poor prognosis group.31

As a consequence of adjuvant treatment guidelines, a substantial proportion of patients inthis validation series (216 of 964 patients, 22%) received adjuvant chemotherapy, with or

without hormonal therapy. Patients classified as poor prognosis by the 70-gene signature

more often received adjuvant chemotherapy (37 versus 10% in the good prognosis group,

respectively; p < 0.001). Tumor characteristics in the poor signature group, that is, more ER-

negative and poorly differentiated, are generally believed to be associated with a higher

likelihood of response to chemotherapy.31 Moreover, Bender and colleagues recently

showed that the benefit of chemotherapy was exclusively seen in patients classified as poor

prognosis by the 70-gene signature.32 This larger efficacy of chemotherapy in combination

with the larger proportion of chemotherapy-treated patients in the poor prognosis

8/10/2019 Stella Mook T3-108


Chapter 6

108

signature group would imply that the prognostic value of the 70-gene signature as shown

in our series is underestimated and would potentially be higher in an untreated group.

In conclusion, our study shows that the 70-gene signature is a strong and independent

prognostic factor for patients with pT1 tumors. In addition, we show that a considerable

proportion of small tumors has metastatic potential, supporting the idea that metastatic

capacity is an early genetic inheritance that can be revealed by the 70-gene signature.

Consequently, selecting patients with pT1 tumors based on the signature will result in a

more accurate allocation of adjuvant systemic therapy in this patient population.

Acknowledgements

We are indebted to Marleen Kok and Rutger Koornstra for providing part of the data used

for our analyses and to Annuska M. Glas and Arno Floore of Agendia BV for hybridization of

all tumor samples. We thank Marjanka K. Schmidt for helpful discussions.


Laura J Van ‘t Veer is named inventor on a 70 gene prognosis-signature patent. Laura J Van

‘t Veer reports holding equity in Agendia BV.

8/10/2019 Stella Mook T3-108



109

6

References

1. Carter CL, Allen C, Henson DE. Relation of tumor size, lymph node status, and survival in 24,740 breast

cancer cases. Cancer 1989; 63: 181-187.

2. Fitzgibbons PL, Page DL, Weaver D, et al . Prognostic factors in breast cancer. College of American

Pathologists Consensus Statement 1999. Arch Pathol Lab Med 2000; 124: 966-978.

3. Koscielny S, Tubiana M, Le MG, et al . Breast cancer: relationship between the size of the primary tumour

and the probability of metastatic dissemination. Br J Cancer 1984; 49: 709-715.

4. Page DL. Prognosis and breast cancer. Recognition of lethal and favorable prognostic types. Am J Surg

Pathol 1991; 15: 334-349.

5. Rosen PP, Groshen S, Saigo PE, Kinne DW, Hellman S. Pathological prognostic factors in stage I (T1N0M0)

and stage II (T1N1M0) breast carcinoma: a study of 644 patients with median follow-up of 18 years. J

Clin Oncol 1989; 7: 1239-1251.

6. Bernards R and Weinberg RA. A progression puzzle. Nature 2002; 418: 823.


cancer. Nature 2002; 415: 530-536.

8. Hanrahan EO, Valero V, Gonzalez-Angulo AM, Hortobagyi GN. Prognosis and management of patients

with node-negative invasive breast carcinoma that is 1 cm or smaller in size (stage 1; T1a,bN0M0): a

review of the literature. J Clin Oncol 2006; 24: 2113-2122.

9. Hanrahan EO, Gonzalez-Angulo AM, Giordano SH, et al . Overall survival and cause-specific mortality of

patients with stage T1a,bN0M0 breast carcinoma. J Clin Oncol 2007; 25: 4952-4960.

10. Eifel P, Axelson JA, Costa J, et al . National Institutes of Health consensus development conference


11. NCCN Clinical Practice Guidelines in Oncology. Breast Cancer V.1.2009. www.nccn.org 2009.


Expert Consensus on the primary therapy of early breast cancer 2009. Ann Oncol 2009; 20:1319–1329.

13. Fracheboud J, Otto SJ, van Dijck JA, et al . Decreased rates of advanced breast cancer due to


14. Connor RJ, Chu KC, Smart CR. Stage-shift cancer screening model. J Clin Epidemiol 1989; 42: 1083-1095.


Health Insurance Plan clinical trial. J Natl Cancer Inst 1988; 80: 1125-1132.16. Van de Vijver MJ, He YD, Van ‘t Veer LJ, et al . A gene-expression signature as a predictor of survival in








8/10/2019 Stella Mook T3-108


Chapter 6

110



2009; 116: 295-302.

21. Mook S, Schmidt MK, Weigelt B, et al . The 70-gene prognosis signature predicts early metastasis in

breast cancer patients between 55 and 70 years of age. Ann Oncol 2010; 21: 717-722.



Lancet Oncol 2007; 8: 1079-1087.

23. Kok M, Koornstra RH, Mook S, et al . Additional value of the 70-gene signature and levels of ER and PR for

the prediction of outcome in tamoxifen-treated ER-positive breast cancer. 2009 submitted .

24. CBO Richtlijn Mammacarcinoom 2008. http://www.cbo.nl/product/richtlijnen/folder20021023121843/rl_mamma_08. pdf



26. Mook S, Bonnefoi H, Pruneri G, et al . Daily clinical practice of fresh tumour tissue freezing and gene

expression profiling; logistics pilot study preceding the MINDACT trial. Eur J Cancer 2009; 45: 1201-1208.






2003; 361: 1590-1596.

30. Andre F, Michiels S, Dessen P, et al . Exonic expression profiling of breast cancer and benign lesions: a

retrospective analysis. Lancet Oncol 2009; 10: 381-390.



2005; 365: 1687-1717.

32. Bender RA, Knauer M, Rutgers EJ, et al . The 70-gene profile and chemotherapy benefit in 1,600 breast

cancer patients. ASCO Annual Meeting Proceedings 2009. J Clin Oncol 2009; 27 (Supplement; abstract no.

512).

8/10/2019 Stella Mook T3-108



111

6


Supplementary Table 1. Univariate Cox-regression analyses for distant metastasis-free survival

and breast cancer-specific survival at 10 years.

Distant metastasesBreast cancer-specific

survival

Variable HR CI P -value HR CI P-value

MammaPrint (poor versus good signature) 2.70 1.88-3.88 <0.001 4.22 2.70-6.60 <0.001

Age (years) 0.98 0.96-0.996 0.02 0.97 0.95-0.99 0.004

Histology

ILC (versus IDC) 0.96 0.49-1.90 0.91 1.07 0.52-2.21 0.85

Other (versus IDC) 0.37 0.09-0.88 0.05 0.45 0.17-1.23 0.12

Tumor size (11-20 mm versus ≤10 mm) 1.49 0.82-2.69 0.19 1.23 0.66-2.29 0.52

Nodal status (negative versus positive) 1.07 0.94-1.21 0.29 0.94 0.77-1.16 0.58

Grade


Grade 3 (versus grade 1) 2.89 1.79-4.67 <0.001 3.82 2.18-6.71 <0.001

ER status (negative versus positive) 1.78 1.19-2.65 0.005 2.86 1.91-4.29 <0.001

HER2/NEU status

Positive (versus negative) 1.67 1.00-2.78 0.05 2.43 1.42-4.14 0.001Unknown (versus negative) 0.89 0.60-1.32 0.56 1.22 0.80-1.87 0.35

Surgery (mastectomy versus BCT) 1.54 1.09-2.20 0.02 1.43 0.96-2.11 0.08

Hormonal therapy (versus no hormonal therapy) 0.73 0.49-1.08 0.11 0.67 0.42-1.05 0.08

Chemotherapy (versus no chemotherapy) 1.04 0.68-1.58 0.88 0.98 0.60-1.59 0.93

IDC, Invasive ductal carcinoma; ILC, invasive lobular carcinoma; ER, estrogen receptor; BCT, breast-conserving

therapy; HR, hazard ratio; CI, 95% confidence interval.

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108



113

6

Supplementary figure 2. Forest plots of hazard ratios (HRs) and 95% confidence intervals (95%

CI) for each series in the poor prognosis versus good prognosis group by the 70-gene signature.

Squares = HRs; Lines = 95% CIs; Diamond = weighted total HR and 95% CI.

A. Distant metastasis-free survival at 10 years.

B. Breast cancer-specific survival at 10 years.

NEJM = 295 series17

; TRANSBIG = TRANSBIG series18

; RASTER = RASTER series23

; JBM validation= Bueno de Mesquita validation series19; LN 1-3 = Mook validation series 121; OVER 55 = Mook

validation series 222; TAM ADJ = Kok series24.

Study or Subgroup

1. NEJM

2. TRANSBIG3. RASTER

4. JBM validation

5. LN 1-3

6. OVER 55

7. TAM ADJ

Total (95% CI)

Heterogeneity: Chi² = 6.18, df = 6 (P = 0.40); I² = 3%

Test for overall effect: Z = 4.95 (P < 0.00001)

IV, Fixed, 95% CI

4.28 [1.97, 9.30]

2.63 [1.18, 5.87]1.44 [0.20, 10.22]

2.74 [0.89, 8.39]

6.56 [1.47, 29.32]

0.90 [0.24, 3.39]

1.99 [0.81, 4.90]

2.69 [1.82, 3.98]

Hazard Ratio Hazard Ratio

IV, Fixed, 95% CI

0.05 0.2 1 5 20

Good prognosissignature better

Poor prognosissignature better

Study or Subgroup

1. NEJM

2. TRANSBIG

3. RASTER

4. JBM validation

5. LN 1-3

6. OVER 55

7. TAM ADJ

Total (95% CI)

Heterogeneity: Chi² = 5.99, df = 6 (P = 0.42); I² = 0%Test for overall effect: Z = 5.29 (P < 0.00001)

IV, Fixed, 95% CI

13.93 [3.31, 58.60]

3.52 [1.53, 8.10]

84.61 [0.00, 988383129.70]

2.84 [0.83, 9.74]

11.03 [1.41, 86.23]

1.84 [0.41, 8.22]

2.59 [0.94, 7.17]

3.71 [2.28, 6.03]

Hazard Ratio Hazard Ratio

IV, Fixed, 95% CI

0.05 0.2 1 5 20Good prognosissignature better

Poor prognosissignature better

A

B

8/10/2019 Stella Mook T3-108


Chapter 6

114

Supplementary figure 3. Kaplan-Meier curves and univariate hazard ratio (HR) by 70-gene

prognosis-signature for patients who did not receive adjuvant systemic therapy (n=552).

A. Distant metastasis-free survival.

B. Breast cancer-specific survival.

Time (years)

1086420


1.0

0.8

0.6

0.4

0.2

0.0

357

195

301

84177 145 131 103

200 171 140 117



risk

Log-rank p < 0.001

96%

78%

86%

70%

HR at 10 yrs: 2.90 (95% CI 1.83-4.79); p < 0.001

1086420


1.0

0.8

0.6

0.4

0.2

0.0

357

195

304

88187 158 140 111

204 178 154 124



risk

Time (years)

Log-rank p < 0.001

99%

85%

69%

91%

HR at 10 yrs: 4.67 (95% CI 2.67-8.18); p < 0.001

A

B

8/10/2019 Stella Mook T3-108



115

6

Supplementary figure 4. Kaplan-Meier curves and univariate hazard ratio (HR) by grade 1/2

versus grade 3 (n=954, tumor grade was missing for 10 patients).

A. Distant metastasis-free survival.

B. Breast cancer-specific survival.

Time (years)

1086420

D i s t a n t m e t a s t a s e s - f r e e s u r v i v a

l

1.0

0.8

0.6

0.4

0.2

0.0

692

262

608

71222 145 121 97

442 380 300 208

Grade 3

Grade 1/2Numbers at

risk

Log-rank p < 0.001

92%

78%

84%

71%

HR at 10 yrs: 2.24 (95% CI 1.54-3.08); p < 0.001

1086420

B r e a s t c a n c e r - s p e c i fi c

s u r v i v a l

1.0

0.8

0.6

0.4

0.2

0.0

692

262

616

75234 162 132 102

456 397 324 221

Grade 3

Grade 1/2Numbers at

risk

Time (years)

Log-rank p < 0.001

97%

85%

72%

86%

HR at 10 yrs: 2.56 (95% CI 1.75-3.74); p < 0.001

A

B

8/10/2019 Stella Mook T3-108


Chapter 7

The predictive value of the 70-gene

signature for adjuvant chemotherapy in

early breast cancer

Michael Knauer

Stella Mook

Emiel J.T. Rutgers

Richard A. Bender

Michael Hauptmann

Marc J. Van de Vijver

Rutger H. Koornstra

Jolien M. Bueno de MesquitaSabine C. Linn


Breast Cancer Res Treat 2010; 120: 655-661

8/10/2019 Stella Mook T3-108


Chapter 7

118

Abstract

Multigene assays have been developed and validated to determine the prognosis of

breast cancer. In this study, we assessed the additional predictive value of the 70-gene

MammaPrint™ signature for chemotherapy (CT) benefit in addition to endocrine therapy

(ET) from pooled study series. For 541 patients who received either ET (n = 315) or ET + CT

(n = 226), breast cancer-specific survival (BCSS) and distant disease-free survival (DDFS) at

5 years were assessed separately for the 70-gene high and low risk groups. The 70-gene

signature classified 252 patients (47%) as low risk and 289 (53%) as high risk. Within the

70-gene low risk group, BCSS was 97% for the ET group and 99% for the ET + CT group at 5

years with a non-significant univariate hazard ratio (HR) of 0.58 (95% CI 0.07–4.98; P = 0.62).

In the 70-gene high risk group, BCSS was 81% (ET group) and 94% (ET + CT group) at 5 years

with a significant HR of 0.21 (95% CI 0.07–0.59; P < 0.01). DDFS was 93% (ET) versus 99% (ET

+ CT), respectively, in the 70-gene low risk group, HR 0.26 (95% CI 0.03–2.02; P = 0.20). In

the high risk group DDFS was 76 versus 88%, HR of 0.35 (95% CI 0.17–0.71; P < 0.01). Results

were similar in multivariate analysis, showing significant survival benefit by adding CT in

the 70-gene high risk group. A significant and clinically meaningful benefit was observed

by adding chemotherapy to endocrine treatment in 70-gene high risk patients. This benefit

was not significant in low risk patients, who were at such low risk for recurrence and cancer-

related death, that adding CT does not appear to be clinically meaningful.

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108


Chapter 7

120

Patients and methods

Patients

A pooled database from seven previously reported studies, including 1637 patients with

known adjuvant treatment status (1637/1696, 97%) was developed. Patients who met the

following criteria were selected: unilateral stage pT1-3, N0-1, M0 invasive breast carcinoma

diagnosed between 1984 and 2006, surgical treatment with either breast-conserving

therapy or mastectomy with sentinel node biopsy or axillary lymph node dissection

followed by radiotherapy, if indicated.15 For this analysis, disease was staged according

to the 2002 UICC TNM-classification, 6th edition. All involved studies had been approved

by the respective institutional review boards. We evaluated all patients who had received

either ET alone or ET plus adjuvant CT (ET + CT). In the whole patient population, 90%

of patients were estrogen receptor (ER) positive and 69% of the study patients were

progesterone receptor (PR) positive. The studies by Van de Vijver et al.,6 Bueno de Mesquita

et al.,14,16 Mook et al.,17,18 and Kok et al. (personal communication) were included, resulting

in the inclusion of 30, 182, 29, 154, 27 and 119 patients from the database, respectively.

Differences in adjuvant CT benefit (CMF or anthracycline +/- taxane regimens) within

the 70-gene low risk and high risk patients were assessed. Of 226 patients treated with

adjuvant CT, 11 patients received CMF, 21 patients received taxane containing regimens,

and the vast majority of 194 patients received different anthracycline-containing regimens.

Time-to-event analyses using updated and centrally verified individual patient data wereperformed using a pooled database (Microsoft Access; Microsoft, Redmond, WA).

Microarray analysis

Frozen tumor samples from each patient were processed at Agendia’s laboratory

(Amsterdam, The Netherlands), for RNA isolation, amplification, and labeling as previously

described.5,19 Samples were eligible for RNA isolation, if they contained at least 30% tumor

cells on hematoxylin/eosin stained sections. To assess the mRNA expression level of the

70 genes, RNA was hybridized to a custom-designed array (MammaPrint™) at Agendia’sISO17025-certified, CLIA accredited, and FDA-cleared laboratory. Tumors were classified

as having a 70-gene high or low risk-signature at the time of initial series as described

previously, and were blinded to clinical data.

8/10/2019 Stella Mook T3-108


Predictive value of the 70-gene signature

121

7


The endpoints evaluated were breast cancer-specific survival (BCSS), defined as time from

surgery to breast cancer-related death and distant disease-free survival (DDFS), defined as

time from surgery to any distant metastasis. For both outcomes, follow-up was censored

at 5 years, because firstly, most of the treatment effect of adjuvant CT is observed within 5

years and secondly to control for differences in median follow-up of the included studies.

Kaplan–Meier survival plots and log-rank tests were used to assess differences in BCSS

and DDFS for the 70-gene profile low and high risk groups. All P -values were two-sided

and considered statistically significant if less than 0.05. Adjusted uni- and multivariate

hazard ratios (HRs) and corresponding 95% confidence intervals (95% CIs) were derived

from Cox proportional hazards models. Co-variates used in adjusted models included age

at diagnosis, tumor size, number of positive lymph nodes, histological grade, ER and PR

status, hormonal therapy, and CT. Relative differences between treatment effects by 70-

gene risk groups were assessed by adding an interaction term to the model. All statistical

analyses were performed with SPSS 15.0 for Windows (SPSS Inc., Chicago, IL) and SAS 9.1

(SAS Institute Inc., Cary, NC).

Results

Five-hundred-forty-one patients with 0–3 positive lymph nodes from the pooled databasewere either treated with ET only or endocrine plus CT and, thus, met the inclusion criteria

for this study. The median follow-up for the study population was 7.1 years (range 0.1–25.2).

At 5 years of follow-up, 52 patients had developed distant metastases and 33 patients had

died of their disease. The 70-gene MammaPrint™ signature classified 252 patients (47%)

as low risk and 289 (53%) as high risk. Detailed patient characteristics are shown in Table 1.

Prognostic value of the 70-gene signature

BCSS and DDFS were significantly better in the 70-gene signature low risk group. The5-year BCSS probabilities were 97% for the low risk group and 87% for the high risk-

signature group, with a univariate HR of 4.81 (95% CI 1.98–11.67; P < 0.01). The probability

of remaining free of distant metastases at 5 years was 95% for the low risk-signature group

and 82% for the high risk-signature group with a univariate HR of 3.88 (95% CI 1.99–7.58;

P < 0.01).

8/10/2019 Stella Mook T3-108


Chapter 7

122

Table 1. Summary of clinico-pathological characteristics of the study population.

Patients (n = 541) Characteristics n (%)

Age ≤ 50 years 231 (43%)

> 50 years 310 (57%)

Tumor size T1 279 (52%)

T2 254 (47%)

T3 7 (1%)

n.a. 1 (0.2%)

Lymph node status N0 265 (49%)

N1 276 (51%)

Histological grade Grade 1 134 (25%)

Grade 2 233 (43%)

Grade 3 163 (30%)

n.a. 11 (2%)

Estrogen receptor status Positive (≥ 10%) 484 (90%)

Progesterone receptor status Positive (≥ 10%) 371 (69%)

Her2-status Positive 59 (11%)

Adjuvant treatment ET only 315 (58%)

ET + CT 226 (42%)

70-gene MammaPrint signature Low risk 252 (47%)

High risk 289 (53%)

Abbreviations: n, number; n.a., not available; ET, endocrine therapy; CT, chemotherapy.

Adjuvant CT benefit for the 70-gene signature risk groups

In order to determine the predictive utility of the 70-gene signature, we assessed

differences in survival between patients who received either ET alone or ET combined withCT, separately within the 70-gene low risk and 70-gene high risk patient groups. Univariate

analysis demonstrated a significantly longer DDFS and BCSS in the 70-gene high risk group

for the patients receiving both CT and endocrine treatment, whereas such a significant

difference was not observed for the 70-gene low risk group. BCSS for the 70-gene low risk

group was 97% for the ET group and 99% for the ET + CT group, with a univariate HR of

0.58 (95% CI 0.07–4.98; P = 0.62). In the 70-gene high risk group, 5-year BCSS was 81% for

the ET group and 94% for the ET + CT group with a HR of 0.21 (95% CI 0.07–0.59, P < 0.01).

The corresponding Kaplan–Meier survival curves are shown in Figure 1A (BCSS) and for DDFS

in Figure 1B. In the 70-gene low risk group, DDFS probabilities at 5-years for the ET and the

8/10/2019 Stella Mook T3-108



123

7

ET + CT groups were 93 versus 99%, respectively, with a HR of 0.26 (95% CI 0.03–2.02; P =

0.20). In the high risk group, survival was 76 versus 88% for the ET and the ET + CT groups,

respectively, with a HR of 0.35 (95% CI 0.17–0.71; P < 0.01).

Figure 1. A. Five-year breast cancer-specific survival by treatment within the 70-gene signature

groups (70-gene low risk on the left, high risk on the right).B. Five-year distant disease-free survival by treatment within the 70-gene signature groups (70-

gene low risk on the left, high risk on the right).

Abbreviations: BCSS, breast cancer-specific survival; DDFS, distant disease-free survival; n,

number; ET, endocrine therapy; ET + CT, endocrine + chemotherapy; HR, univariate hazard ratio.

To further evaluate treatment effects, we compared relative and absolute differences in

survival between patient groups receiving ET or ET + CT for both 70-gene risk groups. The

relative differences as determined by the interaction analysis resulted in a P -value of 0.45.

BCSS: M ammaPrint LOW RISK (n=252)

0 1 2 3 4 5

0

20

40

60

80

100

ET (n=174, 69%)

ET+CT (n=78, 31%)

99%

97%

HR = 0.58 (95% CI 0.07-4.98)

p = 0.62

Time (years)

P e r c e n t s u r v i v a l

BCSS: M ammaPrint HIGH RISK (n=289)

0 1 2 3 4 5

0

20

40

60

80

100

ET (n=141, 49%)

ET+CT (n=148, 51%)

94%

81%

HR = 0.21 (95% CI 0.07-0.59)

p < 0.01

Time (years)


DDFS: M ammaPrint LOW RISK (n=252)

0 1 2 3 4 5

0

20

40

60

80

100

ET (n=174, 69%)

ET+CT (n=78, 31%)

99%

93%

HR = 0.26 (95% CI 0.03-2.02)

p = 0.20

Time (years)


DDFS: MammaPrint HIGH RISK (n=289)

0 1 2 3 4 5

0

20

40

60

80

100

ET (n=141, 49%)

ET+CT (n=148, 51%)

88%

76%

HR = 0.35 (95% CI 0.17-0.71)

p < 0.01

Time (years)

P e r c e n t s u

r v i v a l

A

B

8/10/2019 Stella Mook T3-108


Chapter 7

124

In absolute numbers, the addition of CT for patients in the 70-gene low risk group could

prevent 3 events per 1000 patient years, resulting in a number needed to treat (NNT) of

333 (95% CI 78 harm to 83 benefit, i.e., there is similar chance that adding CT may result

in benefit or harm for this patient group). Adding CT for patients in the 70-gene high risk

group could prevent 33 events per 1000 patient years, resulting in a NNT of 30 (95% CI 19

benefit to 64 benefit).

In multivariate Cox regression analysis adjusted for age, tumor size, number of positive

lymph nodes, grade, ER and PR status, and HER2-expression, the results were similar to the

univariate results, indicating significant benefit in survival for adding CT in the high risk

group (P = 0.02). Details of the multivariate analysis for BCSS are shown separately for the

70-gene high risk and low risk patient groups in Table 2.

Table 2. Multivariate analysis of treatment effects for several prognostic factors. BCSS for the 70-

gene high risk patients is shown above and for the 70-gene low risk patients below.

MammaPrint HR (95% CI) P -value

High risk

Age at diagnosis (by year) 0.96 (0.91–1.02) 0.17

Tumor size (by cm) 1.05 (1.01–1.09) 0.02

No. of positive nodes (0-3) 1.39 (0.95–2.03) 0.09

Grade 1.03 (0.48–2.19) 0.94

ER-positive status 0.48 (0.18–1.34) 0.16

PR-positive status 0.31 (0.09–1.03) 0.06

HER2-positive status 0.72 (0.25–2.10) 0.55

Adjuvant therapy: ET versus ET + CT 0.21 (0.06–0.80) 0.02

Low risk

Age at diagnosis (by year) 1.00 (0.88–1.15) 0.95

Tumor size (by cm) 0.98 (0.89–1.10) 0.77

No. of positive nodes (0-3) 1.09 (0.37–3.16) 0.88

Grade 0.57 (0.12–2.82) 0.49ER-positive status ∞ (0–∞) 0.99

PR-positive status 0.09 (0.01–0.90) 0.04

HER2-positive status ∞ (0–∞) 0.99

Adjuvant therapy: ET versus ET + CT ∞ (0–∞) 0.98

Abbreviations: BCSS, breast cancer-specific survival; HR, hazard ratio; 95% CI, 95% confidence interval;

cm, centimeter; no, number; ER, estrogen receptor; PR, progesterone receptor; ET, endocrine therapy; CT,

chemotherapy.

8/10/2019 Stella Mook T3-108



125

7

Discussion

This is the first study assessing the prediction of adjuvant CT benefit using the 70-gene

MammaPrint™ signature in a pooled analysis of lymph node negative and positive

patients. When grouped by chemo-ET or ET alone, patients in the 70-gene low risk group

derive no significant survival benefit from CT added to ET. Of note, very few events were

observed in this 70-gene low risk patient group, irrespective of type of adjuvant treatment,

confirming their overall good outcome. Indeed, for these patients, a low gene expression

result may indicate a sufficiently low risk of recurrence and cancer-related death at 5 years

to obviate any benefit of adjuvant CT. In contrast, a significant and clinically meaningful

benefit of combined chemo-ET was shown for the 70-gene high risk group. These observed

differences in benefit for the 70-gene low and high risk group were not significant for the

interaction test, comparing the differential in the extent of the benefit between 70-gene

low and high risk patients. Ioannidis et al. have previously indicated that an interaction is not

necessarily required for a predictive score to be useful in therapeutic decisions, especially

when absolute risk in the low risk group is so low that CT would not be recommended. 20

The results from a pooled analysis of individual patient data not only confirm the 70-gene

signature as a validated, independent prognostic tool, but also suggest the assay to be a

predictive tool for the expected benefit of adjuvant CT in patients with early breast cancer

and a high risk 70-gene profile.

One of the strengths of this study is its design using a pooled analysis of centrally reviewed

and updated individual patient data, representing a commonly accepted method of ameta-analysis.21 While this study was not done using retrospective analysis of phase III

clinical trial data, the patients studied represent an unselected early breast cancer cohort,

which can be seen in Table 1. Moreover, the included consecutive series were obtained

from prospectively collected frozen banked tumor material at several leading European

cancer centers. All patients with a cancer diagnosis were accessioned into the tumor banks

consecutively as they presented to the respective institutions. Clinical and pathological

data shown in Table 1 were centrally reviewed and blinded to the microarray analysis. One

of the clear limitations of this study next to limited patient numbers and differences in CT

regimens is its retrospective design. However, it will be several years before survival datafrom ongoing randomized controlled trials such as the MINDACT or TAILORx study22,23 will

be available.

The use of multigene assays such as the 70-gene profile and the 21-gene RS has increased

in recent years and these assays have impacted treatment decisions. In multiple validation

studies it has been demonstrated that the 70-gene signature adds independent prognostic

information to routine clinico-pathologic risk assessment.13,14,17 In a study of 427 breast

cancer patients from 16 community-based Dutch hospitals,16 discordances in risk

stratification between the 70-gene signature and treatment guidelines were noted in up to

41% of patients. This led to an adjustment of the adjuvant treatment regimen in two-third

of the study cohort.

8/10/2019 Stella Mook T3-108


Chapter 7

126

Two studies have evaluated the predictive utility of the 21-gene RS with respect to CT

benefit. The advantage of this assay is its validation on tumor tissue from phase III trials

with uniform CT regimens, although only a subset was available for analysis. Within the

NSABP B20 study, the degree of benefit from adjuvant CT ranged from little in the low and

intermediate RS to 20% absolute benefit in the highest RS group.12 Of note, the control arm

of this study had been used for development of the 21-gene RS which may have resulted

in overinterpretation of the data.7,20 Ioannidis mentioned in his commentary that the poor

performance of the RS in the CT arm of NSABP B20 caused the significant treatment-RS

interaction effect. The second predictive study was done on samples from the Southwest

Oncology Group study S8814 (INT0100) in node positive patients and presented at the

2007 San Antonio Breast Cancer Symposium.24 No benefit in disease-free survival for the

patients with a low RS for added CAF CT concurrent with tamoxifen was shown, whereas

the benefit was significant in the highest RS group.

In the neoadjuvant setting a number of studies using several drugs demonstrated the

predictive value of several gene signatures for CT response. These signatures comprise

known signatures such as the genomic grade index as well as several new classifiers.25-37

Additional data to support the predictive potential for the 70-gene assay comes from the

neoadjuvant study of Straver et al..38 In this study, only patients who had high risk profiles

were likely to achieve a pathologic complete response (pCR) to CMF or anthracycline-

containing CT regimens. In fact, no patient with a low risk profile achieved a pCR and only

two patients (9%) of this group achieved a partial response to therapy compared to 37%

overall response (P = 0.008) in the high risk group including a 20% pCR rate (P = 0.015). Theresults of all these studies support the theory that gene expression profiles can separate

CT-responsive from poorly or non-responsive tumors.

Clinical implications

In about two-third of all hormone receptor-positive cases, clinical and genomic risk

assessment using the 70-gene signature will be concordant. If both methods indicate a

high risk of recurrence, the use of combined chemo-ET seems clinically indicated. If both

methods indicate a low risk of recurrence, then ET alone should be adequate treatment.For the one-third of patients with discordant risk assessment, our findings suggest

consideration of the following approach. If the 70-gene profile indicates a low risk in a

clinically stratified high risk patient, ET alone may be indicated in highly endocrine-

responsive patients, as defined by the St. Gallen consensus panel, as these patients are at

very low risk to recur and will likely gain little or no benefit from additional CT. Conversely,

70-gene high risk and clinically assessed low risk patients will likely benefit from combined

chemo-endocrine treatment. If these patients are highly endocrine-responsive, then

endocrine treatment alone might be the prudent option, however, withholding adjuvant

therapy might not be a prudent option for this group of patients. Furthermore, other

8/10/2019 Stella Mook T3-108



127

7

factors such as age and co-morbidities may influence shared decision-making for adjuvant

systemic therapy. However, generally definitive recommendations cannot be drawn from

retrospective studies and only the ongoing, well designed prospective trials will provide

definitive answers to this important question.

Conclusions

In this study, a statistically significant and clinically meaningful benefit for the addition of

adjuvant CT to endocrine treatment in 70-gene high risk patients in the adjuvant setting

has been shown. There appears to be no evidence for a similar benefit for the 70-gene

low risk patients and these patients are at such a low risk of recurrence and cancer-related

death, that addition of CT may not be justified. ET alone seems to be the optimal treatment

for this group of patients. It seems reasonable to use multigene assays whenever indicated

in hormone receptor-positive patients for improved decision-making regarding the role of

adding adjuvant CT to hormonal treatment.

Acknowledgments

This work was supported by the Austrian Society of Surgery and Agendia BV. Both provided

unrestricted educational grants for the work of M. Knauer. We are indebted to Femke de

Snoo, MD PhD for critically reading the manuscript and providing helpful comments and

to Marleen Kok for providing part of the data used for our analyses.

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108


Chapter 7

130

36. Potti A, Dressman HK, Bild A, et al. Genomic signatures to guide the use of chemotherapeutics. Nat Med

2006; 12: 1294-1300.

37. Thuerigen O, Schneeweiss A, Toedt G, et al. Gene expression signature predicting pathologic complete

response with gemcitabine, epirubicin, and docetaxel in primary breast cancer. J Clin Oncol 2006; 24: 1839-

1845.

38. Straver ME, Glas AM, Hannemann J, et al. The 70-gene signature as a response predictor for neoadjuvant

chemotherapy in breast cancer. Breast Cancer Res Treat 2010; 119: 551-558.

8/10/2019 Stella Mook T3-108


Chapter 8

Calibration and discriminatory accuracy

of prognosis calculation for breast cancer

with the online Adjuvant! program: a

hospital-based retrospective cohort study

Stella Mook*

Marjanka K. Schmidt*

Emiel J.Th. Rutgers

Anthonie O. van de Velde

Otto Visser

Sterre M. Rutgers

Nicola Armstrong

Laura J. Van ’t VeerPeter M. Ravdin

* Contributed equally

Lancet Oncol 2009; 10: 1070-1076.

8/10/2019 Stella Mook T3-108


Chapter 8

134

Summary

Background

Adjuvant! is a web-based program that calculates individualized 10-year survival

probabilities and predicted benefit of adjuvant systemic therapy. The Adjuvant! model

has not been validated in any large European series. The aim of our study was to validate

Adjuvant! in Dutch patients, investigating both its calibration and discriminatory accuracy.

Methods

Patients who were at least partly treated at the Netherlands Cancer Institute for breast

cancer between 1987 and 1998 were included if they met the following criteria: tumor size

T1 (≤2 cm), T2 (2–5 cm), or T3 (>5 cm), invasive breast carcinoma, with information about

involvement of axillary lymph nodes available, no distant metastases, primary surgery,

axillary staging, and radiotherapy according to national guidelines. Clinicopathological

characteristics and adjuvant treatment data were retrieved from hospital records and

medical registries and were entered into the Adjuvant! (version 8.0) batch processor with

blinding to outcome. Endpoints were overall survival and the proportion of patients that

did not die from breast cancer (breast cancer-specific survival [BCSS]).

Findings

5380 patients were included with median follow-up of 11.7 years (range 0.03-21.8). the

10-year observed overall survival (69.0%) and BCSS (78.6%) and Adjuvant! predicted

overall survival (69.1%) and BCSS (77.8%) were not statistically different ( p = 0.87 and p =

0.18, respectively). Moreover, differences between predicted and observed outcomes were

within 2% for most relevant clinicopathological subgroups. In patients younger than 40

years, Adjuvant! overestimated overall survival by 4.2% ( p = 0.04) and BCSS by 4.7% ( p =

0.01). The concordance index, which indicates discriminatory accuracy at the individual

level, was 0.71 for BCSS in the entire cohort.

Interpretation

Adjuvant! accurately predicted 10-year outcomes in this large-scale Dutch validation study

and is of use for adjuvant treatment decision making, although the results may be less

reliable in some subgroups.

8/10/2019 Stella Mook T3-108


Dutch validation of Adjuvant!

135

8

Introduction

Treatment recommendations for early-stage breast cancer are based on prognosis (i.e., the

estimated risk of relapse and death after primary surgery and radiotherapy) and expected

benefit of adjuvant therapy. Treatment guidelines qualitatively incorporate prognosis

and treatment efficacy without quantitative estimates.1-3 Nevertheless, methods that give

quantitative estimates of prognosis exist, such as the Nottingham Prognostic Index and

Adjuvant!.4-6 These quantitative methods include several assumptions, the most crucial

being that the populations for which the models were developed are representative of

others.

Adjuvant! is a computer program that is freely accessible on the internet (www.

adjuvantonline.com). The program provides estimated 10-year survival probabilities and

risk of relapse on the basis of a model incorporating patient’s age, co-morbidity, tumor size,

tumor grade, estrogen-receptor status, and number of involved lymphnodes.6 the program

calculates the expected efficacy of adjuvant therapy (chemotherapy, hormonal therapy, or

both) for different classes of regimens.7-9 The program gives the estimated prognosis and

expected treatment benefit in a comprehensive format and can help to inform patients

and to involve them in decision making about therapeutic options.10-12

Adjuvant! was largely developed with information from the Surveillance, Epidemiology

and End Results (SEER) registry. The SEER registry has data for about 10% of patients with

breast cancer in the USA.6 Olivotto and colleagues13 validated Adjuvant! (version 5.0) in a

population-based series of 4083 early-stage patients with breast cancer registered in theBritish Columbia Breast Cancer Outcomes Unit database. They showed that the Adjuvant!

model was well calibrated - i.e., it accurately predicted the number of breast cancer-related

deaths observed in the whole study cohort and subsets of their population (predicted

and observed outcomes were within 2%). However, because European populations might

differ from those in the USA and Canada, whether outcome predictions of the Adjuvant!

model are applicable in Europe is unknown: differences in incidence of obesity, duration

and type of adjuvant and salvage treatment, ethnic background, and intrinsic tumor

characteristics might affect prognosis.14-19 Furthermore, although Olivotto and colleagues13

showed the goodness of fit of the Adjuvant! model, no information about its discriminatoryaccuracy was given. Because Adjuvant! is used in Europe and the USA to support treatment

decisions in clinical practice and randomized trials,20-22 we aimed to test the validity of

Adjuvant! in a large cohort of Dutch patients with breast cancer, determining its ability to

predict outcomes in groups of patients (calibration) and to distinguish individuals who will

experience different outcomes (discriminatory accuracy).23,24

8/10/2019 Stella Mook T3-108


Chapter 8

136

Methods

Patients

All women who were at least partly treated for breast cancer at the Netherlands Cancer

Institute at the Antoni van Leeuwenhoek Hospital (NKI-AVL) from 1987 to 1998 were

identified in the hospital’s Medical Registry. Patients were included if they had tumor

size T1 (≤2 cm), T2 (2–5 cm), or T3 (>5 cm), unilateral tumors, invasive breast carcinoma,

information about involvement of axillary lymph nodes available, no distant metastases,

primary surgery, axillary staging, and radiotherapy according to national guidelines.

Patients with previous malignant disease and those who received neoadjuvant therapy

were excluded, as were those with unknown tumor size, unknown nodal status, unknown

adjuvant systemic therapy, no definitive axillary surgery (axillary-lymph-node dissection

with fewer than six nodes examined; Figure 1). Information about adjuvant systemic

treatment was derived from the medical registry. Adjuvant treatment was given according

to national guidelines, taking into account patients’ wishes and preferences.17

The study is reported according to the STROBE statement.25 No ethical review was required

according to Dutch legislation.

Procedures

Histology, tumor size, tumor grade, and number of positive lymph nodes were retrievedfrom three sources and entered in the database according to the following hierarchy of

preference for data source: first, personal logbook from pathologists at NKI-AVL containing

pathology revisions of breast cancer diagnosed between 1994–96; second, the PALGA

system (Dutch network and National Database for Pathology); third, medical registry of

the NKI-AVL. Information about estrogen-receptor status was retrieved from three sources

and entered into a database according to the following hierarchy: first, estrogen-receptor

ligand-binding assays (breast cancers diagnosed 1987–95); second, Pathologist logbook

(breast cancers diagnosed 1995–96), and third the PALGA system (Dutch network and

National Database for Pathology; Supplements).Outcome data (date of first local, regional and distant recurrence, second malignancies,

contralateral breast cancer, and date of last follow-up or death) were obtained from the

medical registry. These data were completed by linking patient records to the Dutch

municipal registry, which contains the date of death or emigration if applicable, for all Dutch

citizens. For patients not in this national registry as having died or emigrated, the date of

last follow-up was recorded as Feb 1, 2007 (i.e., 2 months before the date of linkage). Cause

of death was retrieved from the medical registry if available and from individual patients’

files when no cause of death was entered in the registry (n=1090 breast cancer-specific

death; n=188 other causes). If neither the medical registry nor patients’ files contained

8/10/2019 Stella Mook T3-108



137

8

cause of death, we used the presence of distant metastases as a surrogate: patients

without known cause of death (n=875) were assigned to breast cancer-specific mortality

if they were diagnosed with distant metastases during follow-up (n=191 breast cancer-

specific death; n=684 other causes). Patients with registered breast cancer-specific death

and patients who were assigned to breast cancer-specific death based on the presence of

distant metastases were pooled for analyses.

10-year predicted overall survival and breast cancer-specific survival (BCSS) were calculated

for each patient individually. Data on age, co-morbidity, tumor size, tumor grade, number

of positive axillary lymph-nodes, estrogen-receptor status, and adjuvant systemic

treatment were entered in the Adjuvant! (version 8.0) batch processor, with blinding to

patient outcomes. The model’s estimation of prognosis is based on 10-year observed

overall survival of women diagnosed with breast cancer between 1988 and 1992 in the

USA and recorded in the SEER database.6 The estimations of treatment efficacy are mainly

based on the proportional risk reductions derived from the Early Breast Cancer Trialists’

Collaborative Group 1998 meta-analyses and recently updated with the meta-analyses

data from 2005.7-9 Because we could not retrieve reliable data for co-morbidity, we used

the default assumption of minor health problems. For patients with no data on estrogen-

receptor status, the status was entered in the model as unknown.


Overall survival and BCSS were derived from Kaplan-Meier survival analyses of the entiregroup and various subsets.26 For the same datasets, the average predicted overall survival

and BCSS were calculated from individual predicted outcomes by Adjuvant!. To assess the

calibration of the model (goodness of fit), observed and average predicted outcomes were

compared by use of a one-sample t-test for proportions, assuming the Adjuvant! predicted

value to be the population value (under the assumption that the model is true) and thus

fixed. In addition, we plotted averages of observed versus predicted outcomes, grouped by

deciles of predicted outcomes.23 The slope of the fitted line was compared with the slope

of the line indicating a perfect relationship (y=x).

To assess discriminatory accuracy of Adjuvant! (its ability to discern patients havinggood outcomes from those having poor outcomes), we calculated an index of predictive

discrimination, the concordance index (c-index).23 The c-index was corrected for overfitting

by bootstrapping with 200 resamples each. A c-index of 1 means that the model perfectly

ranks patients according to survival (i.e., patients having a better outcome also having a

better predicted outcome), 0.5 means the model does no better than chance. the predictive

accuracy and proportion of explained variation, as defined by Schemper and colleagues,24

was also calculated. SEs were estimated by bootstrapping with 200 resamples each. Known

prognostic factors (i.e., age, tumor size, tumor grade, number of positive lymph-nodes,

histology, estrogen-receptor status, and adjuvant systemic therapy) were used in the Cox

8/10/2019 Stella Mook T3-108


Chapter 8

138

multivariate model. Furthermore, on the basis of backward multivariate Cox regression

analyses, year of diagnosis was added to construct the best predictive model for BCSS and

overall survival in our dataset.27 Analyses were done with SPSS version 15.0 and R statistical

software (www.r-project.org).

Role of the funding source

The funding sources had no role in study design; collection, analysis, or interpretation of

data; writing of the paper; or in decisions relating to publication. SM, MKS, and LVV had full

access to all data. SM, MKS, LVV, and PMR took final responsibility for the decision to submit

the paper for publication.

Results

Our database included 5380 patients, 2604 of whom (48%) received no adjuvant systemic

therapy. The algorithm in Adjuvant! attributes different efficacy estimates depending

on type of chemotherapy and hormonal treatment. Among 1961 patients treated with

endocrine therapy, 1908 (97%) received tamoxifen (2–5 years); therapy was not specified

for 13 (0.7%). 892 (82%) of 1084 patients treated with adjuvant chemotherapy received

cyclophosphamide, methotrexate, and fluorouracil, 122 (11%) received fluorouracil,

epirubicin, and cyclophosphamide, 42 (4%) received high-dose chemotherapy, and 11(1%) received cyclophosphamide and doxorubicin. For the remaining 16 patients (2%),

type of chemotherapy was unspecified. 2276 (42%) of 5380 patients had complete data for

all factors used in the Adjuvant! model to predict outcome. Grade was unknown for 1379

patients (26%), and estrogen-receptor status unknown for 2253 (42%).

During a median follow-up of 11.7 years (range 0.03–21.8), 2153 (40%) of 5380 patients

died; 3032 (94%) of 3227 patients alive at last follow-up had 10 years or more follow-up

(Figure 1). Table 1 shows the distribution of demographic, pathological, and primary treatment

data for our study cohort. For all patients, the 10-year observed overall survival (69 .0%)

and BCSS (78.6%) rates as compared with the 10-year overall survival (69.1%) and BCSS(77.8%) rates predicted by Adjuvant! were within 1% and not significantly different ( p >0.05;

Table 1). In general, Adjuvant! predicted overall survival accurately in the various subsets

of patients (i.e., differences between predicted and observed outcomes were within 2%),

whereas Adjuvant! underestimated BCSS in some subsets (Table 1). Subsets of patients for

whom there was a discrepancy between predicted outcomes by Adjuvant! and actual

observed outcomes included patients under 40 years, for whom both predicted overall

survival and BCSS were overly optimistic (4.2% and 4.7%, respectively; p = 0.04 and p = 0.01).

For patients older than 69 years the program also overestimated overall survival by 3 .4%

( p = 0.05), but BCSS was accurately predicted in this group (predicted–observed –1.7%).

8/10/2019 Stella Mook T3-108



139

8

In subgroups of nodal status, overall survival was accurately predicted by Adjuvant!;

however, the program underestimated BCSS by 3.1% ( p = 0.002) in patients with one to

three positive lymph-nodes.

Although Adjuvant! predicted overall survival accurately for subsets of tumor size, a

discrepancy between predicted and observed BCSS was noted (–5.8% to 2.4%; Table 1). In

particular, predicted BCSS was optimistic for patients with tumors with diameter 11–20

mm, although it was pessimistic in patients with tumors 21–50 mm in diameter. In patients

with estrogen-receptor-negative tumors Adjuvant! underestimated BCSS by 4.1% ( p = 0.02).

This underestimation of outcome by Adjuvant!, although non-significant, was also seen for

overall survival (–3.2%, p = 0.07).

Figure 1. Study profile

5761 patients in tumor registry

5380 patients included in analysis

381 excluded:

− 168 no information on tumor size

− 34 no information on lymph-node status

− 116 N0, fewer than six lymph nodes examined

− 62 N+, fewer than six lymph nodes examined

− 1 no information about adjuvant systemic therapy

3227 alive 2153 dead

1281 breast cancer-specific death 872 other cause death

8/10/2019 Stella Mook T3-108


Chapter 8

140

T a b

l e 1 . B a s e l i n e c h a r a c t e r i s t i c s a n d A d j u v a n t ! p r e d i c t e d v e r s u s o b s e r v e d o v e r a l l s u r v i v a l a n d b r e a s t c

a n c e r - s p e c i fi c s u r v i v a l .

P a t i e n t s

O v e r a l l s u r v i v a l

B r e a s t c a n c e r -

s p e c i fi c s u r v i v a l

N o .

%

A d j u v a n t !

P r e d i c t e d

O b s e r v e d

( S E )

P r

e d i c t e d - O b s e r v e d

( 9 5 % C

I )

P

v a l u e

A d j u v a n t !

P r e

d i c t e d

O b s e r v e d

( S E )

P r e d i c t e d - O b s e r v e d

9 5 % C

I

P

v a l u e

A l l p a t i e n t s

5 3 8 0

1 0 0

6 9 . 1

6 9 ( 0 . 6 )

0 . 1 ( - 1 . 0 8 t o 1 . 2 8 )

0 . 8 7

7 7 . 8

7 8 . 6 ( 0 . 6 )

- 0 . 8 ( - 1 . 9 8 t o 0 . 3 8 )

0 . 1 8

Y e a r o f d i a g n o s i s

1 9 8 7 - 1 9 8 9

1 1 6 9

2 1 . 7

6 7 . 6

6 7 . 7 ( 1 . 4 )

- 0 . 1 ( - 2 . 8 5 t o 2 . 6 5 )

0 . 9 4

7 6 . 4

7 7 . 0 ( 1 . 3 )

- 0 . 6 ( - 3 . 1 5 t

o 1 . 9 5 )

0 . 6 4

1 9 9 0 - 1 9 9 3

2 0 1 7

3 7 . 5

6 9 . 2

6 7 . 8 ( 1 . 0 )

1 . 4 ( - 0 . 5 6 t o 3 . 3 6 )

0 . 1 6

7 8 . 1

7 7 . 7 ( 1 . 0 )

0 . 4 ( - 1 . 5 6 t

o 2 . 3 6 )

0 . 6 9

1 9 9 4 - 1 9 9 7

2 1 9 4

4 0 . 8

6 9 . 8

7 0 . 8 ( 1 . 0 )

- 1 . 0 ( - 2 . 9 6 t o 0 . 9 6 )

0 . 3 2

7 8 . 4

8 0 . 3 ( 0 . 9 )

- 1 . 9 ( - 3 . 6 7 t o - 0 . 1 4 )

0 . 0 4

T y p e o f s u r g e r y

B r e a s t - c o n s e r v i n g s u r g e r y

2 8 8 5

5 3 . 6

7 6 . 5

7 6 ( 0 . 8 )

0 . 5 ( - 1 . 0 7 t o 2 . 0 7 )

0 . 5 3

8 3 . 8

8 2 . 8 ( 0 . 7 )

1 . 0 ( - 0 . 3 7 t

o 2 . 7 3 )

0 . 1 5

M a s t e c t o m y

1 9 7 8

3 6 . 8

5 9 . 1

6 0 ( 1 . 1 )

- 0 . 9 ( - 3 . 0 6 t o 1 . 2 6 )

0 . 4 1

6 9 . 7

7 2 . 6 ( 1 . 0 )

- 2 . 9 ( - 4 . 8 6 t o - 0 . 9 4 )

0 . 0 0 4

U n k n o w n

5 1 7

9 . 6

6 6 . 1

6 4 . 3 ( 2 . 1 )

1 . 8 ( - 2 . 3 3 t o 5 . 7 3 )

0 . 3 9

7 5 . 7

7 7 . 7 ( 1 . 9 )

- 2 . 0 ( - 5 . 7 3 t

o 1 . 7 3 )

0 . 2 9

A g

e ( y e a r s )

< 4 0

5 7 2

1 0 . 6

7 2 . 4

6 8 . 2 ( 2 . 0 )

4 . 2 ( 0 . 2 7 t o 8 . 1 3 )

0 . 0 4

7 3 . 5

6 8 . 8 ( 1 . 9 )

4 . 7 ( 0 . 9 7 t

o 8 . 4 3 )

0 . 0 1

4 0 - 4 9

1 4 4 8

2 6 . 9

7 6 . 6

7 8 . 6 ( 1 . 1 )

- 2 . 0 ( - 4 . 1 6 t o 0 . 1 6 )

0 . 0 7

7 8 . 8

8 1 . 2 ( 1 . 0 )

- 2 . 4 ( - 4 . 3 6 t o - 0 . 4 4 )

0 . 0 2

5 0 - 5 9

1 3 6 9

2 5 . 4

7 2 . 0

7 3 . 6 ( 1 . 2 )

- 1 . 6 ( - 3 . 9 5 t o 0 . 7 5 )

0 . 1 8

7 6 . 8

7 7 . 7 ( 1 . 1 )

- 0 . 9 ( - 3 . 0 6 t

o 1 . 2 6 )

0 . 4 1

6 0 - 6 9

1 1 7 4

2 1 . 8

6 8 . 9

6 8 . 3 ( 1 . 4 )

0 . 6 ( - 2 . 1 5 t o 3 . 3 5 )

0 . 6 7

8 0 . 1

8 0 . 9 ( 1 . 2 )

- 0 . 8 ( - 3 . 1 5 t

o 1 . 5 5 )

0 . 5 1

≥ 7 0

8 1 7

1 5 . 2

4 9 . 1

4 5 . 7 ( 1 . 7 )

3 . 4 ( 0 . 0 6 t o 6 . 7 4 )

0 . 0 5

7 7 . 8

7 9 . 5 ( 1 . 6 )

- 1 . 7 ( - 4 . 8 4 t

o 1 . 4 4 )

0 . 2 9

H i s t o l o g y

M a i n l y D C I S

5 5

1 . 0

8 3 . 8

8 7 . 2 ( 4 . 5 )

- 3 . 4 ( - 1 2 . 4 2 t o 5 . 6 2 )

0 . 4 5

9 0 . 2

9 6 . 2 ( 2 . 6 )

- 6 . 0 ( - 1 1 . 2 1 t o - 0 . 7 9 )

0 . 0 3

I D C

4 0 0 1

7 4 . 4

6 8 . 7

6 8 . 3 ( 0 . 7 )

0 . 4 ( - 0 . 9 7 t o 1 . 7 7 )

0 . 5 7

7 7 . 2

7 7 . 6 ( 0 . 7 )

- 0 . 4 ( - 1 . 7 7 t

o 0 . 9 7 )

0 . 5 7

I L C

6 1 4

1 1 . 4

6 5 . 3

6 6 . 2 ( 1 . 9 )

- 0 . 9 ( - 4 . 6 3 t o 2 . 8 3 )

0 . 6 4

7 5 . 6

7 7 . 8 ( 1 . 7 )

- 2 . 2 ( - 5 . 5 4 t

o 1 . 1 4 )

0 . 2 0

I D / L C

3 1 8

5 . 9

7 0 . 4

7 0 . 1 ( 2 . 6 )

0 . 3 ( - 4 . 8 2 t o 5 . 4 2 )

0 . 9 1

7 8 . 5

7 8 . 5 ( 2 . 4 )

0 . 0 ( - 4 . 7 2 t

o 4 . 7 2 )

1 . 0 0

T u b u l a r

1 1 4

2 . 1

8 8 . 8

9 5 . 6 ( 1 . 9 )

- 6 . 8 ( - 1 0 . 5 6 t o - 3 . 0 4 )

0 . 0 0 0 5

9 5 . 2

1 0 0 ( 0 . 0 )

- 4 . 8 ( - 4 . 8 0 t o - 4 . 7 9 ) < 0 . 0 0 0 1

M u c i n o u s

7 6

1 . 4

7 4 . 0

6 5 . 6 ( 5 . 5 )

8 . 4 ( - 2 . 5 6 t o 1 9 . 3 6 )

0 . 1 3

8 7 . 6

8 8 . 1 ( 4 . 0 )

- 0 . 5 ( - 8 . 4 7 t

o 7 . 4 7 )

0 . 9 0

M e d u l a r

6 8

1 . 3

7 4 . 5

8 0 . 9 ( 4 . 8 )

- 6 . 4 ( - 1 5 . 9 8 t o 3 . 1 8 )

0 . 1 9

7 9 . 4

8 6 . 4 ( 4 . 2 )

- 7 . 0 ( - 1 5 . 3 8 t

o 1 . 3 8 )

0 . 1 0

O t h e r s

1 3 4

2 . 5

6 7 . 5

6 4 . 1 ( 4 . 2 )

3 . 4 ( - 4 . 9 1 t o 1 1 . 7 1 )

0 . 4 2

7 8 . 8

7 6 . 7 ( 3 . 8 )

2 . 1 ( - 5 . 4 2 t

o 9 . 6 2 )

0 . 5 8

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108


Chapter 8

142

P a t i e n t s


B r e a s t c a n c e r -

s p e c i fi c s u r v i v a l

N o .

%

A d j u v a n t !

P r e d i c t e d

O b s e r v e d

( S E )

P r

e d i c t e d - O b s e r v e d

( 9 5 % C

I )

P

v a l u e

A d j u v a n t !

P r e

d i c t e d

O b s e r v e d

( S E )

P r e d i c t e d - O b s e r v e d

9 5 % C

I

P

v a l u e

S y s t e m i c t r e a t m e n t

N o n e

2 6 0 4

4 8 . 4

7 6 . 8

7 5 . 8 ( 0 . 8 )

1 . 0 ( - 0 . 5 7 t o 2 . 5 7 )

0 . 2 1

8 5 . 3

8 4 . 6 ( 0 . 7 )

0 . 7 ( - 0 . 6 7 t o 2 . 0 7 )

0 . 3 2

C h e m o t h e r a p y o n l y

8 1 5

1 5 . 1

6 7 . 4

6 9 . 3 ( 1 . 6 )

- 1 . 9 ( - 5 . 0 4 t o 1 . 2 4 )

0 . 2 4

6 9 . 4

7 1 . 7 ( 1 . 6 )

- 2 . 3 ( - 5 . 4 4 t

o 0 . 8 4 )

0 . 1 5

H o r m o n a l t h e r a p y o n l y

1 6 9 2

3 1 . 4

5 8 . 8

5 8 . 4 ( 1 . 2 )

0 . 4 ( - 1 . 9 5 t o 2 . 7 5 )

0 . 7 4

7 2 . 1

7 3 . 7 ( 1 . 1 )

- 1 . 6 ( - 3 . 7 6 t

o 0 . 5 6 )

0 . 1 5

C h e m o a n d h o r m o n a l

t h e r a p y

2 6 9

5 . 0

6 4 . 5

6 9 . 1 ( 2 . 8 )

- 4 . 6 ( - 1 0 . 1 1 t o 0 . 9 1 )

0 . 1 0

6 7 . 0

7 1 . 4 ( 2 . 8 )

- 4 . 4 ( - 9 . 9 1 t

o 1 . 1 1 )

0 . 1 2

C o

m p l e t e n e s s o f d a t a

M i s s i n g e s t r o g e n - r e c e p t o r

s t a t u s a n d / o r g r a d e

3 1 0 4

5 7 . 7

6 9 . 0

6 8 . 7 ( 0 . 8 )

0 . 3 ( - 1 . 2 7 t o 1 . 8 7 )

0 . 7 1

7 7 . 8

7 8 . 9 ( 0 . 8 )

- 1 . 1 ( - 2 . 6 7 t

o 0 . 4 7 )

0 . 1 7

C o m p l e t e

2 2 7 6

4 2 . 3

6 9 . 2

6 9 . 4 ( 1 . 0 )

- 0 . 2 ( - 2 . 1 6 t o 1 . 7 6 )

0 . 8 4

7 7 . 8

7 8 . 2 ( 0 . 9 )

- 0 . 4 ( - 2 . 1 7 t

o 1 . 3 7 )

0 . 6 6

P - v a

l u e s c a l c u l a t e d w i t h o n e s a m p l e t - t e

s t .

T a b

l e 1 . C o n t i n u e d .

8/10/2019 Stella Mook T3-108



143

8

Figure 2 shows the observed outcome versus the average predicted outcome for the cohort,

grouped by deciles of predicted overall survival or BCSS probabilities. The slope of the

line representing a perfect fit of predicted with observed outcomes (y=x) and the slope

of the actual line fitted to our data for overall survival was not significantly different. This

indicates that the calibration of Adjuvant! is similarly good in patients with poor overall

survival and patients with excellent overall survival. However, for BCSS, the model tended

to underestimate and to overestimate BCSS in the extremes of the distribution of poor and

good survival, respectively (slope was significantly different p < 0.0001).

To assess discriminatory accuracy of the model (i.e., its ability to separate patients who will

die from breast cancer from those who will not), we calculated Harrell’s c-index (0 .71 for

BCSS), as well as the predictive accuracy and explained variation (0 .73 and 13% for BCSS,

respectively). Hence, the predictive accuracy for BCSS increased from 0.69 for a model

without predictors to 0.73 for the Adjuvant! model (Table 2). In various clinical subgroups

(Table 1), the c-index varied from 0.65 to 0.75 (data not shown). The c-index for a multivariate

Cox regression model best fitted to the outcome of the 5380 patients with a backward

approach (model included age, tumor size, tumor grade, number of positive lymph

nodes, estrogen-receptor status, histology, type of adjuvant systemic therapy and year of

diagnosis) was similar to the Adjuvant! model (i.e., 0.72 and 0.71, respectively for BCSS).

Table 2. Discriminatory accuracy of Adjuvant! and a multivariate Cox model fitted to the

outcome.

Overall survival Breast cancer-specific survival

Adjuvant! Cox model* Adjuvant! Cox model*

C index 0.70 0.69 0.71 0.72

Predictive accuracy model without

predictors [1-D0]0.64 0.64 0.69 0.69

Predictive accuracy [1-Dx] (SE) 0.69 (0.008) 0.69 (0.007) 0.73 (0.02) 0.74 (0.02)

Explained variation [(Dx-D0)/D0] (SE) 15% (1%) 15% (1%) 13% (1%) 16% (2%)

For overall survival the model included age, tumor size, tumor grade, number of positive lymph-nodes,histology, adjuvant systemic therapy and year of diagnosis. For breast cancer-specific survival the model

included age, tumor size, tumor grade, number of positive lymph-nodes, histology, estrogen-receptor

status, adjuvant systemic therapy, and year of diagnosis.

C-index = Harrell’s concordance index. Dx = predictive accuracy of model with predictors. D0 = predictive

accuracy of model without predictors.

*Best-fitted multivariate Cox model.

8/10/2019 Stella Mook T3-108


Chapter 8

144

Figure 2. Mean predicted versus observed outcomes by deciles of predicted outcome.

Error bars are SE.

Discussion

Overall projections of overall survival and BCSS with Adjuvant! were within 1% of observed

results and estimates within most subgroups seemed reasonably accurate (within 2% or

not significantly different from observed estimates). The conservative Dutch guidelines for

adjuvant systemic therapy used in the era of the study cohort results in a large proportion

of patients (2604 [48%] of 5380) who received no adjuvant systemic therapy. For this group

of patients, we could assess the prognostic value of Adjuvant! (i.e., the prediction of diseaseoutcome in the absence of adjuvant systemic therapy). Although most patients with early

breast cancer now receive some form of adjuvant systemic treatment, the confirmation

of the prognostic value is important for the decision whether or not to treat. Moreover,

when the program predicts prognosis accurately, the potential benefit of different types of

adjuvant treatment is also predicted more accurately, because the latter depends on the a

priori risk of recurrence. Adjuvant! is commonly used to decide whether patients who will

be treated with endocrine therapy are candidates for additional chemotherapy. Because

a small proportion of patients received chemotherapy with or without endocrine therapy

(815 [15%] and 269 [5%] of 5380, respectively) the prediction of chemotherapy benefit in

0

10

20

30

40

50

60

70

80

90

100

0 10 20 30 40 50 60 70 80 90 100

O b s e r v e d o u t c o m e

Predicted outcome

Overall survival (OS)

Breast cancer-specific survival (BCSS)

R2=0.996

Observed OS= 0.96 + 0.98*Predicted OS

R2=0.992

Observed BCSS= 12.23 + 0.85*Predicted BCSS

8/10/2019 Stella Mook T3-108



145

8

addition to endocrine therapy is less robust. However, the decision to add chemotherapy

to the endocrine treatment regimen depends also on the predicted outcome of a patient

when treated with hormonal therapy only. Adding chemotherapy to endocrine therapy will

be more beneficial for patients who will have an a priori poorer predicted disease outcome.

Our results show that Adjuvant! predicted both overall survival and BCSS accurately in

patients treated with endocrine therapy (1692 [31%] of 5380).

One of the limitations of our study is that in the era of this study standard adjuvant

chemotherapy consisted of cyclophosphamide, methotrexate, and fluorouracil for six

cycles and standard adjuvant endocrine therapy consisted of tamoxifen for which we

evaluated Adjuvant!. Therefore, future studies are required to validate Adjuvant! predictions

of currently used therapies, such as taxane-based chemotherapy and aromatase inhibitors.

The largest discrepancy between subgroups between our study and the Canadian validation

study13 was in patients younger than 35 years (10% for BCSS). As a consequence of this

disagreement and after further review of the SEER registry data, Adjuvant! was modified to

give more pessimistic estimates for estrogen-receptor-positive patients under 35 years of

age. Even after this adjustment (the major difference between Adjuvant! 5.0and Adjuvant!

8.0), the predicted outcomes still seem too optimistic, albeit less so than in the original

validation study (by 5% for BCSS). When results were stratified for estrogen-receptor status

in patients younger than 35 years and age 35–40 years, the overestimation was exclusively

seen in patients with estrogen-receptor-positive tumors in both age groups (Supplements).

This suggests that the correction factor of 1.5 for patients under 35 years is insufficient and

that an additional correction for patients between 35–40 years with estrogen-receptor-positive tumors might be justified.

Both studies showed that the outcomes of ductal and lobular cancers were accurately

predicted, but for other histological subtypes, the predicted outcomes by Adjuvant! are too

pessimistic. At present, histology is not incorporated in Adjuvant!; however, the program

warns the user that some histological subtypes might warrant an adjustment (e.g., medullary

cancers where high grade does not confer high risk).28 Other discrepancies between

observed outcomes and outcomes predicted by Adjuvant! seem modest or inconsistent

between our study and the original validation study. For example, underestimation of

BCSS was seen in one subgroup of tumor size. Olivotto and colleagues13

noted no suchunderestimation, although the distributions of tumor size were similar. This suggests

that the discrepancy in predicted and observed BCSS is not caused by a suboptimum

incorporation of size in the Adjuvant! model.

A second subgroup in which the underestimation of BCSS was significant is patients with

one to three positive lymph nodes. Although we do not have information about the extent of

lymph-node involvement (i.e., isolated tumor cells, micrometastases, or macrometastases),

until the late 1990s lymph nodes that contained only isolated tumor cells were assessed

as positive lymph-nodes in the Netherlands. Consequently, the group of patients with one

to three positive lymph nodes in our database probably includes some patients with only

8/10/2019 Stella Mook T3-108


Chapter 8

146

isolated cells who have a better disease outcome than patients with macrometastases in

one to three lymph-nodes (Supplements).29,30 As for younger patients, the predicted overall

survival by Adjuvant! was too optimistic for patients older than 70 years, which is possibly

caused by the lack of data on co-morbidity status. Co-morbidity was entered as the default

assumption of minor health problems, which is likely to be an underestimation in older

patients and therefore to result in overestimation of overall survival in these patients. The

accurate prediction of BCSS in patients older than 70 years supports this hypothesis.

The proportion of missing data is one of the limitations of our retrospective cohort for

the validation of Adjuvant!; we lacked data on estrogen-receptor status, tumor grade, or

both for 3104 (58%) of 5380 patients. Missing information about tumor grade or estrogen-

receptor status will now be less common. However, patients with incomplete data

had similar disease outcome as patients with complete data, indicating that including

patients with missing data did not induce a selection bias (Table 1). Information on HER2

status was not available in this cohort and will be incorporated in an upcoming version of

Adjuvant!. The program predicted disease outcome accurately for patients with unknown

estrogen-receptor status. By contrast, the model underestimated BCSS in patients with

estrogen-receptor-negative tumors. Detailed analysis of this subgroup revealed that the

underestimation was exclusively seen in patients treated with hormonal therapy (n=250;

Supplements). This particular group had better outcome than predicted, suggesting that these

tumors could have been erroneously scored or coded in the registry as estrogen-receptor-

negative.

Patients who were partly treated at NKI-AVL were mainly referred from regional hospitals toour institute for radiotherapy. All diagnostic information was made available and reviewed

by the NKI-AVL. Adjuvant! predicted overall survival and BCSS accurately in these patients

and those treated at NKI-AVL ( Supplements). As a consequence, the population includes a

much wider representation of patients, and selection bias of our cohort is likely to be less

pronounced than in a single-institute cohort.

This large-scale validation study of Adjuvant! in a hospital-based population of Dutch

patients with breast cancer showed that the calculated predictions by Adjuvant! agreed

with the observed outcomes and that the predictions are applicable to a Dutch population,

and presumably to a European population, corroborating that populations of patientswith European ancestry in different continents have similar disease. Potential differences

between US and European patients with breast cancer could have resulted in deviations

of outcome in both directions and therefore would level out in an overall comparison of

predicted and observed outcomes. The good performance of the model in American and

European settings implies that the prognostic features and disease course are broadly

similar in both settings.

The model’s success in these settings does not ensure success in other uses, because,

for example, time, changes in exogenous exposures (hormone replacement therapy),

diagnostic techniques (types and intensity of screening), and surgical staging could affect

8/10/2019 Stella Mook T3-108



147

8

the prognosis of patients with otherwise similar tumors. Furthermore, although we used a

large validation set, somewhat larger than the validation in British Columbia Breast Cancer

patients published by Olivotto and colleagues (n=4083), some subgroups were small and

findings in those groups should be considered with at least some caution.

To validate the Adjuvant! model in our study population we investigated both the

calibration (goodness of fit) and the discriminatory accuracy of the model. Although the

latter is rarely tested, it is of paramount importance to justify the use of prognostic models

for clinical outcome prediction.23,24 Results of the discriminatory accuracy of the Adjuvant!

model showed that in addition to good calibration, the model was capable of separating

individuals with a poor outcome from those with a good outcome with moderate power

(c-index 0.71). Remarkably, the discriminatory accuracy of a multivariate Cox model fitted

to our dataset was similar to that of Adjuvant!, indicating that the prognostic information

of the variables used in Adjuvant! was incorporated in the model in the best way possible.

Furthermore, the maximum explained variation by clinicopathological variables is about

15%, irrespective of whether they are incorporated in Adjuvant! or a model fitted to

our dataset. that the unexplained variation remains relatively large is supported by the

observation that patients with identical clinicopathological variables can have strikingly

different outcomes and proves that the information captured by these criteria can only

explain part of the differences in outcome. incorporation of biological markers, such as

molecular profiles and germline variants, in the model will likely increase the explained

variation and therefore result in a more rigorous prediction of outcome at the individual

patient level in the near future.

Contributors

SM, MKS, ER, LVV, and PMR designed the study. SM, MKS, AOV, OV, SMR, and PMR collected

data. SM, MKS, and NA analyzed data. SM, MKS, ER, NA, LVV, and PMR interpreted data. SM,

MKS, ER, LVV, and PMR wrote the paper.

Funding

Dutch National Genomics Initiative-Cancer Genomics Center, Dutch Cancer Society-KWF

grant NKI 2007-3839.

Conflicts of interest

PMR owns and is paid in part by Adjuvant Inc, which owns the rights to Adjuvant! online.

The other authors have no conflicts of interest to declare.

8/10/2019 Stella Mook T3-108


Chapter 8

148

Acknowledgments

We thank Hans Peterse (deceased), Otilia Dalesio, and the Medical Registry staff for

providing baseline data, Hans Bonfrer and Tiny Korse for providing the estrogen-receptor

ligand binding assay data, Matti Rookus, Flora van Leeuwen, and Marieke Vollebergh for

helpful discussions. This study was financially supported by the Dutch National Genomics

Initiative-Cancer Genomics Center (SM and SMR) and the Dutch Cancer Society-KWF grant

NKI 2007-3839 (MKS).

8/10/2019 Stella Mook T3-108



149

8

References

1. NCCN Clinical Practice Guidelines in Oncology. Breast Cancer V.1.2009. www.nccn.org. (Last accessed

June 2, 2009).

2. Eifel P, Axelson JA, Costa J, et al. National Institutes of Health consensus development conference


3. Goldhirsch A, Glick JH, Gelber RD, Coates AS, Senn HJ. Meeting Highlights: International consensus

panel on the treatment of primary breast cancer. J Clin Oncol 2001; 19: 3817-3827.

4. Haybittle JL, Blamey RW, Elston CW, et al. A prognostic index in primary breast cancer. Br J Cancer 1982;

45: 361-366.

5. Galea MH, Blamey RW, Elston CE, Ellis IO. The Nottingham Prognostic Index in primary breast cancer.




7. Early Breast Cancer Trialists’ Collaborative Group. Polychemotherapy for early breast cancer: an

overview of the randomised trials. Lancet 1998; 352: 930-942.

8. Early Breast Cancer Trialists’ Collaborative Group. Tamoxifen for early breast cancer: an overview of the

randomised trials. Lancet 1998; 351: 1451-1467.



2005; 365: 1687-1717.

10. Siminoff LA, Gordon NH, Silverman P, Budd T, Ravdin PM. A decision aid to assist in adjuvant therapy

choices for breast cancer. Psychooncology 2006; 15: 1001-1013.

11. Zikmund-Fisher BJ, Fagerlin A, Ubel PA. Improving understanding of adjuvant therapy options by using

simpler risk graphics. Cancer 2008; 113: 3382-3390.

12. Belkora J, Rugo HS, Moore DH, Hutton D, Esserman L. Risk communication with patients with breast

cancer: cautionary notes about printing Adjuvant! estimates. Lancet Oncol 2008; 9: 602-603.



14. Coleman MP, Quaresma M, Berrino F, et al. Cancer survival in five continents: a worldwide population-

based study (CONCORD). Lancet Oncol 2008; 9: 730-756.15. Sant M, Allemani C, Berrino F, et al. Breast carcinoma survival in Europe and the United States. Cancer

2004; 100: 715-722.

16. Chlebowski RT, Aiello E, McTiernan A. Weight loss in breast cancer patient management. J Clin Oncol 2002;

20: 1128-1143.

17. CBO Richtlijn Mammacarcinoom 2008. http://www.cbo.nl/product/richtlijnen/folder20021023121843/

rl_mamma_08. pdf (Last accessed, June 2, 2009).



8/10/2019 Stella Mook T3-108


Chapter 8

150

19. Chlebowski RT, Chen Z, Anderson GL, et al. Ethnicity and breast cancer: factors influencing differences

in incidence and outcome. J Natl Cancer Inst 2005; 97: 439-448.

20. Love N. Management of breast cancer in the adjuvant and metastatic settings: Patterns of care in

medical oncology. http://www.paternsofcare.com/2005/3/adjuvant.htm (Last accessed June 2, 2009).

21. Bogaerts J, Cardoso F, Buyse M, et al. Gene signature evaluation as a prognostic tool: challenges in the


22. Cardoso F, Van ‘t Veer L, Rutgers E, et al. Clinical application of the 70-gene profile: the MINDACT trial.

J Clin Oncol 2008; 26: 729-735.

23. Harrell FE, Jr., Lee KL, Mark DB. Multivariable prognostic models: issues in developing models,

evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med 1996; 15: 361-387.

24. Schemper M and Henderson R. Predictive accuracy and explained variation in Cox regression. Biometrics

2000; 56: 249-255.

25. Von Elm E, Altman DG, Egger M, et al. The Strengthening the Reporting of Observational Studies in

Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet 2007; 370:

1453-1457.

26. Kaplan EL and Meier P. Nonparametric estimation from incomplete observation. J Am Stat Assoc 1958; 53:

457-481.

27. Cox DR. Regression models and lifetables. J R Statist Soc B 1972; 34: 187-220.

28. Pedersen L, Holck S, Schiodt T, Zedeler K, Mouridsen HT. Medullary carcinoma of the breast, prognostic

importance of characteristic histopathological features evaluated in a multivariate Cox analysis. Eur J

Cancer 1994; 30A: 1792-1797.

29. Cote RJ, Fpeterson H, Chaiwun B, et al. Role of immunohistochemical detection of lymph-node

metastases in management of breast cancer. Lancet 1999; 354: 896-900.

30. Reed J, Rosman M, Verbanac K, et al. Prognostic implications of isolated tumor cells and micrometastases

in sentinel nodes of patients with invasive breast cancer: 10-year analysis of patients enrolled in the

prospective East Carolina University/Anne Arundel Medical Center Sentinel Node Multicenter Study.

J Am Coll Surg 2009; 208: 333-340.

8/10/2019 Stella Mook T3-108



151

8


Supplementary Table 1. Different sources of pathology data.

Dept. Clin.

Chem.

Pathologist

logbook*PALGA Med. Reg. Unknown Total

N (%) N (%) N (%) N (%) N (%) N

Histology 482 (9.0) 4692 (87.2) 206 (3.8) 0 5380

Tumor size 429 (8.0) 4339 (80.6) 612 (11.4) 0 5380

Grade 397 (7.5) 3252 (60.4) 352 (6.5) 1379 (25.6) 5380

Number of positive lymph-nodes 486 (9.0) 4478 (83.2) 416 (7.8) 0 5380

Estrogen-receptor status 643 (11.9) 270 (5.0) 2214 (41.2) 0 2253 (41.9) 5380

Dept. Clin. Chemistry, Department of Clinical Chemistry; PALGA, Dutch network and National Database for

Pathology; Med. Reg., Medical Registry of Netherlands Cancer Institute-Antoni van Leeuwenhoek hospital

(NKI-AVL).

* Personal logbook of NKI-AVL pathologist J.L. Peterse.

8/10/2019 Stella Mook T3-108


Chapter 8

152

S u p p l e m e n t a r y T a b l e 2 . A d j u v a n t ! p r e d i c t e d v e r s u s o b s e r v e d O S

a n d B C S S s t r a t i fi e d b y a g e a n d e

s t r o g e n - r e c e p t o r s t a t u s .


B r e a s t c a n c e r - s p e c i fi

c s u r v i v a l

N o .

A d j u v a n t !

p r e d i c t e d

O b s e r v e d

( S

E )

P r e d i c t e d -

O b s e r v e d

P - v a l u e

A d

j u v a n t !

p r e d i c t e d

O b s e r v e d

( S E )

P r e d i c t e d -

O b s e r v e d

P - v a l u e

A g e ≤ 3 5 y r s

2 7 2

6 9 . 5

6 7 . 4 ( 2 . 9 )

2 . 1

0 . 4 7

7 0 . 4

6 7 . 8

2 . 6

0 . 3 5


P o s i t i v e

9 6

7 3 . 8

6 4 . 5 ( 4 . 9 )

9 . 3

0 . 0 6

7 4 . 8

6 4 . 5 ( 4 . 9 )

1 0 . 3

0 . 0 4

N e g a t i v e

5 8

6 6 . 0

6 9 . 0 ( 6 . 1 )

- 3 . 0

0 . 6 2

6 6 . 8

6 9 . 0 ( 6 . 1 )

- 2 . 2

0 . 7 2

U n k n o w n

1 1 8

6 7 . 8

6 9 . 1 ( 4 . 3 )

- 1 . 3

0 . 7 6

6 8 . 7

6 9 . 9 ( 4 . 3 )

- 1 . 2

0 . 7 8

A g e 3 5 - 4 0 y r s

3 0 0

7 5 . 0

6 8 . 9 ( 2 . 7 )

6 . 1

0 . 0 2

7 6 . 3

6 9 . 8 ( 2 . 7 )

6 . 5

0 . 0 2


P o s i t i v e

1 1 9

7 8 . 9

6 8 . 9 ( 4 . 2 )

1 0 . 0

0 . 0 2

8 0 . 2

6 9 . 6 ( 4 . 2 )

1 0 . 6

0 . 0 1

N e g a t i v e

5 8

6 8 . 4

7 0 . 7 ( 6 . 0 )

- 2 . 3

0 . 7 0

6 9 . 5

7 0 . 7 ( 6 . 0 )

- 1 . 2

0 . 8 4

U n k n o w n

1 2 3

7 4 . 4

6 8 . 1 ( 4 . 2 )

6 . 3

0 . 1 4

7 5 . 7

6 9 . 5 ( 4 . 2 )

6 . 2

0 . 1 4

O S , O v e r a l l s u r v i v a l ; B C S S , B r e a s t c a n

c e r - s p e c i fi c s u r v i v a l ; S E , S t a n d a r d e r

r o r .

P - v a l u e s b a s e d o n o n e s a m p l e t - t e s t .

8/10/2019 Stella Mook T3-108



153

8

Supplementary Table 3. Adjuvant! predicted versus observed BCSS stratified by number of

positive lymph-nodes.

Patients Breast cancer-specific survival

No. %Adjuvant!

predicted

Observed

(SE)

Predicted -

Observed P -value

Number of positive lymph-nodes

0 2704 50.3 88.5 87.0 (0.7) 1.5 0.03

1 876 16.3 75.8 81.8 (1.3) -6.0 <0.0001

2 538 10.0 74.7 76.0 (1.9) -1.3 0.49

3 306 5.7 73.6 71.9 (2.7) 1.7 0.53

4 213 4.0 60.6 65.8 (3.4) -5.2 0.13

5 168 3.1 57.9 66.0 (3.8) -8.1 0.03

6 111 2.1 57.4 60.7 (4.9) -3.3 0.50

7 86 1.6 56.9 55.0 (5.5) 1.9 0.73

8 70 1.3 54.5 53.9 (6.2) 0.6 0.92

9 59 1.1 55.3 45.7 (6.8) 9.6 0.16

>9 249 4.6 37.5 37.7 (3.2) -0.2 0.95

BCSS, Breast cancer-specific survival; SE, Standard error.

P-values based on one sample t-test.

8/10/2019 Stella Mook T3-108


Chapter 8

154

Supplementary Table 4. Adjuvant! predicted versus observed BCSS stratified by treatment and

estrogen-receptor status.

Patients Breast cancer-specific survival

No. %Adjuvant!

predicted

Observed

(SE)

Predicted -

Observed P -value

Untreated patients 2604

Estrogen-receptor status

Positive 1143 43.9 87.0 85.0 (1.1) 2.0 0.07

Negative 331 12.7 76.8 77.7 (2.3) -0.9 0.70

Unknown 1130 43.4 86.2 86.2 (1.1) 0.0 1.00

Chemotherapy only 815


Positive 349 42.8 72.0 74.0 (2.4) -2.0 0.41

Negative 129 15.8 62.9 64.0 (4.3) -1.1 0.80

Unknown 337 41.3 69.2 72.3 (2.5) -3.1 0.22

Hormonal therapy only 1692


Positive 787 46.5 77.8 77.4 (1.6) 0.4 0.80

Negative 192 11.3 55.5 64.9 (3.6) -9.4 0.01

Unknown 713 42.1 70.3 72.1 (1.8) -1.8 0.32

Chemo- & hormonal therapy 269


Positive 138 8.2 74.1 73.6 (3.8) 0.5 0.90

Negative 58 3.4 50.7 62.0 (6.4) -11.3 0.08

Unknown 73 4.3 66.3 75.0 (5.1) -8.7 0.09

BCSS, Breast cancer-specific survival; SE, Standard error.


8/10/2019 Stella Mook T3-108



155

8

Supplementary table 5. Adjuvant! predicted versus observed outcomes stratified by location

(within NKI-AVL or elsewhere) of primary surgery.

Patients

No. %Adjuvant!

predictedObserved (SE)

Predicted -

Observed P -value

Overall survival

Location of primary treatment 5380

Primary surgery NKI-AVL 1659 30.8 71.1 72.6 (1.1) -1.5 0.17

Primary surgery elsewhere 3721 69.2 68.2 67.4 (0.8) 0.8 0.32

Breast cancer-specific survival

Location of primary treatment 5380

Primary surgery NKI-AVL 1659 30.8 80.4 80.4 (1.0) 0.0 1.00

Primary surgery elsewhere 3721 69.2 77.1 77.8 (0.7) -0.7 0.32

NKI-AVL, Netherlands Cancer Institute – Antoni van Leeuwenhoek Hospital; SE, Standard error.


8/10/2019 Stella Mook T3-108


Chapter 9

Independent prognostic value of screen

detection in invasive breast cancer

Stella Mook


Emiel J.Th. Rutgers

Peter M. Ravdin

Anthonie O. van de Velde

Flora E. van Leeuwen

Otto Visser

Marjanka K. Schmidt

Accepted for publication in JNCI

8/10/2019 Stella Mook T3-108


Chapter 9

158

Abstract

Background

Mammographic screening has led to a proportional shift toward earlier-stage breast

cancers at presentation. We assessed whether the method of detection provides prognostic

information above and beyond standard prognostic factors and investigated the accuracy

of predicted overall and breast cancer–specific survival by the computer tool Adjuvant!

among patients with screen-detected, interval, and nonscreening-related carcinomas.

Methods

We studied 2592 patients with invasive breast cancer who were treated at the Netherlands

Cancer Institute from January 1, 1990, through December 31, 2000. Overall and breast

cancer–specific survival probabilities among patients with mammographically screen-

detected (n = 958), interval (n = 417), and nonscreening-related (n = 1217) breast

carcinomas were compared. Analyses were adjusted for clinicopathologic characteristics

and adjuvant systemic therapy. Because of gradual implementation of population-based

screening in the Netherlands, analyses were stratified a priori according to two periods of

diagnosis. All statistical tests were two-sided.

Results

Screen detection was associated with reduced mortality (adjusted hazard ratio for all-cause

mortality = 0.74, 95% confidence interval = 0.63 to 0.87, p < 0.001, and adjusted hazard

ratio for breast cancer–specific mortality = 0.62, 95% confidence interval = 0.50 to 0.78, p <

0.001, respectively) compared with nonscreening-related detection. The absolute adjusted

reduction in breast cancer–specific mortality was 7% at 10 years. The prognostic value

of the method of detection was independent of the period of diagnosis and was similar

across tumor size and lymph node status categories, indicating its prognostic value beyond

stage migration. Adjuvant! underestimated breast cancer–specific survival in patients withscreen-detected (-3.2%) and interval carcinomas (-5.4%).

Conclusions

Screen detection was found to be independently associated with better prognosis for

overall and breast cancer– specific survival and to provide prognostic information beyond

stage migration among patients with invasive breast cancer. We propose that the method

of detection should be taken into account when estimating individual prognosis.

8/10/2019 Stella Mook T3-108


Prognostic value of screen detection

159

9

Introduction

Breast cancer mortality has decreased during the last several decades because of both the

introduction of mammographic screening and the improvement and more extensive use

of adjuvant systemic therapy.1-7 Several studies have shown that breast cancer screening

leads to a reduction of breast cancer mortality for the entire population.8-11 However, it is

still unclear how much the method of detection affects the prognosis of individual patients

and whether the method of detection should be used as a prognostic factor to improve

individualized treatment.

Breast cancers detected by screening mammography are often at an earlier stage of

development than those detected after the patient has displayed symptoms of disease.12-17

This stage shift at diagnosis is a reflection of screening-related lead-time bias (i.e., the time

between detection of the tumor by mammography and the moment the tumor would have

been detected in the absence of screening).18-20 Lead-time bias automatically lengthens

survival duration, thereby causing at least part of the observed improved outcome of

patients with screen-detected tumors. Another phenomenon that contributes to the

improved outcome of patients with a screen-detected tumor is length bias.19 Carcinomas

detected by screening are not a random sample of cancers in the population but, instead,

may contain a disproportionately large proportion of slow -growing tumors that tend to be

associated with better survival, even in the absence of screening. If the method of detection

has prognostic value that is independent of known prognostic factors (such as tumor size

and lymph node status), it could potentially improve the prediction of outcome and theselection of patients for adjuvant systemic therapy and should therefore be incorporated

in decision -making tools and guidelines.

Therefore, another important question is whether prognostic tools (such as the web-

based program Adjuvant!) that are based on an unknown mixture of screen-detected

and nonscreening-related carcinomas predict outcome of patients with screen-detected

breast cancer accurately. To our knowledge, the study by Wishart et al.21 was the only study

that has evaluated whether one of the currently available prediction models (i.e., the

Nottingham Prognostic Index) is adequate for screen-detected breast cancers. In addition,

to our knowledge, none of the earlier studies by others8-11

that examined the prognosticvalue of the method of detection on prognosis evaluated both the most important and

least biased outcomes: overall and breast cancer–specific survival.22 Therefore, we assessed

whether the method of detection (i.e., screen-detected carcinomas, interval carcinomas, or

nonscreening-related carcinomas) provided independent prognostic information for the

individual patient in a comprehensive way. In addition, we investigated whether outcomes

predicted by the web-based computer tool Adjuvant! were accurate, independent of the

method of detection.23-25

8/10/2019 Stella Mook T3-108


Chapter 9

160

Patients and Methods

Patient Selection

Women who were treated for invasive breast cancer at the Netherlands Cancer Institute–

Antoni van Leeuwenhoek Hospital (NKI-AVL) from January 1, 1990, through December

31, 2000, and aged 50–69 years were selected from a database that we constructed for

a previous study.23 The following selection criteria were used: 1) a diagnosis of invasive

unilateral breast carcinoma; 2) a known tumor size of T1 (≤ 2 cm), T2 (2–5 cm), or T3 (> 5 cm);

3) a known lymph node status of negative (pN0) or positive (pN1 = 1–3, pN2 = 4–9, or pN3

>9 positive lymph nodes); 4) no distant metastases; 5) primary surgery; 6) complete axillary

lymph node staging; and 7) administration of radiation therapy according to national

guidelines. Patients with previous malignancies and patients who received neoadjuvant

therapy were not included. A total of 2861 patients fulfilled the selection criteria and were

initially included in the analysis. No ethical review was required according to the Dutch

legislation.23

Breast Cancer Screening in the Netherlands

The Dutch screening program started April 1, 1990, in a number of zip code regions, and

all women aged 50–69 years in those regions were invited to participate in the screening

program. Zip code regions were selected on the basis of availability of screening units,and regions were added as soon as a supplementary screening unit became available until

full coverage was achieved in 1997.26-27 Women were invited for biennial mam mography

through a personal letter that included a scheduled appointment for mammography that

could be changed on request. Nonattendants received a reminder after 2–3 months.27

Screening mammograms were performed in independent and (mostly) mobile screening

units (3–8 units per region). No screening mammographies were performed outside the

national screening program. Information about screening mammography or diagnostic

mammography was recorded in separate systems in the screening facility or in the hospital.

Screening was extended to women aged 70–75 years in 1998. The national participationrate of the fully implemented Dutch screening program is between 70% and 80%.26-27

Method of Detection

Information about the method of detection was retrieved from the database of the

Comprehensive Cancer Center Amsterdam and was available for 2592 of 2861 patients.

The Comprehensive Cancer Center Amsterdam is a regional cancer registry that receives

this information from the Dutch national screening facil ities. Patients with an unknown

method of detection (n = 269) were excluded (Figure 1). Breast cancer–specific survival in this

8/10/2019 Stella Mook T3-108



161

9

group was similar to that in the group of patients with a known method of detection (n =

2592) (data not shown).

Figure 1. Flow diagram for patient selection and median follow-up by method of detection.

NKI–AVL = Netherlands Cancer Institute–Antoni van Leeuwenhoek Hospital.

We classified three types of breast carcinomas on the basis of method of detection: 1)

screen-detected carcinomas, defined as carcinomas that were mammographically detected

in the first or subsequent screening rounds (n = 958); 2) interval carcinomas, defined as

symptomatic carcinomas that were diagnosed within 24 months of a negative screening

(n = 417); and 3) nonscreening-related carcinomas, defined as symptomatic carcinomas in

patients who were not participating in the screening program (n = 1217). Among the 958

patients with a screen-detected carcinoma, 510 (53%) were detected in the first screening

round (i.e., prevalent carcinomas) and 443 (46%) in a subsequent screening round (i.e.,

incident carcinomas); this information was missing for five patients. Overall survival and

breast cancer–specific survival among the 510 patients with breast carcinomas that were

Selection NKI-AVL Medical Registry

Patients included in analyses

N = 2861

n = 2592

Median follow-up 11.0 y (range 0.2-19.1 y)

Nonscreening-related

carcinomas

n = 1217

Screen-detected carcinomas

n = 958

Interval carcinomas

n = 417

Alive

n = 689

Deceased

n = 528

Alive

n = 709

Deceased

n = 249

Deceased

n = 123

Alive

n = 294

Excluded

No information about method of detection (n=269)

Median follow-up (range)

14.1 y (0.8-19.1 y)


11.9 y (8.1-18.7 y)


11.2 y (3.8-17.5 y)

8/10/2019 Stella Mook T3-108


Chapter 9

162

detected in the first screening round was similar to those in patients with breast carci-

nomas that were detected in second or subsequent screening rounds ( Supplementary Figure

1 and Supplementary Table 1), so we pooled data from patients with screen-detected prevalent

carcinomas and from patients with screen-detected incident carcinomas. Ninety-six

patients had symptomatic carcinomas detected more than 24 months after a negative

screening (interval range = 25–83 months). Disease outcomes for these 96 patients were

similar to that for patients with nonscreening-related carcinomas, and so we pooled data

from the 96 patients with symptomatic carcinomas detected more than 24 months after

a negative screening and that from the 1121 patients with nonscreening-related cancers.

Because of the stepwise implementation of the screening, the group of patients who were

diagnosed with nonscreening-related breast carcinomas could presumably represent

different groups of patients in each period of diagnosis. That is, there could have been a

larger self-selected group of nonparticipants in the nonscreening group in the later years

of diagnoses (1997–2000) compared with the early years of diagnoses (1990–1996), during

which the nonparticipants were mostly noninvited persons. Therefore, we stratified our

results into two periods of diagnosis: 1990–1996 and 1997–2000.

Pathology Data

Data on histology, tumor size, tumor grade, number of positive lymph nodes, estrogen-

receptor status (Table 1), and HER2 status were retrieved from the NKI-AVL’s Department of

Clinical Chemistry, personal logbook of NKI-AVL pathologist, Dutch Network and NationalDatabase for Pathology, and the Medical Registry of the NKI-AVL, as previously described.23

Tumors were classified into categories of stage according to the International Union Against

Cancer TNM classification and were classified by the differentiation grade according to

methods previously described by Bloom and Richardson.28

Adjuvant Treatment

Information about adjuvant systemic therapy was obtained for each patient in this study

from the NKI-AVL Medical Registry. In general, the use of adjuvant systemic therapy inthe Netherlands increased, especially during the past decade.7 Since the introduction of

a consensus guideline for adjuvant systemic therapy by the Dutch Breast Cancer Platform

(NABON) in 2000, adjuvant systemic therapy was recommended for patients with lymph

node–positive breast cancer and for a selection of patients with lymph node–negative

breast cancer, according to tumor size and grade.29 Before the introduction of this

guideline, adjuvant systemic therapy was mainly recommended for lymph node–positive

disease, tamoxifen was recommended for postmenopausal patients, and chemotherapy

was recommended for premenopausal patients. In our study cohort of 2592 patients, 1150

(44.4%) did not receive adjuvant systemic therapy, 164 (6.3%) received chemotherapy,

1105 (42.6%) received hormonal therapy, and 173 (6.7%) received both chemotherapy and

8/10/2019 Stella Mook T3-108



163

9

hormonal therapy. Trends in the usage of adjuvant systemic therapy in our study cohort are

depicted in Supplementary Figure 2.

Outcome data

Outcome data were obtained from the NKI-AVL Medical Registry (date of first local, regional,

or distant recurrence; second malignancies; and/or contralateral breast cancer and date of

last follow-up or death). These outcome data were further completed by linking patient

records to the Dutch municipal registry, which contains the date of death or emigration

for all Dutch citizens. For patients who were not registered as having died or emigrated,

the date of last follow-up was recorded as February 1, 2009 ( i.e., 2 months before the date

of linkage). Cause of death was partially retrieved from the Medical Registry and partially

from individual patient files, with 418 breast cancer–specific deaths and 94 deaths from

other causes being identified. Patients without a known cause of death (n = 388) were

considered to have died of breast cancer if they were diagnosed with distant metastases

during follow-up (n = 77 breast cancer–specific deaths; n = 311 other causes).

Predicted Outcomes by Adjuvant!

To evaluate the influence of the method of detection on algorithms for outcome prediction,

we compared 10-year observed overall survival and breast cancer–specific survival with

those that had been predicted by Adjuvant! (Batch processor version 8.0; Adjuvant!Incorporation, San Antonio, TX). Adjuvant! is a web-based computer tool that calculates

individual outcomes by entering the patient’s age, co-morbidity, tumor size, tumor grade,

number of positive axillary lymph nodes, estrogen receptor status, and adjuvant systemic

therapy. For this study, predicted outcomes were calculated by entering clinicopathologic

data for each individual patient in the Adjuvant!, version 8.0 batch processor, including

HER2 status. The Adjuvant! processor was run by one of the authors (P. M. Ravdin), while

blinded to patient outcomes. The model’s estimation of prognosis is calculated based on

10-year observed overall survival of women diagnosed with breast cancer between January

1, 1988, and December 31, 1992, in the United States and recorded in the Surveillance,Epidemiology, and End Results database.25 The estimations of treatment efficacy by this

tool are mainly calculated from the proportional risk reductions derived from the Early

Breast Cancer Trialists’ Collaborative Group 1998 meta-analyses, which was recently

updated with the meta-analyses data from 2005.30-31 Because we could not retrieve reliable

data about co-morbidity, we used the default assumption of ‘minor health problems.’ In our

study, 1447 patients had complete data for all factors that were used to predict outcome

by the Adjuvant! model. Grade was unknown for 394 tumors, and estrogen receptor status

was unknown for 931 tumors; for these tumors, grade and estrogen receptor status were

entered in the model as ‘unknown.’ Given that Adjuvant! calculates predicted outcomes at

10 years, we evaluated its accuracy in a subgroup of patients who could have had at least

10 years of follow-up (i.e., the 2329 patients who were diagnosed before the year 2000).

8/10/2019 Stella Mook T3-108


Chapter 9

164

T a b

l e 1 . A s s o c i a t i o n b e t w e e n c l i n i c

o p a t h o l o g i c a l c h a r a c t e r i s t i c s o f p a t i e n t s w i t h b r e a s t c a n c e r a n d m

e t h o d o f d e t e c t i o n * †

N o n s c r e e n i n g - r e l a t e d c a r c i n o m a

S c r e e n - d e t e c t e d c a r c i n o m a

P

I n t e r v a l c a r c i n o m a

P

C h

a r a c t e r i s t i c

N o .

( % )

N o .

( % )

N o .

( % )

Y e a r o f d i a g n o s i s

< 0 . 0 0 1

< 0 . 0 0 1

1 9 9 0 – 1 9 9 6

9 3 2 ( 7 6 . 6 )

5 1 4 ( 5 3 . 7 )

1 6 8 ( 4 0 . 3 )

1 9 9 7 – 2 0 0 0

2 8 5 ( 2 3 . 4 )

4 4 4 ( 4 6 . 3 )

2 4 9 ( 5 9 . 7 )

A g

e , y

< 0 . 0 0 1

0 . 0 2

5 0 – 5 9

7 1 5 ( 5 8 . 8 )

4 9 2 ( 5 1 . 4 )

2 7 2 ( 6 5 . 2 )

6 0 – 7 0

5 0 2 ( 4 1 . 2 )

4 6 6 ( 4 8 . 6 )

1 4 5 ( 3 4 . 8 )

H i s t o l o g y †

0 . 0 3

0 . 1 6

I D C

9 2 8 ( 7 6 . 3 )

7 0 8 ( 7 3 . 9 )

3 0 2 ( 7 2 . 4 )

I L C

1 4 4 ( 1 1 . 8 )

1 0 0 ( 1 0 . 4 )

6 4 ( 1 5 . 3 )

I D / L C

5 4 ( 4 . 4 )

7 2 ( 7 . 5 )

2 5 ( 6 . 0 )

O t h e r s

9 1 ( 7 . 5 )

7 8 ( 8 . 1 )

2 6 ( 6 . 2 )

T u

m o r s i z e ‡

< 0 . 0 0 1

0 . 1 8

p T 1

6 0 9 ( 5 0 . 0 )

7 2 7 ( 7 5 . 9 )

2 2 4 ( 5 3 . 7 )

p T 2

5 6 1 ( 4 6 . 1 )

2 2 4 ( 2 3 . 4 )

1 7 2 ( 4 1 . 2 )

p T 3

4 7 ( 3 . 9 )

7 ( 0 . 7 )

2 1 ( 5 . 0 )

N o

d a l s t a t u s §

< 0 . 0 0 1

0 . 0 5

p N 0

5 6 7 ( 4 6 . 6 )

6 3 5 ( 6 6 . 3 )

1 8 9 ( 4 5 . 3 )

p N 1

4 1 9 ( 3 4 . 4 )

2 1 8 ( 2 2 . 8 )

1 3 6 ( 3 2 . 6 )

p N 2

1 6 9 ( 1 3 . 9 )

7 5 ( 7 . 8 )

5 5 ( 1 3 . 2 )

p N 3

6 2 ( 5 . 1 )

3 0 ( 3 . 1 )

3 7 ( 8 . 9 )

S t a g e

< 0 . 0 0 1

0 . 3 6

I

3 7 9 ( 3 1 . 1 )

5 4 3 ( 5 6 . 7 )

1 3 4 ( 3 2 . 1 )

I I

5 8 5 ( 4 8 . 1 )

3 0 8 ( 3 2 . 2 )

1 8 5 ( 4 4 . 4 )

I I I

2 5 3 ( 2 0 . 8 )

1 0 7 ( 1 1 . 2 )

9 8 ( 2 3 . 5 )

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108


Chapter 9

166

Statistical Analyses

Primary endpoints were overall survival, as measured from the time of diagnosis to death

from any cause, or breast cancer–specific survival, as measured from the time of diagnosis

to breast cancer–specific death. Patients who were still alive or who had died of other causes

were censored on the date of the last follow-up or death. Kaplan–Meier survival analyses,

log-rank tests, and univariate Cox proportional hazard ratios (HRs) were calculated to esti-

mate differences in survival (mortality HRs) among patients with screen-detected, interval,

or nonscreening-related breast carcinomas. To adjust for lead-time and length bias,

multivariable Cox proportional hazard models were used to calculate the independent

prognostic value of the method of detection after adjustment for age, tumor size, axillary

lymph node status, tumor grade, estrogen receptor status, and adjuvant systemic therapy.

In addition, to minimize lead-time bias, we evaluated disease outcome for screen-detected

and nonscreening-related tumors, stratified for lymph node status and for tumor size. The

proportional hazard assumption for the Cox model was evaluated by visual examination of

the log minus log curves. Data are presented as hazard ratios with 95% confidence intervals

(CIs). For the estimation of the absolute difference in survival, directly adjusted Cox survival

curves were generated.

To assess the value of Adjuvant!, we calculated the observed overall survival and breast

cancer–specific survival from Kaplan–Meier survival analyses for each subgroup by the

method of detection and stratified by period of diagnosis. For the same datasets, the

average predicted overall survival and breast cancer–specific survival percentages werecalculated from individual predicted outcomes by Adjuvant!. Observed and average

predicted outcomes were compared with a one-sample t test by assuming that the pre-

dicted outcomes were constant. All P values are two-sided, and a P value of less than .05

was considered statistically significant. Analyses were performed with SPSS, version 15.0

(SPSS, Inc, Chicago, IL) and STATA, version 11.1 (StataCorp, College Station, TX). The study

was reported according to the STROBE statement.32

Results

Baseline Characteristics and Stage Distribution

Analyses included 2592 patients (Figure 1), of whom 1614 were diagnosed with breast cancer

between January 1, 1990, and December 31, 1996, and 978 patients were diagnosed

between January 1, 1997, and December 31, 2000. As a consequence of the stepwise

implementation of breast cancer screening in the Netherlands, breast carcinomas of most

patients who were diagnosed before 1997 were detected outside the screening program.

This group of patients diagnosed between January 1, 1990, and December 31, 1996,

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108


Chapter 9

168

Overall and Breast Cancer–Specific Survival by Method of Detection

Patients with screen-detected carcinomas had statistically significantly better overall

survival and breast cancer–specific survival than patients with nonscreening-related

carcinomas (for all-cause mortality, univariate HR = 0.60, 95% CI = 0.51 to 0.69, p < 0.001;

for breast cancer–specific mortality, univariate HR = 0.43, 95% CI = 0.34 to 0.53, p < 0.001)

(Figure 3, A and B). Similar patterns were observed for all-cause mortality and for breast cancer–

specific mortality in patients diagnosed between 1990 and 1996, and between 1997 and

2000 (Figure 3, C–F ).

In a multivariable model that was adjusted for age at diagnosis, tumor size, tumor grade,

lymph node status, estrogen receptor status, and adjuvant systemic therapy (Table 2), screen

detection was still independently associated with increased survival for patients diagnosed

with breast cancer between 1990 and 1996 (for all-cause mortality, adjusted HR = 0.77, 95%

CI = 0.64 to 0.92, p = 0.005; for breast cancer–specific mortality, adjusted HR = 0.66, 95%

CI = 0.50 to 0.86, p = 0.002). The favorable outcome of screen-detected carcinomas was of

similar magnitude in patients diagnosed more recently (for all-cause mortality, adjusted

HR = 0.73, 95% CI = 0.52 to 1.02, p = 0.07; for breast cancer–specific mortality, adjusted HR

= 0.63, 95% CI = 0.40 to 1.01, p = 0.05) but with less statistical significance. Overall, screen

detection was associated with reduced mortality (adjusted HR for all-cause mortality =

0.74, 95% CI = 0.63 to 0.87, p < 0.001; adjusted HR for breast cancer–specific mortality =

0.62, 95% CI = 0.50 to 0.78, p < 0.001) compared with nonscreening-related detection. The

absolute reduction in breast cancer–specific mortality at 10 years of follow-up betweenthe screen-detected and nonscreening-related carcinomas was 7% (adjusted survival rates

were 86% versus 79%, respectively; unadjusted differences were 13% with survival rates of

89% for screen-detected carcinomas and 76% for nonscreening-related carcinomas).

0.0

0.2

0.4

0.6

0.8

1.0

O v e r a l l s u r v i v a l p r

o b a b i l i t y

0 5 10 15

Time (years)

958

417 27356 217

883 602 152

Interval

Screen-detectedNumbersat

risk 1217 3111004 707 Nonscreening-related

Screen-detected versus nonscreening-related: HR = 0.60, 95% CI = 0.51 to 0.69

Interval versus nonscreening-related: HR = 0.79, 95% CI = 0.65 to 0.96

Log-rank p < 0.001

A

8/10/2019 Stella Mook T3-108



169

9

0.0

0.2

0.4

0.6

0.8

1.0


s u r v i v a l p r o

b a b i l i t y

0 5 10 15

Time (years)

958

417 27356 217

883 602 152

Interval

Screen-detectedNumbers

at


Screen-detectedversus nonscreening-related: HR = 0.43, 95% CI = 0.34 to 0.53


Log-rank p < 0.001

0.0

0.2

0.4

0.6

0.8

1.0

O v e r a l l s u r v i v a l p r o b a b i l i t y

0 5 10 15

Time (years)

514

168 25142 122

472 416 152

Interval


at




Log-rank p < 0.001

B

C

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108



171

9

Figure 3. Kaplan–Meier curves for overall survival and breast cancer-specific survival by method

of detection. Log-rank p-values and univariate hazard ratios (HRs) for all-cause mortality and

breast cancer-specific mortality with corresponding 95% confidence intervals (CIs) at 5, 10, and

15 years are shown (error bars).

A) Overall survival for all patients (n = 2592).B) Breast cancer-specific survival for all patients (n = 2592).

C) Overall survival for patients diagnosed in 1990–1996 (n = 1614).

D) Breast cancer-specific survival for patients diagnosed in 1990–1996 (n = 1614).

E) Overall survival for patients diagnosed in 1997–2000 (n = 978).

F) Breast cancer-specific survival for patients diagnosed in 1997–2000 (n = 978). Interval

carcinomas were diagnosed 24 months or less after a negative screening. Non-screening-related

carcinomas were symptomatic cancer in patients who had not been screened or were screened

more than 24 months before detection of breast cancer. Numbers of patients at risk are shown

below each graph.

Furthermore, diagnosis of an interval carcinoma between 1990 and 1996 was independently

associated with better survival (for all-cause mortality, adjusted HR = 0.71, 95% CI = 0.54

to 0.96, p = 0.02; for breast cancer–specific mortality, adjusted HR = 0.70, 95% CI = 0.49 to

1.01, p = 0.05) compared with nonscreening-related detection. Conversely, diagnosis of an

interval carcinoma between 1997 and 2000 was not associated with survival (for all-cause

mortality, adjusted HR = 0.89, 95% CI = 0.63 to 1.24, p = 0.48; for breast cancer–specific

mortality, adjusted HR = 0.91, 95% CI = 0.59 to 1.40, p = 0.66) (Table 2).

0.0

0.2

0.4

0.6

0.8

1.0


s u r v i v a l p r

o b a b i l i t y

0 5 10

Time (years)

444

249 95209

411 186

Interval


at

risk 285 99244 Nonscreening-related



Log-rank p = 0.72

F

8/10/2019 Stella Mook T3-108


Chapter 9

172

We also found that the interaction term between the method of detection and lymph

node status was statistically significant in multivariable models for all-cause mortality ( p

= 0.045) and breast cancer–specific mortality ( p = 0.009). However, in the model including

the interaction term, screen detection was still associated with all-cause and breast

cancer–specific mortality, with similar HRs for screen-detected cancers (all-cause mortality,

adjusted HR = 0.70, 95% CI = 0.56 to 0.87, p = 0.001; breast cancer–specific mortality,

adjusted HR = 0.51, 95% CI = 0.36 to 0.72, p < 0.001) and somewhat lower HRs for interval

cancers (all-cause mortality, adjusted HR = 0.48, 95% CI = 0.33 to 0.71, p < 0.001; and breast

cancer–specific mortality, adjusted HR = 0.27, 95% CI = 0.14 to 0.51, p < 0.001).

Table 2. Multivariable Cox proportional hazard analyses for all-cause mortality and breast cancer-

specific mortality for all patients, patients who were diagnosed in the period of implementation

of screening (1990-1996), and patients who were diagnosed in the period when screening

reached full coverage (1997–2000)*.

All-cause mortalityBreast cancer-specific

mortality

Characteristics P HR (95% CI) P HR (95% CI)

Year of diagnosis 1990–2000

Method of detection

Screen-detected versus nonscreening-related <0.001 0.74 (0.63 to 0.87) <0.001 0.62 (0.50 to 0.78)

Interval carcinoma versus nonscreening-related 0.009 0.76 (0.62 to 0.93) 0.02 0.73 (0.56 to 0.94)

Age (per year) <0.001 1.05 (1.04 to 1.06) 0.15 1.01 (1.00 to 1.03)

pT†

pT2 (versus pT1) <0.001 1.51 (1.30 to 1.75) <0.001 1.75 (1.43 to 2.14)

pT3 (versus pT1) <0.001 1.80 (1.30 to 2.50) 0.02 1.65 (1.09 to 2.49)

pN‡

pN1 (versus pN0) <0.001 1.44 (1.16 to 1.77) 0.003 1.56 (1.17 to 2.07)

pN2 (versus pN0) <0.001 2.32 (1.82 to 2.97) <0.001 3.01 (2.18 to 4.15)

pN3 (versus pN0) <0.001 4.40 (3.25 to 5.94) <0.001 6.30 (4.36 to 9.10)

GradeII (versus I) 0.01 1.34 (1.06 to 1.69) <0.001 2.31 (1.51 to 3.55)

III (versus I) <0.001 2.31 (1.79 to 2.96) <0.001 4.59 (2.95 to 7.13)

Grade unknown (versus I) <0.001 1.54 (1.19 to 1.99) <0.001 3.03 (1.93 to 4.76)

ER status

ER negative (versus ER positive) 0.04 1.24 (1.01 to 1.53) 0.06 1.29 (0.99 to 1.68)

ER unknown (versus ER positive) 0.15 1.11 (0.96 to 1.29) 0.23 1.13 (0.92 to 1.39)

Chemotherapy (yes versus no) 0.14 0.83 (0.65 to 1.06) 0.02 0.71 (0.53 to 0.95)

Hormonal therapy (yes versus no) 0.02 0.78 (0.65 to 0.95) 0.10 0.80 (0.62 to 1.04)

8/10/2019 Stella Mook T3-108



173

9

Table 2. Continued


mortality


Year of diagnosis: 1990–1996

Method of detection

Screen-detected versus nonscreening-related <0.005 0.77 (0.64 to 0.92) 0.002 0.66 (0.50 to 0.86)


Age (per year) <0.001 1.04 (1.03 to 1.06) 1.0 1.00 (0.98 to 1.02)

pT†

pT2 (versus pT1) <0.001 1.54 (1.30 to 1.82) <0.001 1.86 (1.47 to 2.35)

pT3 (versus pT1) 0.003 1.79 (1.22 to 2.62) 0.04 1.64 (1.01 to 2.67)pN‡

pN1 (versus pN0) 0.04 1.31 (1.02 to 1.69) 0.08 1.37 (0.97 to 1.95)

pN2 (versus pN0) <0.001 2.23 (1.67 to 2.98) <0.001 3.00 (2.06 to 4.37)

pN3 (versus pN0) <0.001 4.13 (2.86 to 5.95) <0.001 6.13 (3.93 to 9.55)

Grade

II (versus I) 0.07 1.28 (0.99 to 1.66) 0.004 2.00 (1.25 to 3.18)

III (versus I) <0.001 2.15 (1.62 to 2.85) <0.001 3.91 (2.43 to 6.30)

Grade unknown (versus I) 0.02 1.40 (1.06 to 1.84) <0.001 2.49 (1.54 to 4.02)

ER status





Continued►

8/10/2019 Stella Mook T3-108


Chapter 9

174

Table 2. Continued


mortality


Year of diagnosis: 1997–2000

Method of detection

Screen-detected versus nonscreening-related 0.07 0.73 (0.52 to 1.02) 0.05 0.63 (0.40 to 1.01)


Age (per year) <0.001 1.06 (1.04 to 1.09) 0.005 1.05 (1.02 to 1.09)

pT†

pT2 (versus pT1) 0.05 1.36 (1.00 to 1.85) 0.22 1.29 (0.86 to 1.93)

pT3 (versus pT1) 0.10 1.71 (0.91 to 3.20) 0.43 1.39 (0.61 to 3.16)pN‡

pN1 (versus pN0) 0.006 1.71 (1.16 to 2.51) 0.02 1.89 (1.12 to 3.18)

pN2 (versus pN0) 0.001 2.27 (1.39 to 3.71) 0.01 2.29 (1.20 to 4.37)

pN3 (versus pN0) <0.001 4.71 (2.75 to 8.07) <0.001 6.04 (3.07 to 11.87)

Grade

II (versus I) 0.06 1.64 (0.97 to 2.75) 0.02 4.24 (1.31 to 13.78)

III (versus I) <0.001 2.86 (1.62 to 5.07) 0.001 7.24 (2.15 to 24.40)

Grade unknown (versus I) 0.02 2.33 (1.12 to 4.83) 0.08 3.68 (0.86 to 15.87)

ER status





* All analyses were done with the use of the Cox proportional hazard model. All statistical tests were two-

sided.

CI = confidence interval; ER = estrogen receptor; HR = hazard ratio.† pT = pT1 ≤2 cm; pT2 = 2–5 cm; pT3 >5 cm. ‡ pN = pN0 = lymph node-negative; pN1 = 1–3 positive lymph

nodes; pN2 = 4–9 positive lymph nodes; pN3 >9 positive lymph nodes.

In 1998, screening in the Netherlands was extended to woman aged 70–75 years. Including

those patients in this analysis resulted in 180 additional breast cancer patients, including

74 (41%) screen-detected breast cancers, 14 (8%) interval breast cancers, and 92 (51%)

nonscreening-related breast cancers. When we included this age group in the survival

analysis, the difference in outcomes between screen-detected and nonscreening-related

8/10/2019 Stella Mook T3-108



175

9

carcinomas was even larger for breast cancer–specific mortality (unadjusted HR = 0.35,

95% CI = 0.23 to 0.52, p < 0.001 and adjusted HR = 0.49, 95% CI = 0.32 to 0.75, p < 0.001)

( Supplementary Table 2).

Screen-detected breast carcinomas were smaller (mean size 17 versus 24 mm; p < 0.001)

and more often (66.3% versus 46.6%) had a lymph node–negative status compared with

nonscreening-related carcinomas, reflecting the well-known stage shift caused by

screening (Table 1). In addition to the multivariable analyses, we compared breast cancer–

specific survival between patients with screen-detected carcinomas and patients with

nonscreening-related carcinomas as stratified by tumor size and by lymph node status.

Because the differences in breast cancer–specific survival between screen-detected and

nonscreening-related carcinomas were similar in both periods of diagnosis (1990–1996

and 1997–2000), we pooled patients with such carcinomas to increase sample sizes for

subgroup analyses by tumor size and lymph node status. Patients with screen-detected

cancers had better breast cancer–specific survival than patients with nonscreening-related

tumors within each stratum of tumor size, with the most pronounced difference in tumors

of 10 mm or less in diameter (for breast cancer–specific mortality, unadjusted HR = 0.28, 95%

CI = 0.11 to 0.71, p = 0.007; adjusted HR = 0.35, 95% CI = 0.13 to 0.96, p = 0.04) ( Supplementary

Table 3 and Supplementary Figure 3, A–D). In analyses stratified by lymph node status, better breast

cancer–specific survival was associated with screen-detected tumors in both patients with

lymph node–negative and patients with lymph node–positive breast cancer (in lymph

node–negative patients, unadjusted HR for breast cancer–specific mortality = 0.40, 95% CI

= 0.28 to 0.56, p < 0.001; in lymph node–positive patients, unad justed HR = 0.59, 95% CI =0.45 to 0.79, p < 0.001). After adjustment for other prognostic factors (including age, tumor

size, grade, estrogen receptor status and adjuvant systemic therapy), screen detection was

strongly and statistically significantly associated with improved breast cancer–specific

survival among patients with lymph node–negative disease, but this association was

weaker and non-statistically significant among patients with lymph node–positive disease

(in lymph node–negative patients, adjusted HR for breast cancer–specific mortality =

0.51, 95% CI = 0.36 to 0.73, p < 0.001; in lymph node–positive patients, adjusted HR = 0.79,

95% CI = 0.59 to 1.06, p = 0.12) ( Supplementary Table 4 and Supplementary Figure 3, E–F ). The observed

survival difference of patients with lymph node–positive screen-detected carcinomas wasto a larger extent associated with stage shift and period of diagnosis; that is, lymph node–

positive patients with screen-detected breast cancer were statistically significantly more

likely to be diagnosed in 1997–2000 and to have smaller and better differ entiated tumors

than lymph node–positive patients whose breast cancer was not detected by screening ( p

< 0.001) (data not shown). Screen detection was also independently associated with breast

cancer–specific survival among systemically untreated patients (adjusted HR = 0.48, 95%

CI = 0.32 to 0.71, p < 0.001).

8/10/2019 Stella Mook T3-108


Chapter 9

176

T a b

l e 3 . A d j u v a n t ! p r e d i c t e d a n d o b

s e r v e d 1 0 - y e a r o v e r a l l s u r v i v a l a

n d b r e a s t c a n c e r - s p e c i fi c s u r v i v a

l f o r a l l p a t i e n t s d i a g n o s e d b e f o r e 2 0 0 0 , f o r

p a t

i e n t s d i a g n o s e d i n t h e p e r i o d o f i m p l e m e n t a t i o n o f s c r e e n i n g ( 1 9

9 0 – 1 9 9 6 ) , a n d f o r p a t i e n t s d i a g n

o s e d i n t h e p e r i o d w h e n s c r e e n i n g r e a c h e d

f u l l

c o v e r a g e ( 1 9 9 7 – 1 9 9 9 ) * .

1 0 - y e a r o v e r a l l s u r v i v a l

1 0 - y e a r b r e a s t c a n c e r - s p e c

i fi c s u r v i v a l

P e

r i o d o f d i a g n o s i s a n d

m e t h o d o f d e t e c t i o n

N o .

P a t i e n

t s ( % )

A d j u v a n t !

p r e d i c t e d ,

%

O b s e r

v e d ,

% (

9 5 %

C I )

P r e d i c t e d –

o b s e r v e d

P

A d

j u v a n t !

p r e d i

c t e d ,

%

O b s e r v e d ,

% (

9 5 % C

I ) P r e d i c t e

d –

o b s e r v e d

P

M e t h o d o f d e t e c t i o n , y e a r o f d i a g n o

s i s : 1 9 9 0 – 1 9 9 9

N

o n s c r e e n i n g r e l a t e d

1 1 4 9

( 6 4 . 9 )

6 9 . 2

6 6 . 6 ( 6 3 . 9 t o 6 9 . 3 )

2 . 6

0 . 0 6

7 6 . 5 7 5 . 7 ( 7 3 . 2 t o 7 8 . 2 )

0 . 8

0 . 5 4

S

c r e e n - d e t e c t e d

8 3 0

( 4 6 . 9 )

7 7 . 6

8 2 . 0 ( 7 9 . 5 t o 8 4 . 5 )

- 4 . 4

< 0 . 0 0 1

8 6 . 1 8 9 . 3 ( 8 7 . 1 t o 9 1 . 5 )

- 3 . 2

0 . 0 0 4

I

n t e r v a l c a r c i n o m a

3 5 0

( 1 9 . 8 )

6 8 . 7

7 2 . 4 ( 6 7 . 7 t o 7 7 . 1 )

- 3 . 7

0 . 1 2

7 5 . 5 8 0 . 9 ( 7 6 . 6 t o 8 5 . 2 )

- 5 . 4

0 . 0 2


s i s : 1 9 9 0 – 1 9 9 6 ( i m p l e m e n t a t i o n o

f s c r e e n i n g )

N


9 3 2

( 5 7 . 7 )

6 8 . 6

6 5 . 3 ( 6 2 . 2 t o 6 8 . 4 )

3 . 3

0 . 0 4

7 6 . 2 7 4 . 3 ( 7 1 . 4 t o 7 7 . 2 )

1 . 9

0 . 2 1

S


5 1 4

( 3 1 . 8 )

7 6 . 7

8 0 . 9 ( 7 7 . 6 t o 8 4 . 2 )

- 4 . 2

0 . 0 1

8 5 . 5 8 7 . 7 ( 8 4 . 8 t o 9 0 . 6 )

- 2 . 2

0 . 1 4

I


1 6 8

( 1 0 . 4 )

6 8 . 9

7 3 . 2 ( 6 6 . 5 t o 7 9 . 9 )

- 4 . 3

0 . 2 1

7 5 . 9 8 0 . 9 ( 7 4 . 8 t o 8 7 . 0 )

- 5 . 0

0 . 1 1


s i s : 1 9 9 7 – 1 9 9 9 ( f u l l c o v e r a g e o f s c r e e n i n g )

N


2 1 7

( 2 2 . 2 )

7 1 . 9

7 2 . 3 ( 6 6 . 2 t o 7 8 . 4 )

- 0 . 4

0 . 9 0

7 8 . 2 8 1 . 7 ( 7 6 . 2 t o 8 7 . 2 )

- 3 . 5

0 . 2 1

S


3 1 6

( 3 2 . 3 )

7 9 . 2

8 3 . 8 ( 7 9 . 7 t o 8 7 . 9 )

- 4 . 6

0 . 0 3

8 7 . 0 9 2 . 1 ( 8 9 . 0 t o 9 5 . 2 )

- 5 . 1

0 . 0 0 2

I


1 8 2

( 1 8 . 6 )

6 8 . 6

7 1 . 7 ( 6 5 . 0 t o 7 8 . 4 )

- 3 . 1

0 . 3 6

7 5 . 3 8 0 . 8 ( 7 4 . 9 t o 8 6 . 7 )

- 5 . 5

0 . 0 0 7

* A n a l y s e s w e r e d o n e w i t h t h e u s e o f o n

e s a m p l e t t e s t s . A l l s t a t i s t i c a l t e s t s w

e r e t w o - s i d e d . C I = c o n fi d e n c e i n t e r v a l .

8/10/2019 Stella Mook T3-108



177

9

Adjuvant! Predictions

To evaluate the influence of the method of detection on an algorithm for outcome

prediction, we compared 10-year observed overall survival and breast cancer–specific

survival with 10-year overall survival and breast cancer–specific survival that had

been predicted by Adjuvant!. Adjuvant! predicted the outcome among patients with

nonscreening-related carcinomas accurately, that is, the predicted survival of patients with

a nonscreening-related carcinoma was within 2% of the observed survival and/or non-

statistically significantly different in all but one group: Adjuvant! overestimated overall

survival in patients with a nonscreening-related carcinoma diagnosed between 1990 and

1996 with 3.3% ( p= 0.04). However, Adjuvant! predictions underestimated overall survival

and breast cancer–specific survival among patients with screen-detected and interval

carcinomas. Prediction of breast cancer–specific survival was underestimated by Adjuvant!

for patients with screen-detected and interval carcinomas by -3.2% and -5.4%, respectively.

Among patients with screen-detected carcinomas, in particular, Adjuvant! underestimated

survival for all periods of diagnosis (Table 3). In addition, Adjuvant! underestimated breast

cancer–specific survival in patients younger than 50 years whose breast cancer was

diagnosed more recently (1997–2000) with -3.0%, reflecting the observed 22% reduction

in breast cancer–specific survival, whereas Adjuvant! predictions for patients younger than

50 years diagnosed between 1990 and 1996 were accurate ( Supplementary Table 5 ).23

Discussion

We found that screen detection was independently associated with better breast cancer–

specific survival, as shown in multivariable analyses and analyses stratified for tumor size

and lymph node status, and provided prognostic information beyond stage migration

for patients with invasive breast cancer. These results are in agreement with previous

studies.9,11,21 Therefore, the method of detection should be taken into account when

selecting patients for adjuvant systemic therapy and withholding chemotherapy for women

with screen-detected carcinoma could be considered. We also analyzed the accuracy ofpredicted disease outcome by the computer tool Adjuvant! as stratified by the method of

detection.25 When we compared outcomes from this study with outcomes predicted by

Adjuvant! for the same patients, we found that predicted breast cancer–specific survival

by Adjuvant! was underestimated for all three groups, with the most pronounced and

statistically significant differences in patients with screen-detected and interval breast

carcinomas.

The true independent prognostic value of the method of detection for individual breast

cancer patients may remain a matter of dispute. This dispute may not be settled unless

a precise method for assessing tumor advancement is developed. Nonetheless, we

8/10/2019 Stella Mook T3-108


Chapter 9

178

have obtained consistent evidence in this study that the method of detection was an

independent prognostic factor beyond stage shift for disease outcome in patients with

invasive breast carcinomas, with increased survival being associated with screen-detected

carcinomas.

However, the question remains whether models and guidelines for adjuvant systemic

therapy that were developed in an (partially) unscreened population are applicable to

patients with screen-detected carcinomas and whether the use of these models and guide-

lines may lead to overtreatment in these patients. Our data indicate that current models for

determining prognosis in breast cancer patients may be improved by including the method

of detection. In our study, the general underestimation of survival outcome by Adjuvant!

for patients diagnosed with breast cancer in 1997 through 1999 may be attributed in part

to improved salvage therapy and adjuvant systemic therapy.33-35 We observed a reduction

in breast cancer–specific mortality for patients younger than 50 years who were recently

diagnosed, which supports this hypothesis. However, the underestimations of overall

and breast cancer–specific survival by Adjuvant! were most pronounced in patients with

screen-detected carcinomas, indicating that although improved therapy will influence the

model’s prediction, the method of detection should be taken into account when selecting

patients for adjuvant systemic therapy.

It is well established and confirmed by our study that screened populations have a larger

proportion of smaller, lymph node–negative, lower grade, and estrogen receptor–positive

tumors than nonscreened populations. Stratification by tumor size reduces the magnitude

of lead-time bias that is caused by stage shift, although it may not completely eliminatelead-time bias because shift within a stage can still occur (e.g., more T1ab tumors with a

diameter of ≤ 1 cm in the pT1 category of ≤ 2 cm tumors). We found that, even within

strata of less than 10-mm tumors and of 10- to 20 mm tumors, screen-detected carcinomas

were associated with better breast cancer–specific survival than non-screened carcinomas.

Stratification by lymph node status showed that, despite the prognostic value of screen

detection being similar among patients with lymph node–negative and lymph node–

positive disease, screen detection was independently associated with breast cancer–

specific survival especially for patients with no lymph node metastases at diagnosis and

that the observed survival difference of patients with lymph node–positive screen-detectedcarcinomas was to a larger extent associated with stage shift and period of diagnosis.

When studying the effect of screening on mortality at the population level, both lead-time

(stage shift) and length bias (less-aggressive tumors) can cause a spurious improvement of

survival in the screened population. When investigating the independent prognostic value

of the method of detection on the prediction of outcome for an individual patient, screen-

detected carcinomas appear to have a more favorable tumor biology (e.g., to be at a low

grade at diagnosis) and are subject to potential overdiagnosis.36 A different natural history

of screen-detected carcinomas has also been postulated by others.37,38 Moreover, we found

that, even after adjustment for known prognostic factors and within strata of tumor size,

8/10/2019 Stella Mook T3-108



179

9

method of detection had independent prognostic value. This result could indicate that

we were not able to completely correct for length bias with the prognostic factors that

were available; however, we argue that it is exactly this remaining unexplained difference

in tumor biology beyond stage shift (length bias) that is important in the prediction of

outcome for the individual patient.

Although the prognostic value of screen detection was similar among patients diagnosed

in 1990–1996 and those diagnosed in 1997–2000, there were several reasons to stratify our

data analyses according to period of diagnosis. First, because the stepwise implementation

of screening in the Netherlands, patients diagnosed with nonscreening-related breast

carcinomas in 1990–1996 were predominantly women who had not been invited for

screening in a certain geographic region, whereas patients diagnosed with nonscreening-

related breast carcinomas after 1996 were more likely to have been a specific subset of

patients who decided not to participate in the breast cancer screening program for various

reasons. Patients diagnosed from 1990 to 1996 were expected to be a random selection

of patients that, consequently, can be viewed as a unique control group of nonscreening-

related cancers. Patients diagnosed after 1996 are subject to selection biases, such as

worse accessibility to adequate treatment facilities or lower socioeconomic status, which

will influence both participation rate and outcome. In addition, as shown by Kalager et

al.,10 organized screening programs also result in a survival benefit for patients outside

the screening program that can be attributed to increased awareness and optimization

of breast cancer care. This effect may dilute the prognostic value of screen detection,

especially in more recent years (1997–2000) when the coverage of screening was complete. The possibility that both screening and improved adjuvant systemic therapy contributed to

the reduction in breast cancer mortality from 1990 to 2010 with similar magnitude further

emphasizes the necessity for comparisons during the same period between patients with

screen-detected carcinomas and those with nonscreening-related carcinomas.39

The observed intermediate survival among patients with interval breast cancers was

consistent with previous studies.40-42 Interval breast carcinomas are a heterogeneous

group of tumors consisting of true interval carcinomas (i.e., rapidly growing tumors), occult

carcinomas, and tumors that were missed on previous screening mammography (i.e., slowly

growing tumors).43

The latter group of tumors could be related to breast density and thusassociated with younger age at diagnosis, which we did observe. Although general baseline

tumor characteristics did not differ much between interval cancers and nonscreening-

related cancers, patients with interval carcinomas received adjuvant systemic therapy

more often. However, we found that, in the adjuvant-untreated group specifically, survival

was similar between patients with interval carcinomas and patients with a screen-detected

carcinoma (data not shown). In addition, increased patient awareness as a result of the

screening program and self-selection could have resulted in a better outcome for interval

carcinomas.

8/10/2019 Stella Mook T3-108


Chapter 9

180

One of the strengths of our study was that information about whether a tumor was

detected in the first screening round ( i.e., was a prevalent carcinoma) or in a subsequent

screening round (i.e., was an incident carcinoma) was available. Another strength was

that both overall survival and breast cancer–specific survival could be evaluated and that

outcome data were collected in a manner that was blinded to the method of detection,

thereby eliminating potential problems of differential bias in ascertainment and coding

of cause of death (i.e., the sticky-diagnosis and slippery-linkage biases, respectively).44 In

addition, patients in this study cohort had a substantial follow-up.

As a consequence of the conservative adjuvant systemic therapy guidelines that were

used during the time patients in this study cohort were treated for breast cancer, a large

proportion of patients received no chemo and/or hormonal therapy ( Supplementary Figure

2). Although most patients with early-stage breast cancer currently receive some form

of adjuvant systemic therapy, the estimated prognosis (i.e., disease outcome without

adjuvant systemic therapy) is important for the decision whether or not to treat patients

with adjuvant systemic therapy. We found that method of detection has independent

prognostic value, with screen detection associated with better survival. Thus, withholding

chemotherapy for women with screen-detected carcinoma could be considered; however,

these results require validation in an independent series of patients.

There have been several randomized clinical screening trials8,9,11 that have investigated the

association between screening and outcome. These studies have identified a reduction of

25%–35% in breast cancer mortality that is associated with screen detection, which was

sustained, though attenuated, after adjustment for prognostic factors. Notwithstandingthe outcome of the randomized clinical trials, these results are not simply applicable to the

general and diverse population that is participating in breast cancer screening. Shen et al.11

studied patients who were included in trials that were conducted in 1963–1966 and 1980–

1985, thus representing different birth cohorts from those of women who are currently at

risk of breast cancer. In addition, they missed important clinicopathologic information (e.g.,

exact tumor size and tumor grade). Although Joensuu et al.9 had detailed information and

could adjust distant disease-free survival for tumor size, number of positive lymph nodes,

tumor grade, and hormone receptor status, there were only 443 patients with a screen-

detected carcinoma. Wishart et al.21

have shown that screen-detected carcinomas wereassociated with better overall survival than symptomatic breast cancer. The prognostic

value of screen detection remained statistically significant after correction for stage shift

defined by the Nottingham Prognostic Index, but no data on breast cancer–specific

survival were included. Although Wishart et al.21 analyzed 5604 patients, the control group

(i.e., symptomatic breast cancer patients) consisted of a mixture of patients with potentially

very different survival rates, that is, true interval carcinomas, patients with interval car-

cinomas diagnosed more than 3 years after a negative mammographic screening, and

patients who did not participate in the screening program.

8/10/2019 Stella Mook T3-108



181

9

This study had several limitations. Information was not available for tumor markers, such

as HER2 (for the majority of patients) or Ki67. However, it is not likely that the remaining

difference in outcome between screen-detected and nonscreening-related tumors can

be explained merely by these factors.45 In general, patients were treated according to

guidelines available at that time; however, other factors (e.g., co-morbidity or HER2 status)

could have influenced choice of treatment. Residual con founding by indication was shown

by the increased risk of death after chemotherapy or hormonal treatment in carcinomas

diagnosed in 1997–2000 (Table 2), although hazard ratios of treatment were non-statistically

significant. Several studies46-48 have shown that gene expression profiles can account for

a substantial part of the unexplained variance in prognosis. Therefore, the independent

prognostic value of method of detection after adjustment for gene expression in a tumor

remains to be determined.

In summary, we have shown that screen detection was consistently associated with disease

outcome and provided prognostic information beyond stage migration among patients

with invasive breast cancer. As a consequence, we propose that method of detection should

be used in combination with traditional markers of tumor burden and aggressiveness

to estimate prognosis for each patient, and to guide their decision to receive adjuvant

systemic therapy.

Funding

This work was supported by The Dutch Cancer Society (grant number NKI 2009-4363)(MKS) and the Dutch National Genomics Initiative-Cancer Genomics Center (grant number

NKI CGC 2008-2012) (LvtV).

Notes

We acknowledge the contribution of the Medical Registry and the Clinical Chemistry

Department of the NKI-AVL (Amsterdam, The Netherlands), all treating oncologists of the

patients concerned, and all radiologists who collected and interpreted the mammographic

screening data provided by the Comprehensive Cancer Center Amsterdam. Dr. Ravdin wasinvolved in the design of and has an interest in Adjuvant!. The funding sources had no role

in study design, collection, analysis, or interpretation of data, writing of the paper, or in

decisions relating to publication.

8/10/2019 Stella Mook T3-108


Chapter 9

182

Reference List

1. Berry DA and Ravdin PM. Breast cancer trends: a marriage between clinical trial evidence and

epidemiology. J Natl Cancer Inst 2007; 99: 1139-1141.

2. Chu KC, Tarone RE, Kessler LG, et al. Recent trends in U.S. breast cancer incidence, survival, and mortality

rates. J Natl Cancer Inst 1996; 88: 1571-1579.

3. Garne JP, Aspegren K, Balldin G, Ranstam J. Increasing incidence of and declining mortality from breast

carcinoma. Trends in Malmo, Sweden, 1961-1992. Cancer 1997; 79: 69-74.

4. Jatoi I and Miller AB. Why is breast-cancer mortality declining? Lancet Oncol 2003; 4: 251-254.

5. Otten JD, Broeders MJ, Fracheboud J, et al. Impressive time-related influence of the Dutch screening

programme on breast cancer incidence and mortality, 1975-2006. Int J Cancer 2008; 123: 1929-1934.

6. Smigal C, Jemal A, Ward E, et al. Trends in breast cancer by race and ethnicity: update 2006. CA Cancer J Clin

2006; 56: 168-183.

7. Vervoort MM, Draisma G, Fracheboud J, van de Poll-Franse LV, de Koning HJ. Trends in the usage of

adjuvant systemic therapy for breast cancer in the Netherlands and its effect on mortality. Br J Cancer

2004; 91: 242-247.

8. De Koning HJ. Mammographic screening: evidence from randomised controlled trials. Ann Oncol 2003;

14: 1185-1189.

9. Joensuu H, Lehtimaki T, Holli K, et al. Risk for distant recurrence of breast cancer detected by

mammography screening or other methods. JAMA 2004; 292: 1064-1073.

10. Kalager M, Haldorsen T, Bretthauer M, et al. Improved breast cancer survival following introduction of

an organized mammography screening program among both screened and unscreened women: a

population-based cohort study. Breast Cancer Res 2009; 11: R44.

11. Shen Y, Yang Y, Inoue LY, et al. Role of detection method in predicting breast cancer survival: analysis of

randomized screening trials. J Natl Cancer Inst 2005; 97: 1195-1203.


Health Insurance Plan clinical trial. J Natl Cancer Inst 1988; 80: 1125-1132.

13. Connor RJ, Chu KC, Smart CR. Stage-shift cancer screening model. J Clin Epidemiol 1989; 42: 1083-1095.

14. Ernst MF, Roukema JA, Coebergh JW, et al. Breast cancers found by screening: earlier detection, lower

malignant potential or both? Breast Cancer Res Treat 2002; 76: 19-25.

15. Feinstein AR, Sosin DM, Wells CK. The Will Rogers phenomenon. Stage migration and new diagnostictechniques as a source of misleading statistics for survival in cancer. N Engl J Med 1985; 312: 1604-1608.

16. Fracheboud J, Otto SJ, van Dijck JA, et al. Decreased rates of advanced breast cancer due to


17. Weaver DL, Rosenberg RD, Barlow WE, et al. Pathologic findings from the Breast Cancer Surveillance

Consortium: population-based outcomes in women undergoing biopsy after screening mammography.

Cancer 2006; 106: 732-742.

18. Hutchison GB and Shapiro S. Lead time gained by diagnostic screening for breast cancer. J Natl Cancer Inst

1968; 41: 665-681.

19. Kramer BS, Gohagan JK, Prorok PC. Cancer screening: theory and practice. New York (NY): Marcel

Dekker; 1999.

8/10/2019 Stella Mook T3-108



183

9

20. Shapiro S, Goldberg JD, Hutchison GB. Lead time in breast cancer detection and implications for

periodicity of screening. Am J Epidemiol 1974; 100: 357-366.

21. Wishart GC, Greenberg DC, Britton PD, et al. Screen-detected vs symptomatic breast cancer: is

improved survival due to stage migration alone? Br J Cancer 2008; 98: 1741-1744.

22. Krzyzanowska MK and Tannock IF. Should screen-detected breast cancers be managed differently?

J Natl Cancer Inst 2005; 97: 1170-1171.

23. Mook S, Schmidt MK, Rutgers EJ, et al. Calibration and discriminatory accuracy of prognosis calculation

for breast cancer with the online Adjuvant! program: a hospital-based retrospective cohort study. Lancet

Oncol 2009; 10: 1070-1076.





26. De Koning HJ, Fracheboud J, Boer R, et al. Nation-wide breast cancer screening in The Netherlands:

support for breast-cancer mortality reduction. National Evaluation Team for Breast Cancer Screening

(NETB). Int J Cancer 1995; 60: 777-780.

27. Fracheboud J, De Koning HJ, Beemsterboer PM, et al. Nation-wide breast cancer screening in The

Netherlands: results of initial and subsequent screening 1990-1995. National Evaluation Team for

Breast Cancer Screening. Int J Cancer 1998; 75: 694-698.

28. Bloom HJ and Richardson WW. Histological grading and prognosis in breast cancer; a study of 1409

cases of which 359 have been followed for 15 years. Br J Cancer 1957; 11: 359-377.

29. Bontenbal M, Nortier JW, Beex LV, et al. [Adjuvant systemic therapy for patients with resectable breast

cancer: guideline from the Dutch National Breast Cancer Platform and the Dutch Society for Medical

Oncology]. Ned Tijdschr Geneeskd 2000; 144: 984-989.

30. Early Breast Cancer Trialists’ Collaborative Group. Polychemotherapy for early breast cancer: an

overview of the randomised trials. Lancet 1998; 352: 930-942.



2005; 365: 1687-1717.

32. Von Elm E, Altman DG, Egger M, et al. The Strengthening the Reporting of Observational Studies in

Epidemiology (STROBE) statement: guidelines for reporting observational studies. Lancet 2007; 370:1453-1457.

33. Coleman MP, Quaresma M, Berrino F, et al. Cancer survival in five continents: a worldwide population-

based study (CONCORD). Lancet Oncol 2008; 9: 730-756.



35. Sant M, Allemani C, Berrino F, et al. Breast carcinoma survival in Europe and the United States. Cancer

2004; 100: 715-722.

36. Jorgensen KJ and Gotzsche PC. Overdiagnosis in publicly organised mammography screening

programmes: systematic review of incidence trends. BMJ 2009; 339: b2587.

8/10/2019 Stella Mook T3-108


Chapter 9

184

37. Esserman L, Shieh Y, Thompson I. Rethinking screening for breast cancer and prostate cancer. JAMA

2009; 302: 1685-1692.

38. Zahl PH, Maehlen J, Welch HG. The natural history of invasive breast cancers detected by screening

mammography. Arch Intern Med 2008; 168: 2311-2316.

39. Berry DA, Cronin KA, Plevritis SK, et al. Effect of screening and adjuvant therapy on mortality from


40. Cowan WK, Angus B, Gray JC, Lunt LG, al Tamimi SR. A study of interval breast cancer within the NHS

breast screening programme. J Clin Pathol 2000; 53: 140-146.

41. Koivunen D, Zhang X, Blackwell C, Adelstein E, Humphrey L. Interval breast cancers are not biologically

distinct--just more difficult to diagnose. Am J Surg 1994; 168: 538-542.

42. Schroen AA, Wobbes T, van der Sluis RF. Interval carcinomas of the breast: a group with intermediate

outcome. J Surg Oncol 1996; 63: 141-144.

43. Porter PL, El Bastawissi AY, Mandelson MT, et al. Breast tumor characteristics as predictors of

mammographic detection: comparison of interval- and screen-detected cancers. J Natl Cancer Inst 1999;

91: 2020-2028.

44. Black WC, Haggstrom DA, Welch HG. All-cause mortality in randomized trials of cancer screening. J Natl

Cancer Inst 2002; 94: 167-173.



46. Paik S, Shak S, Tang G, et al. A multigene assay to predict recurrence of tamoxifen-treated, node-

negative breast cancer. N Engl J Med 2004; 351: 2817-2826.


cancer. Nature 2002; 415: 530-536.



8/10/2019 Stella Mook T3-108



185

9


Supplementary Figure 1. Breast cancer-specific mortality in 953 women diagnosed with a

screen-detected carcinoma between 1990 and 2000. Kaplan–Meier curves for breast cancer-specific survival and the univariate hazard ratio (HR) with its 95% confidence interval (CI) for

prevalent versus incident carcinomas are shown.

Kaplan-Meier survival analysis, log-rank test, and univariate Cox proportional hazard ratio (HRs)

were calculated to estimate differences in survival among patients with prevalent carcinoma ( i.e.,

detected in the first screening round) and patients with incident carcinomas ( i.e., detected in a

second or subsequent screening round).

0.0

0.2

0.4

0.6

0.8

1.0

B r e a s t c a

n c e r - s p e c i fi c

s u r v i v a l p r o b a b i l i t y

0 5 10 15

Time (years)

510

443 15406 213

472 387 137

Incident carcinoma

Prevalent carcinomaNumbers at

risk

Prevalent versus incident carcinoma: HR = 1.02, 95% CI = 0.69 to 1.49

Log-rank p = 0.928

8/10/2019 Stella Mook T3-108


Chapter 9

186

Supplementary Table 1. Multivariable Cox proportional hazard regression analyses for breast

cancer-specific mortality in patients with breast cancer detected in the first screening round

(i.e., prevalent carcinoma) (n = 510) and patients with breast cancer detected in a second or

subsequent screening round (i.e., incident carcinoma) (n = 443)*.

Breast cancer-specific mortality

Characteristics P HR (95% CI)

Incident versus prevalent carcinoma 0.85 0.95 (0.58 to 1.56)

Period of diagnosis (1990–1996 versus 1997–2000) 0.01 0.87 (0.79 to 0.97)

Age (per year) 0.15 1.03 (0.99 to 1.06)

pT†

pT2 (versus pT1) 0.04 1.58 (1.02 to 2.46)

pT3 (versus pT1) 0.04 3.87 (1.04 to 14.32)

pN‡

pN1 (versus pN0) 0.09 1.71 (0.92 to 3.18)

pN2 (versus pN0) 0.002 2.83 (1.45 to 5.55)

pN3 (versus pN0) <0.001 6.21 (2.85 to 13.56)

Grade

II (versus I) 0.002 3.13 (1.51 to 6.49)

III (versus I) <0.001 5.90 (2.65 to 13.16)

Unknown (versus I) 0.41 1.47 (0.59 to 3.70)

ER status

ER negative (versus ER positive) 0.16 1.51 (0.85 to 2.69)

ER unknown (versus ER positive) 0.10 0.65 (0.40 to 1.08)

Chemotherapy (yes versus no) 0.81 0.92 (0.47 to 1.82)

Hormonal therapy (yes versus no) 0.58 0.86 (0.50 to 1.49)

* All statistics were calculated with the use of the Cox proportional hazard model. All statistical tests were

two-sided.

CI = confidence interval; ER = estrogen receptor; HR = hazard ratio.† pT = pT1 ≤ 2 cm; pT2 = 2–5 cm; pT3 > 5 cm. ‡ pN = pN0 = lymph node-negative; pN1 = 1–3 positive lymph

nodes; pN2 = 4–9 positive lymph nodes; pN3 > 9 positive lymph nodes.

8/10/2019 Stella Mook T3-108



187

9

Supplementary Figure 2. Trends in the treatment of patients with adjuvant systemic therapy

in our study cohort of 2592 breast cancer patients aged 50 to 69 years, diagnosed between 1990

and 2000, stratified by nodal status.

CT = chemotherapy; HT = hormonal therapy.

100%

40%

60%

80%

0%

20%

100%

40%

60%

80%

0%

20%

P e r c e n t a g e r e c e i v i n g a d j u v a n t s y s t e m i c t r e a t m e n t

19901995

19941993

19921991

19961997

19981999

2000 1990 19951994199319921991 1996 1997 1998 1999 2000

Lymph node-negative Lymph node-positive

None

NoneCT

CTHT

HT

CT & HT

CT & HT

Year of diagnosis Year of diagnosis

A B

8/10/2019 Stella Mook T3-108


Chapter 9

188


cancer-specific mortality in patients diagnosed between 1997 and 2001, including 180 patients

who were aged 70–75 years*.

Breast cancer-specific mortality

Characteristics P HR (95% CI)

Method of detection

Screen-detected versus nonscreening-related <0.001 0.49 (0.32 to 0.75)

Interval versus nonscreening-related 0.41 0.85 (0.57 to 1.26)

Age (per year) 0.006 1.04 (1.01 to 1.07)

pT†

pT2 (versus pT1) 0.01 1.62 (1.11 to 2.37)

pT3 (versus pT1) 0.05 2.02 (1.01 to4.04)

pN‡

pN1 (versus pN0) 0.005 1.99 (1.24 to 3.20)

pN2 (versus pN0) 0.009 2.16 (1.21 to 3.86)

pN3 (versus pN0) <0.001 5.90 (3.19 to 10.92)

Grade

II (versus I) 0.02 3.56 (1.28 to 9.93)

III (versus I) 0.002 5.61 (1.93 to 16.32)

Unknown (versus I) 0.20 2.28 (0.64 to 8.07)

ER status

ER negative (versus ER positive) 0.002 2.00 (1.28 to 3.11)

ER unknown (versus ER positive) 0.80 1.11 (0.51 to 2.42)

Chemotherapy (yes versus no) 0.66 1.11 (0.69 to 1.80)

Hormonal therapy (yes versus no) 0.42 1.20 (0.77 to 1.87)

* All statistics were calculated with the use of the Cox proportional hazard model. All statistical tests were

two-sided.

CI = confidence interval; ER = estrogen receptor; HR = hazard ratio.† pT = pT1 ≤ 2 cm; pT2 = 2–5 cm; pT3 > 5 cm. ‡ pN = pN0 = lymph node-negative; pN1 = 1–3 positive lymph

nodes; pN2 = 4–9 positive lymph nodes; pN3 > 9 positive lymph nodes.

8/10/2019 Stella Mook T3-108



189

9

S u p p l e m e n t a r y T a b l e 3 . M u l t i v a r i a

b l e C o x p r o p o r t i o n a l h a z a r d r e g r e s s i o n a n a l y s e s f o r b r e a s t c a n c e r - s p e c i fi c m o r t a l i t y s t r a t i fi e d b y t u m o r s i z e * .

T u m o r s i z e o f ≤ 1 0 m m

T u m o r s i z e o f 1 1 – 2 0 m m

T u m o r

s i z e o f 2 1 – 3 0 m m

T u m o r s i z e

3 1 – 5 0 m m

C h

a r a c t e r i s t i c s

P

H R ( 9 5 % C

I )

P

H R ( 9 5 % C

I )

P

H R ( 9 5 % C

I )

P

H R ( 9 5 % C

I )

M e t h o d o f d e t e c t i o n : S c r e e n - d e t e c t e d

v e r s u s n o n s c r e e n i n g - r e l a t e d

0 . 0 4

0 . 3 5 ( 0 . 1 3 t o 0 . 9 6 )

0 . 1 0

0 . 7 4 ( 0 . 5 2 t o 1 . 0 5 )

0 . 0 0 2

0 . 5 3 ( 0 . 3 6 t o 0 . 7 9 )

0 . 0 8

0 . 5 7 ( 0 . 3 1 t o 1 . 0 7 )

A g

e ( p e r y e a r )

0 . 9 2

1 . 0 0 ( 0 . 9 3 t o 1 . 0 9 )

0 . 8 8

1 . 0 0 ( 0 . 9 7 t o 1 . 0 4 )

0 . 7 9

1 . 0 0 ( 0 . 9 8 t o 1 . 0 3 )

0 . 1 4

1 . 0 3 ( 0 . 9 9 t o 1 . 0 8 )

p N

† p N 1 ( v e r s u s p N 0 )

0 . 6 8

1 . 4 9 ( 0 . 2 3 t o 9 . 6 7 )

0 . 9 7

1 . 0 1 ( 0 . 5 8 t o 1 . 7 7 )

0 . 0 6

1 . 5 7 ( 0 . 9 8 t o 2 . 5 1 )

0 . 8 7

0 . 9 4 ( 0 . 4 2 t o 2 . 0 6 )

p N 2 ( v e r s u s p N 0 )

0 . 2 1

4 . 4 3 ( 0 . 4 3 t o 4 6 . 1 6 )

0 . 0 0 7

2 . 3 9 ( 1 . 2 7 t o 4 . 4 8 ) < 0 . 0 0 1

3 . 0 1 ( 1 . 7 7 t o 5 . 1 4 )

0 . 1 3

1 . 9 2 ( 0 . 8 3 t o 4 . 4 8 )

p N 3 ( v e r s u s p N 0 )

0 . 0 2

3 7 . 9 2 ( 1 . 7 5 t o 8 2 2 . 8 5 )

0 . 1 0

2 . 6 4 ( 0 . 8 4 t o 8 . 3 4 ) < 0 . 0 0 1

8 . 6 5 ( 4 . 6 0 t o 1 6 . 2 5 )

0 . 0 0 1

4 . 3 8 ( 1 . 8 5 t o 1 0 . 3 8 )

G r

a d e

I I ( v e r s u s I )

0 . 5 1

1 . 5 1 ( 0 . 4 5 t o 5 . 0 9 )

0 . 0 0 9

2 . 3 3 ( 1 . 2 4 t o 4 . 3 9 )

0 . 2 0

1 . 7 5 ( 0 . 7 5 t o 4 . 0 7 )

0 . 6 9

1 . 5 2 ( 0 . 2 0 t o 1 1 . 6 8 )

I I I ( v e r s u s I )

0 . 6 9

1 . 5 4 ( 0 . 1 9 t o 1 2 . 2 6 ) <

0 . 0 0 1

5 . 9 1 ( 3 . 0 2 t o 1 1 . 5 7 )

0 . 0 1

3 . 1 3 ( 1 . 3 2 t o 7 . 4 1 )

0 . 3 7

2 . 5 3 ( 0 . 3 3 t o 1 9 . 1 4 )

U n k n o w n ( v e r s u s I )

0 . 3 8

1 . 8 4 ( 0 . 4 8 t o 7 . 1 1 )

0 . 0 0 2

2 . 9 7 ( 1 . 4 7 t o 6 . 0 1 )

0 . 0 5

2 . 4 4 ( 1 . 0 1 t o 5 . 8 9 )

0 . 4 1

2 . 3 5 ( 0 . 3 1 t o 1 8 . 0 3 )

E R

s t a t u s

E R n e g a t i v e ( v e r s u s E R p o s i t i v e )

0 . 6 6

1 . 4 6 ( 0 . 2 7 t o 7 . 8 0 )

0 . 4 8

1 . 2 1 ( 0 . 7 2 t o 2 . 0 4 )

0 . 3 8

1 . 2 5 ( 0 . 7 6 t o 2 . 0 6 )

0 . 2 5

1 . 4 9 ( 0 . 7 5 t o 2 . 9 5 )

E R u n k n o w n ( v e r s u s E R p o s i t i v e )

0 . 4 9

0 . 6 8 ( 0 . 2 3 t o 2 . 0 5 )

0 . 1 3

1 . 3 5 ( 0 . 9 2 t o 1 . 9 7 )

0 . 8 1

1 . 0 4 ( 0 . 7 3 t o 1 . 4 9 )

0 . 2 1

1 . 4 1 ( 0 . 8 3 t o 2 . 4 1 )

C h

e m o t h e r a p y ( y e s v e r s u s n o )

0 . 6 7

1 . 5 3 ( 0 . 2 2 t o 1 0 . 8 4 )

0 . 8 1

0 . 9 2 ( 0 . 4 8 t o 1 . 7 7 )

0 . 0 3

0 . 5 5 ( 0 . 3 2 t o 0 . 9 4 )

0 . 1 6

0 . 5 6 ( 0 . 2 6 t o 1 . 2 5 )

H o

r m o n a l t h e r a p y ( y e s v e r s u s n o )

0 . 6 4

0 . 6 7 ( 0 . 1 3 t o 3 . 4 9 )

0 . 9 3

0 . 9 8 ( 0 . 5 9 t o 1 )

0 . 3 4

0 . 8 0 ( 0 . 5 1 t o 1 . 2 6 )

0 . 0 7

0 . 5 5 ( 0 . 2 9 t o 1 . 0 5 )

* A l l s t a t i s t i c s w e r e c a l c u l a t e d u s i n g t h e

C o x p r o p o r t i o n a l h a z a r d m o d e l a n d

w e r e t w o - s i d e d .

C I =

c o n fi d e n c e i n t e r v a l ; E R = e s t r o g e n r

e c e p t o r ; H R = h a z a r d r a t i o .

† p N : p N 0 = l y m p h n o d e - n e g a t i v e ; p N 1

= 1 – 3 p o s i t i v e l y m p h n o d e s ; p N 2 = 4

– 9 p o s i t i v e l y m p h n o d e s ; p N 3 > 9 p o

s i t i v e l y m p h n o d e s .

8/10/2019 Stella Mook T3-108


Chapter 9

190

0.0

0.2

0.4

0.6

0.8

1.0


s u r v i v a l p r o b a b

i l i t y

0 5 10 15

Time (years)

238

116 36109 72

225 163 46


Screen-detectedNumbers at

risk


Log-rank p = 0.004

0.0

0.2

0.4

0.6

0.8

1.0


s u r v i v a l p

r o b a b i l i t y

0 5 10 15

Time (years)

476

484 139432 320

450 305 74



risk


Log-rank p = 0.009

A

B

8/10/2019 Stella Mook T3-108



191

9

0.0

0.2

0.4

0.6

0.8

1.0


s u r v i v a l p r o b a b

i l i t y

0 5 10 15

Time (years)

164

384 89290 204

144 90 19



risk


Log-rank p = 0.007

0.0

0.2

0.4

0.6

0.8

1.0


s u r v i v a l

p r o b a b i l i t y

0 5 10 15

Time (years)

55

172 35128 82

41 31 10



risk


Log-rank p = 0.07

C

D

8/10/2019 Stella Mook T3-108


Chapter 9

192

Supplementary Figure 3. Breast cancer-specific survival by method of detection. Kaplan–Meier

curves for breast cancer-specific survival and univariate hazard ratios (HRs) for breast cancer-

specific mortality.

A) Patients with tumors of 10 mm or less.

B) Patients with tumors of 11–20 mm.

C) Patients with tumors of 21–30 mm.

D) Patients with tumors of 31–50 mm.

0.0

0.2

0.4

0.6

0.8

1.0


s u r v i v a l p r o b a

b i l i t y

0 5 10 15

Time (years)

635

567 164503 362

608 430 118



risk


Log-rank p < 0.001

0.0

0.2

0.4

0.6

0.8

1.0


s u r v i v a l p r o b a b i l i t y

0 5 10 15

Time (years)

323

650 147501 345

275 172 34


Screen-detectedNumbers atrisk


Log-rank p = 0.001

E

E

8/10/2019 Stella Mook T3-108



193

9

E) Patients with lymph node-negative breast cancer.

F) Patients with lymph node-positive breast cancer.

NRS = Non-screening-related carcinomas, symptomatic cancer in patients who had not been

screened or were screened more than 24 months before detection of breast cancer. SD = Screen-

detected carcinomas.

The number of patients at risk is shown below each graph. For 32 patients, no exact tumor size

was available (only pT stage), and so they were excluded from the analyses shown in Figure

3, A–D. Kaplan-Meier survival analyses, log-rank tests, and univariate Cox proportional hazard

ratios (HRs) were calculated to estimate differences in survival among patients with screen-

detected carcinoma and patients with non-screening-related carcinoma.


cancer-specific mortality stratified by lymph node status*.

Lymph node-negative Lymph node-positive


Method of detection: screen-detected

versus non-screening-related<0.001 0.51 (0.36 to 0.73) 0.12 0.79 (0.59 to 1.06)

Age (per year) 0.26 1.02 (0.99 to 1.05) 0.77 1.00 (0.98 to 1.03)

pT†

pT2 (versus pT1) <0.001 1.87 (1.32 to 2.65) <0.001 2.00 (1.51 to 2.67)

pT3 (versus pT1) 0.08 3.62 (0.88 to 14.95) 0.009 2.05 (1.20 to 3.49)

Grade

Grade II (versus grade I) 0.002 2.45 (1.38 to 4.35) 0.05 1.98 (0.99 to 3.94)

Grade III (versus grade I) <0.001 4.08 (2.11 to 7.87) <0.001 4.62 (2.31 to 9.25)

Grade unknown (versus grade I) <0.001 3.35 (1.82 to 6.16) 0.001 3.19 (1.57 to 6.49)

ER status





* All statistics were calculated using the Cox proportional hazard model and were two-sided.

CI = confidence interval; ER = estrogen receptor; HR = hazard ratio.

† pT: pT1 ≤ 2 cm; pT2 = 2–5 cm; pT3 > 5 cm.

8/10/2019 Stella Mook T3-108


Chapter 9

194

Supplementary Table 5. Adjuvant! 10-year observed and predicted breast cancer-specific

survival for patients younger than 50 years by period of diagnosis*.

10-year breast cancer-specific survival

Period of diagnosisNo. patients

(%)

Adjuvant!

predicted, %

Observed,

% (95% CI)

Predicted –

observed P

1990–1996 1381 (70.0) 77.3 77.5 (75.3 to 79.7) 0.2 0.86

1997–2000 592 (30.0) 79.3 82.3 (79.2 to 85.4) -3.0 0.06

* All statistics were calculated by one sample t tests and were two-sided.

CI = confidence interval.

8/10/2019 Stella Mook T3-108


Chapter 10

General discussion

and future prospects

8/10/2019 Stella Mook T3-108


Chapter 10

198

8/10/2019 Stella Mook T3-108


Discussion & future prospects

199

10

Over the last decades, breast cancer management has changed dramatically. Primary local

treatment has evolved from the extensive Halsted mastectomy to less invasive breast

conservative surgery followed by radiotherapy, which is currently the standard treatment

for approximately 2/3 of the breast cancer patients.1 In addition, staging of the axilla by

the sentinel node procedure is now widely practiced and the standard of care for patients

with clinically negative nodal status.2,3 Patients with a negative sentinel node can be spared

complete axillary lymph node dissection and its associated side-effects.4,5 The final analysis

of the AMAROS trial show will show whether in case of sentinel lymph node involvement

axillary lymph node dissection can be abandoned and radiotherapy of the axilla provides a

safe and equivalent alternative with less morbidity. In the meantime, even more provocative

results of the ASCO Z0011 trial by Giuliano and colleagues were published.6 Patients with

invasive breast carcinomas ≤ 5 cm treated with breast conserving therapy and adjuvant

systemic therapy (in 96% of the patients) who had 1-3 positive sentinel nodes were

randomized between axillary lymph node dissection and no further axillary treatment.

Remarkably, no difference in local and regional recurrence was observed, suggesting that

for a selected group of patients with macrometastases in the sentinel lymph node further

axillary treatment can be safely omitted.

All the above mentioned changes in breast cancer management touch upon the ultimate

goal to optimize and tailor treatment by reducing side-effects without jeopardizing

survival and were guided by a conceptual change in the theory on breast cancer

etiology and progression.7 Traditionally, breast cancer management was based on the

Halsted theory. Halsted stated that breast cancer is a localized disease, spreading in anorderly and consecutive manner from local tissue, to regional lymph nodes and then to

distant sites.8 This theory justified the use of extensive loco-regional surgery (i.e. radical

mastectomy) to remove all local and regional disease, thereby improving survival. In 1968,

Fisher introduced an alternative hypothesis, which has led to a number of changes in

breast cancer management.9,10 Fisher postulated that breast cancer is primarily a systemic

disease, with the presence of circulating cancer cells already at an early stage, thus requires

treatment of the entire patient (systemic treatment). As a consequence, according to the

Fisher theory, local recurrence should be considered an indicator of metastatic disease, and

the development of distant metastases is a result of both tumor and patient characteristicsand the interaction between them. This hypothesis and the knowledge of the incurable

nature of metastatic breast cancer with the emanating fear of undertreatment have caused

a substantial increase in the use of adjuvant systemic therapy.11,12,13

8/10/2019 Stella Mook T3-108


Chapter 10

200

Selection of patients for adjuvant systemic therapy; current practice

Currently, guidelines recommend adjuvant systemic therapy (AST) for all patients with

lymph node positive disease and for over 80% of breast cancer patients with lymph node

negative disease. 13,14 These guidelines base their recommendations for the use of AST

on clinicopathological prognostic characteristics, such as age, tumor size, tumor grade,

lymph node status and estrogen receptor status. 14-17 These clinicopathological factors are

used to identify subgroups of patients with a poor prognosis who are expected to benefit

more from adjuvant systemic therapy. However, many patients in these subgroups are

overtreated since they do not have micrometastases at diagnosis and thus are likely to

remain free from distant metastases.18 This overtreatment is particularly poignant in lymph

node negative breast cancer where over 80% is treated with AST, whereas approximately

70% of the patients are free of distant metastases at 10 years and likely to be cured with

locoregional treatment alone.18 Conversely, patient selection based on clinicopathological

criteria can also cause undertreatment. According to some current guidelines, AST is often

not recommended for patients with small tumors of less than 1 cm; however, a proportion of

these small tumors may have spread before detection and should consequently be treated

with AST. It is clear that patients who suffer from an apparently similar tumor with regard

to pathological characteristics can have remarkably different disease outcomes. Therefore,

patient selection for adjuvant systemic therapy by traditional prognostic factors has its

limitations and will lead to both over- and undertreatment. Breast cancer treatment with

as little as possible side effects and optimal survival requires a patient-tailored approachbased on appropriate patient selection. This shift from a ‘one size fits all’ approach to a more

personalized approach uncovers the need for better prognostic markers or tools in breast

cancer and is the rational for this thesis.

Multigene prognosis signatures

The introduction of high-throughput microarray technology facilitated the development of

gene-expression profiles or signatures that can measure the expression of multiple genesin a single test.19-21 The 70-gene prognosis signature (MammaPrint™) is one of the new

prognostic markers that can accurately discriminate between breast cancer patients at low

and high risk of developing distant metastases, based on the expression level of 70 selected

genes.22 Validation studies confirmed that the signature can accurately predict disease

outcome in premenopausal, lymph node negative breast cancer patients.23,24 In addition,

the prognostic value of the signature was independent of known clinicopathological

prognostic factors.23-27 These studies, as well as studies described in this thesis have led to

the inclusion of the 70-gene signature in current guidelines.14,15 Soon after the development

of the 70-gene signature, the Recurrence Score (OncotypeDX™) was developed and

8/10/2019 Stella Mook T3-108



201

10

validated.28-31 This test classifies tumors based on the expression of 16 genes into a low

Recurrence Score (RS), an intermediate RS or a high RS. The RS has been validated in several

patient series and has been incorporated into the St. Gallen recommendations and NCCN

guidelines.15,32 The recently conducted Trial Assigning IndividuaLized Options for Treatment

(Rx) (TAILORx), will address whether patients who are assigned to the intermediate RS have

benefit from adjuvant chemotherapy in addition to endocrine therapy. Patient inclusion for

the TAILORx was finished end 2010.33

Prognostic value of the 70-gene signature in breast cancer subgroups

For the studies presented in this thesis independent retrospective patient series were

selected to assess the prognostic value of the 70-gene signature in postmenopausal breast

cancer patients and in patients with 1-3 positive lymph nodes. For the validation of the

signature in small tumors patients were selected from a pooled database of previously

published studies and studies described in chapter 4 and 5 of this thesis.23-25,34,35

The majority of breast cancer patients are postmenopausal women.36 Although former

treatment guidelines recommended adjuvant endocrine therapy only, there is a strong

increase in the use of chemotherapy for postmenopausal patients.37-39 This more extensive

use of adjuvant treatment is intuitively in contrast to the more favorable tumor characteristics

and good disease outcome observed in many postmenopausal patients.37-39 Besides, the

overview data show that the benefit of chemotherapy diminishes as age increases.18 Inaddition, the benefit of chemotherapy in postmenopausal patients seems to occur mainly

in the first five years.18 In chapter 4 we show that even though the 70-gene signature was

developed in premenopausal patients, it has independent prognostic value and utility

in postmenopausal women. The signature identified poor prognosis patients who are at

high risk of developing distant metastases early in the disease course and are therefore

likely to benefit more from chemotherapy. Postmenopausal patients who were classified

as low risk were likely to remain free of early disease recurrence; however, a proportion of

those low risk patients did develop late metastases. This low risk subgroup consisted of

estrogen receptor positive tumors and only 1 patient received endocrine therapy. Sincethe beneficial effect of chemotherapy in postmenopausal patients is limited to the first five

years, patients classified as low risk who are at risk of late breast cancer related events are

more likely to benefit from endocrine therapy.

Historically, lymph node status is considered to be the most powerful prognostic factor

in breast cancer, with the presence and number of involved nodes being associated with

poor disease outcome.40-42 As a consequence, patients with lymph node positive disease

are currently offered chemotherapy, regardless of other clinicopathological characteristics.

However, a subset of patients (approximately 25-30%) will remain free of distant metastases

for at least 10 years without AST and are presumably cured by locoregional treatment

8/10/2019 Stella Mook T3-108


Chapter 10

202

alone.18,43 So far, no prognostic factor that can identify this subset of good prognosis lymph

node positive patients has been identified. New prognostic markers or signatures will only

have medical utility when they can identify a clinically relevant subset of patients with

potential treatment consequences. For example, the identification of a subset of patients

who have a low risk of recurrence will only be relevant when the risk is sufficiently small

to justify withholding chemotherapy. In chapter 5 we have demonstrated that the 70-

gene signature was able to identify a subset of patients with 1-3 positive lymph nodes

who are at sufficiently low risk to consider withholding chemotherapy. The appreciation

of the existence of low risk lymph node positive patients who might not benefit from

chemotherapy was also confirmed for the Recurrence Score by Albain and colleagues. 28

Based on these results the currently conducted randomized MINDACT trial has extended

its eligibility criteria to patients with up to 3 positive nodes. Results of this prospective

trial might end the long existing and persistent idea that all patients with lymph node

involvement will be confronted with metastatic disease and should receive adjuvant

chemotherapy.

In addition to lymph node status, tumor size is a traditional prognostic factor that is taken

into account when selecting patients for adjuvant systemic therapy. 40-42 Current guidelines

are inconsistent in the systemic treatment recommendations for patients with small breast

tumors.14,15,17,32 In the study described in chapter 6 we show that the 70-gene signature can

select patients with pT1c tumors (pT1c: 11-20cm) who are at low risk of distant recurrence

and therefore could be safely spared chemotherapy. Moreover, the 70-gene signature

could identify patients with small breast tumors (pT1ab ≤10mm) who do have a substantialrisk of developing distant metastases despite the small tumor size. The above mentioned

results support the Fisher hypothesis that metastatic capacity is an early inheritance, and

that lymph node involvement is an indicator and not instigator of distant disease.9,44 In

addition, our studies suggest that the metastatic capacity depends (at least partially) on

the genetic makeup of a tumor and thus can be identified by the measurement of tumor

gene expression.

Retrospective validation

Retrospective studies are of indispensable value to identify potential biomarkers that

deserve further evaluation; however, there are some drawbacks that one needs to be aware

of.

In retrospective series that were not part of a randomized trial, patients have received

treatment according to the guidelines present at that time. Therefore, clinicopathological

markers will have influenced treatment decision, which will complicate extrapolation of

the results. However, recommendations for the use of adjuvant systemic therapy in the

Netherlands have been conservative for a long time, as described in the introduction of

8/10/2019 Stella Mook T3-108



203

10

this thesis.14,45,46 As a result, retrospective series of Dutch breast cancer patients who were

diagnosed before 2000 will consist of a relatively large proportion of untreated patients.

For instance, the first validation study of the 70-gene signature included a true consecutive

series of lymph node negative patients of whom only 10% received AST.24 In addition, the

postmenopausal patient series described in chapter 4 was a true consecutive series and

no patients were excluded because of chemotherapy treatment. One might argue that the

prognostic value of a marker cannot be assessed in a (partially) treated population. However,

selecting patients who did not receive adjuvant systemic therapy will introduce selection

bias. Markers that can define the residual risk of recurrence when a patient will be treated

with endocrine therapy alone are of utmost importance in determining the potential value

and necessity of additional chemotherapy. Including patients who have only received

adjuvant endocrine therapy seems to be a reasonable compromise. The evaluation of the

70-gene signature in postmenopausal patients (chapter 4) was performed in a consecutive

series of patients who did not receive AST or were only treated with endocrine therapy.

The Recurrence Score (OncotypeDX™) provides an estimate of the additive effect of adjuvant

chemotherapy in combination with 5 years of endocrine treatment with Tamoxifen.28-31

However, for patients who are assigned to the intermediate RS, the additional benefit of

chemotherapy remains uncertain while having a considerable risk of recurrence and result

of the TAILORx needs to be awaited.33

To avoid selection bias as mentioned above, we included patients with lymph node positive

disease regardless of adjuvant systemic therapy for the validation study described in

chapter 5. As a consequence, 56% of the patients received adjuvant chemotherapy. Thesepatients had more often estrogen receptor negative and poorly differentiated tumors, which

in general are believed to have more benefit from chemotherapy.18 In addition, patients

treated with chemotherapy were more often classified as poor prognosis by the 70-gene

signature. In the study presented in chapter 7, we analyzed 541 patients who had received

adjuvant systemic therapy and who were classified by the 70-gene signature. Patients with

a 70-gene poor prognosis signature treated with chemotherapy followed by endocrine

therapy had a significantly better distant disease-free survival compared with poor

prognosis patients who were treated with endocrine therapy alone. Conversely, patients

with a low risk 70-gene signature who were treated with chemotherapy and endocrinetherapy had similar disease outcomes as low risk patients treated with endocrine therapy

alone. This study provides evidence that patients with a high risk 70-gene signature are

more likely to benefit from adjuvant chemotherapy, whereas a low risk 70-gene signature

indicates limited benefit from adjuvant chemotherapy, in addition to a low recurrence

risk to begin with. Additionally, in a recently published study the predictive value of the

70-gene signature was assessed in patients treated with neoadjuvant chemotherapy.47

Although patients included in this study were considered as clinically high risk patients

(95% of the tumors were > 2 cm and 72% had lymph node positive disease), the 70-gene

signature identified 14% good prognosis tumors. Among patients with a good prognosis

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108



205

10

bank for future research, including proteomics, temporarily preservation of tissue in

preservation fluid might not be suitable and fresh frozen tissue is probably a more reliable

source for future research. Therefore we conducted a pilot study preceding the MINDACT

trial, in which we have tested and optimized the comprehensive logistics to obtain good-

quality fresh frozen tumor tissue (chapter 3). The feasibility of performing gene expression

profiling in daily practice is further reflected by the fast accrual of the MINDACT trial.50

8/10/2019 Stella Mook T3-108


Chapter 10

206

Clinicopathological risk assessment and individualized prognostication

The prognostic tool Adjuvant!

In the MINDACT trial, the clinicopathological risk will be assessed by Adjuvant! to provide an

internationally used and standardized method for predicting outcome. Adjuvant! predicts

10 year disease outcome with and without the use of adjuvant systemic therapy based

on age, co-morbidity, tumor size, tumor grade, estrogen receptor status and lymph node

status.16 The model was developed in an American breast cancer population and previously

validated in Canadian breast cancer patients.16,51 In chapter 8 a Dutch validation study of

the model is described. The results show that in general Adjuvant! can be used for Dutch

breast cancer patients; however, predictions in patients under 40 years should be carefully

judged. In this validation study we assessed both the accuracy of the model to predict

disease outcome in subgroups of breast cancer patients (i.e. the goodness of fit or calibration)

and the model’s ability to distinguish individuals who will experience different outcomes

(discriminatory accuracy). In the era of personalized treatment, the discriminatory accuracy

of a prognostic tool or marker is of paramount importance.52-54 A model can predict disease

outcome very accurately in the whole group or in clinically relevant subgroups, without

identifying the correct patients who are at high risk of recurrence. For instance, when 30%

of the patients will suffer from recurrence and the model indeed predicts a recurrence

of 30% in this group, its calibration is excellent. However, the model could still identify

the wrong patients as poor prognosis as is depicted in figure 1, hence resulting in a poordiscriminatory accuracy and limited value for the individual patient. In addition to the

good calibration of the Adjuvant! tool in Dutch patients (differences between predicted

and observed outcomes were within 2% for most clinically relevant subgroups), the model

showed moderate discriminatory accuracy. This observation was expected since we know

that patients with identical clinicopathological characteristics can have different outcomes.

Consequently, the prognostic information that is captured by these characteristics can

only explain part of the differences in outcome. Considering the results described in this

thesis, the 70-gene signature as a measurement of tumor biology will be the most obvious

marker to incorporate in the Adjuvant! model. The signature will probably explain part ofthe residual variation and increase the discriminatory accuracy of the model and therefore

could provide the opportunity to improve personalized treatment. In addition, other

(new) prognostic factors could be added to the model and potentially improve outcome

prediction, such as Her2 status or Ki67.55,56 In the MINDACT trial, an adapted version of

Adjuvant! including Her2 is used.

8/10/2019 Stella Mook T3-108



207

10

Figure 1. Calibration and discriminatory accuracy of a hypothetical prognostic marker.

The predicted distant metastases-free survival of 70% is in good agreement with the actual

observed DMFS of 70%, reflecting the good calibration of the marker. However, the 30% who

were identified by the prognostic marker as patients who are at risk of developing distantmetastases are completely different from the patients in whom distant metastases were

observed, depicting the poor discriminatory accuracy.

The white figures represent patients who were predicted to remain free of distant metastases

by the marker and/or in whom no distant metastases were observed. The gray figures represent

patients who are predicted to develop metastases according to a prognostic marker. The black

and white striped figures are patients in whom distant metastases were observed.

Method of detection

A new marker that is currently ignored when selecting patients for AST is method of

detection, i.e. whether a tumor is detected by screening mammography or by the cause

of symptoms. Adding method of detection to models such as Adjuvant! could improve

the individual prediction of disease outcome. In chapter 9 we show that Adjuvant!

underestimated disease outcome in patients with screen-detected tumors. Furthermore,

results of our study show that, even after adjusting for factors associated with tumor

advancement and aggressiveness, patients with screen-detected tumors have a better

survival compared with patients with nonscreening-related tumors. This suggests that

Group level: Good calibration

Individual patient level: poor discriminatory accuracy

Predicted outcome Observed outcome

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108



209

10

value seems limited since the vast majority of these patients will be classified as high risk

according to the signature.

Future prospective

The introduction of microarray technology will increasingly impact the management of

breast cancer. It will increase our understanding about breast cancer biology and further

elucidate its heterogeneity. The studies described in this thesis show that the 70-gene

signature, which was developed in a well-defined subset of breast cancer patients, has

prognostic value in several other breast cancer subgroups, suggesting that one prognostic

test fits all patients. However, there is still room for improvement and with the currently

increasing knowledge about tumor biology it is likely that new markers will be developed

in more specific subgroups, such as in estrogen receptor negative or triple negative

tumors. In the future we might be able to perform one assay in a certain subgroup in

order to determine a patient’s prognosis but also the likelihood of response to different

therapies and the presence of drug targets. These developments will also influence clinical

trial design, in which patients will be stratified by both prognostic and predictive markers,

thereby identifying targeted therapy that will be highly effective in a (small) subgroup of

patients who indeed need additional systemic treatment.

8/10/2019 Stella Mook T3-108


Chapter 10

210

Reference List

1. Veronesi U, Saccozzi R, Del VM, et al. Comparing radical mastectomy with quadrantectomy, axillary

dissection, and radiotherapy in patients with small cancers of the breast. N Engl J Med 1981; 305: 6-11.

2. Veronesi U, Viale G, Paganelli G, et al. Sentinel lymph node biopsy in breast cancer: ten-year results of a

randomized controlled study. Ann Surg 2010; 251: 595-600.

3. Veronesi U, Paganelli G, Viale G, et al. A randomized comparison of sentinel-node biopsy with routine

axillary dissection in breast cancer. N Engl J Med 2003; 349: 546-553.

4. Bergkvist L, de Boniface J, Jonsson PE, et al . Axillary recurrence rate after negative sentinel node biopsy

in breast cancer: three-year follow-up of the Swedish Multicenter Cohort Study. Ann Surg 2008; 247: 150-

156.

5. Giuliano AE, Haigh PI, Brennan MB, et al. Prospective observational study of sentinel lymphadenectomy

without further axillary dissection in patients with sentinel node-negative breast cancer. J Clin Oncol 2000;

18: 2553-2559.

6. Giuliano AE, McCall L, Beitsch P, et al. Locoregional recurrence after sentinel lymph node dissection with

or without axillary dissection in patients with sentinel lymph node metastases: the American College

of Surgeons Oncology Group Z0011 randomized trial. Ann Surg 2010; 252: 426-432.

7. della Rovere GQ and Benson JR. Ipsilateral local recurrence of breast cancer: determinant or indicator

of poor prognosis? Lancet Oncol 2002; 3: 183-187.

8. Halsted WS. I. The results of operations for the cure of cancer of the breast performed at the Johns

Hopkins Hospital from June, 1889, to January, 1894. Ann Surg 1894; 20: 497-555.

9. Fisher B. The evolution of paradigms for the management of breast cancer: a personal perspective.

Cancer Res 1992; 52: 2371-2383.

10. Fisher B and Anderson SJ. The breast cancer alternative hypothesis: is there evidence to justify replacing

it? J Clin Oncol 2010; 28: 366-374.

11. Van Steenbergen LN, Van de Poll-Franse LV, Wouters MW, et al . Variation in management of early breast

cancer in the Netherlands, 2003-2006. Eur J Surg Oncol 2010; 36 Suppl 1: S36-S43.

12. Sukel MP, van de Poll-Franse LV, Nieuwenhuijzen GA, et al . Substantial increase in the use of adjuvant

systemic treatment for early stage breast cancer reflects changes in guidelines in the period 1990-2006

in the southeastern Netherlands. Eur J Cancer 2008; 44: 1846-1854.

13. Harlan LC, Clegg LX, Abrams J, Stevens JL, Ballard-Barbash R. Community-based use of chemotherapyand hormonal therapy for early-stage breast cancer: 1987-2000. J Clin Oncol 2006; 24: 872-877.

14. CBO Richtlijn Mammacarcinoom 2008. http://www.cbo.nl/product/richtlijnen/folder20021023121843/rl_mamma_08.pdf.


Expert Consensus on the primary therapy of early breast cancer 2009. Ann Oncol 2009; 20:1319–1329.



17. Todd JH, Dowle C, Williams MR, et al. Confirmation of a prognostic index in primary breast cancer. Br J

Cancer 1987; 56: 489-492.

8/10/2019 Stella Mook T3-108



211

10



2005; 365: 1687-1717.

19. Brown PO and Botstein D. Exploring the new world of the genome with DNA microarrays. Nat Genet 1999;

21: 33-37.

20. DeRisi J, Penland L, Brown PO, et al. Use of a cDNA microarray to analyse gene expression patterns in

human cancer. Nat Genet 1996; 14: 457-460.

21. Schena M, Shalon D, Davis RW, Brown PO. Quantitative monitoring of gene expression patterns with a

complementary DNA microarray. Science 1995; 270: 467-470.


cancer. Nature 2002; 415: 530-536.







26. Ishitobi M, Goranova TE, Komoike Y, et al . Clinical utility of the 70-gene MammaPrint profile in a Japanese

population. Jpn J Clin Oncol 2010; 40: 508-512.

27. Wittner BS, Sgroi DC, Ryan PD, et al. Analysis of the MammaPrint breast cancer assay in a predominantly


28. Albain KS, Barlow WE, Shak S, et al . Prognostic and predictive value of the 21-gene recurrence score

assay in postmenopausal women with node-positive, oestrogen-receptor-positive breast cancer on

chemotherapy: a retrospective analysis of a randomised trial. Lancet Oncol 2010; 11: 55-65.

29. Habel LA, Shak S, Jacobs MK, et al. A population-based study of tumor gene expression and risk of breast

cancer death among lymph node-negative patients. Breast Cancer Res 2006; 8: R25.

30. Paik S, Shak S, Tang G, et al. A multigene assay to predict recurrence of tamoxifen-treated, node-negative


31. Paik S, Tang G, Shak S, et al. Gene expression and benefit of chemotherapy in women with node-

negative, estrogen receptor-positive breast cancer. J Clin Oncol 2006; 24: 3726-3734.

32. NCCN Clinical Practice Guidelines in Oncology. Breast Cancer V.1.2009. www.nccn.org 2009.33. National Cancer Institute The TAILORx Breast Cancer Trial Available at: http://www.cancer.gov/

clinicaltrials/digestpage/TAILORx.



Lancet Oncol 2007; 8: 1079-1087.

35. Kok M, Koornstra RH, Mook S, et al. Additional value of the 70-gene signature and levels of ER and PR for

the prediction of outcome in tamoxifen-treated ER-positive breast cancer. Submitted.

36. Ries LAG, Melbert D, Krapcho M, et al. SEER Cancer Statistics Review, 1975-2005 , National Cancer Institute. Bethesda,

MD. http://seer.cancer.gov/csr/1975_2005/, based on November 2007 SEER data submission, posted to

the SEER web site.

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108



213

10



56. Soerjomataram I, Louwman MW, Ribot JG, Roukema JA, Coebergh JW. An overview of prognostic

factors for long-term survivors of breast cancer. Breast Cancer Res Treat 2008; 107: 309-330.

57. Knauer M, Cardoso F, Wesseling J, et al. Identification of a low-risk subgroup of HER-2-positive breast

cancer by the 70-gene prognosis signature. Br J Cancer 2010; 103: 1788-1793.

8/10/2019 Stella Mook T3-108


Chapter 11

Summary

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108


Chapter 11

218

clinicopathological criteria (adjusted hazard ratio (HR) for distant metastases as a first event

was 4.6; 95% confidence interval (CI) 2.3–9.2; p < 0.001). These results were confirmed by

a second and independent validation study in 302 lymph node negative patients who did

not receive AST and were diagnosed in 5 European hospitals. Furthermore, the so-called

RASTER (MicroarRAy PrognoSTics in Breast CanceR) study, showed the feasibility of using

the signature for adjuvant treatment decision-making in 16 community based hospitals in

the Netherlands. In a European pilot study, which is also described in detail in chapter 3,

the logistics for the prospective MINDACT (Microarray In Node-negative and 1-3 positive

lymph node disease may Avoid ChemoTherapy) study were tested and optimized. This

study showed that it is feasible to collect good quality fresh frozen tissue in different

European hospitals and that frozen samples can be shipped to a central microarray facility

on a real-time basis. The success rate of the 70-gene signature was 77% (46/60) when

a tumor sample could be obtained. The main reason for exclusion from profiling was a

non-representative sample of the tumor; 18% (11/60) of the samples contained < 50%

tumor cells. Based on these results and the experience gained in this pilot study standard

operating procedures, which are currently used in the MINDACT trial, were developed. The

prospective MINDACT trial is discussed in more detail in the appendix of this thesis. This

international randomized trial will evaluate whether patients who are considered high

risk according to the currently available prognostic tool Adjuvant! but classified as low

risk by the 70-gene signature can be spared chemotherapy without jeopardizing disease

outcome. Recruitment of the trial is anticipated to be completed mid 2011.

Up to now, the 70-gene signature has been developed and validated in a selected group

of patients: predominantly premenopausal patients with lymph node negative disease.

In order to assess the potential improvement of prognostication by using the 70-gene

signature thereby broaden its application, we assessed the prognostic value of the 70-gene

signature in several clinically relevant breast cancer subgroups. In Chapter 4 we report on

the prognostic value and clinical utility of the signature in 148 postmenopausal patients

who were aged between 55 and 70 years and diagnosed with lymph node negative breast

cancer. Patients classified as good prognosis by the signature had a 5-year breast cancer-

specific survival (BCSS) of 99% (Standard error (SE) 1%), compared with 80% (SE 3%) inpatients with a poor prognosis signature respectively ( p = 0.036). Furthermore, the 70-

gene prognosis-signature was a significant and independent predictor of BCSS, especially

during the first 5 years of follow-up with an adjusted HR of 14.4 (95% CI 1.7-122.2; p = 0.01).

The benefit of chemotherapy in postmenopausal patients seems to be most pronounced

in the first 5 years after diagnosis, therefore results of this study indicated a more accurate

allocation of AST using the signature. In Chapter 5 we describe the validation of the 70-gene

signature in an independent retrospective series of breast cancer patients with 1-3 positive

lymph nodes. The aim of this study was to identify patients with 1-3 positive nodes who are

likely to remain free of distant metastases. Among the 241 patients, 99 (41%) were classified

8/10/2019 Stella Mook T3-108


Summary

219

11

as good prognosis by the 70-gene signature, whereas 142 (59%) patients were classified as

poor prognosis. The 10-year distant metastasis-free (DMFS) and BCSS probabilities were

91% (SE 4%) and 96% (SE 2%) for patients with a low risk 70-gene signature, respectively

and 76% (SE 4%) and 76% (SE 4%), respectively for patients with a high risk signature. The

signature was associated with disease outcome independent of traditional prognostic

factors, with an adjusted HR of 7.17 (95% CI 1.81-28.43; p = 0.005). In contrast to the 70-

gene signature, Adjuvant! Classified only 32 patients (13%) as clinical low risk and 209

patients (87%) as clinical high risk, resulting in discordant risk assessments in 32% (72

patients). Remarkably, among the patients who were classified as high risk by Adjuvant! the

signature could identify 72 patients (34%) who had a low risk signature and indeed a good

disease outcome (10-year BCSS of 94% (SE 3%)). Furthermore, the signature was associated

with disease outcome in patients who did not receive adjuvant chemotherapy. The results

of this study showed that the 70-gene signature could identify patients with an excellent

disease outcome, even among patients with lymph node positive disease who might be

safely spared chemotherapy. Based on these results the MINDACT trial has extended its

eligibility criteria to include patients with up to 3 positive nodes.

The aim of the study described in chapter 6 was to evaluate the accuracy of the 70-

gene signature in patients with tumors less than 2.1 cm (pT1). With the introduction

of mammographic screening the proportion of small breast tumors has increased

tremendously. In a pooled database of 964 patients the signature accurately distinguished

patients with a good outcome from those with a poor outcome; this prognostic value was

independent of clinicopathological characteristics with an adjusted HR of 3.25 (95% CI 1.92-5.51; p < 0.001) for BCSS at 10 years. The results of this study emphasize that a considerable

proportion of small tumors metastasize, supporting the idea that metastatic capacity is

an early inheritance that can be identified by the 70-gene signature (28% distant relapse

rate at 10 years in tumors classified as poor prognosis by the signature). Therefore, the 70-

gene signature can be useful to optimize and individualize treatment decision-making in

patients with pT1 tumors.

In Chapter 7 we analyzed 541 patients from a retrospective pooled database who had

received adjuvant systemic therapy and who were classified by the 70-gene signature.

Among the 541 patients who received either endocrine therapy alone (n=315) or incombination with chemotherapy (n=226) the 70-gene signature classified 252 patients as

low risk and 289 patients as high risk. Patients with a high risk 70-gene signature treated

with chemotherapy followed by endocrine therapy had a significantly better 5-year distant

disease-free survival (DDFS) compared with high risk patients who were treated with

endocrine therapy alone (88% versus 76%, respectively; p < 0.01). Conversely, patients with

a low risk 70-gene signature who were treated with chemotherapy followed by endocrine

therapy had similar disease outcomes as low risk patients treated with endocrine therapy

alone (5-year DDFS 99% versus 93%, respectively; p = 0.62). This suggests that patients

classified as high risk by the signature do benefit from adjuvant chemotherapy in addition

8/10/2019 Stella Mook T3-108


Chapter 11

220

to endocrine therapy. Moreover, the benefit of chemotherapy appears to be absent in

patients with a low risk signature, which will further justify withholding chemotherapy in

these patients.

In Chapter 8 we describe a validation study of the computer tool Adjuvant! in 5,380 Dutch

breast cancer patients. Adjuvant! is a web-based tool that predicts disease outcome and

treatment benefit for the individual patient, based on clinicopathological characteristics

such as age, co-morbidity, tumor size, tumor grade, lymph node status and estrogen receptor

status. The program has been developed and validated on American and Canadian breast

cancer patients. The aim of this study was first to assess the accuracy of predicted outcome

by Adjuvant! in (subgroups of) Dutch breast cancer patients. In addition, we investigated

its ability to discern patients having good outcomes from those having poor outcomes

(discriminatory accuracy). Results showed that the model could accurately predict outcome

on group level (differences between predicted and observed outcomes were within 2%

for most clinically relevant subgroups) and could be applied to most patients, with the

exception of patients younger than 40 years. Adjuvant! overestimated outcome in these

patients by approximately 4.5% and predictions of Adjuvant! in patients less than 40 years

should be treated with caution, especially in patients with an estrogen receptor positive

tumor. The discriminatory accuracy of Adjuvant! was only moderate, suggesting that the

model’s predictions could be improved by adding additional prognostic information, such

as provided by the 70-gene signature.

As we have shown in the previous chapter, models such as Adjuvant! can predict disease

outcome but are still suboptimal. Therefore, we investigated whether method of detection

has additional prognostic value that could improve the estimation of prognosis. This question

is addressed in chapter 9, where we studied the accuracy of prediction by Adjuvant! in

patients with a screen-detected carcinoma as well as assessed the independent prognostic

value of screen-detection in a retrospective patient cohort of 2,592 breast cancer patients

aged 50-69 years, with invasive breast cancer. Method of detection was classified as (1)

screen-detected carcinomas, defined as carcinomas that were mammographically detected

in the first or subsequent screening rounds (n = 958); (2) interval carcinomas, defined assymptomatic carcinomas that were diagnosed within 24 months of a negative screening

(n = 417); and (3) nonscreening-related carcinomas, defined as symptomatic carcinomas

in patients who were not participating in the screening program (n = 1,217). Adjuvant!

predicted the outcome among patients with nonscreening-related carcinomas accurately

(predicted survival was within 2% of the observed survival and/or not significantly different

in all but one group), whereas Adjuvant! predictions underestimated overall survival

and breast cancer-specific survival among patients with screen-detected and interval

carcinomas. Prediction of breast cancer-specific survival was underestimated by Adjuvant!

for patients with screen-detected carcinomas by -3.2%. Screen-detected carcinomas were

8/10/2019 Stella Mook T3-108


Summary

221

11

associated with a significant reduced mortality compared with nonscreening-related

carcinomas. The prognostic value of screen-detection was independent of the well-known

stage shift that is caused by screening (i.e. earlier stage at diagnosis), with an adjusted HR of

0.62 (95% CI 0.50-0.78; p < 0.001). In addition, the prognostic value of method of detection

was similar across tumor size and lymph node status categories, again indicating its

prognostic value beyond stage migration. As a consequence of these results, we propose

that method of detection should be used in combination with traditional markers of tumor

burden and aggressiveness to estimate prognosis for each patient, and to guide their

decision to receive adjuvant systemic therapy.

In chapter 10 the major results presented in this thesis are discussed and put in

perspective of current clinical practice. In general, the 70-gene signature could improve the

prediction of disease outcome in several subgroups. Most likely the combination of (new)

clinicopathological factors and gene expression signatures could even further improve

accurate estimation of prognosis.

8/10/2019 Stella Mook T3-108


Chapter 12

Nederlandse samenvatting

List of publications

Dankwoord

Curriculum vitae

8/10/2019 Stella Mook T3-108


Chapter 12

224

8/10/2019 Stella Mook T3-108



225

12

Borstkanker is de meest voorkomende maligniteit bij vrouwen in de wereld. In 2008 werden

er in Nederland 13.005 vrouwen gediagnosticeerd met een invasief mammacarcinoom en

overleden er 3.327 patiënten aan de gevolgen van borstkanker. Hoewel de incidentie van

borstkanker stijgt, wordt er, ten gevolge van de invoering van screening op borstkanker

en het toegenomen gebruik van adjuvant systemische therapie (AST), een afname van

de mortaliteit gezien. De nieuwere adjuvante systemische therapieën zijn bovendien

effectiever en leiden tot een verdere reductie van de mortaliteit. Volgens de huidige

richtlijnen komt het merendeel van de borstkankerpatiënten in aanmerking voor een vorm

van AST. Hoewel de behandeling met AST de overleving in de borstkankerpopulatie in zijn

geheel verbetert, zullen er ook patiënten zijn die geen baat van deze behandeling hebben

omdat ze door alleen locoregionale behandeling genezen zijn. Voorts kan chemotherapie

ernstige acute en late bijwerkingen hebben, waardoor in een bepaalde groep patiënten

AST dan ook meer schadelijk dan nuttig zal zijn.

Aanvullende systemische therapie wordt geadviseerd op grond van prognostische en

predictieve kenmerken van het primaire tumorproces. Van oudsher wordt met behulp

van klinische en pathologische factoren een inschatting gemaakt van de prognose

van een patiënt. Patiënten met een ongunstige prognose, d.w.z. een hoog risico op

het ontwikkelen van afstandsmetastasen en/of het overlijden aan de gevolgen van

borstkanker, komen in aanmerking voor AST. Laagrisico patiënten zijn patiënten die met

alleen locoregionale behandeling een grote kans op genezing hebben en waarbij AST

naar verwachting geen of slechts een geringe verbetering van de overleving zal geven.De klinisch-pathologische kenmerken die gebruikt worden voor de voorspelling van de

prognose van de individuele patiënt blijken echter maar van beperkte waarde, vooral

omdat patiënten met morfologisch identieke tumoren een heel verschillend ziektebeloop

kunnen hebben. Het gebrek aan nauwkeurige identificatie van patiënten met een laag

risico op het ontwikkelen van afstandsmetastasen resulteert in overbehandeling, gepaard

gaande met onnodige toxiciteit, terwijl een onjuiste selectie van hoogrisico patiënten

onderbehandeling kan veroorzaken en daarmee de overlevingskansen van patiënten

kan verminderen. Behandeling op maat, de zogenaamde “patient-tailored treatment”, zal

over- en onderbehandeling terugdringen, maar kan alleen worden gerealiseerd indiener nieuwe en betere prognostische en predictieve markers worden geïdentificeerd.

Dit proefschrift richt zich op het nut van een nieuwe prognostische test (het 70-genen

profiel of MammaPrint™; hoofdstuk 2-7), een reeds veelgebruikt prognostisch model

(Adjuvant!; hoofdstuk 8) en een potentieel nieuwe prognostische factor (methode van

detectie; hoofdstuk 9). Bovendien hebben we de toepasbaarheid van deze prognostische

markers in verschillende subgroepen geëvalueerd.

8/10/2019 Stella Mook T3-108


Chapter 12

226

Hoofdstuk 1 geeft naast een algemene inleiding over de behandeling van borstkanker en

het gebruik van traditionele prognostische markers en genexpressieprofielen, een beknopt

overzicht van het onderzoek beschreven in dit proefschrift en de klinische vraagstellingen

die er aan ten grondslag liggen.

Het eerste deel van dit proefschrift richt zich op de toepasbaarheid en de klinische waarde

van het 70-genen profiel. In hoofdstuk 2 geven we een overzicht van de ontwikkeling

en eerste validatie van het 70-genen profiel. Dit profiel is ontwikkeld met behulp van een

retrospectieve serie van 78 patiënten, jonger dan 55 jaar en gediagnosticeerd met een

invasief mammacarcinoom kleiner dan 5.1 cm (pT1-2) zonder axillaire lymfekliermetastasen.

Vierenveertig van de 78 patiënten hadden een ziektevrije overleving van minimaal 5 jaar

(goede prognose), terwijl 34 patiënten binnen 5 jaar na diagnose afstandsmetastasen

ontwikkelden (slechte prognose). Met behulp van gesuperviseerde analyses werden 70

genen geselecteerd die verschillend tot expressie kwamen in de 2 prognostische groepen en

die het sterkst correleerden met het ontwikkelen van afstandsmetastasen. Het profiel werd

vervolgens gevalideerd in een consecutieve serie van 151 patiënten met lymfekliernegatief

en 144 patiënten met lymfeklierpositief mammacarcinoom. Deze validatiestudie

toonde aan dat het profiel in staat was om patiënten met een goede prognose (n=115)

nauwkeurig te kunnen onderscheiden van patiënten met een slechte prognose (n=180).

Bovendien was de voorspellende waarde van het profiel onafhankelijk van traditionele

klinisch-pathologische criteria (gecorrigeerde hazard ratio (HR) voor afstandsmetastasen

als eerste event was 4.6; 95% confidence interval (CI) 2.3–9.2; p < 0.001). In een volgendeinternationale en onafhankelijke validatiestudie werd in 302 lymfekliernegatieve patiënten

uit 5 Europese ziekenhuizen de prognostische waarde van het profiel bevestigd. Deze

patiënten waren geen van allen adjuvant systemische behandeld. Naast de validatiestudies

worden in hoofdstuk 2 ook twee studies naar de uitvoerbaarheid van het 70-genen profiel

in de dagelijkse praktijk beschreven. De eerste studie, de zogenaamde RASTER (MicroarRAy

PrognoSTics in kanker van de borst) demonstreerde de haalbaarheid van het gebruik

van het profiel voor de besluitvorming betreffende AST in 16 algemene ziekenhuizen in

Nederland. In een Europese pilot studie, welke ook in detail beschreven staat in hoofdstuk

3, werd de logistiek voor de prospectieve gerandomiseerde MINDACT (Microarray InNode-negative and 1-3 positive lymph node disease may Avoid ChemoTherapy) studie

getest en geoptimaliseerd. Deze studie toonde aan dat het goed mogelijk is om vers

gevroren tumorweefsel van voldoende kwaliteit voor microarray analyse te verzamelen

in verschillende Europese ziekenhuizen en daarvandaan te verzenden naar een centrale

microarray faciliteit. Indien de patholoog een tumorsample kon verkrijgen was het in 77%

(46/60) van de gevallen mogelijk om een 70-genen profiel te bepalen. In 18% (11/60) was

het tumorsample niet representatief voor de tumor (< 50% tumorcellen) en kon er geen

70-genen profiel worden bepaald. Op basis van deze resultaten en de opgedane ervaring

in deze pilot studie werden zogeheten “standard operating procedures” ontwikkeld, welke

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108


Chapter 12

228

voor patiënten met een hoogrisico profiel. De prognostische waarde van het profiel was

opnieuw onafhankelijk van traditionele prognostische factoren, met een gecorrigeerde

HR van 7.17 (95% CI 1.81-28.43; p = 0.005). In tegenstelling tot het 70-genen profiel

classificeerde het Adjuvant! model (op basis van traditionele klinisch-pathologische

kenmerken) slechts 32 patiënten (13%) als laag risico en 209 patiënten (87%) als klinisch

hoog risico wat resulteerde in een discordante risico-inschatting in 32% van de patiënten

(72 patiënten). Opmerkelijk is dat binnen de patiënten die werden geclassificeerd als hoog

risico door Adjuvant! het profiel nog 72 patiënten (34%) kan identificeren die een laagrisico

profiel en inderdaad een goede overleving hebben (10-jaars borstkankerspecifieke

overleving van 94% (SE 3%)). Het profiel bleek eveneens geassocieerd met overleving

van de patiënten die niet waren behandeld met adjuvante chemotherapie. De resultaten

van deze studie tonen aan dat het 70-genen profiel patiënten weet te identificeren die

ondanks 1-3 lymfekliermetastasen een uitstekende prognose hebben en waarbij adjuvante

chemotherapie wellicht overbodig is. Op basis van deze resultaten zijn de inclusiecriteria

voor de MINDACT studie uitgebreid tot patiënten met maximaal 3 positieve lymfeklieren.

Hoofdstuk 6 beschrijft de waarde van het 70-genen profiel voor patiënten met een

mammacarcinoom kleiner dan 2.1 cm. Door de introductie van het bevolkingsonderzoek

naar borstkanker, waar vrouwen van 50 jaar en ouder middels mammografie worden

gescreend, is het percentage kleine mammacarcinomen dat gediagnosticeerd wordt de

afgelopen jaren sterk toegenomen. Hoewel een kleine tumordiameter over het algemeen

beschouwd wordt als een indicator voor een goede prognose, blijkt het 70-genen profiel ook

in deze groep patiënten met een goede prognose nauwkeurig te kunnen onderscheidenvan patiënten met een slechte prognose. In 964 patiënten met een tumor kleiner dan 2.1

cm (pT1) had het profiel opnieuw prognostische waarde onafhankelijk van de klinisch-

pathologische kenmerken, met een gecorrigeerde HR van 3.25 (95% CI 1.92-5.51; p < 0.001)

voor 10-jaars borstkankerspecifieke overleving. De resultaten van deze studie benadrukken

dat een aanzienlijk deel van de kleine tumoren metastaseert, en ondersteunen daarmee

het idee dat het vermogen tot metastaseren al vroeg in de tumorontwikkeling wordt

bepaald. Met behulp van het 70-genen profiel kan dit vermogen tot metastaseren

nauwkeuriger worden bepaald (van patiënten geclassificeerd door het profiel als hoog

risico recidiveert 28% binnen 10 jaar) en kan de behandeling van borstkankerpatiëntenmet relatief kleine tumoren (pT1) worden geoptimaliseerd en geïndividualiseerd.

In Hoofdstuk 7 hebben we 541 adjuvant systemisch behandelde patiënten geanalyseerd

die zijn geselecteerd uit een retrospectieve gepoolde database. Van de patiënten

die behandeld waren met endocriene therapie alleen (n = 315) of in combinatie met

chemotherapie (n = 226) werden 252 patiënten geclassificeerd als laag risico en 289 patiënten

als hoog risico door het 70-genen profiel. Patiënten met een hoogrisico 70-genen profiel

die behandeld waren met endocriene therapie gevolgd door chemotherapie hadden een

significant betere 5-jaars afstandsmetastasen-vrije overleving vergeleken met hoogrisico

patiënten die behandeld werden met endocriene therapie alleen (respectievelijk 88% en

8/10/2019 Stella Mook T3-108



229

12

76%; p < 0.01). Omgekeerd hadden patiënten met een laagrisico profiel die behandeld

werden met endocriene therapie gevolgd door chemotherapie een vergelijkbare

uitkomst als laagrisico patiënten die behandeld werden met alleen endocriene therapie

(5-jaars afstandsmetastasen-vrije overleving van respectievelijk 99% en 93%; p = 0.62).

Dit suggereert dat patiënten geclassificeerd als hoog risico door het profiel baat zullen

hebben bij additionele chemotherapie, terwijl voor patiënten met een laagrisico profiel het

toevoegen van chemotherapie aan adjuvante endocriene therapie geen overlevingswinst

geeft. Deze resultaten ondersteunen de conclusies van bovengenoemde studies dat voor

patiënten met een laagrisico profiel het onthouden van chemotherapie gerechtvaardigd

is, enerzijds omdat de prognose van deze patiënten zeer goed is en anderzijds omdat de

winst van chemotherapie zeer beperkt lijkt te zijn.

In hoofdstuk 8 wordt de validatiestudie van het computerprogramma Adjuvant! in 5.380

Nederlandse borstkankerpatiënten gepresenteerd. Adjuvant! is een online beschikbaar

computerprogramma dat met behulp van klinisch-pathologische kenmerken de

prognose en de te verwachten winst van AST voor de individuele patiënt voorspeldt.

Voor het berekenen hiervan maakt het model gebruik van leeftijd en comorbiditeit

van de patiënt, tumor grootte en graad, aantal positieve lymfeklieren en de oestrogeen

receptorstatus. Het programma is ontwikkeld met behulp van gegevens van Amerikaanse

borstkankerpatiënten en werd reeds gevalideerd in een Canadese borstkankerpopulatie.

Het doel van deze studie was allereerst om te beoordelen of de voorspellingen van prognose

en behandelwinst door Adjuvant! ook toepasbaar zijn in (subgroepen van) Nederlandseborstkankerpatiënten. Bovendien is onderzocht of Adjuvant! in staat was om patiënten

met een goede prognose te onderscheiden van patiënten met een slechte prognose; deze

zogenaamde discriminatory accuracy zegt meer over de waarde van een marker of test

voor de individuele patiënt. De voorspelling van overleving op groepsniveau blijk zeer

nauwkeurig (verschil tussen voorspelde en geobserveerde overleving was < 2% voor de

meeste klinisch relevante subgroepen), met uitzondering van de voorspellingen voor

patiënten onder de 40 jaar. Adjuvant! overschat de overleving in deze groep patiënten met

ongeveer 4.5% waardoor enige voorzichtigheid is geboden bij het gebruik van Adjuvant!

in deze patiëntengroep, voornamelijk in geval van oestrogeen receptorpositieve tumor.De nauwkeurigheid van de voorspellingen op individueel niveau (discriminatory accuracy)

blijkt niet optimaal en kan worden verbeterd door het toevoegen van prognostische

informatie, zoals die van het 70-genen profiel.

Zoals hierboven beschreven zijn de voorspellingen van prognose door modellen als

Adjuvant! nog altijd suboptimaal. Het toevoegen van nieuwe prognostische factoren kan

deze predictie wellicht verbeteren. In Hoofdstuk 9 hebben we onderzocht of Adjuvant! de

prognose van patiënten met een screen-detected mammacarcinoom accuraat voorspelde

en of methode van detectie (d.w.z. carcinomen gedetecteerd in het kader van screening

8/10/2019 Stella Mook T3-108


Chapter 12

230

versus symptomatische carcinomen gediagnosticeerd buiten screening) zelf onafhankelijke

prognostische waarden had. Hiertoe zijn 2.592 patiënten in de leeftijd van 50-69 jaar,

met een invasief mammacarcinoom geselecteerd uit de database die is gebruikt voor

de validatie van Adjuvant! zoals beschreven in hoofdstuk 8. Methode van detectie is als

volgt gedefinieerd (1) screen-detected carcinomen, niet-symptomatische carcinomen

die werden gediagnosticeerd op basis van het mammogram gemaakt in het kader van

screening (n = 958); (2) interval carcinomen, gedefinieerd als symptomatische carcinomen

die werden gediagnosticeerd binnen 24 maanden na een negatieve screening (n = 417); en

(3) niet-screeninggerelateerde carcinomen, gedefinieerd als symptomatische carcinomen

gediagnosticeerd in vrouwen die niet aan de screening deelnamen (n = 1.217). Screen-

detected carcinomen werden geassocieerd met een aanzienlijk betere overleving in

vergelijking met niet-screeningsgerelateerde carcinomen. De prognostische waarde van

screen-detectie was onafhankelijk van de verschuiving naar een vroegtijdiger stadium

bij diagnose zoals die wordt gezien bij screening (stage shift), met een gecorrigeerde HR

van 0.62 (95% CI 0.50-0.78; p < 0.001). Bovendien was de voorspellende waarde van de

methode van detectie gelijk in patiëntgroepen onderverdeeld op basis van tumorgrootte

en lymfeklierstatus, wat nogmaals bevestigt dat de prognostische waarde van methode

van detectie niet afhankelijk is van de verschuiving naar een vroeger stadium bij diagnose.

Gezien bovenstaande lijkt het gebruik van methode van detectie als marker in combinatie

met traditionele prognostische factoren tot een betere voorspelling van prognose te

kunnen leiden, en zal het op die manier de keuze voor wel of geen AST betrouwbaarder

maken.

In hoofdstuk 10 worden de belangrijkste resultaten van dit proefschrift besproken

en gerelateerd aan de huidige klinische praktijk. In het algemeen zal het gebruik van

het 70-genen profiel de voorspelling van ziektebeloop verbeteren, niet alleen in de

patiëntengroep waarin het profiel is ontwikkeld, maar ook in andere subgroepen zoals

postmenopauzale patiënten. Waarschijnlijk zal het combineren van (nieuwe) klinisch-

pathologische factoren met genexpressieprofielen de voorspelling van prognose verder

kunnen verbeteren.

8/10/2019 Stella Mook T3-108


Chapter 12



Dankwoord

Curriculum vitae

8/10/2019 Stella Mook T3-108


Chapter 12

234

Mook S, Van ’t Veer LJ, Rutgers EJ, Ravdin PM, Van de Velde AO, Van Leeuwen FE, Visser O,

Schmidt MK. Independent prognostic value of screen detection in invasive breast cancer.

JNCI, accepted for publication.

Knauer M, Mook S, Rutgers EJ, Bender RA, Hauptmann M, Van de Vijver MJ, Koornstra RH,

Bueno-de-Mesquita JM, Linn SC, Van ‘t Veer LJ. The predictive value of the 70-gene signature

for adjuvant chemotherapy in early breast cancer. Breast Cancer Res Treat 2010; 120: 655-661.

Mook S, Knauer M, Bueno-de-Mesquita JM, Retel VP, Wesseling J, Linn SC, Van ‘t Veer LJ,

Rutgers EJ. Metastatic potential of T1 breast cancer can be predicted by the 70-gene

MammaPrint signature. Ann Surg Oncol 2010; 17: 1406-1413.

Mook S, Schmidt MK, Weigelt B, Kreike B, Eekhout I, Van de Vijver MJ, Glas AM, Floore A,

Rutgers EJ, Van ‘t Veer LJ. The 70-gene prognosis signature predicts early metastasis in

breast cancer patients between 55 and 70 years of age. Ann Oncol 2010; 21: 717-722.

Mook S, Schmidt MK, Rutgers EJ, van de Velde AO, Visser O, Rutgers SM, Armstrong N, Van

‘t Veer LJ, Ravdin PM. Calibration and discriminatory accuracy of prognosis calculation for

breast cancer with the online Adjuvant! program: a hospital-based retrospective cohort

study. Lancet Oncol 2009; 10: 1070-1076.

Bedard PL, Mook S, Piccart-Gebhart MJ, Rutgers ET, Van ‘t Veer LJ, Cardoso F. MammaPrint70-gene profile quantifies the likelihood of recurrence for early breast cancer. Expert Opinion on

Medical Diagnostics 2009; 3: 193-205.

Mook S, Bonnefoi H, Pruneri G, Larsimont D, Jaskiewicz J, Sabadell MD, MacGrogan G, Van

‘t Veer LJ, Cardoso F, Rutgers EJ. Daily clinical practice of fresh tumour tissue freezing and

gene expression profiling; logistics pilot study preceding the MINDACT trial. Eur J Cancer 2009;

45: 1201-1208.

Mook S, Schmidt MK, Viale G, Pruneri G, Eekhout I, Floore A, Glas AM, Bogaerts J, CardosoF, Piccart-Gebhart MJ, Rutgers ET, Van ‘t Veer LJ. The 70-gene prognosis-signature predicts

disease outcome in breast cancer patients with 1-3 positive lymph nodes in an independent

validation study. Breast Cancer Res Treat 2009; 116: 295-302.

Reyal F, van Vliet MH, Armstrong NJ, Horlings HM, de Visser KE, Kok M, Teschendorff AE,

Mook S, Van ‘t Veer L, Caldas C, Salmon RJ, van de Vijver MJ, Wessels LF. A comprehensive

analysis of prognostic signatures reveals the high predictive capacity of the proliferation,

immune response and RNA splicing modules in breast cancer. Breast Cancer Res 2008; 10: R93.

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108


Chapter 12



Dankwoord

Curriculum vitae

8/10/2019 Stella Mook T3-108


Chapter 12

238

8/10/2019 Stella Mook T3-108


Dankwoord

239

12

Zo, dit is het dan. Het is bijna klaar. Rest mij alleen nog het schrijven van het meest gelezen

gedeelte van een proefschrift, het dankwoord. De afgelopen vijf jaar heb ik, steeds een

beetje meer, naar dit moment uitgekeken. Dankzij geweldige collega’s, vrienden en familie

is het uiteindelijk dan toch gelukt. Een aantal personen wil ik graag in het bijzonder

bedanken.

Allereerst gaat mijn dank uit naar mijn beide promotores.

Prof. dr. Laura J van ’t Veer, beste Laura, dat mijn lidmaatschap van U.S.R. Triton zou leiden

tot dit proefschrift en tot jouw eerste promovenda hadden we allebei nooit kunnen

bedenken. Dank voor je vertrouwen in mij. Ik bewonder jouw pioniersmentaliteit en heb

van je geleerd dat gewoon beginnen vaak het beste is. Dank ook voor je enthousiasme

en optimisme. Dat optimisme heeft iemand zoals ik, met een licht pessimistische inslag,

wel nodig af en toe. Zoals bij de TRANSBIG vergadering waar jij me, tijdens een verhitte

discussie, influisterde dat ik wel moest blijven lachen. Daarnaast wist jij altijd als geen ander,

op de momenten dat ik het allemaal even niet meer overzag, orde en rust te scheppen. De

traditie om mooie momenten (de dag dat alle samples voor de pilot studie uit Milaan in

een keer arriveerden, het tekenen van het MINDACT contract, de eerste MINDACT patiënt)

te vieren (met een flesje bubbels of geheel in stijl met een Turkse massage in Istanbul) vind

ik geweldig. Helaas ben je inmiddels vertrokken naar San Francisco, maar ik weet zeker dat

in de toekomst door een mooie samenwerking onze wegen weer zullen kruisen.

Prof. dr. Emiel J Th Rutgers, beste Emiel, bij onze eerste ontmoeting stond jij verstopt achtereen deur en maakte je een goede grap. Jij kunt je dat waarschijnlijk niet meer herinneren,

maar het heeft zeker een rol gespeeld in mijn komst naar het NKI. De afspraken met jou

waren constructief, stimulerend en gezellig. Ik denk dat we minstens de helft van de tijd

praatten over andere dingen in het leven. De paar momenten waarop ik echt vond dat

de dingen anders moesten en dat wilde bespreken, was jij me net voor en had je al over

een oplossing nagedacht. Een unieke en kostbare gave, zeker naarmate er meer vrouwen

in het vak komen! Dank voor de mogelijkheden die je me hebt gegeven. Dank ook voor

de waardevolle input en kritische klinische blik op mijn werk, maar zeker ook voor alle

gezelligheid!

Mijn dank gaat ook uit naar de leden van de promotiecommissie, Prof. dr. R. Bernards, Prof.

dr. J.W. Coeberg, Prof. dr. C.C.E. Koning, Prof. dr. J.W.R. Nortier, Prof. dr. S. Rodenhuis, Prof. dr.

M.J. van de Vijver en Dr. J.H.G. Klinkenbijl, voor het beoordelen van mijn manuscript.

Lieve Marjanka, dat jij mijn paranimf moest worden stond vast. Onze samenwerking begon

in Aspen, bij de AACR cursus. Onze vriendschap begon ook daar, bij de ontmoeting met

een bruine beer, geweldig! Ik heb veel van je geleerd op het gebied van epidemiologie

en statistiek, maar ook van jouw rust en kwaliteit om altijd dicht bij jezelf te blijven. Jouw

8/10/2019 Stella Mook T3-108


Chapter 12

240

kritische blik als ik er eigenlijk helemaal klaar mee was, en andersom, hebben tot 2 mooie

publicaties geleid. Door de drukte spreken we elkaar veel te weinig. Onze volgende

afspraak moet er een in een restaurant of op de rug van een paard zijn!

Lieve Inge, zonder jou had dit proefschrift zeker langer op zich later wachten. Jouw komst

bij de MINDACT was geweldig. Samen langs de Nederlandse ziekenhuizen, nieuwsbrieven

schrijven en de logistieke obstakels oplossen, we waren een goed team! Het was heerlijk

om de MINDACT werkzaamheden op een gegeven moment aan jou over te dragen en

me meer op mijn promotie te kunnen richten. Ik heb goede herinneringen aan onze

congresbezoeken. Zo weinig voorbereid als ik onderweg ging, zo goed voorbereid was jij.

In elke stad regelde jij een tafeltje in een leuk en trendy restaurant en zorgde je voor een

kamer in dat geweldige hotel. Lieve Ing, ik ken geloof ik weinig mensen die zo zorgzaam en

warm zijn als jij. Laten we snel weer eens een hapje gaan eten. Dan zoek ik een restaurant

uit…

Marleen, lieve Lena, allebei in Utrecht gestudeerd, maar pas in Amsterdam elkaar echt

gevonden. Wat hebben we een goede tijd gehad! Veel gepraat, gelachen, gehuild,

geklaagd en gedronken. Gegeten op de een of andere manier dan weer minder, we

kwamen vaak niet verder dan een portie bitterballen. Ik wil je bedanken voor de vele

discussies over het onderzoek: wat is nou de beste statische methode, wat is nou de

optimale patiëntenselectie. Met jouw gedrevenheid, enthousiasme en brede kennis ga

je ongetwijfeld een geweldige internist worden. Het enige minpuntje van jou als roomiewaren de met exotische zwammen begroeide koffiekopjes... Lieve Lena, ik mis je. Het wordt

tijd voor weer een blauwe-schoenen-borrel (zonder laptop), en dan beloof ik dat ik niet de

hele dag op jouw bank blijf hangen!

En dan natuurlijk Marieke, lieve dokteur Fulliburghhhh! Wat jammer dat je weg bent uit

het NKI. Miami Beach met jou was geweldig. De voorpret (en jouw angst dat ik een week

lang niets zou eten) ook. Behalve dat je een ster bent in relativeren en analyseren (en dat

dan ook nog razendsnel) ben je vooral ook een heel lief mens met een briljant gevoel voor

humor! Het komt allemaal goed!

Lieve Olga, hoe heerlijk is het om bij jou in je stoel neer te ploffen om gewoon even lekker

te kletsen! En hoe heerlijk ook dat jij, als ik bijna helemaal in de stress schoot, mij een

mailtje stuurde dat me deed schaterlachen of weer eens iets voor me regelde. Dank, je

bent een held!

Prof. dr. F.E. van Leeuwen en dr. M. Rookus, beste Floor en Matti, graag wil ik jullie bedanken

voor jullie kritische commentaar op manuscripten en presentaties. Otto Visser en Tony van

de Velde, wil ik bedanken voor het elke keer weer nauwkeurig aanleveren van data. Sterre,

8/10/2019 Stella Mook T3-108


Dankwoord

241

12

dank voor het eindeloos en nauwkeurig invoeren van data, wat heeft geleid tot een mooie

publicatie, één met twee Rutgers’s in de auteurslijst.

Dank aan iedereen op de afdeling pathologie, waar ik de eerste 2 jaar van mijn onderzoek

heb doorgebracht. Lieve Jolien en Valesca, roomies op de pathologie, dank voor alle steun

en gezelligheid. Jolien, wat hebben we een hoop meegemaakt! Laten we snel weer eens

echt bijkletsen!

Marieke, Leonie en Roelien, van jullie heb ik de kunst kunnen afkijken. Wat is het lang

geleden!

Ook wil ik graag alle H6ers en/of C2ers bedanken. Thea, dank voor je gezelligheid, ik kom

snel weer eens buurten. Annegien, Richard, Linde, Sjoerd B, Lorenza, Hans, Renske, Astrid

en de rest, dankzij jullie hoorde ik eindelijk eens echt ergens bij…

Huug, roomie van het laatste uur, wat goed dat ook jouw proefschrift binnenkort klaar is!

Dear Michael, thank you for the collaboration!

Sjoerd E, beste Sjoerd, straks heb ik eindelijk tijd om aan ons project te werken. Ik kijk er

naar uit.

Veel artikelen in dit proefschrift zijn mede mogelijk gemaakt door Agendia. In het bijzonder

wil ik Annuska, Arno, Bas, Femke, Guido, Ilja, Iris en Lisette bedanken voor de plezierige

samenwerking.

It has been an honor to be part of TRANSBIG. Dear Fatima, thank you for all the great

conversations and for your valuable input. Thanks also for taking the time to discuss my

future plans in Bordeaux. Radiation Oncology is such a great profession.

It has been a great pleasure and honor to work with Peter Ravdin. Dear Peter, thank you for

sharing all the Adjuvant! data with us. Our meetings were always stimulating. I admire your

broad knowledge about epidemiology and the clinical aspects of breast cancer, but also

your great sense of humor! Even at 4 o’ clock in the morning!

Thanks to professor Giuseppe Viale and Giancarlo Pruneri we were able to collaborate with

the European Institute of Oncology in Milan. Dear Beppe and Giancarlo, thank you and

hopefully we will meet somewhere soon at an international conference!

Dear Mahasti, working with you has always been a pleasure. We were not the most efficient

team, but definitely the team who had the best laughs.

I would also like to thank the EORTC, especially Jillian Harrison, Frederic Henot and Gaston

Demonty for the great collaboration. Hope to see you soon at an EORTC meeting!

8/10/2019 Stella Mook T3-108


Chapter 12

242

Medewerkers van de afdeling radiotherapie in het NKI-AVL, van stafleden tot laboranten,

van secretaresses tot doktersassistenten, het is een voorrecht mijn opleiding te mogen

volgen op een afdeling waar de sfeer zo goed is!

En dan natuurlijk de assistent van de radiotherapie. Lieve collega’s, dank voor alle

gezelligheid. Ik kan me geen betere collega’s wensen. Brian, dank voor het beantwoorden

van al mijn computervragen. Gerben, dank voor het beantwoorden van ongeveer alle

andere vragen (wanneer moet nou precies het manuscript naar de drukker??). Brenda, jouw

liefde voor de ani heeft heel wat gezellige en productieve avondjes op het NKI opgeleverd.

Laten we snel weer borrelen!

Lieve Utrechtse meiden, wat heb ik vaak verstek laten gaan… Het is altijd weer gezellig om

elkaar te zien. Ik ga er meer tijd voor vrij maken, dat beloof ik.

Lieve Anne & Sander, jullie zijn een goed stel, en niet in de laatste plaats omdat het ontstaan

is bij een kampvuurtje in Frankrijk, zonder jullie aanwezigheid... Ik hoop dat jullie snel weer

naar Haarlem komen, zodat we weer wat makkelijker een (of een paar) drankje(s) kunnen

drinken. Geniet van de kleine.

Lieve Baselga-gangers, dank voor de prachtweek, in het heetst van de strijd, toen ik het zo

nodig had!

Lieve Jantien, geweldig dat jouw werk op de cover van mijn proefschrift staat. Minstens zo

leuk is het om je weer te spreken! Je hebt een etentje van me te goed!

Lieve Jaap en Rina, met het afronden van mijn proefschrift komt er hopelijk weer meer tijd

om te fietsen. Hoewel het best frustrerend is om eraf gefietst te worden door 2 mannen

‘op leeftijd’ heb ik het er graag voor over. Zeker ook omdat het naborrelen bij jullie altijd zo

gezellig is!

Lieve Juul, dank voor je interesse en betrokkenheid. En voor al je lieve kaartjes op bijzonderemomenten!

Lieve Saakje, ook jij bedankt voor je interesse in mij en in mijn werk. We komen snel weer

eens jouw kant op! En wat heb je toch een leuke zoon!

Lieve El(les), super dat je mijn paranimf bent! Wat is er een hoop gebeurt de laatste jaren,

ongelooflijk. Ik heb enorme bewondering voor hoe je je staande houdt. Je bent een

supervriendin (en dat al zoo lang) en mooi mens!

8/10/2019 Stella Mook T3-108


Dankwoord

243

12

Lieve Myra, schoonzussie en pechvogel. Ik hoop dat de komende jaren meer geluk met zich

mee brengen! Dank voor je gastvrijheid, ik vind het heerlijk om bij jullie te zijn. Dank ook

voor het vertrouwen in Marcel en mij als oom en tante!

Lieve, lieve Merle & Luke, wat een rijkdom dat jullie er zijn. Als niets me meer vrolijk maakt,

laten jullie de zon weer schijnen! Komen jullie snel weer eens logeren?

Lieve Matijs, grote broer, wat ben ik blij met je. Hoewel we elkaar niet heel veel zien is het

altijd goed. Groter wederzijds respect en waardering is denk ik niet mogelijk.

Pap & mam, dank voor het grenzeloze vertrouwen en de onvoorwaardelijke liefde.

Lieve mam, geweldig om weer een week met je te schaatsen in Baselga, wat hebben we

gelachen (het was tenslotte allemaal voor de l..). Geweldig ook dat we zo veel kunnen

delen.

Lieve Pap, onze schaatszondag is heilig, vooral omdat het zo leuk is om iets samen met jou

te doen. Dank dat ik nooit iets hoef uit te leggen en dat je me af en toe komt redden! Die

deuk in mijn waguh is me dierbaar!

Ik hou van jullie.

Lieve, lieve Marcel (ik durf geen Marce meer te schrijven), mijn grote liefde. Dat ik jou 15 jaar

geleden toch ben tegen gekomen! Jouw optimisme, nuchterheid, humor en liefde maken

het leven tot een feestje. Dank voor alle steun en hulp, zonder jou was dit proefschrift erecht niet gekomen. Dank ook dat je bent wie je bent! Ik hoop dat de komende tijd je heel

veel goeds gaat brengen. Vanaf nu gaan we weer leuke dingen doen in het weekend! Ik

hou zielsveel van je.

8/10/2019 Stella Mook T3-108


Chapter 12



Dankwoord

Curriculum vitae

8/10/2019 Stella Mook T3-108


Chapter 12

246

8/10/2019 Stella Mook T3-108


Curriculum vitae

247

12

Stella Mook is geboren op 21 februari 1977 te Hoorn. Na het behalen van haar VWO

diploma aan de Openbare Scholengemeenschap te Hoorn begon zij in 1996 met de studie

geneeskunde aan de Universiteit van Utrecht. Tijdens haar studie deed zij gedurende 1 jaar

onderzoek naar het metabolisme van vrije vetzuren op de afdeling vasculaire geneeskunde

van het Universitair Medisch Centrum Utrecht, onder supervisie van dr. Manuel Castro

Cabezas. Na het behalen van het artsenexamen in 2003 werkte zij tot eind 2004 als

assistent-niet-in-opleiding op de afdeling interne geneeskunde van het Diakonessenhuis

te Zeist/Utrecht.

In mei 2005 begon de auteur als research fellow bij het Nederlands Kanker Instituut-Antoni

van Leeuwenhoek Ziekenhuis met de coördinatie van een Europese pilot studie. Deze

aanstelling werd een jaar later omgezet in een promotietraject, onder supervisie van prof.

dr. Laura J van ’t Veer en prof. dr. Emiel J.Th. Rutgers. Een gedeelte van dit onderzoek deed

zij in het kader van haar TRANSBIG fellowship.

In 2009 is zij begonnen aan de opleiding tot radiotherapeut in het Nederlands Kanker

Instituut-Antoni van Leeuwenhoek Ziekenhuis te Amsterdam, onder supervisie van

opleider dr. R.L.M. Haas en plaatsvervangend opleider en afdelingshoofd prof. dr. M. Verheij.

8/10/2019 Stella Mook T3-108


Appendix

Gene signature evaluation as a prognostic

tool: challenges in the design of

the MINDACT trial

Jan Bogaerts

Fatima Cardoso

Marc Buyse

Sofia Braga

Sherene Loi

Jillian A Harrison

Jacques Bines

Stella Mook Nuria Decker

Peter Ravdin

Patrick Therasse

Emiel Rutgers

Laura J Van ’t Veer

Martine Piccart

on behalf of the TRANSBIG consortium

Nat Clin Pract Oncol 2006; 3: 540-551.

8/10/2019 Stella Mook T3-108


Chapter 12

250

Summary

This Review describes the work conducted by the TRANSBIG consortium in the development

of the MINDACT (Microarray In Node negative Disease may Avoid ChemoTherapy) trial. The

goal of the trial is to provide definitive evidence regarding the clinical relevance of the 70-

gene prognosis signature, and to assess the performance of this signature compared with

that of traditional prognostic indicators for assigning adjuvant chemotherapy to patients

with node-negative breast cancer. We outline the background work and the key questions

in node-negative early-stage breast cancer, and then focus on the MINDACT trial design

and statistical considerations. The challenges inherent in this trial in terms of logistics,

implementation and interpretation of the results are also discussed. We hope that this

article will trigger further discussion about the difficulties of setting up and analyzing trials

aimed at establishing the worth of new methods for better selection of patients for cancer

treatment.

Review criteria

A formal literature search for this review was not performed; this review includes a summary

of the authors’ own work and knowledge, which covers various fields relating to oncology

and molecular biology.

8/10/2019 Stella Mook T3-108


Appendix. Design of the MINDACT trial

251

12

Introduction

In the past 20 years, little progress has been made in identifying new prognostic markers

that can assist oncologists in treatment decision-making relating to node-negative

early-stage breast cancer. As a result, considerable differences exist worldwide in the

selection of women who require adjuvant chemotherapy based on their risk of breast

cancer recurrence. The breast cancer TNM (tumor–node–metastases) staging system is

based on anatomical extent (e.g. the size and lymph-node status) of the tumor, but this

classification gives little insight into breast cancer biology. Clinicians have long recognized

the heterogeneity of human breast cancers, not only in terms of their diverse natural

histories despite identical morphological features, but also in their variation in response

to treatment.1 These differences are also evident in the small ( i.e. <2 cm), node-negative

tumors that would generally be associated with a good clinical outcome. Attempts have

been made to identify good and poor prognosis groups based on pathological features

such as tumor grade, lymphatic invasion and S-phase fraction,2 which might better reflect

tumor biology. In recent years, numerous molecular prognostic and predictive markers

in oncology have been reported (Box 1). These tumor markers have had little impact in

routine clinical practice. Studies are often based on small, heterogeneous retrospective

series that have not been reported in a rigorous enough fashion to provide sufficient

information, particularly with regard to their methodology.3 Many follow-up studies have

shown inconsistent data compared with original results, which has been attributed to a

lack of statistical power, different patient populations, and technical limitations associatedwith such studies. There is also a paucity of well-designed, prospective assessments of the

clinical value of these tumor markers. As a result, the value of many promising prognostic

markers is still uncertain. We have yet to fully translate our increased understanding of

breast cancer biology into improved outcome for those with this heterogeneous disease.

The prognostic factors accepted by the NIH 2000 Consensus Development Conference

on Adjuvant Therapy for Breast Cancer did not include any molecular markers relevant

to breast cancer biology apart from the hormone receptors.4 The most recent St Gallen

consensus panel (2005)5 established three risk categories: minimal, intermediate and high.

Hormone receptors, tumor size, tumor grade and age remain key discriminating factors,and HER2 status, lymphatic or vascular invasion, or both in the primary tumor are new

accepted prognostic factors. In the UK, the Nottingham Prognostic Index is commonly used

to predict clinical outcome; this index is based on tumor size, tumor grade and lymph-node

status, and has a key role in discriminating node-negative patients for whom chemotherapy

should or should not be considered.6 All these consensus recommendations, however,

have important limitations and, in this era of evidence-based medicine, it is not possible to

reliably identify a group of women with excellent long-term clinical outcome.

By using gene-expression profiling, the Netherlands Cancer Institute developed a 70-

gene prognostic signature for node-negative breast cancer.7 The signature was developed

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108



253

12

criteria and the Adjuvant! software,11 which can calculate a 10-year survival probability

based on the patient’s age, tumor size, grade and ER status. 12 A recent evaluation of the

Adjuvant! software found that known clinical prognostic factors were able to predict overall

survival (OS), breast-cancer-specific survival, and event-free survival quite accurately in

4,083 patients diagnosed with breast cancer in British Columbia from 1989 to 1993, with

the exception of very young patients diagnosed under the age of 35.13 This independent

validation of the software reinforced its credibility as an accurate clinical tool to evaluate

breast cancer prognosis, making the ability of the 70-gene signature to outperform this

tool all the more notable.

The 70-gene signature remained a significant prognostic indicator of time to distant

metastasis and OS even after adjustment for all clinicopathologic factors known to have

prognostic value in this disease. The consortium decided that the low clinical risk group

would consist of patients with a 10-year breast cancer survival probability of at least 88% if

their tumors were 1% or greater positive for expression of ER using immunohistochemistry,

and of at least 92% if they were not. These two cutoffs were chosen to reflect the fact

that patients with ER-positive tumors now receive adjuvant endocrine therapy (with an

estimated absolute 10-year benefit of about 4% overall), whereas patients in the validation

series were all untreated regardless of their ER status. When adjusted for clinical risk based

on 10-year survival probability using the Adjuvant! software, the gene-signature adjusted

hazard ratios (Box 1) were 2.13 (95% CI 1.19–3.82) for time to distant metastasis, 2.66 (95% CI

1.46–4.84) for OS, and 1.36 (95% CI 0.91–2.03) for disease-free survival. Similar hazard ratios

were found in Cox multivariate regression analysis. These results indicate that the genesignature adds independent prognostic information to that provided by a risk assessment

based solely on clinicopathologic factors. Central pathology review of ER and tumor grade

and an independent source verification of all data by external auditors give these findings

significant strength. Furthermore, within each gene-signature risk group, the Kaplan–Meier

estimates of 10-year OS were almost identical to the two clinical risk groups as assessed by

the Adjuvant! software: patients classified as gene-signature low risk had 10-year survival

rates of 88% and 89%, respectively, for low and high clinical risk as defined by Adjuvant!,

while for patients classified as gene-signature high risk, the 10-year survival rate was 69%

for both clinical risk groups. The external and independent validation therefore confirmed the original findings that the

gene signature added significant independent prognostic information to that produced

by current clinicopathologic factors, and provided strong support for the initiation of the

MINDACT (Microarray In Node negative Disease may Avoid ChemoTherapy) trial.

8/10/2019 Stella Mook T3-108


Chapter 12

254

The MINDACT study

The MINDACT trial is an international prospective, randomized study comparing the

70-gene signature classifier with commonly used clinicopathologic criteria for selecting

node-negative breast cancer patients for adjuvant chemotherapy. The trial is intended to

address whether a tool such as this can improve on existing methods of risk assessment

and treatment decision-making by assisting oncologists to select between node-negative

women who need adjuvant chemotherapy and those who do not. We will discuss

the challenges that arose in incorporating this question into a suitable design for the

prospective clinical trial. Studies similar in purpose to MINDACT might become more

frequent in the future as more prognostic and/or predictive signatures require validation

before their use in clinical practice.

The need for a randomized trial

Given the available retrospective validation data of the 70-gene prognostic signature, is

there any need to perform a large randomized trial? Although the available validation data

are compelling, we believe that before being accepted as standard practice, new biological

diagnostic tools must go through the same strict validation process as, for example, a new

drug or treatment approach. While phase II results may be promising, a new therapy might

only become a standard of care after being evaluated in at least one large prospective

randomized phase III trial. This is especially true with this technology given its high cost- €2,000 per patient - and the complexity and costs associated with the collection of

frozen tumor samples. The 21-gene recurrence score, developed by the National Surgical

Adjuvant Breast and Bowel Project and Genomic Health, and based solely on retrospective

validations, is marketed under the name Oncotype DX® (Genomic Health, Inc., Redwood

City, CA) and has not yet been approved by the FDA. The scientific community shares our

belief that a full prospective validation must be performed before this tool can be accepted

as standard of care, and such a validation is about to start via a large phase III prospective

trial - Tailor x - which is a joint effort between several American groups, funded by the

National Cancer Institute. The design of Tailor x is quite similar to that of the MINDACT trial,and collaborations and discussions are underway between the two consortia utilizing the

two trials. Only this type of large, prospective, biologically driven phase III trial can provide

the necessary level 1 evidence (Box 1) for any new biological marker or tool.

The huge research efforts in the development of microarray-based gene signatures are

often weakened by restricted numbers of patients. The supervised analysis of expression

data of thousands of genes for a limited number of patients has well-known pitfalls.14,15

For the external validation data set of the 70-gene signature, while the hazard ratios were

smaller than the previously published series, the result was still more powerful than any

other available covariate for this data set, providing evidence that the prognosis provided

8/10/2019 Stella Mook T3-108



255

12

by the gene signature was robust and the technology reproducible. Nonetheless, as

retrospective patient series can also be biased by unknown factors leading to patient

selection, a prospective evaluation is vital. Furthermore, the use of the signature to guide

chemotherapy decision-making has not been tested by external validation, which dealt

only with the prognostic (as opposed to the predictive) value of the signature (Box 1).

It can also be argued that the patients selected for participation in a randomized trial will

be an investigator-selected subset of the population under consideration. This is indeed an

important point, and to evaluate selection bias during accrual an additional step has been

incorporated into the MINDACT trial design. Following enrollment of 800 patients (termed

the ‘pilot’ stage; see below), the data will be examined not only for logistical problems, but

also for potential bias of investigators and compliance with the randomization.

Some questions have arisen over whether it is too early to initiate this trial, given the

current rapid evolution of the technology, the previous methodological criticisms, and

the real possibility that the technology may be outdated at the completion of the trial.

Undoubtedly, there are and will be many predictive and prognostic signatures derived

from high-throughput technology reported in the future. It is clear that gene signatures

must be independently and externally validated before they proceed to prospective clinical

assessment and widespread use. Furthermore, like all diagnostic approaches, the ultimate

diagnostic gene signature may need refining as our knowledge advances. In our opinion,

prospective studies are the only way to provide level 1 evidence (Box 1) about the clinical

relevance of genomic signatures, and therefore the only way to endorse their widespread

clinical use, thereby allowing patients to benefit from these advances. Participation in thistrial should be strongly encouraged so that this issue can be addressed as soon as possible.

8/10/2019 Stella Mook T3-108


Chapter 12

256

Box 1. Common statistical terms relevant to theMINDACT trial.

Prognostic

Defines patient outcome based on overall survivalor relapse-free survival in a group of patients

independent of exposure to therapy

PredictivePredictive factors should define sensitivity of a

tumor to a distinct therapeutic agent

Hazard ratiosIn survival analysis, the hazard ratio is an indication

of the difference between two survival curves,

representing the reduction in the risk of death withtreatment compared with control, over the period

of follow-up; the hazard ratio is a form of relative

risk

Level 1 evidence

Evidence arising from a randomized controlledclinical trial

Specificity The percentage of patients with a negative testresult who were not diagnosed with malignancy

Sensitivity

The number of patients with a true positive testresult (positive test result and tumor) divided by

the

total number of patients diagnosed withmalignancy

Key Points

In the past 20 years, little progress has beenmade regarding new prognostic markers that

can assist oncologists in treatment decision-making for node-negative early-stage breast

cancer

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108


Chapter 12

258

axillary clearance. No patients treated with previous chemotherapy or radiotherapy will be

enrolled.

Determining the control arm

The trial objective is to prove that the 70-gene signature will safely assign fewer node-

negative patients to chemotherapy, and is directly related to the control criteria. For these

control criteria, two conditions are essential: the criteria should reflect current practice,

and they should be applied as homogeneously as possible in the trial. Satisfying both

conditions at the same time is a real challenge. Currently, oncologists decide whether

to prescribe chemotherapy according to several methods and guidelines, and it is

reasonable to assume that many modified versions of these criteria are being applied

in practice. Thus, there is no straightforward way of deciding on one set of rules as the

standard chemotherapy assignment method. One practical approach could be to allow the

participating investigational sites to apply their own predefined set of rules to represent

the current standard. Such an approach, however, would lead to considerable variability,

and the rate of chemotherapy assignment of such ‘standard’ criteria would depend on

the accrual at each participating site, rather than on the characteristics of the population

studied.

As the Adjuvant! software uses information from the San Antonio database, the SEER

(Surveillance, Epidemiology and End Results) database, the Overviews of clinical trials,

individual clinical trial results, and the literature in general, it is considered appropriatefor analysis of available patient prognosis data. It should be noted that the whole risk

assignment method is largely prognosis-based, in contrast to other attempts to base

the method on predictions of chemotherapy effect. Calculation of prediction has only

been applied to some level for ER status, where the method acknowledges the effect of

endocrine therapy in ER-positive patients, and the possibly greater effect of chemotherapy

in ER-negative patients, leading to a sizable benefit even for good-prognosis ER-negative

patients.

Also of note is the fact that the current version of the Adjuvant! software does not include

HER2 status, a marker that many believe has considerable data supporting its prognosticvalue.20 In addition, preliminary data indicate that HER2 status may have important

predictive implications particularly related to endocrine therapy.21 To tackle this issue, a

new version of the Adjuvant! software is being developed that will incorporate HER2

status. An additional consideration for HER2-positive patients is the need to administer

adjuvant trastuzumab, which, with its very important efficacy results, could interfere with

the detection of any difference between the chemotherapy and the endocrine therapy

arms of the trial. How this effect will influence treatment decisions will be evaluated in the

pilot phase of the first 800 patients.

8/10/2019 Stella Mook T3-108



259

12

Design of the trial

The initial concept of the trial design was straightforward. Each enrolled node-negative

patient would be randomized to chemotherapy treatment decision according to either

clinicopathologic criteria (control) or gene signature (experimental). The trial would then

aim to prove that a lower rate of chemotherapy administration in the experimental arm did

not result in inferior efficacy. With this approach, only half of the patients would need to

have their microarray analysis performed on a real-time basis.

There were two major and interrelated objections to this design. The first objection was

that, while this design tested the two approaches (experimental and control) against each

other in an overall fashion, it did not take into account the fact that for more than half

of the patients both approaches were in agreement. From a methodological perspective,

it is clear that any benefit of either approach would be greatly diluted in such overall

comparison, because a majority of patients could achieve the same result regardless of the

arm of randomization. Additionally, since it is impossible to ‘blind’ the investigators to the

clinicopathologic prognostic factors, in practice this design would actually compare the

combination of the methods with the clinicopathologic risk assessment alone because of

selection bias. This issue is related to the second limitation of this approach. Apart from the

discussion of defining an appropriate noninferiority threshold (delta), for the time being,

let us say we want to reject an inferiority null hypothesis (H 0) on the overall population

of H 0 = θE ÷ θC ≥ 1.25, where θE and θC represent the experimental and control hazards,

respectively, for some time to event endpoint (DMFS or OS). For any reasonably sized trial,it would be very hard to come up with a credible scenario wherein such a noninferiority

test would be likely to fail. In a noninferiority testing situation in which one can assume a

hazard ratio between two randomized arms, a one-sided 95% confidence noninferiority

test for the above null hypothesis would require about 512 events to perform the analysis

with 80% power. We performed simulations for such a trial and primary test, and showed

that in a situation in which the clinical criteria would perform very well (identifying

90% of patients who will metastasize), and the gene signature would select patients for

chemotherapy at random, the study would still yield powers of up to 50%. For less extreme,

but equally unacceptable situations, in which the gene signature should be identified asinferior, the power of ‘proving’ noninferiority would be even higher.

A trial using an overall (i.e. using all patients) noninferiority test would need to have an

inferiority threshold hazard ratio of at least 0.90, and probably at least 0.95, to convincingly

exclude performances of a gene signature that would still be considered to be very

poor. Thus, a noninferiority trial that would reliably exclude the poor performance of a

gene signature would need to be huge or of extremely long duration. Even under such

conditions, it is likely that a scenario of underperforming gene tests that would yield

reasonably nice hazard ratios in an overall comparison would arise. The addition of a large

fraction of observations that are the same irrespective of the arm they are randomized

8/10/2019 Stella Mook T3-108


Chapter 12

260

to makes any equivalence test procedure highly suspect. Such observations increase the

likelihood of ‘random noise’, which makes statistically significant rejection of inferiority

more likely. Such a situation, in general, is not acceptable for equivalence or noninferiority

testing.22 This design, therefore, was not selected, and other possibilities were considered.

Two other options that could be considered were the following: to assign chemotherapy

according to the 70-gene classifier risk model while assessing the clinical risk, or to assign

chemotherapy according to clinical criteria while assessing the 70-gene risk. The first

option is a much too big leap forward and the second would not test the 70-gene signature

appropriately.

As an outcome of the above considerations, attention and discussion started focusing on

discordant patients, and considered such patients as the core group. Using the original

approach, this group would have been identifiable only in patients randomized to use

the gene signature, as the signature would not have been performed for the other arm. It

became clear that one should perform the gene signature on all patients entered into the

trial, in order to identify all patients who would be treated differently by the two methods.

This group of patients would consist of two distinct subgroups: those who are at low

risk according to the gene signature and high risk according to clinicopathologic criteria

(stratum A), and those who are at high risk according to the gene signature and low risk

according to clinicopathologic criteria (stratum B; Figure 1). Having a lower chemotherapy

assignment rate with the gene signature is equivalent to saying that stratum A should be

larger than stratum B, because more patients would be at low risk according to the gene

signature.Since the main objective of the trial is to put the gene signature to the test, we concluded

that its design should randomize patients who have discordant risk assessments to one

of the two methods to be used for chemotherapy decision-making. In fact, such a course

is equivalent to randomizing such patients to receive chemotherapy or not. For the

other patients, who have either both risk assessments as high risk, or both as low risk, a

randomization for this type of risk assessment will not have a true value. These patients,

therefore, will not be randomized, but will be treated or not with chemotherapy according

to the concordant risk assessments, and followed further. All patients with hormone-

sensitive disease will also receive endocrine therapy.

8/10/2019 Stella Mook T3-108



261

12

Figure 1. Randomization of discordant cases in the MINDACT trial.

Abbreviation: CT, chemotherapy.

Sample size

At the time of the first discussions regarding the MINDACT trial design, the noninferiority

design was proposed and a sample size of 5,000 was envisaged, with microarray analysis

performed on half of these patients. As discussions evolved, it was decided that only the

discordant cases would be randomized to treatment decision-making using either thegenomic tool or the clinicopathologic criteria. In order to be able to formulate and answer

questions with regard to the core group of discordant cases, the sample size was increased

from 5,000 to 6,000 patients.

Testing

From the available data on the gene signature and Adjuvant! software, there is a strong

belief that the gene signature will produce a larger fraction of patients designated as low

risk than the clinicopathologic method described above. Assignment fractions accordingto both methods will be evident from the trial design, and both methods will be available

for all patients, so paired data are available for analyzing this endpoint. After the preceding

discussions, it would be fair to conduct a noninferiority test on the selected group of

patients with discordant risk data. Unfortunately, with an expected proportion of discordant

patients of 30–35%, the number of patients needed for the trial would become prohibiting,

particularly considering the high costs associated with the microarray technology and the

complex trial logistics.

Let us revisit the question of what the real objective of such a trial is. We can first clarify what

the objective is not. It is not a trial to find the fractions of high-risk and low-risk patients

Randomized:

decide with clinical tool

Randomized:

Decide with genomic tool

High clinical risk

(low genomic risk)

Low clinical risk

(high genomic risk)

High genomic risk

(low clinical risk)

Low genomic risk

(high clinical risk)

A1: CT

B1: no CT

B2: CT

A2 no CT

Discordant cases-randomized

Stratum A: genomic low, clinical high

Stratum B; genomic high, clinical low

8/10/2019 Stella Mook T3-108


Chapter 12

262

according to the gene signature, because these can be found with an observational trial.

In this context, an observational trial would involve treating the patients according to the

established clinicopathologic guidelines and the patients’ genomic risk, but not making a

decision based on the treatment outcome. Indeed, for this endpoint, the present trial will

function as a very large prospective observational trial. Also, it is not a trial to determine

the effect of chemotherapy in specific subgroups. Such an issue may be of interest, but it is

not the primary goal of the trial to address it. A priori assumption (before having the outcome

of the trial) may be that chemotherapy does have an effect (in terms of a hazard ratio)

for some of the patients that have a good gene signature, but that for these patients the

prognosis is so good that it is not acceptable to treat them all with a toxic treatment. This

is the same rationale whereby not all node-negative patients are given chemotherapy. In

mathematical terms, the hazard ratio may still be different from 1 (and may be more or less

constant for the whole population), but because of the small event rate in an identifiable

good-prognosis group, the absolute effect is outweighed by the acute and long-term

toxicity of chemotherapy.

If we consider a gene signature (or some other set of criteria) as a diagnostic test to detect

those patients who will have recurrent disease that can no longer be treated with curative

intent (i.e. metastasis), we can discuss its performance in terms of specificity and sensitivity

(Box 1). We cannot expect the gene signature to be perfect (i.e. to have 100% sensitivity and

specificity), but we can try to prove that it is good enough in the clinical situation—that is,

sufficient to prevent undertreatment of patients, which relates to the sensitivity of the gene

signature. To address this requirement, we incorporated the following primary test intothe trial design. In the set of patients who have a low-risk gene-signature prognosis and

high-risk clinicopathologic criteria, and who will be randomized to use the gene-signature

prognosis and thus receive no chemotherapy (group A2 in Figure 1), a null hypothesis of a

5-year DMFS of 92% will be tested. With 6,000 patients accrued overall, and based on the

available validation data estimates, this set has an expected size of 672 patients. With an

accrual of 3 years, and a total duration of 6 years ( i.e. 3–6 years’ follow-up for each patient),

a one-sided test at 97.5% confidence level has 80% power to reject this hypothesis if the

true 5-year DMFS is 95%.

The major criticism of this primary test is that it is not a test that compares the randomizedgroups. If the above test is statistically significant, however, and the gene signature does

select fewer patients to be treated with chemotherapy while not adversely affecting DMFS,

then this can be taken to be equivalent to proving that the signature has a very good

sensitivity, as well as a specificity that is better than the clinicopathologic method. In our

opinion, therefore, a significant primary test as described above would establish the role

of such a signature in chemotherapy treatment decision-making in node-negative breast

cancer patients.

8/10/2019 Stella Mook T3-108



263

12

The pilot phase

All patients must have an available and high-quality frozen tumor sample to be eligible

for the MINDACT trial. The desired number of patients might become impractical because

of logistics problems such as transport problems, insufficient material, insufficient quality

of RNA, and so on. This issue is currently being assessed in a Pilot Logistics Study, run in

seven centers in seven different European countries. Additionally, all these issues and all

the assumptions made will be assessed and corrected if necessary during the pilot phase

of MINDACT, composed of the first 800 patients.

This pilot phase should ensure that the complex logistical framework put in place for

this trial is feasible for the patients, physicians and laboratories involved. Assessment by

Agendia of the quality and quantity of RNA samples will be part of this first phase. The

second aim of the pilot phase is to ascertain that the patient population recruited upfront

for MINDACT is not a biased one. This will be done by checking whether the ratio of low-

risk to high-risk patients is as expected. Third, there is a concern that clinicians might not

comply with the randomization of the treatment decision, violating the protocol. A patient

who is clinically high-risk and has a low genomic risk, who is randomized by use of the

genomic tool, should not be given chemotherapy. By contrast, a patient who is clinically

low-risk and has a high genomic risk and is randomized to decision with the genomic tool

should be given chemotherapy. The compliance with this randomization is to be assessed

in the pilot phase. Following the treatment randomization, there will be two additional

randomizations: chemotherapy and endocrine therapy. The fourth objective of the pilotphase is to check whether at least 66% of those women assigned to chemotherapy are

subsequently randomized in the chemotherapy question. The final aim of the pilot study

is to ensure that there is a statistically significant difference between the percentage of

patients that have a high clinicopathologic risk and those with a high genomic risk, thus

reflecting the expected reduction in chemotherapy administration.

Further randomizations

Since the primary randomization is the most complex and innovative part of this trial, thebulk of this article has focused on this element, but it should be noted that the trial will

also have two further randomizations. Patients who are to receive chemotherapy may be

randomized to receive either an anthracycline-based regimen or a docetaxel–capecitabine

regimen.23,24 This randomization (designated R-C) will ask whether a docetaxel–

capecitabine regimen can safely replace an anthracycline-based regimen in high-risk

node-negative women, with the potential advantage of a reduction in the two long-term

toxicities associated with anthracyclines: cardiac toxicity and secondary leukemia. The

docetaxel–capecitabine combination is currently being evaluated in the adjuvant setting

in the US. As short-term toxicity is of some concern with this regimen, in MINDACT the

8/10/2019 Stella Mook T3-108


Chapter 12

264

first 40–120 patients randomized to this treatment will be closely monitored. Several

commonly used anthracycline-based regimens will be accepted within the trial that have

adequate anthracycline dose intensity, three or more drug combinations, and six cycles of

administration.

Patients eligible for endocrine therapy can participate in the endocrine therapy

randomization stage of the trial (designated R-E), which consists of a randomization to 7

years of letrozole, or to 2 years of tamoxifen followed by 5 years of letrozole.25–27 There will

be stratifications for HER2 status, ER-positive and/or progesterone-negative, and gene-

signature risk. These further randomizations will answer clinically relevant questions by

taking advantage of the power that the large sample size used in MINDACT offers. As the

associated biological material will be collected, there will be also ample opportunity to

develop and identify predictive gene signatures, as well as important genes and proteins

influencing response to administered agents (Figure 2).

Figure 2. Biological material flowchart in the MINDACT trial. (1) Tissue for RNA extraction. (2) Anyremaining tissue and RNA. (3) Paraffin blocks for central pathology review and tumor microarray

production. (4) Serum. (5) Tissue and/ or serum for proteomics.

Abbreviations: EIO, European Institute of Oncology; TBMB, TRANSBIG biological materials bank;

Univ Wales; University of Wales.

Center A

Center B

Center C

Agendia

TBMB

EIO

5

Univ Wales

1

3

4

2

8/10/2019 Stella Mook T3-108



265

12

Predictive power of the 70-gene signature

At this time, there are no data on the predictive power of the 70-gene signature. It would be

interesting to have predictive power data on anthracycline–taxane-based chemotherapy,

tamoxifen, and aromatase inhibitors, because these are the treatments administered

in the MINDACT trial. The European Organisation for Research and Treatment of Cancer

(EORTC) Breast Cancer Group is now undertaking a project in which the 70-gene signature

is being evaluated in some of the patients enrolled in the EORTC 10994/BIG 00-01 ‘p53’

trial.28 The p53 trial randomized patients with locally advanced or large operable tumors

to one of two neo adjuvant chemotherapy regimens: six cycles of fluorouracil, epirubicin

and cyclophosphamide (epirubicin 100 mg/m2) or three cycles of docetaxel (100 mg/

m2) followed by three cycles of docetaxel (75 mg/m2) combined with epirubicin (90 mg/

m2). Results are not yet available, but the use of standardized chemotherapy regimens

and the availability of enough good-quality frozen material will probably yield data on the

predictive power of the signature. These results will help adjustment of estimates for the

MINDACT trial, if needed.

The MINDACT trial will provide the setting for prospectively assessing the predictive power

of the 70-gene signature and any other signatures currently being developed in response

to the chemotherapy regimens and endocrine therapy used.

Logistics

The logistics of MINDACT have been one of the most challenging and expensive parts of the

trial. All RNA extraction, quality control and microarray analysis for samples in this trial will

be performed at Agendia in Amsterdam. Indeed, at this stage this technology is probably

too immature for even the RNA extraction to be performed in external laboratories; operator

and technical variability is well known to influence the results of microarray experiments.

Upon diagnosis of a clinically node-negative invasive breast cancer, patients are eligible to

sign the first informed consent prior to surgery to allow for a sample of frozen tumor tissue

obtained during surgery to be sent to Agendia for RNA extraction (screening informed

consent). At this time, only RNA extraction and quality control check will be done. Oncethe local pathology report confirms node negativity, the genomic risk assessment will

be performed. This process will hopefully avoid much unnecessary hybridization and

hence reduce the cost. All tumor specimens will be couriered to Agendia by a specifically

contracted courier agent specialized in global express transport and storage at –80 °C. If a

patient is ineligible for the trial, her tumor material will be returned, stored in the TRANSBIG

biological materials bank, or destroyed.

Simultaneously, the investigator responsible for patient care will assess the

clinicopathologic risk using the Adjuvant! software embedded in a web-based platform

designed specifically for the MINDACT randomization. In Informed consent form 1, the

8/10/2019 Stella Mook T3-108


Chapter 12

266

patient will consent to have her risk assessed by the genomic and clinicopathologic

methods and to enter randomization for treatment (R-T) if she belongs to the discordant

group (Figure 3). If the patient is assigned to receive chemotherapy, she will be proposed to

enter the chemotherapy randomization (R-C) and to sign the Informed consent form 2.

After chemotherapy (if applicable) and radiotherapy, the endocrine therapy randomization

(R-E) will be proposed to endocrine responsive patients in the Informed consent form 3.

Paraffin tumor blocks will be sent every 6 months to the European Institute of Oncology in

Milan, for Construction of tissue arrays and for central pathology review to be performed.

Proteomic analysis of the tumor and serum samples will also be performed in the MINDACT

trial, in collaboration with the University of Wales in Aberystwyth. Additionally, since one

of the aims of MINDACT is to create a biological materials bank, frozen tumor samples (as

well as whole genome microarray data and paraffin-embedded tissue) will be collected for

all patients. The TRANSBIG biological materials bank will be located in Brussels under the

guardianship of TRANSBIG, and hence this trial will have great potential for the identification

and validation of additional gene signatures with prognostic and predictive value in early

breast cancer, as well as other markers and technologies. Figures 2 and 3 summarize some of

the logistics involved in this trial.

Figure 3. Logistics of MINDACT—tumor biopsy collection, shipment, RNA extraction and

eligibility check.

Abbreviations: ICF1, informed consent form 1; ICF2, informed consent form 2; R-T, randomization

to treatment.

Unicentric node-negative tumor

ICF1 Donation of biopsy to research

Surgery

Shipment to Agendia

RNA extractionLocal pathology

Node-positive Poor-quality RNA

Microarray analysis

ICF2 R-T

IneligibleIneligible

Node-negative Good-quality RNA

8/10/2019 Stella Mook T3-108


8/10/2019 Stella Mook T3-108


Chapter 12

268

References



2. Hutchins L , Green S, Ravdin P, et al. CMF versus CAF with and without Tamoxifen in high-risk node-

negative breast cancer patients and a natural history follow-up study in low-risk node-negative

patients: First results of Intergroup trial INT 0102 [abstract #2]. Proc Am Soc Clin Oncol 1998; 17: a1

3. McShane LM, Altman DG, Sauerbrei W, et al. REporting recommendations for tumor MARKer prognostic

studies (REMARK). Nat Clin Pract Oncol 2005; 2: 416-422.

4. National Institutes of Health Consensus Development Conference Statement: Adjuvant Therapy for

Breast Cancer, November 1-3, 2000. J Natl Cancer Inst 2001; 93: 979-989.

5. Goldhirsch A, Glick JH, Gelber RD, et al. Meeting highlights: International expert consensus on the

primary therapy of early breast cancer 2005. Ann Oncol 2005; 16: 1569-1583.

6. D’Eredita’ G, Giardina C, Martellotta M, Natale T, Ferrarese F. Prognostic factors in breast cancer: the

predictive value of the Nottingham Prognostic Index in patients with a long-term follow-up that were

treated in a single institution. Eur J Cancer 2001; 37: 591-596.


cancer. Nature 2002; 415: 530-536.



9. Piccart MJ, Loi S, Van ‘t Veer LJ, et al. Multi-center external validation study of the Amsterdam 70-gene

prognostic signature in node negative untreated breast cancer: are the results still outperforming the

clinical-pathological criteria? [abstract] Breast Cancer Res Treat 2004; 88 (Suppl 1): aS38.

10. Cardoso F et al. Multi-centre independent validation of a gene prognostic signature for patients with

node-negative breast cancer (BC)—final results [abstract]. Breast Cancer Res Treat 2005; 95 (Suppl 1): aS301.

11. Adjuvant! Online [http://www.adjuvantonline.com].





14. Simon R, Radmacher MD, Dobbin K, McShane LM. Pitfalls in the Use of DNA Microarray Data forDiagnostic and Prognostic Classification. J Natl Cancer Inst 2003; 95: 14-18.

15. Dudoit S et al. Comparison of discrimination methods for the classification of tumors using gene

expression data. JASA 2002; 97: 77–87.

16. De Vita Jr VT et al. (Eds; 2006) Cancer. Principles and Practice of Oncology. Philadelphia: Lippincott

Williams & Wilkins.

17. Cardoso F et al. Multi-centre independent validation of a gene prognostic signature for patients with

node-negative breast cancer (BC)—final results [abstract]. Breast Cancer Res Treat 2005; 95 (Suppl 1): aS301.

18. Buyse M, Loi S, Van ‘t Veer L, et al. Validation and clinical utility of a 70-gene prognostic signature for


8/10/2019 Stella Mook T3-108



269

12



2005; 365: 1687-1717.

20. Ross JS, Fletcher JA, Linette GP, et al. The Her-2/neu gene and protein in breast cancer 2003: biomarker

and target of therapy. Oncologist 2003; 8: 307-325.

21. Smith IE, Dowsett M, Ebbs SR, et al. Neoadjuvant treatment of postmenopausal breast cancer with

anastrozole, tamoxifen, or both in combination: the Immediate Preoperative Anastrozole, Tamoxifen,

or Combined with Tamoxifen (IMPACT) multicenter double-blind randomized trial. J Clin Oncol 2005; 23:

5108-5116.

22. ICH Harmonised Tripartite Guideline: statistical principles for clinical trials. International Conference on

Harmonisation E9 Expert Working Group. Stat Med 1999; 18: 1905–1942.

23. O’shaughnessy J, Miles D, Vukelja S, et al. Superior survival with capecitabine plus docetaxel

combination therapy in anthracycline-pretreated patients with advanced breast cancer: phase III trial

results. J Clin Oncol 2002; 20: 2812-2823.

24. Jones S et al. Three year results of a prospective randomized trial of adjuvant chemotherapy for

patients (pts) with stage I-III operable, invasive breast cancer comparing 4 courses of doxorubicin/

cyclophosphamide (AC) to 4 courses of docetaxel/ cyclophosphamide (TC) [abstract]. Proc Am Soc Clin Oncol

2003; a59.

25. Goss PE, Ingle JN, Martino S, et al. A randomized trial of letrozole in postmenopausal women after five

years of tamoxifen therapy for early-stage breast cancer. N Engl J Med 2003; 349: 1793-1802.

26. Coombes RC, Hall E, Gibson LJ, et al. A randomized trial of exemestane after two to three years of

tamoxifen therapy in postmenopausal women with primary breast cancer. N Engl J Med 2004; 350: 1081-

1092.

27. Thurlimann BJ et al. BIG 1–98: randomized double-blind phase III study to evaluate letrozole (L) vs.

tamoxifen (T) as adjuvant endocrine therapy for postmenopausal women with receptor-positive

breast cancer [abstract #511]. Proc Am Soc Clin O ncol 2005; 23: 16S.

28. Bonnefoi H, ebold-Berger S, Therasse P, et al. Locally advanced/inflammatory breast cancers treated

with intensive epirubicin-based neoadjuvant chemotherapy: are there molecular markers in the

primary tumour that predict for 5-year clinical outcome? Ann Oncol 2003; 14: 406-413.

8/10/2019 Stella Mook T3-108


Stellingen

behorend bij het proefschrift

Prognostic factors in breast cancerOne fits all?

1. Voor optimale moleculaire analyse van een carcinoom zou het invriezen

van vers tumorweefsel routine moeten zijn Dit is logistiek haalbaar (o a dit

Stella Mook T3-108

Documents

Transcript of Stella Mook T3-108