Proefschrift Uijtendaal

M

onitorin

g drug th

erapy in

hosp

italized patien

ts Esth

er Uijten

daal 2014

Monitoring drug therapy in hospitalized patients

Esther Uijtendaal

UITNODIGING

Voor het bijwonen van deopenbare verdedigingvan het proefschriftt

Monitoring drug therapy in hospitalized

patients

Dinsdag 2 december 2014om 16:15 uur

in het Academiegebouw,Universiteit Utrecht

Domplein 29 te Utrecht

Feestelijke borrel na afloop van de promotie in het

Academiegebouw

Esther UijtendaalOverboslaan 20

3722 BL [email protected]

Paranimfen:

Marcel [email protected]

06-52803009

Hanneke den [email protected]

06-52097419

M

onitorin

g drug th

erapy in

hosp

italized patien

ts Esth

er Uijten

daal 2014


Esther Uijtendaal

UITNODIGING



patients





Academiegebouw



Paranimfen:


06-52803009


06-52097419

Cover design by: Marieke Velthuis

Lay-out by: Gildeprint – The Netherlands

Printed by: Gildeprint – The Netherlands

The work in this thesis was performed at the department of Clinical Pharmacy in

close collaboration with the department of Clinical Chemistry and Haematology

(Division of Laboratory and Pharmacy) and the department of Intensive Care

(Division of Vital Functions) of the University Medical Center Utrecht, the

Netherlands.

E.V. Uijtendaal


Thesis Utrecht University – with ref. – with summary in Dutch

ISBN/EAN: 978-90-393-6237-2

©2014 E.V.Uijtendaal


Het monitoren van geneesmiddeltherapie bij in het ziekenhuis opgenomen patiënten

(met een samenvatting in het Nederlands)

PROEFSCHRIFT

ter verkrijging van de graad van doctor aan de Universiteit Utrecht

op gezag van de rector magnificus, prof.dr. G.J. van der Zwaan,

ingevolge het besluit van het college voor promoties

in het openbaar te verdedigen op

dinsdag 2 december 2014 des middags te 4.15 uur

door

Esther Véronique Uijtendaalgeboren op 1 april 1967

te ‘s-Hertogenbosch

Promotoren: Prof. dr. A. C. G. Egberts

Prof. dr. W.W. van Solinge

Copromotor: Dr. J.E.F. Zwart- van Rijkom

Voor mijn ouders

Table of Contents

Chapter 1 General introduction, aims, and outline of the thesis 9

Chapter 2 Drug-drug interactions in hospitalized patients

2.1 Frequency and nature of drug-drug interactions in a Dutch 21

university hospital

2.2 Analysis of potential drug-drug interactions in medical 37

intensive care unit patients

Chapter 3 Monitoring drug therapy in hospitalized patients

3.1 Frequency and determinants of laboratory measurements for 53

serum potassium, sodium and serum creatinine in hospitalized

patients

3.2 Frequency of laboratory measurement and hyperkalaemia 73

in hospitalised patients using serum potassium concentration

increasing drugs

3.3 Serum potassium influencing interacting drugs: 91

Risk-modifying strategies also needed at discontinuation

3.4 Influence of a strict glucose protocol on serum potassium 103

levels in intensive care patients

Chapter 4 General discussion and summary 123

Samenvatting 151

List of co-authors 159

List of publications 161

Dankwoord 163

Curriculum vitae 165

1General introduction, aims and

outline of the thesis

Chapter 1 | General introduction, aims, and outline of the thesis

10

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

11

1R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Background

Medical errors have increasingly been acknowledged as a major public health

problem, since these errors received worldwide attention due to the publication

of the US Institute of Medicines report “To Err is Human: Building a Safer Health

System”1. The report describes that adverse events (AEs), defined as injuries due

to medical management (rather than the underlying disease), occur in 2.9 - 3.7

percent of all hospitalized patients in the US and that AEs lead or significantly

contribute to death in 6.6-13.6% of these patients2,3,4. The leading cause of AEs was

the use of drugs, accounting for 19.4% of the AEs3. Based on these findings, Bates

et al. started research on the relation between the use of drugs, AEs and their

preventability. Bates defined adverse drug events (ADEs) as injuries resulting from

medical intervention related to a drug, encompassing adverse drug reactions (i.e.

noxious and unintended effects resulting from appropriate use of drugs) as well

as medication errors (i.e. adverse drug events resulting from inappropriate drug

use). He reported that ADEs and preventable ADEs occurred in 6.5% and 5.5% of the

hospitalized patients respectively and that 28% of all ADEs were preventable5. The

preventable ADEs (i.e. medication errors) were also associated with an additional

length of stay of 4.6 days. The post event costs attributable to a preventable ADE was

estimated at $46856.

Also ADEs occurring in ambulatory care contribute to the medical error public

health problem. The Dutch HARM (Hospital Admissions Related to Medication)

study showed for example that 5.6% of all unplanned admissions were medication

related and that 46.5% of these were judged preventable7. Of the patients admitted

due to a preventable medication related problem, 6.3% died and 9.3% experienced

a disability after discharge.

The available evidence indicating that an important part of health damage and

associated costs due to ADEs can be prevented, led to awareness among health

care professionals, as well as policy makers, that strategies have to be developed

to prevent these ADEs and associated medical consequences and costs. In order to

develop better strategies for the prevention of ADEs, both causes and the process

of pharmaceutical care (figure 1) were analyzed8,9. The most frequently identified

underlying causes of ADEs are lack of dissemination of drug knowledge and

inadequate and timely availability of relevant patient information such as laboratory

test results9. Within the process of pharmaceutical care, errors associated with

preventable ADEs most frequently occur during the prescription and monitoring

stages8. Monitoring errors were defined as inadequate monitoring (biomarker or

clinical) of drug therapies or a delayed response or failure to adequately respond to

signs and symptoms or laboratory evidence of drug toxicity.


12

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

prescribingtranscription &verification

dispensing administration& use

monitoring evaluation

Figure 1: the stages of the pharmaceutical care process

As such, risk mitigation strategies to prevent ADEs should focus primarily on

the prescribing and monitoring stage and facilitate the dissemination of drug

knowledge and the availability of patient information.

Drug-drug interactionsWithin the stages of prescribing and monitoring, drug-drug interactions (DDIs)

are of special interest as ADEs resulting from DDIs are often predictable and

preventable10,11. ADEs resulting from DDIs may, for example, be prevented if DDIs

are recognized timely. Generation of alerts by a computerized order entry (CPOE)

system at stage of prescribing may be helpful in this. Prescribers indeed believe that

DDI alerting improves their prescribing, especially when these alerts are evidence-

based12. The Dutch G-standard is such an evidence based DDI guideline which is

implemented in most CPOE systems of Dutch hospitals13. The G-standard does

not only support the alerting of potential DDIs, but also provides comprehensive

background information on the mechanism and level of evidence of the interaction

and advises for mitigation of the patient risk related to the interaction for the

individual patient. This risk usually consists of an increased risk of side effects,

and less often a risk of a decreased efficacy. Data, however, on both occurrence

in hospitals and clinical consequences for patients are scare. Several studies have

investigated the frequency and nature of DDIs in community setting14,15,16, but

results from these studies cannot be simply extrapolated to the hospital setting

as patients, medications and monitoring practices may differ. Risk mitigation

strategies that are often advised are discontinuation/substitution of one of the

drugs, dose adjustment, adding a risk modifying drug or monitoring biomarkers17.

However, it was shown that physicians override the majority of the alerts18. It is not

known, whether prescribers did follow the advice of the risk modifying strategy

before overriding the alert. It is therefore unknown if patients are still at risk for

an ADE resulting from these DDIs. To determine the most effective risk mitigation

strategies for prevention of ADEs resulting from DDIs in hospitalized patients, it is

of interest to investigate which DDIs are responsible for most of the alerts.

Within the hospital setting, patients admitted to the Intensive Care Unit (ICU)

constitute a special vulnerable group. Compared to patients admitted to general

hospital departments, ICU patients are more severely ill and their capacity to

counteract and deal with (patho)physiological disturbances is often diminished19,20.

Moreover, the medications used are often classified as “high-risk”21, which means

13

1R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

that drugs may have a small therapeutic index. Medications involved in DDIs on

the ICU differ therefore from medications involved in DDIs of the general hospital

setting. Consequently, ADE resulting from these DDIs may also differ. The high

technological environment of the ICU, however, enables caregivers to monitor

patients intensively and detect disturbances earlier. Moreover, due to the parenteral

access for medications, deviations can also be corrected faster. Despite, Cullen et al

found that ADEs were likely to be more severe in ICUs than in non-ICUs22, although

the duration of the injury was short. To what extent DDIs contributed to these ADEs

was not reported.

Monitoring drug therapy Monitoring drug therapy is a process of checking in time to see whether a drug

works, while protecting the patient from adverse drug effects23. Monitoring

may consist of clinical monitoring and/or biomarker monitoring. Biomarker

monitoring may support clinical monitoring when treatment goals cannot be

directly observed. This biomarker monitoring may consist of measurements such

as physical parameters like blood pressure and ECG or measurement of laboratory

parameters like drug concentrations or endogenous substances such as potassium.

To mitigate drug related problems, drug labels often advise to measure laboratory

markers such as international normalized ratio (INR), serum creatinine and

serum electrolytes23,24. Drug labels, however, often lack clarity on monitoring. Of

all drug labels that advise to monitor laboratory values, only 17% of the provided

information on ‘what to monitor’, ‘critical value’ and ‘how to respond’24,25. This

lack of clarity may hamper the adherence to monitoring advices. Raebel showed

that the proportion of drug dispensing’s without all recommended laboratory

monitoring varied from 29-46% in ambulatory care26. Comparable rates were found

in vulnerable patient groups, kidney function for example, was determined in only

66% of the elderly outpatients patients with a renally cleared cardiovascular drug 24.

Rates do not seem to be better in hospitalized patients. Drenth et al. showed that

serum creatinine levels were measured in only 59% of the hospitalized elderly

patients, although this rate was not specific for patients using renally cleared

drugs27.

Inadequate monitoring may have serious consequences for patients. For example,

inadequate monitoring of drug therapy was an important cause of drug related

hospitalizations7,28,29. The exchange of information on a limited set of seven

laboratory markers, e.g. drug serum levels, INR, serum creatinine, potassium,

sodium levels and pharmacogenetic markers, therefore became obliged by the

Dutch law in January 201230. Due to practical implementation problems this law

was amended again in July 2013, now only demanding to communicate laboratory


14

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

values on kidney function when these deviate from normal limits31. Whether

health care providers adhere to this obligation, however, is not known.

Measurement of laboratory markers is not only frequently advised in drug labels to

evaluate drug therapy, but is also one of the most frequently advised risk mitigation

strategy of DDIs. It was shown that 33% of all clinical significant drug-drug

combinations advise to monitor laboratory makers, and that at least one laboratory

test was required in 37% of the patients with a DDI32. Tests on renal function and

serum potassium were most frequently advised as preventive measures in DDIs,

serum creatinine accounted for 42% and serum potassium for 20% of the tests.

Of these markers, serum potassium has special attention in medication management.

Not only is serum potassium one of the most frequently involved laboratory markers

in DDIs, but the clinical consequences of deviating serum potassium levels can also

be severe. For example, serious hyperkalemia may occur at any time in patients

treated for heart failure with spironolactone combined with an angiotensin

converting enzyme inhibitor and evolve rapidly to cause lethal arrhythmia. It was

shown that patients admitted to the hospital with hyperkalemia were twenty times

more likely to use an ACE inhibitor combined with spironolactone than when the

ACE inhibitor was not combined with spironolactone10. Moreover, it was shown that

publication of the Randomized Aldactone Evaluation Study (RALES) that advised

to add spironolactone to standard treatment of patients with severe heart failure,

led to an increased rate of prescription of spironolactone but also to an increased

rate of hyperkalemia, and associated mortality33. Whether serum potassium levels

of these patients were adequately monitored is not known, but literature suggests

that adherence to monitoring guidelines for serum potassium increasing drugs are

generally poorly met34,35. Monitoring of serum potassium levels therefore deserves

extra attention.

Thesis objectivesThe objective of this thesis is to describe the frequency and potential clinical

relevance of drug therapy monitoring with laboratory markers in hospitalized

patients with a special focus on DDIs, potassium and ICU patients.

Thesis outlineChapter 2 describes the frequency and nature of DDIs in both hospitalized patients

and patients admitted to a general intensive care unit (ICU). Hospitalized patients

are considered as a more vulnerable patient group as compared to outpatients, they

may be more severely ill and their capacity to counteract with disturbances may be

diminished. The frequency and nature of DDIs occurring in hospitalized patients

may differ from DDIs occurring in outpatients as different medications are used

15

1R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

and drug therapy may change more frequently (chapter 2.1). Special attention is

drawn to ICU patients (chapter 2.2). Due to the technological environment, ICU

patients are intensively monitored and management advices may not have the

same relevance as for patients admitted to general hospital departments. Insight in

the frequency of DDIs and the most frequently advised management strategies, may

lead to customization of DDI signaling for these specific vulnerable patient groups.

Chapter 3 focuses on laboratory monitoring in hospitalized patients. In chapter

3.1, the frequency of laboratory monitoring of serum potassium, serum sodium

and serum creatinine values is described for different kinds of hospitalized patient

groups.

DDIs are recognized as a risk factor to develop an ADE. The DDI between two or

more serum potassium increasing drugs, for example, is recognized as a risk factor

to develop life threatening hyperkalemia. The DDIs between serum potassium

increasing drugs are therefore included in the G-standard and computerized

physician order entry systems (CPOEs) using this standard will signal and advice to

monitor laboratory values when this DDI occurs. It is not known whether physicians

adhere to the monitoring advice. Chapter 3.2 therefore examines the frequency of

serum potassium measurements and hyperkalemia in hospitalized patients using

one or more potassium increasing drugs and the determinants thereof.

Serum potassium levels may not only alter due to the start of potassium influencing

medication, but also due to discontinuation. CPOEs, however, generally do not signal

when a drug is stopped. To investigate whether monitoring is also relevant after

discontinuation, serum potassium levels before discontinuation are compared to

serum potassium levels after discontinuation in patients who stopped a potassium

increasing drug and patients who stopped a potassium lowering drug (chapter 3.3).

Many ICU’s have implemented a tight glucose control (TGC) protocol to keep

serum glucose levels between target levels. However, not only elevated glucose

levels are associated with an increased mortality. Hypoglycemia is also associated

with mortality and TGC may increase the risk on hypoglycemia. Moreover, TGC

may also influence serum potassium levels and both hyper and hypokalemia

are also associated with mortality. Outcomes, however, seem to depend on

accurate monitoring of glucose levels. Chapter 3.4 describes the influence of the

implementation of a TGC protocol on both serum glucose and serum potassium

levels.

In the final chapter, chapter 4, the results of the different studies are discussed

and put into a broader perspective of current need of monitoring drug therapy.

Furthermore, directions for future research are given.


16

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

References

1 Kohn LT, Corrigan JM, Donaldson MS. To Err is Human: Building a Safer Health System. Washington,

DC: Committee on Quality of Health Care in American, Institute of Medicine. Washington, DC :

National Academy Press; 1999

2 Brennan TA, Leape LL, Laird NM, Hebert L, Localio AR, Lawthers AG, Newhouse JP, Weiler PC, Hiatt HH.

Incidence of adverse events and negligence in hospitalized patients: Results of the Harvard Medical

Practice Study I. N Engl J Med 1991;324:370–376

3 Leape LL, Brennan TA. Laird NM, Lawthers AG, Localio AR, Barnes BA, Hebert L, Newhouse JP, Weiler

PC, Hiatt H. The nature of adverse events in hospitalized patients: Results of the Harvard Medical

Practice Study II. N Engl J Med 1991;324:377–384

4 Thomas EJ, Studdert DM, Burstin HR, Orav, EJ, Zeena T, Williams EJ, Howard KM, Weiler PC, Brennan

TA. Incidence and Types of Adverse Events and Negligent Care in Utah and Colorado. Med Care

2000;38:261-271

5 Bates DW, Cullen DJ, Laird N, Petersen LA, Small SD, Servi D, Laffel G, Sweitzer BJ, Shea BF, Hallisey

R, vander Vliet M, Nemeskal R, Leape LL. Incidence of adverse drug events and potential adverse drug

events. Implications for prevention. ADE Prevention Study Group. JAMA 1995;274:29-34

6 Bates DW, Spell N, Cullen DJ, Burdick E, Laird N, Petersen LA, Small SD, Sweitzer BJ, Leape LL. The costs

of adverse drug events in hospitalized patients. Adverse Drug Events Prevention Study Group. JAMA

1997;277:307-311

7 Leendertse AJ, Egberts AC, Stoker LJ, van den Bemt PM. HARM Study Group. Frequency of and risk

factors for preventable medication-related hospital admissions in the Netherlands. Arch Intern Med

2008;168:1890-1896

8 Gurwitz JH, Field TS, Harrold LR, Rothschild J, Debellis K, Seger AC, Cadoret C, Fish LS, Garber L,

Kelleher M, Bates DW. Incidence and preventability of adverse drug events among older persons in the

ambulatory setting. JAMA 2003;289:1107-1116

9 Leape LL, Bates DW, Cullen DJ, Cooper J, Demonaco HJ, Gallivan T, Hallisey R, Ives J, Laird N, Laffel G,

Nemeskal R, Petersen LA, Porter K, Servi D, Shea BF, Small SD, Sweitzer BJ, Thompson BT, Vander Vliet

M. Systems analysis of adverse drug events. ADE Prevention Study Group. JAMA 1995;274:35-43

10 Juurlink DN, Mamdani M, Kopp A, Laupacis A, Redelmeier DA. Drug-Drug Interactions among elderly

patients hospitalized for drug toxicity JAMA 2003;289:1652-1658

11 Bagheri H, Michel F, Lapeyre-Mestre M, Lagier E, Cambus JP, Valdiguie P, Montastruc JL. Detection and

incidence of drug-induced liver injuries in hospital: a prospective analysis from laboratory signals. Br

J Clin Pharmacol 2000;50:479–484

12 Glassman PA, Simon B, Belperio P, Lanto A. Improving recognition of Drug Interactions benefits and

barriers to using automated drug alerts. Medical Care 2002;40:1161–1171

13 Van Roon EN, Flikweert S, le Comte M, Langendijk PNJ, Kwee-Zuiderwijk WJM, Smits P, R.B.J. Brouwers

JRBJ. Clinical relevance of Drug-Drug Interactions; A structured assessment procedure. Drug Saf

2005;28:1131-1139

14 Merlo J, Liedholm H, Lindblad U, Bjorck-Linne A, Falt J, Lindberg G,Melander A. Prescriptions with

potential drug interactions dispensed at Swedish pharmacies in January 1999: cross sectional study.

BMJ 2001;323:427–428

15 Peng CC, Glassman PA, Marks IR, Fowler C, Castiglione B, Good CB. Retrospective drug utilization

review: incidence of clinically relevant potential drug–drug interactions in a large ambulatory

population. J Manag Care Pharm 2003;9:513–522

16 Buurma H, de Smet PA, Egberts AC. Clinical risk management in Dutch community pharmacies: the

case of drug–drug interactions. Drug Saf 2006;29:723–732

17 Mallet L, Spinewine A, Huang A. Prescribing In Elderly People 2; The challenge of managing drug

interactions in elderly people. Lancet 2007;370:185–191

18 Sijs H, Mulder A, van Gelder T, Aarts J, Berg M, Vulto A. Drug safety alert generation and overriding in

a large Dutch university medical centre. Pharmacoepidem Drug Saf 2009;18:941–947

19 Spriet I, Meersseman W, de Hoon J, von Winckelmann S, Wilmer A, Willems L. Mini series II-clinical

aspects: clinically relevant CYPP450-mediated drug interactions in the ICU. Intensive Care med

2009;35:603-612

17

1R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

20 Krishnan V, Murray P. Pharmacologic issues in the critically ill. Clin Chest Med 2003;24:671–688

21 Institute of Safe Medication Practices: ISMP’s list of high alert medications 2012. http://www.ismp.

org/Tools/highalertmedications.pdf (Accessed at March 1th, 2014)

22 Cullen DJ, Sweitzer BJ, Bates DW, Burdick E, Edmondson A, Leape LL. Preventable adverse drug events

in hospitalized patients: A comparative study of intensive care and general care units. Crit Care Med

1997;25:1289-1297

23 Pirmohamed M, Ferner RE. Monitoring drug treatment. BMJ 2003;327:1179-1181

24 Geerts AFJ, de Koning GHP, van Solinge WW, de Smet PGAM, Egberts ACG. Instructions on laboratory

monitoring in 200 drug labels. Clin Chem Lab Med 2012;50:1351-1358

25 Ferner RE, Coleman J, Pirmohamed M, Constable SA, Rouse A. The quality of information on

monitoring for haematological adverse drug reactions. Br J Clin Pharmacol 2005;60:448-451

26 Raebel MA, Lyons EE, Andrade SE, Chan KA, Chester EA, Davis RL, Ellis JL, Feldstein A, Gunter MJ,

Lafata JE, Long CL, Magid DJ, Selby JV, Simon SR, Platt R. Laboratory Monitoring of Drugs at Initiation

of Therapy in Ambulatory Care. J Gen Intern Med 2005;20:1120-1126

27 Drenth van Maanen AC, van Marum RJ, Jansen PAF, Zwart JEF, Solinge WW, Egberts TCG. Adherence

with dosing guideline in patients with impaired renal function at hospital discharge. Thesis 2013 -

under review

28 Van den Bemt PMLA, Egberts ACG, Leendertse A. Hospital admissions related to medication (HARM).

Een prospectief, multicenter onderzoek naar geneesmiddel gerelateerde ziekenhuisopnames. Utrecht:

Division of Pharmacoepidemiology & Pharmacotherapy, Utrecht Institute for Pharmaceutical

Sciences, 2006

29 Warlé-van Herwaarden MF, Kramers C, Sturkenboom MC, van den Bemt PMLA, De Smet PAGM.

Targeting outpatient drug safety: recommendations of the Dutch HARM-Wrestling task force. Drug

Saf 2012;35:245-259

30 Kingdom of the Netherlands. Regulation of the Ministry of Health, Welfare and Sport of 9

December 2011, amending the regulation on the Dutch Medicines Act (in Dutch) https://zoek.

officielebekendmakingen.nl/stcrt-2011-22913.html (Accessed at March 1th, 2014)

31 Kingdom of the Netherlands. Regulation of the Ministry of Health, Welfare and Sport of 2 July 2013,

amending the regulation on the Dutch Medicines Act (in Dutch) https://zoek.officielebekendmakingen.

nl/stcrt-2013-18918.html (Accessed at March 1th, 2014)

32 Geerts AFJ, de Koning GHP, de Smet PGAM , van Solinge WW, Egberts ACG. Laboratory Tests in the

Clinical Risk Management of Potential Drug-Drug Interactions. Drug Saf 2009;32:1189-1197

33 Juurlink DN, Mamdani MM, Lee DS, Kopp A, Austin PC, Laupacis A, Redelmeier DA. Rates of

hyperkalemia after publication of the Randomized Aldactone Evaluation Study. N Engl J Med

2004;351:543–551

34 Schiff GD, Aggarwal HC, Kumar S, McNutt RA . Prescribing Potassium Despite Hyperkalemia:

Medication Errors Uncovered by Linking Laboratory and Pharmacy Information Systems. Am J Med

2000;109:494-497

35 Bootsma JE, Warle-van Herwaarden MF, Verbeek AL, Füssenich P, De Smet PAGM , Olde Rikkert MG,

Kramers C. Adherence to biochemical monitoring recommendations in patients starting with renin

angiotensin system inhibitors: a retrospective cohort study in the Netherlands. Drug Saf 2011;34:605-

614

2Drug-drug interactions in

hospitalized patients

2.1Frequency and nature of drug-drug

interactions in a Dutch university

hospital

Jeannette E.F. Zwart-van Rijkom

Esther V. Uijtendaal

Maarten J. ten Berg

Wouter W. van Solinge

Antoine C.G. Egberts

Br J Clin Pharmacol 2009:68:187–193

Chapter 2.1 | Frequency and nature of drug-drug interactions in a Dutch university hospital

22

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Abstract

Aim

Drug-drug interactions (DDIs) may lead to often preventable adverse drug events

and health damage. Especially within hospitals this might be an important

factor, as patients are severely ill and multiple medications may be prescribed

simultaneously. The objective of this study was to measure the frequency and

nature of DDI alerts in a Dutch university hospital.

Methods

All patients hospitalized in the University Medical Center Utrecht in 2006 who

were prescribed at least one medication were included. The frequency of DDIs

was calculated as: i) the percentage of patients experiencing at least one DDI, and

ii) the percentage of prescriptions generating a DDI alert. Based on the national

professional guideline, DDIs were classified into categories of potential clinical

outcome, management advice, clinical relevance (A-F) and available evidence (0-4).

Results

Of the 21,277 admissions included, 5,909 (27.8%) encountered at least one DDI.

Overall, the prescribing physician received a DDI alert in 9.6 % of all prescriptions.

The most frequently occurring potential clinical consequence of the DDIs was an

increased risk of side-effects such as increased bleeding risk (22.0%), hypotension

(14.9%), nephrotoxicity (12.6%) and electrolyte disturbances (10.5%). Almost half

(48.6%) of the DDIs could be managed by monitoring laboratory values.

Conclusions

Computerized DDI alerts may be a useful tool to prevent adverse drug events within

hospitals, but they may also result in ‘alert fatigue’. The specificity of alerts could

significantly improve by the use of more sophisticated clinical decision support

systems taking into account, for example, laboratory values.

23

2.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Introduction

Adverse drug events are increasingly acknowledged as an area of major concern

in medical care. The HARM-study recently reported that 2.4% of all hospital

admissions and 5.6% of all emergency admissions in the Netherlands were related

to adverse drug events, of which almost half were considered preventable1. Drug-

drug interactions (DDI) constitute one of the potential mechanisms leading to

often preventable adverse drug events and health damage2,3. Several studies have

investigated the frequency and nature of DDIs in community settings4-7.

Within hospitals, the problem of DDIs may deserve extra attention, as more

medications may be prescribed simultaneously, more complex schemes and

compounds may be used, and the number of (dose) changes may be larger.

Moreover, compared with outpatients, hospitalized patients are more severely ill,

and the capacity to counter-act and deal with disturbances is often diminished.

DDIs may therefore occur more frequently within hospitals than in an outpatient

setting and their consequences may be more severe. Krähenbühl et al. estimated

that 17% of all adverse drug events in hospitalized patients are caused by DDIs

and that approximately 1% of the patients will experience an adverse drug event

during hospitalization due to a DDI8. At discharge about 60% of patients were

found to have at least one potentially interacting drug combination9,10. Two studies

performed in patients visiting an emergency room, found that when medications

were added, a potential adverse DDI was introduced in 5-10% of cases11,12. In two

internal medicine wards in Finland potentially serious interactions were detected

in 1.4% of all prescriptions, as such possibly affecting 6.8% of patients13. Overall,

data on the occurrence and consequences of DDI alerts within hospitals is scarce.

Therefore, the objective of this study was to measure the frequency and nature of

DDI alerts in a large university hospital in the Netherlands and to make suggestions

for improving the management of DDIs.

Methods

SettingThe University Medical Center Utrecht (UMCU) is a 1042-bed academic medical

centre located in the centre of the Netherlands. It consists of an adults’ hospital

and a children’s hospital. In 2006 28,888 clinical hospitalizations took place.

All medications for hospitalized patients are prescribed using a computerized

physician order entry (CPOE) system. For research purposes all prescriptions are

routinely exported in the Utrecht Patient Oriented Database (UPOD). UPOD is an

infrastructure of relational databases comprising data on patient demographics,


24

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

hospital discharge diagnoses, medical procedures, medication orders and

laboratory tests for all patients treated at the UMCU since 2004, and has been

described in detail elsewhere14.

Study populationAll patients hospitalized in the UMCU in 2006 for >24 hours and who were

prescribed at least one medication were included. If a patient was admitted during

the study period to the hospital more than once, the individual hospitalizations

contributed to the study multiple times. Prescriptions from the intensive care (IC)

units were not included in this study, because in the IC units a different CPOE

system is used and the prescription data are not (yet) stored in the UPOD database.

DDIsIn the Netherlands, a working group of the Scientific Institute of Dutch Pharmacists

developed and maintains an evidence base and professional guideline for the

management of DDIs, described in detail elsewhere15. In this professional guideline,

the G-standard, DDIs are classified on a six-point potential clinical relevance scale

ranging from not very serious to potentially lethal (categories A-F), and on a five-point

evidence scale ranging from not proven to very well proven (categories 0-4). This

classification in evidence-relevance categories is described in brief in Appendix 1.

Based on these categories, 331 DDIs have been classified as ‘relevant’, meaning

that they should generate a direct pop-up DDI alert at the moment of prescribing

and that assessment of these combinations of drugs is considered necessary.

Since the generated DDI alerts had not been logged during the study period, they

were reconstructed by combining the G-standard from December 2006 with the

prescription data in UPOD.

In addition, the DDIs were classified in ‘potential clinical outcome’ categories and

‘management advice’ categories. The Dutch professional guideline, the G-standard,

provides textual information about background, mechanism and advice regarding

each DDI. This professional guideline text was converted into ‘potential clinical

outcome’ categories and ‘management advice’ categories separately by two hospital

pharmacist (J.E.F.Z-v.R. and E.V.U). Differences in classification were subsequently

discussed and settled. One DDI could have multiple outcomes or management

advices.

Data analysisThe frequency of DDIs was calculated as: i) the percentage of patients treated with

medication that experienced at least one potential DDI, and ii) the percentage

of all medication orders generating a DDI alert. The frequencies were calculated

25

2.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

separately for each clinical specialism in the children’s and the adults’ hospital.

The nature of the DDIs was described by listing the 10 most frequent alerts for

the children’s and the adults’ hospital. The percentage of alerts in each clinical

outcome and management advice category was calculated.

Finally, the percentages of alerts in each of the G-standard evidence-relevance

categories in our hospital were compared with the percentages found in Dutch

community pharmacies by Buurma et al.4.

Results

In total, 21,277 patients who had been prescribed at least one medication were

admitted to the hospital in 2006. Of these patients, 5909 (27.8%) encountered at

least one DDI; on average these 5909 patients experienced 20,058/5909=3.4 DDIs.

The percentage of patients experiencing at least one DDI during the hospital

admission varied from 3.4% in the paediatric surgery department to 62.0 % in the

adults’ nephrology department (see Table 1).

In 2006, a total of 208,187 prescriptions was registered for clinical patients of which

20,058 (9.6%) resulted in a DDI alert. The percentage of prescriptions generating

an alert varied from 1.6% in the paediatric surgery department to 17.2 % in the

adults’ cardiology department. Although the professional guideline identifies 331

different relevant DDIs, overall only 10 DDIs made up >50% of all DDI alerts, and

50 DDIs made up > 90% of all DDI alerts. The 10 most frequently occurring DDIs in

the adults’ hospital and the children’s hospital respectively are shown in Table 2.

In Table 3 it is shown that an increased risk of side-effects was the most frequently

occurring potential clinical consequence of the DDIs in our patient population

(81.2% of alerts). This risk encompassed an increased bleeding risk (22.0%),

hypotension (14.9%), nephrotoxicity (12.6%) and electrolyte disturbances (10.5%).

The DDI could also lead to a decreased effectiveness of the medication in 25.2%

of cases. This may have severe consequences, for example in the case of antibiotics

(ongoing infection) or immunosupressives (rejection of the transplant). With

respect to the management almost half (48.6%) of all DDI alerts gave the advice to

monitor laboratory values. In 36.5% of the occurring DDIs it was advised to avoid

the combination, in 35.7% to apply a risk-modifying strategy (such as the addition

of a proton pump inhibitor for gastric protection) and in 17.2% tot adjust the dose.


26

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40 Tab

le 1

Fr

equ

ency

of

dru

g-d

rug

inte

ract

ion

ale

rts

aler

ts

Ad

mis

sio

ns

Pres

crip

tio

ns

nu

mb

er o

f ad

mis

sio

ns

wit

h

≥1 a

lert

s*

nu

mb

er o

f ad

mis

sio

ns*

%av

erag

e n

um

ber

of

aler

ts p

er a

dm

issi

on

w

ith

≥1

aler

ts

aler

tsp

resc

rip

tio

ns

%

TOTA

L5,

909

21,2

7727

.8%

3.4

20,0

5820

8,18

79.

6%

Ad

ult

’s h

osp

ital

5,38

415

,389

32.7

%3.

418

,373

167,

757

11.0

%

Car

dio

logy

an

d c

ard

iosu

rger

y1,

363

2,33

658

.3%

3.7

5,02

429

,175

17.2

%

Ger

iatr

ics

172

279

61.6

%4.

576

75,

043

15.2

%

Nep

hro

logy

304

490

62.0

%3.

41,

026

7,15

014

.3%

Gen

eral

in

tern

al m

edic

ine

566

1,39

540

.6%

3.5

1,99

514

,813

13.5

%

Lun

g d

isea

ses

403

798

50.5

%4.

01,

610

12,2

7513

.1%

On

colo

gy a

nd

hae

mat

olog

y33

495

934

.8%

4.5

1,49

812

,759

11.7

%

Der

mat

olog

y59

200

29.5

%5.

029

72,

920

10.2

%

Neu

rolo

gy a

nd

neu

rosu

rger

y69

82,

193

31.8

%2.

81,

958

22,4

158.

7%

Surg

ery

1,49

76,

668

22.5

%2.

74,

024

55,4

467.

3%

Psyc

hia

try

6454

911

.7%

2.7

174

5,76

13.

0%

Ch

ild

ren

’s h

osp

ital

525

5,85

310

.9%

3.2

1,68

540

,430

4.2%

Paed

iatr

ic h

aem

atol

ogy

& n

eph

rolo

gy18

083

421

.6%

4.4

797

9,95

48.

0%

Paed

iatr

ic n

euro

logy

9367

913

.7%

2.8

261

4,43

45.

9%

Paed

iatr

ic c

ard

iolo

gy42

292

14.4

%2.

189

2,88

63.

1%

Gyn

aeco

logy

an

d o

bste

tric

s17

223

897.

2%2.

033

812

,697

2.7%

Gen

eral

pae

dia

tric

s50

656

7.6%

2.1

107

4,69

62.

3%

Paed

iatr

ic s

urg

ery

3710

933.

4%2.

593

5,76

31.

6%

* N

umbe

rs d

o no

t add

up

as p

atie

nts

can

stay

at d

iffe

rent

dep

artm

ents

dur

ing

one

adm

issi

on.

27

2.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Table 2 Frequency and nature of the 10 most frequently encountered drug-drug interactions

% of alerts

Cumu-lative %

Evidence-relevance category

Adult’s hospital

1 ACE-Inhibitors + diuretics 12.1% 12.1% 3D

2 RAS-inhibitors + potassium (saving agents) 8.5% 20.7% 2F

3 Coumarins + antibiotics (excl. co-trimoxazole/metronidazole/ cefamandole)

7.5% 28.2% 3D

4 NSAIDs (excl. COXIB’s) + corticosteroids 6.0% 34.2% 3D

5 QT-prolongators + QT-prolongators(excl. clarithromycin/erythromycin/ voriconazole)

4.0% 38.2% 1E

6 Bisphosphonates + antacids/iron/calcium 3.6% 41.9% 0A

7 Betablockers selective + insulin 3.6% 45.4% 3B

8 Diuretics + NSAIDs 3.0% 48.5% 3D

9 Betablockers + NSAIDs 2.5% 51.0% 3C

10 Betablockers + oral antidiabetics 2.3% 53.3% 3B

Children’s hospital

1 Ciclosporin + nephrotoxic compounds 9.3% 9.3% 3C

2 Ciclosporin + CYP3A4-inhibitors 8.7% 18.0% 3D

3 Ciclosporin + cotrimoxazol/trimethoprim 8.0% 25.9% 3D

4 NSAIDs (excl. COXIB’s) + corticosteroids 7.8% 33.8% 3D

5 QT-prolongators + QT-prolongators(excl. clarithromycin/erythromycin/ voriconazole)

6.2% 39.9% 1E

6 QT-prolongators + clarithromycin/ erythromycin/voriconazole

4.7% 44.6% 3E

7 Midazolam/alprazolam + enzyme inductors 4.5% 49.1% 3C

8 Midazolam/alprazolam + CYP3A4-inhibitors 3.8% 52.9% 3B

9 Diuretics + NSAIDs 3.6% 56.5% 3D

10 Phenytoin + valproic acid 3.4% 59.9% 3A

ACE, angiotensin converting enzyme; RAS, renin-angiotensin system; NSAID, nonsteroidal anti-inflammatory drug


28

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Table 3 Mechanism and advice of drug-drug interactions

Potential Clinical Outcome No. of alerts

% * No. of alerts

% *

Increased risk of side effects/toxicity 16,280 81.2%

Bleeding risk (incl. gastrointestinal ulcer risk) 4,410 22.0%

Hypotension 2,982 14.9%

Hephrotoxicity 2,535 12.6%

Electrolyte disturbances 2,115 10.5%

Cardiac arrhythmias (incl. QT-prolongation) 1,696 8.5%

Masking of hypoglycaemia 1,226 6.1%

Miscellaneous antiepileptics side-effects 415 2.1%

Risk of serotonin syndrome 232 1.2%

Other 669 3.3%

Risk of decreased effectiveness 5,054 25.2%

Anti-hypertensive drugs 1,665 8.3%

Antibiotics and antimycotics 722 3.6%

Biphosphonates 688 3.4%

Thyreomimetics 441 2.2%

Immunomodifiers 396 2.0%

Anti-epileptic drugs 229 1.1%

Other 913 4.6%

Advised Management Strategy No. of alerts

% * No. of alerts

% * No. of alerts

% *

Monitoring 16,268 81.1%

Clinical monitoring of toxicity/effectiveness 3,823 19.1%

Blood pressure monitoring 1,539 7.7%

ECG monitoring 1,151 5.7%

Monitoring of laboratory values 9,755 48.6%

Kidney function (serum creatinin) 2,463 12.3%

Blood clotting time (International Normalised Ratio)

2,408 12.0%

Potassium 2,059 10.3%

Drugs (Therapeutic Drug Monitoring) 1,651 8.2%

Glucose 802 4.0%

Differential blood count 163 0.8%

Liver function 147 0.7%

Sodium 62 0.3%

Avoid combination 7,330 36.5%

Risk modifying strategy 7,169 35.7%

Taking medication when sitting or laying down 2,982 14.9%

Separate moments of oral administration 2,305 11.5%

Add gastric protection (protonpump-inhibitor) 1,789 8.9%

Other 88 0.4%

Adjust dose / titrate dose slowly 3,429 17.1%

Other 889 4.4%

* Percentages do not add up to 100% as one alert could encompass multiple outcomes or management advice.

29

2.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

The evidence for the vast majority (77 %) of the DDI alerts was of relatively high

quality (categories 3-4). The potential clinical relevance was high in 21% (categories

E-F), medium in 58% (categories C-D) and low or not classified in 21% of the DDIs

(categories A-B). This spreading over the different evidence-relevance categories is

presented in Figure 1 and is similar to that found in Dutch community pharmacies

by Buurma et al.4.

0%

10%

20%

30%

40%

0A

1A, 1B, 1

C, 1D 1E 1F 2D 2E 2F 3A 3B 3C 3D 3E 3F 4A 4C 4D

not cla

ssifie

d

Evidence-relevance category

Perc

enta

ge o

f DD

I ale

rts

university hospital (this study)

community pharmacies (Buurma et al.)

Figure 1 Percentage of drug-drug interaction (DDI) alerts by evidence-relevance category in our study compared with the percentage in Dutch community pharmacies in the study of Buurma et al.4

Discussion

We found that of all patients admitted to the hospital and being prescribed

medication, exclusive of IC patients, 28% experienced at least one potential DDI.

On average this group of patients experienced 3.4 DDIs. Overall, the prescribing

physician received a DDI alert in 9.6 % of all prescriptions; this percentage largely

varied between clinical specialisms. The most frequently occurring potential

clinical consequence of the DDIs was an increased risk of side-effects such as

increased bleeding risk (22.0%), hypotension (14.9%), nephrotoxicity (12.6%) and

electrolyte disturbances (10.5%). Almost half (48.6%) of the DDIs could be managed

by monitoring laboratory values.

All admissions in 2006, including re-admissions within the same year, were

included in our study as separate cases. Compared with first admissions, the risk

for DDIs may be increased during subsequent admissions, as these patients may be

more severely ill. To check whether or not this was the case, we also performed the

analysis including for each patient only the first admission in 2006. The outcomes


30

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

of both analyses were similar (29% of first admissions experiencing at least one

potential DDI; average number of DDIs 3.2). We consider it therefore justified to

combine both first and subsequent admissions in the analysis.

The 10 most frequently encountered DDIs in the adults’ hospital did not consist

of specific ‘high-tech’ hospital medications, but of medications that are used and

initiated on a large scale in the community setting as well. This suggests that

home medication caused a substantial part of the DDIs. In contrast to the adults’

hospital, the DDIs in the children’s hospital do seem to be caused by specific

‘hospital medication’, such as, for example, ciclosporin and midazolam. It can be

expected that children, in general, will be on less (or no) home medication on (first)

admittance to the hospital than adults. Another explanation for our findings in the

adults’ hospital may be found in the fact that we used the professional guideline,

the G-standard. The G-standard is in fact the national standard at this moment

and it is used in all hospitals in the Netherlands. However, it has primarily been

developed for its use within community pharmacies. DDIs concerning specific

hospital medications, e.g. anaesthetics and cytostatic agents, were largely missing

in the G-standard in 2006. It may be worthwhile considering if our results would

be different with the inclusion of the intensive care units and operating rooms and

with the use of a specific clinical DDI reference database. However, to our knowledge

such a specific clinical DDI database does not yet exist. Dawson and Karalliedde16

published an overview of DDIs that are important to the clinical anesthetist. This

publication could be a good starting point.

Another kind of interactions that is not included in the professional guideline

used but is of high importance in the hospital setting is the chemical or physical

incompatibility of two intravenous medications when mixed in the same infusion

bag or syringe or when administered simultaneously through the same catheter.

These types of interactions were not included in our study. Ideally, to improve

patient safety, a CPOE system should also signal this kind of DDI and make proposals

for its management.

A limitation may be that the number of separate DDIs may be overestimated, as the

same alert may be generated several times, each time the prescription is altered (re-

start, dose change). On the other hand, the number of ‘DDI risk moments’ may be

underestimated, as a DDI alert was generated only when a DDI interaction started.

A warning may also be in place when the DDI stops. For example, in the case of an

enzyme induction DDI, the plasma concentration of a drug may decrease and a

dose increase may be required. As a consequence, when the interacting medication

is stopped, the dose may need to be decreased again. In common practice of our

hospital, this alert is not generated, and therefore, and to avoid double-counting,

was not included in our study.

31

2.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Patient safety may be improved by decreasing the frequency of preventable adverse

drug events Computerized alerts may be a useful tool to signal DDIs, but they may

also result in ‘alert fatigue’ and the overriding of important signals, especially when

the system produces an overload of signals that are of minor clinical relevance17,18.

Although 79% of the alerts were of high or medium potential clinical relevance,

the actual relevance of some signals may still be low. Currently, the computerized

DDI surveillance systems that are routinely used in the Netherlands only take into

account the concomitant use of two drugs. Until now they have been unable to check

for additional relevant data such as laboratory values, times of administration,

absence of gastric protection, age, etc, which could increase the specificity of the

signal. For example, the concomitant use of a renin-angiotensin system inhibitor

and potassium always results in an alert warning for hyperkalaemia, even when

the serum potassium level is low. In this study we did not gather the information

to assess the actual relevance of the generated DDI alerts; this may be the topic of

our further research.

From Table 3, it can be expected that the specificity of alerts could significantly

improve by the use of more sophisticated clinical decision support systems

(CDSS)19,20, taking into account, for example, laboratory values (48.6% of alerts),

times of administration (11.5%) and the absence of gastric protection 8.9%). This

may be important knowledge in finding strategies to combat ‘alert-fatigue’.

Another important finding in this respect is that a relatively small number of

combinations (10 different DDIs) is responsible for a large number of alerts (>50%).

As these are very well-known DDIs, it may be worthwhile to discuss with the medical

board if and when these alerts have indeed to be shown. Turning-off some of these

top 10 DDI alerts may substantially decrease the ‘alert overload’ and in this way

increase the attention to dangerous and less known DDIs. In a series of qualitative

interviews, Van der Sijs et al.23 found that indeed the majority of respondents wanted

to turn off DDI alerts to reduce alert overload. Since the top 10 of alerts varies

among clinical specialism and since also the knowledge and monitoring practices

vary between clinical specialisms within the hospital, it may be wise to suppress

DDI alerts differentially between prescribers. This also requires more sophisticated

CDSS than we have available at the moment.

In conclusion, 28% of all patients admitted to our hospital are exposed to at least

one potential DDI. An increased risk of side-effects (e.g. increased bleeding risk,

hypotension, nephrotoxicity, electrolyte disturbances) was the most prevalent

potential clinical consequence. Almost half of the DDIs could be managed by

monitoring laboratory values. Prescribing physicians receive an automated DDI

alert in nearly 10% of all prescriptions. Further research is needed to investigate the

clinical relevance of these DDIs and to develop methods to increase the specificity

of automated CDSS alerts.


32

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Acknowledgements

The authors are grateful to Hanneke den Breeijen for the data-analysis and to their colleagues

at the Utrecht Institute for Pharmaceutical Sciences and the UMC Utrecht for their support in

establishing and maintaining UPOD.

33

2.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Appendix 1 Classification of drug-drug interactions according to the professional guideline by WINAp15

Category Description

Potential clinical relevance

A No inconvenience, insignificant effect

B Short-lived inconvenience (<24-48 hours) without residual symptoms

C Long-lived inconvenience (48-168 hours) without residual symptoms

D Long-lived inconvenience (>168 hours) with residual symptoms or handicap

E Potential failure of life-saving therapy, increased risk of pregnancy (without risks concerning mother and/or fetus), cardiac arrhythmia, rhabdomyolysis, malignant hypertension, pseudopheochromocytome, multi-organ failure

F Death, torsade de pointes, ventricular arrhythmia, myocardial infarction, serotonin syndrome, hyperpyrexia (42°C)

Quality of evidence

4 Controlled, published study with clinically relevant end-points

3 Controlled, published study with relevant surrogate end-points

2 Well-documented, published case reports; analysis of case series

1 In-complete, published case reports

0 Animal studies, in vitro studies, data on file

- No evidence

Not classified


34

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

References

1. Leendertse AJ, Egberts AC, Stoker LJ, van den Bemt PM. Frequency of and risk factors for preventable

medication-related hospital admissions in the Netherlands. Arch Intern Med 2008;168:1890-1896

2. Juurlink DN, Mamdani M, Kopp A, Laupacis A, Redelmeier DA. Drug-drug interactions among elderly

patients hospitalized for drug toxicity. JAMA 2003;289:1652-1658

3. Bagheri H, Michel F, Lapeyre-Mestre M, Lagier E, Cambus JP, Valdiguie P, Montastruc JL.. Detection and

incidence of drug-induced liver injuries in hospital: a prospective analysis from laboratory signals. Br

J Clin Pharmacol 2000;50:479-484

4. Buurma H, De Smet PA, Egberts AC. Clinical risk management in Dutch community pharmacies: the

case of drug-drug interactions. Drug Saf 2006;29:723-732.

5. Merlo J, Liedholm H, Lindblad U, Bjorck-Linne A, Falt J, Lindberg G, Melander A. Prescriptions with

potential drug interactions dispensed at Swedish pharmacies in January 1999: cross sectional study.

BMJ 2001;323:427-428

6. Buurma H, Schalekamp T, Egberts AC, De Smet PA. Compliance with national guidelines for the

management of drug-drug interactions in Dutch community pharmacies. Ann Pharmacother

2007;41:2024-2031

7. Peng CC, Glassman PA, Marks IR, Fowler C, Castiglione B, Good CB. Retrospective drug utilization

review: incidence of clinically relevant potential drug-drug interactions in a large ambulatory

population. J Manag Care Pharm 2003;9:513-522

8. Krahenbuhl-Melcher A, Schlienger R, Lampert M, Haschke M, Drewe J, Krahenbuhl S. Drug-related

problems in hospitals: a review of the recent literature. Drug Saf 2007;30:379-407

9. Egger SS, Drewe J, Schlienger RG. Potential drug-drug interactions in the medication of medical

patients at hospital discharge. Eur J Clin Pharmacol 2003;58:773-778

10. Kohler GI, Bode-Boger SM, Busse R, Hoopmann M, Welte T, Boger RH. Drug-drug interactions

in medical patients: effects of in-hospital treatment and relation to multiple drug use. Int J Clin

Pharmacol Ther 2000;38:504-513

11. Heininger-Rothbucher D, Bischinger S, Ulmer H, Pechlaner C, Speer G, Wiedermann CJ. Incidence and

risk of potential adverse drug interactions in the emergency room. Resuscitation 2001;49:283-288

12. Beers MH, Storrie M, Lee G. Potential adverse drug interactions in the emergency room. An issue in

the quality of care. Ann Intern Med 1990;112:61-64

13. Gronroos PE, Irjala KM, Huupponen RK, Scheinin H, Forsstrom J, Forsstrom JJ. A medication database

- a tool for detecting drug interactions in hospital. Eur J Clin Pharmacol 1997;53:13-17

14. ten Berg MJ, Huisman A, van den Bemt PMLA, Schobben AFAM, Egberts ACG, van Sollinge WW.

Linking laboratory and medication data: new opportunities for pharmacoepidemiological research.

Clin Chem Lab Med 2007;45:13-19

15. Van Roon EN, Flikweert S, Le Comte M, Langendijk PNJ, Kwee-Zuiderwijk WJM, Smits P, Brouwers

JRBJ. Clinical relevance of drug-drug interactions: a structured assessment procedure. Drug Saf

2005;28:1131-1139

16. Dawson J, Karalliedde L. Drug interactions and the clinical anaesthetist. Eur J Anaesthesiol

1998;15:172-189

17. van der Sijs H, Aarts J, Vulto A, Berg M. Overriding of drug safety alerts in computerized physician

order entry. J Am Med Inform Assoc 2006;13:138-147

18. Kuperman GJ, Bobb A, Payne TH, Avery AJ, Gandhi TK, Burns G, Classen DC, Bates DW. Medication-

related clinical decision support in computerized provider order entry systems: a review. J Am Med

Inform Assoc 2007;14:29-40

19. Chertow GM, Lee J, Kuperman GJ, Burdick E, Horsky J, Seger DL, Lee R, Mekala A, Song J, Komaroff

AL, Bates DW. Guided medication dosing for inpatients with renal insufficiency. JAMA 2001;286:2839-

2844

20. Schiff GD, Klass D, Peterson J, Shah G, Bates DW. Linking laboratory and pharmacy: opportunities for

reducing errors and improving care. Arch Intern Med 2003;163:893-900

35

2.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

21. Raschke RA, Gollihare B, Wunderlich TA, Guidry JR, Leibowitz AI, Peirce JC, Lemelson L, Heisler MA,

Susong C. A computer alert system to prevent injury from adverse drug events: development and

evaluation in a community teaching hospital. JAMA 1998;280:1317-1320

22. Judge J, Field TS, DeFlorio M, Laprino J, Auger J, Rochon P, Bates,DW, Gurwitz JH. Prescribers’

responses to alerts during medication ordering in the long term care setting. J Am Med Inform Assoc

2006;13:385-390

23. van der Sijs H, Aarts J, van Gelder T, Berg M, Vulto A. Turning off frequently overridden drug alerts:

limited opportunities for doing it safely. J Am Med Inform Assoc 2008;15:439-448

2.2Analysis of potential drug-drug

interactions in medical intensive

care unit patients


Lieke L.M. van Harssel

Gerard W.K. Hugenholtz

Emile M. Kuck

Jeannette E.F. Zwart- van Rijkom

Olaf L. Cremer

Toine C.G. Egberts

Pharmacotherapy 2014;34:213-219

Chapter 2.2 | Analysis of potential drug-drug interactions in medical intensive care unit patients

38

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Abstract

Objective

To describe the frequency and type of potential drug–drug interactions (pDDIs) in

a general intensive care unit (ICU) and to make recommendations to improve the

management of these pDDIs.

Design

Retrospective observational study.

Setting

General ICU of a tertiary care hospital.

Subjects

All patients admitted for more than 24 hours between May 2009 and December

2010, who were prescribed at least one medication.

Measurement and Main Results

Based on the G-Standaard, the Dutch national drug database, pDDIs were identified

and classified into categories of potential clinical outcome and management advice.

In total, 35,784 medication episodes were identified, resulting in 2,887 pDDIs

(8.1%). These 2,887 pDDIs occurred in 1,659 patients for a mean frequency of 1.7

(95% confidence interval [CI], 1.6-1.9) pDDIs per patient. Overall, 54% of the patients

experienced at least one pDDI with pDDIs present during 27% of all ICU admission

days. All pDDIs could be reconstructed using 81 of the 358 (23%) relevant unique

pDDI-pairs described in the G-Standaard. The most frequently occurring potential

clinical consequence was an increased risk of side effects or toxicity (91% of the

pDDIs), such as electrolyte disturbances and masking of hypoglycemia. The most

important advised management strategy was monitoring (81%), consisting of

monitoring of laboratory values (52%), clinical monitoring of toxicity or effectiveness

(48%), or monitoring of physical parameters such as electrocardiogram and blood

pressure (11%).

Conclusion

Potential drug-drug interactions occur in 54% of all ICU patients, which is two times

more than the rate seen in patients on general wards. A limited set of 20 pDDI-pairs

is responsible for more than 90% of all pDDIs. Therefore, it is worthwhile to develop

guidelines for the management of these specific pDDIs. As the vast majority of the

interactions can be managed by monitoring, advanced clinical decision support

systems linking laboratory data to prescription data may be an effective risk

management strategy.

39

2.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Introduction

Given their medical and societal impact1, 2, adverse drug events (ADEs) are

increasingly acknowledged as an area of major concern in medical care. Potential

drug–drug interactions (pDDIs) constitute one of the often preventable causes of

ADEs and patient harm. One review estimates that 17% of all preventable adverse

drug events in hospitalized patients are caused by DDIs and that approximately 1%

of patients will experience an adverse drug event during hospitalization due to a

DDI3. Other research suggests that among patients in intensive care unit (ICUs), 10-

16% of all preventable ADEs are related to a DDI and that approximately 5% of all

ICU patients will experience an ADE during hospitalization due to a pDDI4,5.

Because approximately half of the ADEs resulting from DDIs are predictable and

preventable 6, it is important to detect those pDDIs that are relevant for patients in

the ICU. Several studies have investigated the frequency and nature of pDDIs in a

general hospital setting7, 8. The occurrence of pDDIs in an ICU may, however, differs

from general wards, because a larger number of medications may be used, (dose)

changes may occur more often, and medications can often be classified as high-risk9.

Moreover, compared with patients in other departments, ICU patients are generally

more severely ill, which can interfere with their ability to metabolize drugs and

their capacity to counteract and deal with physiological disturbances. The larger

number of medications may cause pDDIs to occur more frequently in ICU patients,

and their altered capacity to counteract with physiologic disturbances may cause

the consequences to be more severe than in a general hospital setting10. However,

the technical environment of the ICU -- including bed-site diagnostic testing and

continuous monitoring -- allows for an effective risk management strategy and the

possibility of quick adjustment.

Data on the occurrence and management of pDDIs on ICUs are still limited. In some

smaller studies, the percentage of ICU patients exposed to at least one pDDI ranged

from 45% to 73%11,12,13. However, the number of patients included in these studies

was only 100-400. One recent large study of 9644 admissions found that 3892

(40%) experienced at least one pDDI. Potential clinical outcomes and management

advices were not discussed14.

Therefore, the objective of this study was to describe the type and frequency of pDDIs

occurring on a large general ICU and to make recommendations for improving the

management of these pDDIs.


40

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Materials and Methods

Setting and study populationThis retrospective observational study was carried out at the 32-bed ICU of the

University Medical Centre Utrecht (UMCU), which is a tertiary care teaching hospital

in the Netherlands. All medications and clinical data for patients hospitalized at

the ICU are registered in an electronic Patient Data Management System (ePDMS;

Metavision; iMDsoft, Sassenheim, the Netherlands). No computerized medication

surveillance system has been implemented in this ePDMS. However, a clinical

pharmacist reviews each patient’s medication each day and discusses relevant

pDDIs with the clinicians. All adult-patients (18 years or older) admitted to the

ICU for more than 24 hours between May 2009 and December 2010 and who were

prescribed at least one medication were eligible for inclusion. If a patient was

admitted more than once to the ICU during this time period either within the same

hospitalization period or during a subsequent admission, only the first admission

was included. Because only routinely documented patient data were used, the

ethics board waived the need for approval.

Potential drug-drug interactionsA pDDI was defined as two potentially interacting drugs that were administered

concomitantly. Relevant pDDIs were reconstructed by linking the G-Standaard

from May 2009 with the medication records from the ePDMS research database. In

the Netherlands, a working group of the Scientific Institute of Dutch Pharmacists

developed and maintains the G-Standaard, an evidence-based professional

guideline for the management of pDDIs, described in detail elsewhere15. In this

professional guideline, pDDIs are classified into either ‘relevant’ pDDIs requiring

an interaction alert (358 pDDIs) or ‘less/not relevant’ pDDIs that do require an alert.

Based on the G-Standaard’s textual information on background, mechanism, and

advice, the pDDIs were classified into ‘potential clinical outcome’ categories were

described in detail elsewhere7. One pDDI could potentially have multiple outcomes

or management advice.

In the ePDMS, medication is not registered as a medication record with a start and

stop date; each administration is registered separately. For this study, medication

records were reconstructed retrospectively. Administrations for a specific drug were

attributed to one medication record if the time gap did not exceed 12 hours for

continuously administered medications (e.g., dobutamine 2.5 mg/kg/min) and 36

hours for discontinuously administered medications (e.g. erythromycin two times/

day 250 mg).

41

2.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Data Analysis and Statistical AnalysisThe frequency of pDDIs was calculated as the number of pDDIs per patient; the

number of patients that experienced at least one pDDI; and the number of ICU

days with at least one pDDI. The frequencies were calculated separately for baseline

patient characteristics such as length of ICU stay, predicted hospital mortality,

observed hospital mortality, mechanical ventilation upon admission, admitting

service, number of comorbidities, and the number of different drugs used. Predicted

hospital mortality was calculated during the first 24 hours of ICU admission, using

the Acute Physiologic and Chronic Health Evaluation IV model16.

The type of pDDIs was described by listing the 20 most frequently occurring pDDIs.

The percentage of pDDIs in each potential clinical outcome category and in the

management advice category was listed.

Data were analyzed using descriptive statistics from the SPSS software v. 17.0 for

Windows (SPSS, Chicago, IL).

Results

During the study period, 3489 patients 18 years of age or older were hospitalized

in the ICU; 1793 were admitted to the ICU for less than 24 hours and 37 patients

did not receive any medication during their stay in the ICU. This resulted in a total

of 1659 patients, accounting for 10252 ICU days and a median length of stay in the

ICU of 2.9 (Interquartile range 1.7-6.8) days. Among the 1659 patients, 84% were

mechanically ventilated; the observed hospital mortality was 20% (Table 1).

Frequency of pDDIsIn total, 35,784 medication records were identified, resulting in 2,887 pDDIs (8.1%).

The total study population was 1,659 patients, for a mean frequency of 1.7 pDDIs

per patient (95% confidence interval, [CI] 1.6-1.9) (Table 2). During ICU admission,

900 patients (54%) experienced at least one pDDI. Within this subgroup of patients,

the frequency was 3.2 pDDIs per patient (95% CI, 3.0-3.4). One or more pDDIs were

present during 27% of all ICU admission days. Table 2 lists these three outcomes for

the total study population stratified for various baseline characteristics such as age,

gender, predicted mortality upon ICU admission and admitting service.


42

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Table 1. Characteristics of the study population

Population characteristics (n=1659)

Male gender; %, n 60% (n=994)

Age; median, interquartile range 62 years (50-72)

Observed ICU Length of stay; median, interquartile range 2.9 days (1.7-6.8)

Observed hospital mortality; %, n 20% (n=324)

APACHE IV predicted mortalitya; mean %, 95% CI 23.0% 22.7-25.0%

Mechanical ventilation at admission; %, n 84% (n=1340)

Elective admission to ICU; %, n 36% (n=581)

Admitting medical specialism; %, n

Internal medicine 18% (n=293)

General surgery 26% (n=431)

Cardiology or cardiothoracic surgery 34% (n=569)

Neurology or neurosurgery 22% (n=366)

Comorbidities; %, n

Cardiovascular 9.9% (n=164)

Diabetes mellitus 13% (n=212)

Renal disease b 6.0% (n=99)

Hepatic disease c 1.0% (n=16)

Respiratory disease 14% (n=230)

Number of different drugsd;; %, n

1-12 31% (n=507)

13-20 38% (n=623)

≥21 32% (n=529)

pDDI = potential drug-drug interaction; CI = confidence interval; ICU = intensive care unit; APACHE = Acute Physiologic and Chronic Health Evaluation Three patients were prescribed only one medicationa calculated within the first 24 hours of ICU admission, using the APACHE IV model;b Renal diseases: serum creatinine > 177micromol/L combined with a medical history of chronic renal disease or renal dialysis c Hepatic disease: including portal hypertension combined with a positive liver biopsy or bleeding in the upper part of the gastro-intestinal tract, hepatic encephalopathyd Sum of unique drugs per patient (active ingredient and dosage form) during the entire ICU stay

43

2.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Table 2. Main Outcome Parameters

pDDIs per patient Patients with ICU days with

(mean; 95%CI a) ≥ 1 pDDI ≥ 1 pDDI % (n)

total DDIs total patients total ICU days

n = 2,887 n = 1,659 n = 10,252

Total 1.7 (1.6-1.9) 54% (n=900) 27% (2,779/10,252)

Gender

Male 1.8 (1.6-2.0) 56% (n=557/994) 27% (n=1,774/6,471)

Female 1.6 (1.4-1.9) 52% (n=343/665) 27% (n=1,005/3,781)

Age category, yrs

18-39 1.2 (0.9-1.6) 42% (n=95/228) 18% (n=284/1,590)

40-59 1.6 (1.4-1.9) 51% (n=251/497) 25% (n=772/3,129)

- 60-79 1.9 (1.7-2.1) 59% (n=472/804) 29% (1,399/4,742)

≥ 80 2.4 (1.7-3.0) 63% (n=82/130) 41% (324/792)

Length of stay ICU, days

1-2 0.6 (0.5-0.7) 39% (n=332/853) 10% (138/1,436)

3-4 1.5 (1.2-1.7) 58% (n=148/256) 22% (223/1016)

5-6 2.4 (2.0-2.8) 68% (n=113/165) 29% (293/998)

≥ 7 4.2 (3.7-4.6) 80% (n=307/385) 31% (2,126/6,802)

Apache IV predicted mortality, %

0-33 1.3 (1.2-1.5) 51% (n=578/1,145) 24% (1,395/5,803)

34-66 3.0 (2.5-3.4) 65% (n=192/294) 32% (887/2,796)

67-100 2.7 (2.0-3.4) 62% (n=76/123) 35% (335/967)

Unknown 1.7 (1.1-2.3) 56% (n=54/97) 23% (161/686)

Observed hospital mortality

Yes 2.5 (2.1-2.9) 57% (n=185/324) 32% (874/2,694)

No 1.6 (1.4-1.7) 54% (n=694/1,288) 25% (1,799/7,093)

Unknown 1.8 (0.9-2.72) 45% (n=21/47) 23% (106/465)

Mechanical ventillation at admission

Yes 1.8 (1.6-1.9) 56% (n=744/1,340) 27% (2,198/8,169)

No 1.9 (1.5-2.3) 51% (n=127/251) 29% (563/1,969)

Unknown 1.0 (0.37-1.54) 43% (n=29/68) 16% (18/114)

Elective admission at ICU

Yes 1.3 (1.2-1.5) 55% (n=319/581) 25% (577/2,300)

No 2.0 (1.8-2.2) 54% (n=556/1,026) 28% (2,187/7,860)

Unknown 1.2 (0.4-1.9) 48% (n=25/52) 16% (15/92)

Admitting medical specialty

Internal medicine 2.2 (1.8-2.7) 51% (n=150/293) 31% (845/2,747)

General surgery 1.6 (1.3-1.8) 43% (n=186/431) 23% (711/3,132)

Cardiology or cardiothoracis surgery 2.0 (1.7-2.2) 70% (n=396/569) 33% (822/2,504)

Neurology or neurosurgery 1.3 (1.0-1.5) 46% (n=168/366) 21% (398/1,863)

Number of comorbidities

0 1.5 (1.4-1.7) 50% (n=522/1,045) 24% (1,545/6,367)

1 2.0 (1.7-2.3) 61% (n=253/415) 29% (760/2,628)

≥ 2 2.8 (2.1-3.5) 69% (n=97/141) 39% (452/1,149)

Unknown 1.12 (0.43-1.81) 48% (n=28/58) 19% (21/109)

Number of different drugs usedb

1-12 0.3 (0.3-0.4) 25% (n=129/507) 8% (81/989)

13-20 0.9 (0.8-1.0) 51% (n=315/623) 15% (344/2,229)

≥ 21 4.0 (3.7-4.4) 86% (n=456/529) 33% (2,353/7,034)

pDDI = potential drug-drug interaction; CI = confidence interval; ICU = intensive care unit; APACHE = Acute Physiologic and Chronic Health Evaluationa 95%CI = 95% Confidence Intervalb sum of unique drugs per patient (active ingredient and dosage form) during the entire ICU stay


44

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Type of pDDIsOnly 81 (23%) of the 358 relevant unique pDDI-pairs described in the G-Standaard

occurred in our study population. Of the 81 different pDDI pairs observed, 20

pDDI pairs accounted for 93% of all observed pDDIs (Table 3). The most frequently

occurring potential clinical consequence of the pDDIs was an increased risk of side

effects or toxicity (91% of all pDDIs); masking of hypoglycemia (21%) and electrolyte

disturbances (20%) were the most common (Table 4). Another potential clinical

consequence is decreased effectiveness of the medication. This was potentially

the case in 10% of all encountered pDDIs, mainly due to interactions with

antihypertensive drugs (4.6%).

Table 3. Twenty most frequently occurring pDDIs in ICU patients

pDDI Frequency n

β-blockers selective + insulin 15.9% 459

Midazolam/alprazolam + CYP3A4- inhibitors 11.0% 317

QT-prolongators + QT-prolongators 10.7% 310

RAS inhibitors + potassium salts/potassium saving agents 9.7% 280

ACE-inhibitors + diuretics 8.5% 245

Acetazolamide + diuretics (potassium losing) 5.7% 164

β-blockers nonselective + insulin 5.0% 143

Coumarins + antibiotics 4.3% 125

Potassium salts + potassium saving agents 4.2% 120

β-blockers nonselective + β-mimetics 4.0% 116

NSAIDs (excl.COXIBs) + corticosteroids 3.7% 107

Statins + CYP3A4 inhibitors 2.1% 60

Midazolam/Alprazolam + enzyme inductors 1.5% 42

Corticosteroids + enzyme inductors 1.4% 40

Tacrolimus + CYP3A4 inhibitors 1.2% 34

Coumarins + Salicylates 1.0% 30

Haloperidol + enzyme inductors 0.9% 26

Calcium antagonists + CYP3A4 inhibitors 0.8% 22

Digoxine + Amiodarone 0.6% 18

Ciclosporine + CYP3A4 inhibitors 0.5% 15

Total 20 most frequently occurring pDDIs 92.6% 2673

45

2.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Table 4. Type of pDDIs: Possible Clinical Outcome Categories and Advised Management Strategy

Potential outcome category pDDIs % (n); total n=2,887

Increased risk of side-effects/toxicity 90.6a(2,617)

Masking hypoglycemia 20.9 (604)

Electrolyte disturbance 19.6 (565)

Cardiac arrhythmias (including QT-prolongation) 11.7 (338)

Increased effect benzodiazepines 11.0 (317)

Bleeding risk (incl gastrointestinal ulcer risk) 10.3 (296)

Hypotension 8.8 (253)

Nephrotoxicity 2.4 (70)

Myopathy 2.1 (60)

Other 3.1 (104)

Risk of decreased efficacy 10.7a(310)

Antihypertensive drugs 4.6 (134)

Immunomodulators 1.5 (43)

Benzodiazepines/opioids 1.5 (42)

Antipsychotic drugs (haloperidol) 0.9 (26)

Other 2.3 (65)

Advised Management strategy pDDIs % (n); total n=2,887

Monitoring 80.8b (2,330)

Monitoring of laboratory values 51.6c (1489)

Glucose 20.9 (602)

Potassium 19.6 (566)

Drugs (therapeutic drug monitoring) 5.0 (144)

Blood clotting time (international normalized ratio) 4.6 (134)

Kidney function (serum creatinine) 1.2 (36)

Liver function 0.2 (5)

Sodium 0.1 (2)

Clinical monitoring of toxicity/effectiveness 48.4c (1,398)

ECG monitoring 10.7c (310)

Blood pressure monitoring 0.6c (16)

Avoid combination 40.3b (1,163)

Adjust dose/titrate dose slowly 25.0b (722)

Risk-modifying strategy 19.3b (558)

Taking medication when sitting or laying down 8.8 (253)

Potassium or potassium sparing diuretic 5.7 (166)

Add gastric protection (proton pump inhibitor) 4.6 (134)

Separate moments of oral administration 0.2 (5)

Other 1.9b (54)

pDDI = potential drug-drug interaction; ECG = electrocardiogram.a Percentages add up to more than 100% because one alert could potentially have multiple outcomesb Percentages add up to more than 100% because one alert could potentially have multiple management advicec Percentages add up to more than 100% because one alert could potentially have multiple advice for monitoring


46

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

With respect to the management of pDDIs, monitoring was the most important

advised management strategy (81%) (Table 4) and included the monitoring of

laboratory values (52%), toxicity or effectiveness (48%) and physical parameters

such as electrocardiogram and blood pressure (11%). Other frequently advised

strategies were avoiding combination (40%), adjusting or titrating dose (25%) and

modifying risk strategy (19%). Comparable results for potential clinical outcome

and management strategies were obtained when the data were analyzed for the

number of patients instead of the number of pDDIs (data not shown).

Discussion

At least one relevant pDDI occurs in 54% of all ICU patients. Only a limited set of

drug-drug pairs (n=20) is responsible for more than90% of all pDDIs. Most pDDIs

can be managed by monitoring either clinical symptoms or laboratory values.

The percentage of patients on the ICU experiencing at least one pDDI in our study

is comparable with the percentages found in other ICU studies (40-73%) 11-14. The

percentage is twice as high as the percentage on the general wards of the same

hospital as previously reported (28%; n=5,909) 7. However, when the number of

pDDIs is considered in relation to the number of medication records, the difference

between the ICU and general hospital departments disappears (9.6% vs. 8.1%). The

average number of pDDIs calculated for ICU-patients with at least one pDDI (3.2

pDDIs per patient experiencing at least one pDDI), is also comparable to the average

number of pDDIs found on general wards (3.4 pDDIs). As such, the average number

of medications per patient is higher for ICU patients than for other hospitalized

patients, resulting in a higher percentage of patients experiencing at least one

pDDI; the percentage of pDDIs per medication prescription is about equal.

Only 20 pDDI pairs make up to more than 90% of all pDDI alerts. The 10 most common

pDDI pairs in ICU patients differs substantially from the most common pairs found

in the general wards7. This difference can be explained by the differences in the

medical needs of ICU patients compared with general ward patients. For example,

the pDDI between midazolam/alprazolam and CYP3A4 inhibitors seldom occurs in

a general ward. On the ICU however, this is the second most frequently occurring

pDDI, because midazolam is frequently prescribed to ICU patients for sedation.

The pDDI between nonsteroidal antiinflammatory drugs and corticosteroids or

bisphosphonates and antacids are among the top 10 pDDIs on general wards, but

they seldom occurred in our ICU study population.

The limited set of 20 pDDI pairs being responsible for more than 90% of all pDDIs

enables caregivers to establish local guidelines for the handling of these pDDIs.

For example, local caregivers may agree to discontinue alerts for some of these

47

2.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

frequently occurring combinations because they are familiar with these pDDIs and

feel that the potential benefit of these combinations outweighs the potential risk

in their patient population. This is the case in our ICU setting for the combination

of insulin and selective β-blockers. Moreover, most (81%) of the pDDIs occurring at

the ICU can be managed by monitoring. Because ICU patients are, by definition,

intensively monitored, deviations from normal values may be noticed quickly,

even without a pDDI alert. As such, it may not be very useful to generate standard

pDDI alerts in the electronic prescribing system for these 20 well-known and easy

to monitor pDDIs. Instead, advanced clinical decision support systems linking

laboratory data to prescription data, and signaling only when values fall out of

normal ranges, may be much more valuable. Furthermore, when laboratory data

are linked to prescription data, changes may also be detected when one of the

interacting drugs is discontinued. For other, less frequently occurring pDDIs, pDDI

alerts at the moment of prescribing can still be considered.

In this study, the G-Standaard was used to reconstruct the pDDIs. Because the

G-Standaard was developed primarily for community pharmacies, some advice (

e.g. taking medication when sitting upright) are not applicable to ICU patients.

Moreover, pDDIs involving ICU specific medications, such as sedatives or inotropics,

might be missing. For example, both the Lexi-Interact (Wolter Kluwer Health,

Baltimore, Maryland)17 and Micromedex (Truven Health Analytics Inc., Greenwood

Village, Colorado) 18 databases mention several relevant pDDIs with norepinephrine

(Table 5), while no pDDIs were mentioned in the G-Standaard. However, the

G-Standaard may include pDDIs that do not occur in the Lexi-Interact17 and

Micromedex18 databases, such as the major pDDI between sotalol and telaprevir. It

is well known that there are major differences between different DDI databases11,12.

Ideally, we would have used a pDDI database especially designed for ICU patients.

Because such a database does not exist, we used the G-Standaard, the national

professional guideline used in all Dutch hospitals. For further research it would be

worthwhile to develop a specific ICU pDDI database.

The key strength of this study is that it was performed in a general ICU of an

academic hospital with a large and diverse patient population, which increases

the generalizability of the results to other ICUs. Moreover, potential outcomes

and management strategies were analyzed. A limitation of this study may be

that potential adverse drug events (ADEs) resulting from pDDIs have not been

studied. Although theoretically it would be worthwhile to measure the clinical

consequences of the pDDIs, in practice it is almost impossible to attribute clinical

outcomes to pDDIs in complex and severely ill ICU patients.


48

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Table 5. Drug-Drug Interactions Involving Five Specific Intensive Care Unit Medicationsa Missing in The G-Standaard

Drug-drug pair EffectPatients% (n)

Propofol + midazolam Serum concentration of propofol/midazolam may increase

16% (273)

Dopamine + sympathicomimetics

Sympathicomimetic may increase the effect of other sympathicomimetics

12% (202)

propofol + CYP2D6 inhibitors

Decreased metabolism of propofol 7.2% (120)

Epinephrine + sympathicomimetics

Sympathicomimetic may increase effect of other sympathicomimetics

4.0% (67)

Dobutamine + calcium salts Effect of dobutamine may decrease 3.5% (58)

Norepinephrine + β-adrenergic blockers

Effect norepinephrine may increase 3.0% (50)

(nor)epinephrine + spironolacton

Vascular effect of (nor)epinephrine may decrease

2.3% (38)

Dobutamine + β-adrenergic blockers

Effect of dobutamine may decrease 2.1% (35)

(nor)epinephrine + Carbonic anhydrase inhibitors

Excretion of (nor)epinephrine may decrease

1.9% (32)

Epinephrine + antacids Excretion of epinephrine may decrease 1.3% (21)

Norepinephrine + antidepressant, tricyclic

Effect norepinephrine may increase 0.8% (13)

Epinephrine + rocuronium Increased risk on postoperatieve paralysis

0.8% (13)

Epinephrine + haloperidol Decreased effect of epinephrine 0.1% (2)

afive specific ICU medications: propofol, dopamine, dobutamine, epinephrine, norepinephrine

Conclusion

Potential drug-drug interactions occur in 54% of all ICU patients. A limited set of

20 pDDI-pairs is responsible for more than 90% of all pDDIs occurring on the ICU.

Therefore it is very worthwhile to develop guidelines for the management of these

specific pDDIs. Because the vast majority of the interactions can be managed by

monitoring, advanced clinical decision support systems linking laboratory data

to prescription data may be an important tool for effective management of risks

associated with pDDIs.

Acknowledgment

The authors are grateful to Hanneke den Breeijen for the data analysis.

49

2.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

References

1 Leape LL, Brennan TA, Laird N et al. The nature of adverse events in hospitalized patients: results of

the Harvard Medical Practice Study II. N Engl J Med 1991;324:377-384

2 Nebeker JR, Hoffman JM, Weir CR, Bennett CL, Hurdle JF. High rates of adverse drug events in a highly

computerized hospital. Arch Intern Med 2005;165:1111-1116

3 Krahenbuhl-Melcher A, Schlienger R, Lampert M, Haschke M, Drewe J, Krahenbuhl S. Drug-related

problems in hospitals: a review of the recent literature. Drug Saf 2007;30:379–407

4 BJ Kopp, Erstad BL, Allen ME, Theodorou AA, Priestley G. Medication errors and adverse drug events

in an intensive care unit: direct observation approach for detection. Crit Care Med 2006;34:415-425

5 Reis AMM, de Bortoli Cassiani SH. Adverse drug events in an intensive care unit of a university

hospital. Eur J Clin Pharmacol 2011;67: 625-632

6 Bertsche T, Pfaff J, Schiller P et al. Prevention of adverse drug reactions in intensive care patients by

personal intervention based on an electronic clinical decision support system. Intensive Care Med

2010;36:665–672

7 Zwart-van Rijkom JE, Uijtendaal EV, ten Berg MJ, van Solinge WW, Egberts ACG. Frequency and nature

of drug-drug interactions in a Dutch university hospital. Br J Clin Pharmacol 2009;68:187-193

8 Reimche L, Forster AJ, van Walraven C. Incidence and contributors to potential drug-drug interactions

in hospitalized patients. J Clin Pharmacol 2011;51:1043-1050

9 Institute of Safe Medication Practices: ISMP’s list of high alert medications 2008. http://www.ismp.

org/Tools/highalertmedications.pdf. (Accessed at March 1th, 2013)

10 Spriet I, Meersseman W, de Hoon J et al. Mini series: II-clinical aspects: clinically relevant CYPP450-

mediated drug interactions in the ICU. Intensive Care med 2009;35:603-612

11 Smithburger PL, Kane-Gill SL, Seybert AL. Drug-drug interactions in cardiac and cardiothoracic

intensive care units. Drug Saf 2010;33:879-888

12 Smithburger PL, Kane-Gill SL, Seybert AL. Drug-drug interactions in the medical intensive care unit:

an assessment of frequency, severity and the medications involved. Int J Pharm Pract 2012;20:402-408

13 Lima REF, Cassiani, SHB. Potential drug interactions in intensive care patients at a teaching hospital.

Rev LatAm Enfermagem 2009;17:222-227

14 Askari M, Eslami S, Louws M, et al. Frequency and nature of drug-drug interactions in the intensive

care unit. Pharmacoepidem Drug Safe 2013;22:430–437

15 van Roon EN, Flikweert S, le Comte M. Clinical relevance of drug-drug interactions : a structured

assessment procedure. Drug Saf 2005;28:1131-1139

16 Cerner, Kansas City: The APACHE IV equations: benchmarks for mortality and resource use. https://

apachefoundations.cernerworks.com/apachefoundations/resources/APACHE%20IV%20White%20

Paper%20Version%201.0.pdf (Accessed at July 1th, 2013)

17 Up-to-Date, Inc.: Lexi-interact online. http://www.uptodate.com/crlsql/interact/frameset.jsp (Accessed

at July 1th, 2013)

18 Thomson Reuters (Healthcare) Inc.: Micromedex healthcare series. Interactions. Updated periodically.

Available from http://micromedexsolutions.com/micromedex2/librarian. (Accessed at March 1th,

2013)

3Monitoring drug therapy in


3.1Frequency and determinants of

laboratory measurements for serum

potassium, sodium and creatinine in





Toine C.G. Egberts

Maarten J. ten Berg

Chapter 3.1 | Frequency and determinants of laboratory measurements for serum potassium, sodium and serum creatinine in hospitalized patients

54

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Abstract

BackgroundAdequate laboratory monitoring is considered important in the prevention of adverse drug events (ADEs). Information on laboratory markers is also essential in medication reconciliation at discharge. There is limited knowledge, however, on how often and when general biomarkers that are relevant to pharmacotherapy, such as serum potassium, sodium and creatinine are measured in daily clinical hospital practice.

ObjectiveThis study aimed to estimate the frequency and timing of measurement of serum creatinine, potassium and sodium in hospitalized patients and determinants thereof. Determinants of special interest were the use of potassium- and sodium influencing medications and the use of drugs in which renal function needs to be known (DRF drugs).

MethodsA retrospective observational study was conducted using the Utrecht Patient Oriented Database. All patients aged 18 years or older hospitalised for ≥24 hours were included. The proportion of patients with at least one measurement of potassium, sodium or creatinine respectively was calculated. The stratified estimates per medication-group were expressed as percentage and as odds ratios (ORs) with 95% confidence intervals. Also the time between the last measurement and discharge was determined. Potassium- and sodium influencing medications and DRF drugs were determined in accordance with the Dutch professional medication guideline on clinical risk management.

ResultsPotassium, sodium and creatinine were measured at least once during hospital admission in about 50% of the patients. The percentages of patients with a measurement was associated with patient- and admission related factors, i.e. percentages varied largely among different age groups and medical specialties, but percentages were also associated with the use of specific medications, i.e. potassium levels were measured in 99% of the patients using a potassium supplement. The use of a DRF drug did not influence laboratory monitoring of serum creatinine. Relevant laboratory markers were not measured within the last 24 hours before discharge in the majority of the patients using medication that requires laboratory monitoring, i.e. ranging from 72% of the patients using a potassium decreasing drug to 88% of the patients using a DRF drug.

ConclusionThis study provided insight into the frequency and determinants of monitoring electrolytes and creatinine at our institution. The results suggest that required laboratory monitoring in pharmacotherapy for detecting ADEs is low. In addition, we observed a low frequency of laboratory monitoring at discharge. Effort should be made to improve monitoring for ADEs.

55

3.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Introduction

The prevention of adverse drug events (ADE) has become a major public health

issue since it was shown that these occur frequently, are to a substantial part

potentially preventable and contribute majorly to health care costs1,2. It has been

estimated that 6.5 ADEs and 5.5 potential ADEs occur per 100 admissions and that

approximately 30-50% could have been prevented1,3.

Part of the ADEs may be prevented by adequately monitoring the clinical effects

during the course of drug therapy4. This monitoring may consist of careful direct

patient observation (e.g. skin reactions), but monitoring may also concern the

measurement of physical parameters like blood pressure and biochemical markers

like serum creatinine. Adequate monitoring of drug therapy is considered one of

the most important strategies to prevent ADEs; 61% of potentially preventable ADEs

occurred due to inadequate monitoring of drug therapy or a delayed response or

failure to respond to signs and symptoms or laboratory evidence of drug toxicity3.

Inadequate monitoring was also the most frequently reported category (22-45%)

of drug therapy problems leading to preventable hospital admissions5,6. The most

frequently involved drug-related problems leading to hospital admission were

bleeding, disturbances of serum electrolyte levels and kidney insufficiency5,7.

Monitoring is also the most frequently advised mitigation strategy in clinical

guidelines for risks related to drug-drug interactions8. A limited set of laboratory

markers, i.e. international normalized ratio (INR), serum creatinine and serum

electrolytes are important to monitor for the majority of these problems7,9-11. This

knowledge has led to an amendment of the Dutch Drug Law, obligating the exchange

of information between health care professionals on a limited set of laboratory

markers namely drug serum levels, INR, serum creatinine, potassium, sodium

levels and pharmacogenetic markers12. Due to practical implementation problems

this law was amended again in July 2013, now only demanding to communicate

laboratory values on kidney function when these deviate from normal limits13.

Whether health care professionals adhere to this obligation, however, is not known.

The exchange of these laboratory data between health care providers aims to

contribute to a continuity of care14. Discontinuity of care has been found to lead

to significant harm to patients, including harm due to ADEs7,15. Because laboratory

facilities are easily accessible and intensely used in hospitals, values for these markers

will be generally available for hospitalized patients. Information on these markers

can than also be provided upon discharge. However, there is limited knowledge on

how often these biomarkers relevant to pharmacotherapy are measured in daily

clinical practice. A recent study showed that serum creatinine levels were measured

in 59% of the hospitalized elderly patients16, but this frequency is not known for other


56

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

relevant drug-monitoring-biomarkers or for other groups of hospitalized patients. It

is also not known whether the decision to measure these laboratory values depends

on patient, medication and hospitalization characteristics. Moreover, it is not

known when this measurement is taken, shortly after admission and/or just before

leaving the hospital. If a measurement is performed shortly after admission, values

may represent the patients’ illness rather than the patients’ condition at discharge.

This may be relevant when laboratory data are communicated to the next health

care provider after discharge. This study therefore aims to estimate the frequency of

measurement of three general laboratory markers, serum potassium, sodium and

creatinine, in hospitalized patients and determinants thereof. Of special interest is

the association between the use of potassium- and sodium influencing medication

and drugs needing dose adjustment in case of renal impairment and measurement

of these biomarkers. Finally, this study also aims to determine the timing of these

laboratory measurements during hospitalization.

Methods

Design and setting This hospital-based retrospective observational study was performed in the

University Medical Centre Utrecht (UMCU), a 1042-bed academic medical centre

located in the Netherlands. At the UMCU all medications are prescribed using a

computerized physician order entry system. For research purposes all prescriptions

are routinely exported in the Utrecht Patient Oriented Database (UPOD). UPOD is

an infrastructure of relational databases comprising data on patient demographics,

hospital discharge diagnoses, medical procedures, medication orders and laboratory

tests for all patients treated at the UMCU since 2004, and has been described in

detail elsewhere17.

Study populationAll patients aged 18 years or older hospitalized in the UMCU between October 1st

2011 and October 1st 2013 for ≥24 hours were included. Per patient only the first

episode of hospital admission during the study period was included and only the

time spent on the nursing wards was included, the days that patients stayed at the

intensive care unit (ICU) were excluded from the analysis.

OutcomesPer patient all laboratory values of serum potassium, sodium and creatinine

measured during admission were collected. The proportion of patients with at least

one measurement during hospitalization was assessed for potassium, sodium and

57

3.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

creatinine. In addition, the percentage of patients in which all three biomarkers

were measured at least once was determined.

DeterminantsPatient-, admission, and medication related determinants for laboratory marker

measurement were studied. Patient-related determinants evaluated were gender,

age, treatment specialism at discharge, urgent admission (defined as admitted

through the emergency department). Admission related determinants were ICU

stay during admission, length of stay, and number of medication orders during

hospital stay. Medication-related determinants were the use of a serum potassium

increasing drug, a serum potassium decreasing drug, a serum sodium decreasing

drug or the use of a drug that is renally cleared and requires dose modification in

case of decreased renal function (DRF-drug). A drug was classified as a potassium

increasing, potassium decreasing and sodium decreasing drug as the drug was

known to affect serum potassium or serum sodium levels respectively18-21 and if

monitoring was advised according to the Dutch professional medication guideline

on clinical risk management22. The drugs that are classified into one of the potassium

increasing, potassium decreasing or sodium decreasing subgroups are represented

in Table 1. DRF-drugs were defined following the classification in the Z-Index of

October 2013: dose reduction for decreased renal function was determined relevant

for 250 of the 606 (41%) medications that are renally cleared.

Table 1: Drugs defined as potassium increasing, serum potassium decreasing and serum sodium decreasing drugs

Serum potassium increasing drugs

Serum potassium decreasing drugs

Serum sodium decreasing drugs

Potassium supplements diuretics (lis- and thiazides) diuretics (lis- and thiazides)

RAAS-inhibitors acetazolamide (ox)carbamazepine

potassium saving diuretics mineralocorticosteroids SSRI’s

potassium-binders venlafaxine and duloxetine

Data analysisThe proportions of patients with at least 1 measurement during hospitalization were

stratified per biomarker for patient characteristics, hospitalization characteristics

and medication-related factors. The stratified estimates per medication-group

were also expressed as odds ratios (ORs) with 95% confidence intervals. ORs were

adjusted for gender, age, treatment specialty at discharge, IC stay during admission,

admission due to an emergency, length of hospital stay and the number of

medication orders during hospital stay.


58

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

For the three biomarkers the number of hours between the time of discharge and

the time of last measurement was calculated per patient. In addition, the proportion

of patient with at least one measurement within the last 24 hours of admission was

calculated per biomarker. These proportions were stratified for medication-groups.

Data analyses were performed using SPSS release 20 (SPSS release 20 (SPSS, Inc.,

Chicago, Illinois).

Results

During the study period, 31,982 patients ≥= 18 years were hospitalized ≥ 24 hours in

the UMCU. Mean age was 52.7 years (95%CI = 52.5-52.9 years) and median length of

stay was 3.2 days (interquartile range = 1.9-7.1 days). Approximately one third of the

patients were hospitalized for surgical specialties and 10% of the patients stayed

one or more days at the ICU during hospital admission (Table 2).

In total, 48% of the patients had at least one potassium measurement, 48% had at

least one sodium measurement and 49% had at least one creatinine measurement

during hospitalization. Most patients with at least one measurement had

measurement for all three laboratory markers during hospital stay, i.e. all three

biomarkers were measured in 45% of the patients (Figure 1a).

The proportion of patients with at least one measurement of the biomarkers stratified

for patient- and admission-related factors is presented in Table 2. Proportions

increased with increasing age, length of stay, and number of medication orders, e.g.

potassium was measured in 25% of the patients of 35 years and younger and this

percentage increased to 70% for patients older than 70 years. The percentages of

measurements were also higher in males, for patients who had been admitted due

to an emergency and for patients who stayed at the ICU during hospital stay. With

regard to the treatment specialism, percentages of measurements were highest

for cardio-pulmonary patients (73-78%) and lowest for gynecology patients (7-15%).

Figure 1 shows which measurements were made in patients using a medication

from one of the medication groups.

59

3.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40Tab

le 2

: Pe

rcen

tage

of

pat

ien

ts w

ith

at

leas

t on

e m

easu

rem

ent

for

seru

m p

otas

siu

m,

sod

ium

an

d c

reat

inin

e p

er c

ateg

ory

of p

atie

nt

char

acte

rist

ics

n

um

ber

(to

tal)

K+

mea

sure

men

t (%

)N

a+ m

easu

rem

ent

(%)

Cre

at m

easu

rem

ent

(%)

Gen

der

M15

480

55.4

55.2

55.4

F

1650

241

.040

.842

.4

age

cate

gory

(yr)

≤35

7735

24.9

24.8

28.1

35-5

588

6741

.441

.241

.8

55-7

087

1759

.158

.958

.2

>7

066

6369

.168

.869

.3

trea

tmen

t sp

ecia

lty

at d

isch

arge

gyn

ecol

ogy

4872

6.5

6.4

15.3

surg

ery

1049

432

.132

.030

,0

onco

logy

1886

59.3

59.0

56.5

inte

rnal

3951

69.9

69.0

72.9

brai

n53

2671

.371

.264

.9

ca

rdio

-pu

lmon

ary

5453

73.3

73.2

78.3

IC a

dm

issi

on d

uri

ng

hos

pit

aliz

atio

nN

o28

738

43.3

43.1

44.3

Ye

s32

4489

.589

.387

.5

urg

ent

adm

issi

onN

o23

608

36,0

35.8

37.2

Ye

s83

7481

.881

.781

.2

len

gth

of

stay

, tot

al (d

ays)

1-2

8931

20.5

20.3

24.0

3-4

1130

337

.136

.937

.6

>=

511

748

79.4

79.2

78.1

nu

mbe

r of

med

icat

ion

ord

ers

(du

rin

g h

osp

ital

sta

y)<=

510

310

24.3

24.2

28.6

6-13

1089

941

.241

.141

.5

>=

1410

773

77.5

77.1

75.2


60

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Figure 1: Distribution of serum potassium, sodium and creatinine measurements per stratum of medication group

Figure 1: distribution of serum potassium, sodium and creatinine measurements per stratum of medication group

1.a: distribution for all patients

1.b: Patients using a potassium increasing drug (n=8739 (25%)

1.c: Patients using potassium decreasing drug n=5881 (18%)

1.d: Patients using sodium decreasing drug n=6670 (21%)

1.e: Patients using a using a drug that is renal cleared or affects renal function n=24542 (77%)







1.a: distribution for all patients 1.d: Patients using sodium decreasing drug n=6670 (21%)













1.b: Patients using a potassium 1.e: Patients using a using a drug that isincreasing drug (n=8739 (25%) renal cleared or affects renal function n=24542 (77%)














61

3.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Table 3 shows the percentage of patients with at least one measurement for serum

potassium, sodium or creatinine per stratum of medication group. Serum potassium

was measured more often in patients using a serum potassium increasing (82% vs

35%, ORadjusted = 2.4 (95%CI 2.2-2.6)) or a potassium decreasing drug (82% vs 40%,

ORadjusted = 1.8 (95%CI 1.6-2.0)) compared to patients not using such a drug. The

highest percentage of patients with at least one potassium measurement was seen

in patients using potassium supplements, this percentage was 99% (ORadjusted = 52

(95%CI 35-77)), numbers not mentioned in tables) or in patients using a potassium

binding drug (99%, ORadjusted = 18 (95%CI 5.4-60).

At least one serum sodium level was measured in 79% of the patients using a

serum sodium decreasing drug and in 40% of the patients not using such a drug

(ORadjusted=1.7 (95%CI 1.5-1.8)). The use of a selective serotonin reuptake inhibitor

(SSRI) did not contribute to the percentage of patients with at least one serum

sodium measurement. Serum sodium levels were measured in 64% of the SSRI

users 47% of the non-users (ORadjusted = 1.11 (95%CI 0.93-1.32)). Within the group

of diuretics, only the use of a lisdiuretic contributed to the measurement of at least

one serum sodium level (ORadjusted = 2.75 (95%CI 2.43-3.12) for users of lisdiuretics

and 1.06 (95%CI 0.93-1.21) for users of thiazide diuretics). The majority of the patients

(77%) used a drug that is renally cleared and requires dose modification in case

of decreased renal function (DRF drug) during hospitalization. Serum creatinine

was measured in 55% of these patients and in 30% of the patients not using such

drugs, but the use of a DRF-drug did not contribute to the percentage of patients

with at least one serum creatinine measurement. After correction for patient and

admission related factors, the percentage of patients with at least serum creatinine

measurement was even slightly decreased compared to patients not using a DRF-

drug (ORadjusted = 0.89 (0.81-0.97)).

62

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Tab

le 3

: N

um

ber

of p

atie

nts

wit

h a

mea

sure

men

t fo

r se

rum

pot

assi

um

, ser

um

sod

ium

an

d s

eru

m c

reat

inin

e p

er m

edic

atio

n c

ateg

ory

nu

mb

er

tota

l (u

sers

)

nu

mb

er

tota

l (n

on

u

sers

)

% K

+ m

easu

red

w

hen

usi

ng

% K

+ m

easu

red

w

hen

no

t u

sin

gO

R a

dj

e

% N

a+

mea

sure

d

wh

en u

sin

g

% N

a+

gem

eten

w

hen

no

t u

sin

gO

R a

dj

e

%

Cre

atin

ine

mea

sure

d

wh

en u

sin

g

%

Cre

atin

ine

mea

sure

d

wh

en

no

t u

sin

gO

R a

dj

e

K i

ncr

easi

ng

dru

g a

8739

2324

381

.835

.32.

37 (2

.18-

2.57

)81

.435

.12.

33 (2

.15-

2.52

)80

.336

.82.

14 (1

.98-

2.32

)

K d

ecre

asin

g d

rug

b58

8126

101

81.7

40.4

1.81

(1.6

4-1.

98)

81.3

40.2

1.78

(1.6

2-1.

96)

80.3

41.6

1.66

(1.5

1-1.

82)

Na

dec

read

ing

c66

7025

312

79.4

39.7

1.66

(1.5

2-1.

81)

79.1

39.5

1.67

(1.5

3-1.

82)

77.9

41.0

1.57

(1.4

5-1.

71)

DR

F d

rug

d

2454

274

4055

.523

.21.

31 (1

.19-

1.45

)55

.223

.31.

27 (1

.15-

1.40

)54

.529

.50.

89 (0

.81-

0.97

)

all

pat

ien

ts (t

otal

)

48

.0

47

.8

4

8.7

a pot

assi

um

in

crea

sin

g d

rugs

: pot

assi

um

su

pp

lem

ents

, Ren

ine

An

giot

ensi

ne

Ald

oste

ron

in

hib

itor

s, p

otas

siu

m s

avin

g d

iure

tics

b pot

assi

um

dec

reas

ing

dru

gs: d

iure

tics

(lis

an

d t

hia

zid

es),

acet

azol

amid

e, m

iner

aloc

orti

cost

eroi

ds,

pot

assi

um

-bin

din

g d

rugs

c sod

ium

dec

reas

ing

dru

gs: d

iure

tics

(lis

an

d t

hia

zid

es),

(ox)

carb

amaz

epin

e, S

SRI’

s, v

enla

faxi

ne,

du

loxe

tin

ed D

RF:

dru

gs t

hat

are

ren

ally

cle

ared

an

d r

equ

ire

dos

e m

odifi

cati

on i

n c

ase

of a

dec

reas

ed r

enal

fu

nct

ion

e ad

just

ed f

or s

exe,

age

, sp

ecia

lity

at

dis

char

ge, I

CU

sta

y d

uri

ng

hos

pit

al s

tay,

urg

ent

adm

issi

on, l

engt

h o

f st

ay a

nd

nu

mbe

r of

med

icat

ion

ord

ers

du

rin

g h

osp

ital

sta

y

63

3.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

The timing of the last serum potassium, sodium and creatinine measurement prior

to discharge is presented in Figure 2 and Table 4. Of all patients, 11-12% had at

least one measurement of potassium, sodium or creatinine within the last 24 hours

before discharge. At least one serum potassium level, was obtained for 26%-28% of

the patients using a serum potassium influencing drug within 24 hours before

discharge. These percentages increased to 46-48% for the time window of 48 hours

prior to discharge. Twenty three percent of the hospitalized patients using a serum

sodium decreasing drug had at least one measurement of serum sodium within the

24 hours prior to discharge. This percentage increased to 42% when a time window

of 48 hours prior to discharge was taken. Twelve percent of the patients using a DRF

drug had a serum creatinine measurement within 24 hours prior to discharge. This

percentage increased to 25% for the time window of 48 hours prior to discharge.

Perc

ent o

f pat

ient

s

14

12

10

8

6

4

2

0

Time of last serum potassium measurement (hours before discharge)936864792720648576504432360288216144720

Page 1

Figure 2a: Time of last serum potassium measurement before discharge (all patients)


64

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Time of last serum sodium measurement (hours before discharge)936864792720648576504432360288216144720

Perc

ent o

f pat

ient

s

14

12

10

8

6

4

2

0

Page 1

Figure 2b: Time of last serum sodium measurement before discharge (all patients)

Time of last serum creatinine measurement (hours before discharge)936864792720648576504432360288216144720

Perc

ent o

f pat

ient

s

14

12

10

8

6

4

2

0

Page 1

Figure 2c: Time of last serum creatinine measurement before discharge (all patients)

65

3.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40Tab

le 4

: Pe

rcen

tage

of p

atie

nts

wit

h a

t le

ast

one

mea

sure

men

t fo

r se

rum

pot

assi

um

, sod

ium

or

crea

tin

ine

per

tim

e fr

ame

of 2

4, 4

8 an

d

72 h

ours

res

pec

tive

ly b

efor

e d

isch

arge

Pota

ssiu

m m

easu

rem

ent

Sod

ium

mea

sure

men

tC

reat

inin

e m

easu

rem

ent

Dru

g u

seN

um

ber

to

tal

wit

hin

24

ho

ur

bef

ore

d

isch

arge

(%)

wit

hin

48

ho

ur

bef

ore

d

isch

arge

(%)

wit

hin

72

ho

ur

bef

ore

d

isch

arge

(%)

wit

hin

24

ho

ur

bef

ore

d

isch

arge

(%)

wit

hin

48

ho

ur

bef

ore

d

isch

arge

(%)

wit

hin

72

ho

ur

bef

ore

d

isch

arge

(%)

wit

hin

24

ho

ur

bef

ore

d

isch

arge

(%)

wit

hin

48

ho

ur

bef

ore

d

isch

arge

(%)

wit

hin

72

ho

ur

bef

ore

d

isch

arge

(%)

K i

ncr

easi

ng

dru

g87

3925

.846

.158

.923

.343

.156

.321

.240

.252

.9

no

K i

ncr

easi

ng

dru

g23

243

6.6

15.2

20.6

6.5

14.9

20.3

8.0

17.1

22.5

K d

ecre

asin

g d

rug

5881

27.7

48.2

60.4

25.5

45.3

57.8

24.3

44.0

56.2

no

K d

ecre

asin

g d

rug

2610

18.

318

.124

.47.

817

.523

.98.

718

.825

.1

Na

dec

reas

ing

dru

g66

7024

.944

.456

.123

.242

.254

.121

.840

.652

.3

no

Na

dec

reas

ing

dru

g25

312

8.4

18.1

24.4

7.9

17.5

23.8

8.9

18.9

25.1

dru

g w

ith

dos

age

dep

end

ent

on k

idn

ey f

un

ctio

n24

542

13.8

26.8

35.6

12.8

25.5

34.5

12.3

24.8

33.5

no

dru

g w

ith

dos

age

dep

end

ent

on k

idn

ey f

un

ctio

n74

405.

313

.015

.85.

213

.015

.89.

318

.721

.8

tota

l31

982

11.9

23.6

31.0

11.1

22.6

30.1

11.6

23.4

30.8


66

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Discussion

In this study we observed that at our institution potassium, sodium and creatinine,

are measured at least once during hospital admission in about half of the patients.

In addition, we observed that monitoring of these biomarkers, which are relevant

biomarkers in clinical risk management of ADEs, is associated with the use of

specific medication. Finally, we observed that in the majority of the patients there is

no laboratory measurement for these markers in the last 24 hours before discharge

(88-89%), however in patients treated with medication that require laboratory

monitoring, this percentage is lower (72-88%).

Patients with at least one measurement, most often had laboratory measurements

for all three biomarkers during hospitalization (45% of all patients). The fact that

not one but all three biomarkers are measured during hospitalization may indicate

that patient characteristics are more important determinants for monitoring than

the use of drugs. Percentages of patients with a measurement indeed varied strongly

depending on patient and admission related factors, e.g. 7% for patients treated for

gynecology specialties to 90% for patients with an ICU stay during hospitalization.

The association between laboratory monitoring and the use of specific medications

was of special interest in this study. The percentage of patients with at least

one measurement was higher when patients were using a serum potassium

increasing, serum potassium decreasing or sodium decreasing drug, but these

percentages varied less compared to the percentages found for the different

patient characteristics. However, when the medication was intended to influence

serum levels, these levels were almost always monitored (e.g. potassium levels

were measured in 99% of the patients using potassium supplements or potassium

binding drugs). When the influence on serum levels was a side effect of drug

treatment, these levels were measured less frequently (e.g. potassium levels were

measured in 75% of the patients using renin-angiotensin system (RAS) inhibitors).

From this observation we can conclude that required laboratory monitoring is

suboptimal in our institution, putting patients at risk for ADEs.

The use of the considered medications was found be related to certain medical

specialties, i.e. potassium increasing drugs like RAS inhibitors and spironolactone

are used more often in patients hospitalized for cardiac specialties than in patients

hospitalized for other medical specialties. After adjustment for patient factors

such as medical specialty, however, the adjusted ORs for the association between

drug use and laboratory monitoring was still significant, suggesting that the use of

medication influences monitoring as well.

The use of a DRF-drug did not contribute to the percentage of patients with at

least one serum creatinine measurement. The OR for at least one serum creatinine

67

3.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

measurement was even lower for patients using a DRF-drug than for patients not

using such a drug (OR=0.89; 95%CI = 0.81-0.97). The CPOE, however, did not signal to

verify kidney function when a DFR-drug was prescribed, signaling would probably

have improved the adherence to serum creatinine monitoring. Since the majority

of the patients (77%) were using a DRF drug, signaling only would have led to a high

burden of signals, causing alert fatigue and the risk that physicians override both

important and unimportant signals23,24. The specificity and clinical relevance of

these signals should therefore be high, ie, only be generated when kidney function

drops below the threshold value for a specific DRF drug. An advanced clinical

decision support system in which prescription data are linked to laboratory data is

necessary for this25,26. Additionally, if serum creatinine values are not measured in

advance, a corollary order for serum creatinine measurement may further improve

adherence to monitoring protocols27. The extent to which corollary orders may

increase monitoring frequency and may prevent ADEs, however, is not known and

needs further research.

As exchange of information on medication between health care providers may

contribute to continuity of care, and laboratory values are often involved in the

monitoring of pharmacotherapy, these values should also be transferred at

discharge. Values measured during hospitalization, however, are only suitable

when they represent the patients’ health status at the moment of discharge. Of all

hospitalized patients, only 11-12% had at least one potassium, sodium or creatinine

measurement within the last 24 hours before discharge. For patients using a serum

potassium influencing drugs, 26-28% had at least one potassium measurement

within 24 hours before discharge. This percentage increased to 46-48% when the

last 48 hours before discharge was considered. Comparable percentages were found

for serum sodium measurements in patients using serum sodium decreasing

drugs. Percentages were considerably lower for creatinine measurements in

patients a DRF drug. Only 12% of the patients using a DRF drug had at least one

serum creatinine value within the last 24 hours before discharge. This percentage

increased to 25% when the last 48 hours before discharge was considered,

although, a serum creatinine measurement within 48 hours prior to discharge

was missing in 75% of the patients using a DRF drug. As the clinical relevance of

many DDIs can only be assessed when information on serum electrolytes and/or

serum creatinine is available, these percentages indicate that there is still room

for improvement. E.g. the DDI between RAS inhibitors and spironolactone is one

of the most frequently occurring DDIs in hospitals and patients using these drugs

concomitantly are especially at risk developing hyperkalemia when their kidney

function is decreased. Moreover, a decreased kidney function has also shown one of

the most frequently involved drug-related problems leading to hospital admission7.


68

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Transition of information on laboratory levels should therefore be part of the

medication reconciliation at discharge. The time frame, however, which must be

held to consider a laboratory value valid for transfer at discharge may be discussed.

Guidelines on transfer of medication data also lacks clarity on which laboratory

data can be considered valid28-30. As clinical pharmacists and clinical chemist

share responsibility with physicians in the transition of care, measurement of

laboratory markers should not be reserved to physicians only. When pharmacists,

for example, notice that a valid value is missing at discharge, they should have

the possibility to order or perform a measurement. Even point of care testing has

shown feasible in this31. Finally, when information on pharmacy and laboratory

data are transferred together with a monitoring advise at discharge, unnecessarily

repeated measurements may be prevented in primary care

Conclusion

Measurement of the three biomarkers potassium, sodium and creatinine, occurs

on average in 50% of the patients, but varies strongly across patient-, admission-

and medication-related factors. The use of electrolyte influencing medication

seems to influence laboratory monitoring, the use of a DRF drug did not seem to

influence laboratory monitoring of serum creatinine. In addition, the percentage

of patients in which these biomarkers are measured at the end of hospitalization

is low. As information on these laboratory markers is essential in medication

reconciliation at discharge, efforts should be given to improve the availability of

relevant laboratory monitoring data.

69

3.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

References

1 Bates DW, Cullen DJ, Laird N, Petersen LA, Small SD, Servi D, Laffel G, Sweitzer BJ, Shea BF, Hallisey

R, vander Vliet M, Nemeskal R, Leape LL. Incidence of adverse drug events and potential adverse drug

events. Implications for prevention. ADE Prevention Study Group. JAMA 1995;274:29-34

2 Bates DW, Spell N, Cullen DJ, Burdick E, Laird N, Petersen LA, Small SD, Sweitzer BJ, Leape LL. The costs

of adverse drug events in hospitalized patients JAMA 1997;277:307-311

3 Gurwitz JH, Field TS, Harrold LR, Rothschild J, Debellis K, Seger AC, Cadoret C, Fish LS, Garber L,

Kelleher M, Bates DW. Incidence and preventability of adverse drug among older persons in the


4 Coleman JJ, Ferner RE, Evans SJW. Monitoring for adverse drug reactions. Br J Clin Pharmacol

2006;61:371–378

5 Howard RL, Avery AJ, Slavenburg S, Royal S, Pipe G, Lucassen P, Pirmohamed M. Which drugs cause

preventable admissions to hospital? A systematic review. Br J Clin Pharmacol 2007;63:136-147

6 Thomsen LA, Winterstein AG, Søndergaard B, Haugbølle LS, Melander A. Systematic review of

the incidence and characteristics of preventable adverse drug events in ambulatory care. Ann

Pharmacother 2007;41:1411-1426

7 Leendertse AJ, Egberts AC, Stoker LJ, van den Bemt PM; HARM Study Group. Frequency of and risk

factors for preventable medication-related hospital admissions in the Netherlands. Arch Intern Med

2008;168:1890-1896

8 Zwart-van Rijkom JE, Uijtendaal EV, ten Berg MJ, van Solinge WW, Egberts AC. Frequency and nature


9 van der Hooft CS, Dieleman JP, Siemes C, Aarnoudse AJ, Verhamme KM, Stricker BH, Sturkenboom MC.

Adverse drug reaction-related hospitalisations: a population-based cohort study. Pharmacoepidemiol

Drug Saf 2008;17:365-371

10 Ministry of Health, Welfare and Sports. Den Haag, 2008. HARM-WRESTLING. A proposal by the

expert group on medication safety for concrete interventions that can improve mediation safety

in ambulatory care at short term (in Dutch) http://www.medicatieveiligheid.info/websites/nvza_

remedie/docs/Harmwrestling_rapport_20091.pdf (Accessed at May 1th, 2014)

11 Geerts AF, De Koning FH, De Smet PA, Van Solinge WW, Egberts TC. Laboratory tests in the clinical risk

management of potential drug-drug interactions: a cross-sectional study using drug-dispensing data

from 100 Dutch community pharmacies. Drug Saf 2009;32:1189-1197

12 Kingdom of the Netherlands. Regulation of the Ministry of Health, Welfare and Sports of 9

December 2011, amending the regulation on the Dutch Medicines Act (in Dutch) https://zoek.

officielebekendmakingen.nl/stcrt-2011-22913.html (Accessed at May 1th, 2014)

13 Kingdom of the Netherlands. Regulation of the Ministry of Health, Welfare and Sport of 2 July 2013,

amending the regulation on the Dutch Medicines Act (in Dutch) https://zoek.officielebekendmakingen.

nl/stcrt-2013-18918.html (Accessed at March 1th, 2014)

14 Abdullah-Koolmees H, Gerbranda T, Deneer VHM, Tjoeng MM & De Ridder AJM Gardarsdottir H,

Heerdink ER. Discontinuation of anticoagulant care during admission to a psychiatric hospital. Eur J

Clin Pharmacol 2013;69:1025–1029

15 Forster AJ, Murff HJ, Peterson JF, Gandhi TK, Bates DW. The incidence and severity of adverse events

affecting patients after discharge from the hospital. Ann Intern Med 2003;138:161-167

16 Drenth van Maanen AC, van Marum RJ, Jansen PAF, Zwart JEF, Solinge WW, Egberts TCG. Adherence

with dosing guideline in patients with impaired renal function at hospital discharge. Thesis 2013 -

under review

17 ten Berg MJ, Huisman A, van den Bemt PM, Schobben AF, Egberts AC, van Solinge WW. Linking

laboratory and medication data: new opportunities for pharmacoepidemiological research. Clin

Chem Lab Med 2007;45:13-19

18 Perazella MA. Drug-induced hyperkalemia: old culprits and new offenders. Am J Med 2000;109:307-

314

19 Gennari FJ. Hypokalemia. N Engl J Med 1998;339:451-458

20 Liamis G, Milionis H, Elisaf M. A review of drug-induced hyponatremia. Am J Kidney Dis 2008;52:144-

153


70

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

21 Nederlandse Internisten Vereniging. Guideline electrolyte disturbances (in Dutch) http://www.

internisten.nl/uploads/PH/-u/PH-uQr_fQ7huJ5eVGYBBVA/richtlijn_2012_elektrolytstoornissen.pdf

(Accessed at May 1th, 2014)

22 Scientific Institute of Dutch Pharmacists. The Hague, Dutch Surveillance Guideline (G-standard) (in

Dutch). http://www.knmp.nl/ (Accessed at May 1th, 2014)

23 Glassman PA, Simon B, Belperio P, Lanto A. Improving recognition of drug interactions: benefits and

barriers to using automated drug alerts. Med Care 2002;40:1161-1171

24 Sijs H, Mulder A, van Gelder T, Aarts J, Berg M, Vulto A. Drug safety alert generation and overriding in


25 Schiff GD, Klass D, Peterson J, Shah G, Bates DW. Linking Laboratory and Pharmacy Opportunities for

Reducing Errors and Improving Care. Arch Intern Med 2003;163:893-900

26 Seger AC, Jha AK, Bates DW. Adverse drug event detection in a community hospital utilizing

computerized medication and laboratory data. Drug Saf 2007;30:817-824

27 Overhage JM, Tierney WM, Zhou XH, McDonald CJ. A randomized trial of “corollary orders” to prevent

errors of omission. J Am Med Inform Assoc 1997;4:364-375

28 Policy document: transfer of information on medication [in Dutch] 2008. http://www.

medicatieoverdracht.nl/artikelen/raadplegen.asp?display=2&atoom=9008&atoomsrt=2&actie=

2&menuitem=189 (Accessed at May 1th, 2014)

29 The American Medical Association. The physician’s role in medication reconciliation: issues, strategies

and safety principles 2007. http://bcpsqc.ca/documents/2012/09/AMA-The-physician%E2%80%99s-role-

in-Medication-Reconciliation.pdf (Accessed at May 1th, 2014)

30 Australian Pharmaceutical Advisory Council. Guiding principles to achieve continuity in medication

management 2005. http://www.commcarelink.health.gov.au/internet/main/publishing.nsf/Content/

5B47B202BBFAFE02CA257BF0001C6AAC/$File/guiding.pdf (Accessed at May 1th, 2014)

31 Geerts AF, De Koning FH, Vooght KM, Egberts TC, De Smet PA, Van Solinge WW. Feasibility of point-of-

care creatinine testing in drug therapy of ambulatory elderly in community pharmacy. J Clin Pharm

Ther 2013;38:416-422

71

3.1

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

3.2Frequency of laboratory measurement

and hyperkalaemia in hospitalised

patients using serum potassium

concentration increasing drugs


Jeannette E.F. Zwart-van Rijkom,


Toine C.G. Egberts

Eur J Clin Pharmacol 2011;67:933–940

Chapter 3.2 | Frequency of laboratory measurement and hyperkalaemia in hospitalised patients using serum potassium concentration increasing drugs

74

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Abstract

Purpose

Although, drug-drug interactions (DDIs) between potassium increasing drugs

(PIDs) are known risk factors for developing hyperkalaemia, not much is known

about their risk and management strategies during hospitalisation. This study

examines the frequency of serum potassium measurements and hyperkalaemia in

hospitalised patients, based on the use of one or more PIDs, and the determinants

thereof.

Methods

Adult patients hospitalised in the University Medical Centre Utrecht between 2006-

2008 were included in this cross-sectional study. The frequency of serum potassium

measurement and of hyperkalaemia were compared between patients using only

one PID at a time (monotherapy group) and patients using two or more PIDs

concomitantly (interaction group). The determinants studied were renal failure,

diabetes mellitus, use of diuretics, type of DDI, start of the PIDs within the hospital

versus continued home medication, and medical speciality.

Results

Serum potassium was measured more frequently in the interaction group than in

the monotherapy group [67% vs. 56%; relative risk (RR) 1.19, 95% confidence interval

(CI) (1.14-1.24)] and the risk of hyperkalaemia was also increased in the interaction

group (9.9% vs. 5.9%, RR=1.7, 95% CI 1.3-2.1). The combination of potassium-sparing

diuretics plus a potassium supplement, start of the PID within the hospital and

hospitalisation in non-internal medicine departments was associated with higher

relative risk estimates for hyperkalaemia.

Conclusions

Among our patient cohort, even when physicians received a direct pop-up to

monitor serum potassium levels when prescribing two PIDs concomitantly, serum

potassium levels were not measured in 33% of patients and 10% of patients developed

hyperkalaemia. Improved management strategies and/or clinical decision-support

systems are needed to decrease the frequency of hyperkalaemia following DDIs.

75

3.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Introduction

Drug-drug interactions (DDI) often lead to preventable adverse drug events and

contribute to the burden of drug-related health damage1,2. The interaction between

serum potassium concentration increasing drugs (PIDs) is one of the most frequently

occurring DDI and has been reported to occur in 8-10% of all hospitalized and non-

hospitalized patients3,4.

Hyperkalaemia is a serious and potentially life threatening electrolyte disorder that

follows from an imbalance in potassium homeostasis5. The magnitude of the risk

of hyperkalaemia caused by the combined use of two or more PIDs was highlighted

after the findings of the Randomized Aldactone Evaluation Study (RALES) study

were implemented in regular patient care. In 1999, the RALES demonstrated

significantly improved outcomes for congestive heart failure patients treated with

spironolactone in addition to treatment with diuretics and angiotensin-converting

enzyme (ACE)-inhibitors6. Five years later, in 2004, it was found that this publication

had resulted in a clear increase in the prescription rates for spironolactone as

well as a marked increase in hyperkalaemia-associated morbidity and mortality7.

The risk does not only exist for renin-angiotensin system (RAS) inhibitors and

spironolactone, but also for other combinations of PIDs including potassium

supplements8-11.

Many studies have investigated the risk of developing hyperkalaemia from DDIs

between PID’s, but most studies have been performed in an outpatient setting,

upon hospital admission, in a specific patient group or with a specific combination

of PID’s. Moreover, in these studies, serum potassium levels were often measured

per protocol and were closely monitored. No published data are available about the

risk and management strategies of combining PIDs during hospitalisation where

the average stay is only a few days and the access to laboratory measurement is

more easy. Therefore, the objective of this study is to examine the frequency of

serum potassium measurement and of hyperkalaemia for hospitalised patients

using one or more PIDs, and determinants there of.

Methods

Setting This cross-sectional study was performed in the University Medical Centre

Utrecht (UMCU), a 1042-bed academic medical centre located in the centre of the

Netherlands. Between 2006-2008, approximately 30,000 clinical hospitalisations

took place annually. All medications for hospitalised patients are prescribed

using a computerized physician order entry (CPOE) system. In the Netherlands,


76

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

a working group of the Scientific Institute of Dutch Pharmacists developed and

maintains an evidence based and professional guideline for the management of

DDIs, the G-standard, which is described in detail elsewhere12. For clinical risk

management purposes, the G-standard is incorporated in the CPOE system13. All

drugs in the G-standard are coded according to the Anatomical Therapeutical

Chemical Classification (ATC) index. The ATC code system is an international

standard for drug utilization studies determined by a World Health Organisation

(WHO) International Working Group14. Each DDI in this guideline was assessed by

four core parameters: (1) evidence on the interaction; (2) clinical relevance of the

potential adverse reaction resulting from the interaction; (3) risk factors identifying

patient, medication or disease characteristics for which the interaction is of special

importance; (4) incidence of the adverse reaction. Based on the information in

the G-standard, the CPOE generates an alert when a combination of two or more

interacting drugs are prescribed concomitantly. In the case the concomitant

prescribing of two PIDs, the physician receives a direct pop-up that there is a risk of

hyperkalaemia and that the serum potassium levels should be closely monitored.

For research purposes all prescriptions are routinely exported in the Utrecht Patient

Oriented Database (UPOD), which is an infrastructure of relational databases

comprising data on patient demographics, hospital discharge diagnoses, medical

procedures, medication orders and laboratory tests for all patients treated at the

UMCU since 2004. This database is described in detail elsewhere15.

Study populationAll patients aged >18 years who were hospitalised in the UMCU in 2006-2008 for ≥24

h and who were prescribed at least one PID were enrolled in the study. If a patient

was admitted during the study period to the hospital more than once, only the

first admission was included. Patients admitted to the dialysis department were

excluded because patients at end-stage renal disease lack the renal compensation

mechanism for potassium homeostasis and large variations in serum potassium

levels may occur due to dialysis. Patients admitted to the intensive care (IC) units

were also excluded because a different CPOE system is used in the IC units and the

prescription data are not (yet) stored in the UPOD database.

ExposurePatients who were prescribed one PID (monotherapy group) were compared

to patients who were prescribed two or more PIDs concomitantly (interaction

group). In this study, PIDs were defined as: (1) RAS inhibitors, (2) potassium

sparing diuretics or (3) potassium supplements. The professional DDI clinical risk

management guideline has defined a combination of these drugs as a risk for

77

3.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

developing hyperkalaemia. Other combinations of potentially PIDs mentioned in

literature such as nonsteroidal anti-inflammatory drugs, β-adrenergic blocking

drugs, heparin and trimethoprim are not considered to be sufficiently relevant by

the professional DDI clinical risk management guideline and were therefore not

included.

OutcomeTwo different outcome measures were studied: (1) frequency of serum potassium

measurement and (2) frequency of hyperkalaemia. The frequency of serum

potassium measurement was defined as the percentage of patients for which

a serum potassium level was measured at least once during the hospitalisation

period. The frequency of hyperkalaemia was defined as the percentage of patients

with any serum potassium measurement ≥ 5.5 mEq/L during the hospitalisation

period.

DeterminantsIn addition to age and gender, the following determinants were studied:

- Renal function: glomerular filtration rates were estimated (eGFR) with the

Modification of Diet in Renal Disease (MDRD) equation16,17 using the highest

serum creatinine level measured during the hospitalisation period. A

decreased renal function is a known risk factor for hyperkalaemia.

- Diabetes mellitus: a patient was considered to have diabetes mellitus when

one or more medication orders for any anti-diabetic medication had been

prescribed during hospitalisation. Anti-diabetic medication was defined

as medication with an ATC code “A10”. Diabetes mellitus is also a known

risk factor for hyperkalaemia.

- Diuretics: any use of diuretics. This was defined as the use of lis- or thiazide

diuretica or a combination of drugs containing thiazide diuretics during

hospitalisation (ATC C03C, C03A). This determinant was studied, because

both the use of thiazide- and lisdiuretics have been associated with a

protective effect against the development of hyperkalaemia18

- Type of combination: in the interaction group three combinations are

possible, namely (1) RAS-inhibitor + potassium-sparing diuretic, (2) RAS-

inhibitor + potassium supplement, (3) potassium sparing diuretic +

potassium supplement.

- Start at home or start during hospitalisation: in the monotherapy group

start at home was defined as the first PID prescribed in the CPOE system

within 24 h of arrival in the hospital. In the interaction group start at

home was defined as the second PID prescribed in the CPOE system within


78

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

24 h of arrival in the hospital. PIDs prescribed >24 h after admission were

considered to be newly started within the hospital.

- Department: internal medicine departments versus non-internal medicine

departments. Internal medicine departments were defined as general

internal medicine departments, (namely Heart & Lung departments and

the departments of Geriatrics, Nephrology and Oncology). Non-internal

medicine departments were mainly the Surgery, Neurology, Psychiatry

and Gynaecology departments.

Data analysisRelative risks (RR) with corresponding 95% confidence intervals (CI) were calculated

for both outcome measures. The SPSS Package for Social Sciences software (SPSS)

version 15.1 for windows (SPSS., Chicago, IL) was used to analyse the collected data.

Results

A total of 9,441 patients using at least one PID were included in this study of whom

1,396 patients (14.7%) used two or more PIDs concomitantly at any time during

their hospitalisation (interaction group). On average, patients using ≥2 PIDs were

older, had a worse kidney function, and used anti-diabetic drugs and diuretics

more often than patients using only one PID (monotherapy group) (Table 1). Of

the patients in the monotherapy group, 35% started PID therapy during their stay

in the hospital; of those using > 2 PIDs, 50.4% were first prescribed the second PID

during their hospital stay.

Serum potassium was measured more frequently in the interaction group than

in the monotherapy group (67 vs. 56%, respectively RR 1.19, 95% CI 1.14-1.24), and

the risk of hyperkalaemia was also higher in the interaction group (9.9 vs. 5.9%,

respectively, RR 1.7, 95% CI 1.3-2.1) (Table 2). For patients whose potassium was

measured at least once, a median of 0.67 measurements per hospital admission

day were performed in the interaction group and 0.50 in the monotherapy group.

There were no no differences in the relative risk estimates between the monotherapy

and interaction groups when age, gender, renal function, diabetes and use of

diuretics were stratified. The absolute risk of hyperkalaemia, however, was higher in

patients with an eGFR ≤ 50 mL/min (p<0.01 for both monotherapy and interaction

groups).

79

3.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Table 1. Baseline characteristics for monotherapy group (users of 1 PID) versus interaction group (users of ≥2 PIDs)

Monotherapy:1 PIDN=8,045

Interaction:≥2 PIDsN=1,396

p valuea

Median length of stay, days (interquartile range)

6.18 (3.0-12.0) 8.9 (4.2-16.1) <0.001

Mean age (95% CI) 62.2 (61.8-62.5) 67.3 (66.6-68.1) <0.001

Male gender (%) 3,981 (49.5%) 751 (53.8%) <0.01

Renal functioneGFR ≥80 mL/mineGFR 50-80 mL/mineGFR ≤50 mL/minUnknown

1,479 (18.4%)1,876 (23.3%)1,150 (14.3%)3,540 (44.0%)

123 (8.8%)375 (26.9%)422 (30.2%)476 (34.1%)

<0.001

Diabetes mellitus 1,582 (19.7%) 340 (24.4%) <0.001

Diuretics 3,106 (38.6%) 1,097 (78.6%) <0.001

Drug type (first monotherapy)RAS inhibitorPotassium sparing diureticPotassium supplement

4,819 (59.9%)622 (7.7%)2,604 (32.4%)

n.a.

Interaction typeRAS-inhibitor + potassium sparing

diureticRAS-inhibitor + potassium

supplementPotassium sparing diuretic +

potassium supplement

707 (50.6%)

501 (35.9%)188 (13.5%)

n.a.

Start drug/interactionAt homeDuring hospitalisation

5,192 (64.5%)2,853 (35.5%)

692 (49.6%)704 (50.4%)

<0.001

DepartmentInternal medicine specialitiesNon-internal medical specialities

3,694 (45.9%)4,351 (54.1%)

977 (70.0%)419 (30.0%)

<0.001

Data are given as the number (n) of patients, with the percentage in parenthesis, unless indicated otherwise

PID, Potassium-increasing drug; CI, confidence interval; eGFR, estimated glomerular filtration rate; RAS,

renin-angiotensin system, n.a., not applicable

a) Chi square test (categorical variables) or t-test (continuous variables) as appropriate.

Comparison of the risk of hyperkalaemia for the different interaction types,

revealed that the risk was higher for the combination of a potassium supplement

plus a potassium-sparing diuretic (RR 3.0, 95% CI 2.0-4.4) than for the two other

combinations: a potassium-sparing diuretic plus RAS-inhibitor (RR 1.5, 95% CI 1.1-

2.1) and a potassium supplement plus RAS-inhibitor (RR 1.3, 95% CI 0.9-2.0).


80

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Tab

le 2

. Pe

rcen

tage

of

pat

ien

ts i

n w

hom

ser

um

pot

assi

um

was

mea

sure

d a

nd

per

cen

tage

of

pat

ien

ts w

ith

hyp

erka

laem

ia (

defi

ned

as

seru

m p

otas

siu

m l

evel

≥ 5

.5 m

Eql/

L) f

or u

sers

of

1 PI

D v

ersu

s u

sers

of

≥2 P

IDs

Pota

ssiu

m m

easu

red

Hyp

erk

alae

mia

Mon

oth

erap

y1

PID

(n=8

,045

)

Inte

ract

ion

≥PID

s (n

=1,3

96)

Rel

ativ

e ri

sk

(95%

CI)

Mon

oth

erap

y1

PID

(n

=4,5

20)

Inte

ract

ion

≥2 P

IDs

(n=9

30)

Rel

ativ

e ri

sk

(95%

CI)

Ove

rall

56.2

% (4

,520

/8,0

45)

66.6

% (9

30/1

,396

)1.

19 (1

.14-

1.24

)5.

9% (2

67/4

,520

)9.

9% (9

2/93

0)1.

68 (1

.34-

2.10

)

Age

18

-50

50-7

0 70

-80

>= 8

0

56.8

% (9

89/1

,742

)55

.3%

(1,9

21/3

,473

)54

.7%

(964

/1,7

61)

60.4

% (6

46/1

,069

)

70.8

% (1

14/1

61)

64.3

% (3

69/5

74)

68.5

% (2

63/3

84)

66.4

% (1

84/2

77)

1.25

(1.1

2-1.

39)

1.16

(1.0

9-1.

24)

1.25

(1.1

6-1.

36)

1.10

(1.0

-1.2

1)

6.2%

(61/

989)

5.6%

(107

/1,9

21)

5.2%

(50/

964)

7.6%

(49/

646)

12.3

% (1

4/11

4)8.

4% (3

1/36

9)9.

5% (2

5/26

3)12

.0%

(22/

184)

1.99

(1.1

5-3.

44)

1.51

(1.0

3-2.

21)

1.83

(1.1

6-2.

90)

1.58

(0.9

8-2.

54)

Gen

der

Mal

eFe

mal

e58

.4%

(2,3

26/3

,981

))54

.0%

(2,1

94/4

,064

)67

.4%

(506

/751

)65

.7%

(424

/645

)1.

15 (1

.09-

1.22

)1.

22 (1

.14-

1.30

)6.

3% (1

47/2

,326

)5.

5% (1

20/2

,194

)9.

5% (4

8/50

6)10

.4%

(44/

424)

1.50

(1.1

0-2.

05)

1.90

(1.3

7-2.

64)

Ren

al f

un

ctio

neG

FR ≥

80 m

L/m

ineG

FR 5

0-80

mL/

min

eG

FR ≤

50 m

L/m

inU

nkn

own

97.2

% (1

,438

/1,4

79)

97.5

% (1

,829

/1,8

76)

99.3

% (1

,142

/1,1

50)

3.1%

(111

/3,5

40)

99.2

% (1

22/1

23)

99.7

% (3

74/3

75)

99.8

% (4

21/4

22)

2.7%

(13/

476)

1.02

(1.0

0-1.

04)

1.02

(1.0

1-1.

03)

1.01

(1.0

0-1.

01)

0.87

(0.4

9-1.

54)

2.5%

(36/

1,43

8)2.

6% (4

8/1,

829)

15.8

% (1

81/1

,142

)1.

8% (2

/111

)

4.1%

(5/1

22)

4.0%

(15/

374)

16.9

% (7

1/42

1)7.

7% (1

/13)

1.64

(0.6

5-4.

10)

1.53

(0.8

7-2.

70)

1.06

(0.8

3-1.

37)

4.27

(0.4

2-43

.91)

Dia

bete

s N

on-d

iabe

tes

54.0

% (8

55/1

,582

)56

.7%

(3,6

65/6

,463

)69

.1%

(235

/340

)65

.8%

(695

/1,0

56)

1.28

(1.1

8-1.

39)

1.16

(1.1

1-1.

22)

10.1

% (8

6/85

5)4.

9% (1

81/3

,665

)12

.8%

(30/

235)

8.9%

(62/

695)

1.27

(0.8

6-1.

87)

1.81

(1.3

7-2.

38)

Diu

reti

cs

Thia

zid

eLo

opTh

iazi

de

+ lo

opN

o d

iure

tics

55.5

% (1

,723

/3,1

06)

45.0

% (5

97/1

328)

62.9

% (1

,036

/1,6

48)

69.2

% (9

0/13

0)56

.6%

(2,7

97/4

,939

)

66.5

% (7

30/1

,097

)63

.4%

(92/

145)

67.3

% (5

88/8

74)

64.1

% (5

0/78

)66

.9%

(200

/299

)

1.20

(1.1

4-1.

26)

1.41

(1.2

3-1.

62)

1.07

(1.0

1-1.

14)

0.93

(0.7

6-1.

13)

1.18

(1.0

9-1.

28)

7.7%

(133

/1,7

23)

1.5%

(9/5

97)

10.5

% (1

09/1

036)

16.7

% (1

5/90

)4.

8% (1

34/2

,797

)

10.8

% (7

9/73

0)4.

3% (4

/92)

11.2

% (6

6/58

8)18

.0%

(9/5

0)6.

5% (1

3/20

0)

1.40

(1.0

8-1.

83)

2.88

(0.9

1-9.

17)

1.07

(0.8

0-1.

42)

1.08

(0.5

1-2.

29)

1.36

(0.7

8-2.

35)

Inte

ract

ion

*T

ype1

Typ

e 2

Typ

e 3

56.2

% (4

,520

/8,0

45)

56.2

% (4

,520

/8,0

45)

56.2

% (4

,520

/8,0

45)

65.6

% (4

64/7

07)

65.9

% (3

30/5

01)

72.3

% (1

36/1

88)

1.17

(1.1

0-1.

24)

1.17

(1.1

0-1.

25)

1.29

(1.1

8-1.

41)

5.9%

(267

/4,5

20)

5.9%

(267

/4,5

20)

5.9%

(267

/4,5

20)

9.1%

(42/

464)

7.9%

(26/

330)

17.6

% (2

4/13

6)

1.53

(1.1

2-2.

10)

1.33

(0.9

1-1.

96)

2.99

(2.0

4-4.

37)

81

3.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Star

t d

rug

or i

nte

ract

ion

At

hom

eD

uri

ng

hos

pit

alis

atio

n52

.1%

(2,7

06/5

,192

)63

.6%

(1,8

14/2

,853

)63

.0%

(436

/692

)70

.2%

(494

/704

)1.

21 (1

.14-

1.29

)1.

10 (1

.04-

1.17

)5.

9% (1

59/2

,706

)6.

0% (1

08/1

,814

)6.

7% (2

9/43

6)12

.8%

(63/

494)

1.13

(0.7

7-1.

66)

2.14

(1.6

0-2.

88)

Dep

artm

ent

Inte

rnal

med

icin

e sp

ecia

liti

esN

on-in

tern

al m

edic

al

spec

iali

ties

66.4

% (2

,453

/3,6

94)

47.5

% (2

,067

/4,3

51)

68.6

% (6

70/9

77)

62.1

% (2

60/4

19)

1.03

(0.9

8-1.

08)

1.31

(1.2

0-1.

42)

8.2%

(202

/2,4

53)

3.1%

(65/

2,06

3)10

.4%

(70/

670)

8.5%

(22/

260)

1.27

(0.9

8-1.

64)

2.69

(1.6

9-4.

29)

* T

ype

1, R

AS-

inh

ibit

or +

pot

assi

um

sp

arin

g d

iure

tic,

typ

e 2,

RA

S-in

hib

itor

+ p

otas

siu

m s

up

ple

men

t, t

ype

3, p

otas

siu

m s

par

ing

diu

reti

c +

pot

assi

um

su

pp

lem

ent


82

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

The relative risk estimate was also found to be higher when the PIDs were first

prescribed during hospitalisation (RR 2.14, 95%CI 1.60-2.88) than when they were

already started at home (RR 1.13, 95% CI 0.77-1.66). Serum potassium levels were

measured more frequently when PIDs were started during hospitalisation in both

the interaction (63.0 vs 70.2%, p<0.01) and the monotherapy group (52.1 vs 63.6%,

p<0.01).

When stratifying for type of department, the relative risk estimate for developing

hyperkalaemia was found to be higher for patients hospitalised in non-internal

medicine departments (RR 2.7, 95% CI 1.7-4.3) than in those hospitalised in internal

medicine departments (RR 1.3, 95% CI 1.0-1.6). Serum potassium measurements

were made more frequently for patients of both groups hospitalised in internal

medicine departments, than in non-internal medicine departments (monotherapy:

66.4 vs. 47.5%, p<0.01; interaction groups 68.6 vs. 62.1%, p=0.018).

Discussion

In the patients of our hospital who participated in this study, serum potassium levels

were measured in only 56-67% of patients using one or more PIDs. Hyperkalaemia,

defined as a serum potassium concentration ≥ 5.5 mEq/L, occurred in 6-10% of

patients. The absolute risk for developing hyperkalaemia was highest for patients

with a eGFR≤ 50 and for patients with diabetes mellitus. However, the relative risk

estimates for comparing interaction to monotherapy were not increased for age,

gender, kidney function and diabetes. When using two or more PIDs concomitantly,

the risk of hyperkalaemia was elevated in patients using a combination of a

potassium-sparing diuretic with a potassium supplement, in patients starting their

PID therapy during their hospitalisation period and in patients hospitalised in non-

internal medicine departments.

Serum potassium measurements were made more frequently for patients in the

interaction group than for those in the monotherapy group. This may suggest

that the interaction alerts, namely, direct pop-ups advising physicians to monitor

potassium levels, were indeed (partly) effective. However, the higher frequency

may also be due to the longer hospital stay of patients in the interaction group

and differences in patients characteristics between the interaction group and

monotherapy group. To check our assumption that the higher frequency of serum

potassium measurement was indeed caused by a drug interaction signal, we

compared the interaction and the monotherapy groups for the measurement of

two ‘neutral markers’, that is leukocyte count and haemoglobin level. We found

that both markers were measured more frequently in the interaction group than

in the monotherapy group (66 versus 58%, p < 0.01 for haemoglobin level, and 63

83

3.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

versus 51%, p < 0.01 for leukocyte count, respectively, results not shown). This result

indicates that the higher frequency of potassium measurements in the interaction

group may not be (entirely) due to warnings about the drug-drug interaction, but

may also be explained by differences in patient characteristics between patients

using only one PID and patients using ≥2 PIDs. Apparently, more laboratory

measurements were required in the interaction group, which may suggest that

these patients were more severely ill or that they had more complex diseases or

more co-morbidity.

As shown in Table 2, serum creatinine and serum potassium were always almost

measured simultaneously. When renal function was known, serum potassium was

measured in 97-99% of patients, while if it was unknown serum potassium was

measured in only about 3% of patients. It is likely that the same holds true for the

measurement of, for example, sodium. As such, serum potassium measurements

may not always be carried out with the intention to monitor potassium, but simply

as an adjunct.

As expected, the frequency of hyperkalaemia was much higher in patients with

an estimated eGFR ≤ 50 mL/min than in patients with an estimated eGFR > 50 mL/

min. Patients with chronic renal failure are known to bet at risk of developing

hyperkalaemia19. It is also known that the use of a PID may contribute to the risk

in patients with renal failure20. In our patient cohort, however, this risk did not

further increase for patients using ≥ 2 PIDs (15.8 and 16.9% for the monotherapy

and interaction group, respectively). It is possible that monotherapy already

requires the maximum capacity of renal autoregulation. If so, this may not leave

much space for a further increase in risk when ≥ 2 PIDs are used concomitantly.

A comparable result was seen in diabetic patients: diabetes is a known risk factor

for hyperkalaemia10,19,21. In our study however, the introduction of a second PID did

not increase the risk, suggesting that risk factors for developing hyperkalaemia due

to a PID are not the same as those for developing hyperkalaemia after a DDI.

The frequency of hyperkalaemia in patients using a thiazide diuretic seemed to be

lower than in patients without diuretics, which is suggestive for a protective effect

in both monotherapy and interaction groups. This protective effect was not seen

for the use of loop diuretics or for the combination of a thiazide and a loop diuretic

in the co-medication group. The use of diuretics is a known protective factor for

the development of hyperkalaemia18. Weinberg reported a 59% reduction in the

probability of hyperkalaemia with the use of diuretics; however, thiazide diuretics,

not loop diuretics, were found to contribute the most to this effect18.

A comparison of the different combinations of PIDs, revealed that hospitalised

patients using potassium-sparing diuretics combined with potassium supplements

had the highest risk of hyperkalaemia. Of course, the risk of hyperkalaemia


84

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

is obvious for this combination and some authors have even argued that an

alert is superfluous in this situation22. Indermitte et al. found that the use of

potassium supplements contributed most strongly to the velocity of developing

hyperkalaemia23, while others found that a combination of PID with RAS- inhibitors

was the strongest predictor to develop hyperkalaemia in hospitalised patients10.

Patients who were started on interacting PIDs during hospitalisation were at

higher risk developing hyperkalaemia than patients who were started taking

their interacting PID therapy at home. One possible explanation may be when

the interacting medication was started at home, prior to hospitalisation, serum

potassium levels may have been monitored before the patient was hospitalised.

Adjustments may have been made, such as by decreasing the dose of the PIDs or

adding medicine to the therapeutic regimen for decreasing the potassium serum

level24. It may also be possible that the PIDs caused only a temporary rise of serum

potassium levels. Moreover, one might expect that physicians are more inclined

to monitor the effects of newly added medication than of continued home-

medication25. Indeed, the serum potassium levels were measured somewhat more

often in the group of patients who started the PIDs in the hospital than in the

group of patients who continued their PID home-medication. This was seen in both

the monotherapy and interaction group (63.6 vs. 52.1%, p< 0.01 and 70.2 vs. 63.0%,

p< 0.01 respectively).

Finally, the relative risk estimate for developing hyperkalaemia was higher for

patients hospitalised in non-internal medicine departments (RR 2.7, 95% CI 1.7-

4.3) than in patients hospitalised in internal medicine departments (RR 1.3, 95%

CI 1.0-1.6). The frequency of serum potassium measurements was also increased for

patients with an interaction (47.5 vs. 62.1% RR 1.3, 95% CI 1.2-1.4), but the absolute

frequency of serum potassium measurements was remarkably low for patients

with monotherapy in the non-internal medicine departments. This may once again

suggest that warning signals were followed. The frequency of the measurement of

the two neutral markers, however, were also different (54.2 vs. 40.2%, p< 0.01 for

leucocytes and 62.3 vs. 50.3%, p< 0.01 for haemoglobin), which means that differences

in patient characteristics may have played a role as well. Moreover, the frequency

of serum potassium measurements in the monotherapy group of the non-internal

departments is relatively low, which may reflect the fact that physicians working

in non-internal medicine departments are less preoccupied with serum electrolyte

levels than their counterparts working in internal medicine departments. As

such, they may be less aware of the risk for developing of hyperkalaemia when

prescribing a PID. Therefore, special attention is needed for patients using PIDs at

non-internal medicine departments.

85

3.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

There are a number of limitations to our study. First, one potential limitation

may be that nonsteroidal anti-inflammatory drugs, β-adrenergic blocking drugs,

heparin and trimethoprim were not included in our definition of a PID. The

exclusion of these drugs was deliberate as they increase serum potassium levels to

a far lesser extent than RAS-inhibitors, potassium-sparing diuretics and potassium

supplements. As such, the former are not included in the G-standard as DDIs and

the CPOE does not generate a pop-up to warn for hyperkalaemia.

Second, the local hospital setting may limit the generalisability of the results.

Finally, as this is an observational study, patients were not randomised to either

the monotherapy or the interaction group and serum potassium levels were not

measured per protocol. The differences in baseline characteristics make it very

clear that the patients in the monotherapy and the interaction group are not the

same. In addition, serum potassium levels were not measured systematically in

all patients, but were known for only 56-66% of patients, meaning that there is

a testing bias26. As such, this study does not aim to answer the etiologic question

about the contribution of DDIs to the risk of developing hyperkalaemia. The

objective was to study the consequences of a frequently encountered DDI in a daily-

hospital practice and to gain insight in the determinants thereof. Consequently,

the adjusted relative risk estimates have consciously been omitted from Table 2, and

only uni-variate analysis were performed. The message is not so much that these

DDIs cause hyperkalaemia but that, in daily practice, patients in the interaction

group deserve extra attention and serum potassium measurements should be made

more frequently.

At our hospital, when two PIDs are prescribed concomitantly, the physicians

directly receive a pop-up that there is a risk of hyperkalaemia and that the serum

potassium levels should be monitored closely. However, we found that this advice

was not followed in 33% of patients. One of the reasons for this may be that the

warnings are not always appropriate. The DDI signal always appears when a DDI

occurs, even if the serum potassium level is actually low which may cause alert

fatigue27,28.

Another reason may be that the alert to monitor potassium serum levels only

occurs when the medication is started. There are no subsequent reminders, while

in fact it will take at least a few days before a rise in serum potassium levels can

actually be determined.

More advanced methods of clinical decision support may be suitable to improve

the management of this DDI and to reduce the frequency of hyperkalaemia. For

example, it may be worthwhile to generate a reminder every 3-4 days to monitor

serum potassium for all hospitalised patients using two or more PIDs. Another

possibility would be to link laboratory and medication data and to generate a


86

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

warning when serum potassium levels increase above a certain limit for patients

using one or more PIDs.

Acknowledgement

The authors are grateful to Hanneke den Breeijen for the data analysis and to their colleagues



87

3.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

References

1 Leendertse AJ, Egberts AC, Stoker LJ, van den Bemt PM. Frequency of and risk factors for preventable

medication-related hospital admissions in the Netherlands. Arch Intern Med 2008;168:1890-1896

2 Juurlink DN, Mamdani M, Kopp A, Laupacis A, Redelmeier DA. Drug-drug interactions among elderly

patients hospitalized for drug toxicity. JAMA 2003;289:1652-1658

3 Buurma H, De Smet PA, Egberts AC. Clinical risk management in Dutch community pharmacies: the

case of drug-drug interactions. Drug Saf 2006;29:723-732

4 Zwart-van Rijkom JE, Uijtendaal EV, ten Berg MJ, van Solinge WW, Egberts AC. Frequency and nature


5 Perazella MA, Mahnensmith RL. Hyperkalaemia in the elderly: drugs exacerbate impaired potassium

homeostasis. J Gen Intern Med 1997;12:646-656

6 Pitt B, RALES study group. Effectiveness of spironolactone added to an angiotensin-converting enzyme

inhibitor and a loop diuretic for severe chronic congestive heart failure (the Randomized Aldactone

Evaluation Study [RALES]). Am J Cardiol 1996;78:902-907

7 Juurlink DN, Mamdani MM, Lee DS, Kopp A, Austin PC, Laupacis A, Redelmeier DA. Rates of

hyperkalaemia after publication of the Randomized Aldactone Evaluation Study. N Engl J Med

2004;351:543-551

8 Cruz CS, Cruz AA, Marcilio de Souza CA. Hyperkalaemia in congestive heart failure patients using

ACE inhibitors and spironolactone. Nephrol Dial Transplant 2003;18:1814-1819

9 Desai A. Hyperkalaemia associated with inhibitors of the renin-angiotensin-aldosterone system:

balancing risk and benefit. Circulation 2008;118:1609-1611

10 Henz S, Maeder MT, Huber S, Schmid M, Loher M, Fehr T. Influence of drugs and comorbidity on

serum potassium in 15 000 consecutive hospital admissions. Nephrol Dial Transplant 2008;23:3939-

3945

11 Johnson ES, Weinstein JR, Thorp ML, Platt RW, Petrik AF, Yang X, Anderson S, Smith DH. Predicting the

risk of hyperkalaemia in patients with chronic kidney disease starting lisinopril. Pharmacoepidemiol

Drug Saf 2010;19:266-272

12 van Roon EN, Flikweert S, le Comte M. Langendijk PN, Kwee-Zuiderwijk WJ, Smits P, Brouwers

JR. Clinical relevance of drug-drug interactions : a structured assessment procedure. Drug Saf

2005;28:1131-1139

13 Z-index (2008). Z-index .http://www.z-index.nl/ (Accessed at Jan, 2009)

14 WHO Collaborating Center for Drug Statistics Methodology (2009) Anatomical Therapeutic Chemical

(ATC) Classification Index. http://www.whocc.no/ (Accessed at Jan, 2009)

15 ten Berg MJ, Huisman A, van den Bemt PM, Schobben AF, Egberts AC, van Solinge WW. Linking

laboratory and medication data: new opportunities for pharmacoepidemiological research. Clin

Chem Lab Med 2007;45:13-19

16 Levey AS, Bosch JP, Lewis JB, Greene T, Rogers N, Roth D. A more accurate method to estimate

glomerular filtration rate from serum creatinine: a new prediction equation. Modification of Diet in

Renal Disease Study Group. Ann Intern Med 1999;130:461-470

17 Levey AS, Coresh J, Greene T, Stevens LA, Zhang YL, Hendriksen S, Kusek JW, van Lente LF. Using

standardized serum creatinine values in the modification of diet in renal disease study equation for

estimating glomerular filtration rate. Ann Intern Med 2006;145:247-254

18 Weinberg JM, Appel LJ, Bakris G, Gassman JJ, Greene T, Kendrick CA, Wang X, Lash J, Lewis JA, Pogue

V, Thornley-Brown D, Philips RA. Risk of hyperkalaemia in nondiabetic patients with chronic kidney

disease receiving antihypertensive therapy. Arch Intern Med 2009;169:1587-1594

19 Einhorn LM, Zhan M, Hsu VD, Walker LD, Moen MF, Seliger SL, Weir MR, Fink JC. The frequency of

hyperkalaemia and its significance in chronic kidney disease. Arch Intern Med 2009;169:1156-1162

20 Bakris GL, Siomos M, Richardson D, Janssen I, Bolton WK, Hebert L, Agarwal R, Catanzaro D. ACE

inhibition or angiotensin receptor blockade: impact on potassium in renal failure. VAL-K Study

Group. Kidney Int 2000;58:2084-2092

21 Stevens MS, Dunlay RW. Hyperkalaemia in hospitalized patients. Int Urol Nephrol 2000;32:177-180


88

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

22 Ponce SP, Jennings AE, Madias NE, Harrington JT. Drug-induced hyperkalaemia. Medicine 1985;64:357-

70

23 Indermitte J, Burkolter S, Drewe J, Krähenbühl S, Hersberger KE. Risk factors associated with a high

velocity of the development of hyperkalaemia in hospitalised patients. Drug Safety 2007; 30:71-80

24 Vereijken TL, Bellersen L, Groenewoud JM, Knubben L, Baltussen L, Kramers C. Risk calculation for

hyperkalaemia in heart failure patients. Neth J Med 2007;65:208-211

25 van der Sijs H, Mulder A, van GT, Aarts J, Berg M, Vulto A. Drug safety alert generation and overriding

in a large Dutch university medical centre. Pharmacoepidemiol Drug Saf 2009;18:941-947

26 Velthove KJ, Leufkens HG, Souverein PC, Schweizer RC, van Solinge WW. Testing bias in clinical

databases: methodological considerations. Emerg Themes Epidemiol 2010;7:2-9

27 van der Sijs H, Aarts J, Vulto A, Berg M Overriding of drug safety alerts in computerized physician

order entry. J Am Med Inform Assoc 2006;13:138-147

28 Weingart SN, Toth M, Sands DZ, Aronson MD, Davis RB, Phillips RS. Physicians’ decisions to override

computerized drug alerts in primary care. Arch Intern Med 2003;163:2625-2631

89

3.2

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

3.3Serum potassium influencing

interacting drugs: Risk modifying

strategies also needed at

discontinuation




Toine C.G. Egberts

Chapter 3.3 | Serum potassium influencing interacting drugs: Risk modifying strategies also needed at discontinuation

92

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Abstract

Background

Although the discontinuation of a medication may have important clinical

consequences, there is generally much less attention given to medication

surveillance when a prescription is stopped than when it is started.

Objective

To investigate the consequences on serum potassium levels of discontinuing a drug

that increases the serum potassium level (PID↑) and drug that decreases the serum

potassium level (PLD↓) in patients taking both

Methods

Patients who were hospitalized in the University Medical Centre Utrecht (UMCU) in

2004-2009 and were using both a PID↑ and a PLD↓ were included when one of these

drugs was discontinued during hospitalization. Serum potassium levels measured

before (potassium1) and after (potassium

2) discontinuation were compared in

patients who stopped the PLD↓ and in patients who stopped the PID↑.

Results

In the group of patients who stopped the PLD↓, (i.e., continued the PID↑),

mean serum potassium levels increased 0.19 mEq/L (range -0.9–1.8 mEq/L). After

discontinuation of the PLD↓, serum potassium levels increased in 91 (59%) patients.

Five patients (3.2%) developed hyperkalemia (potassium2>5.5 mEq/L).

In the group of patients who stopped the PID↑, (i.e. continued the PLD↓), mean

serum potassium levels decreased 0.40 mEq/L (range -2.6-0.7 mEq/L). Serum

potassium levels decreased in 61 (70%) patients after discontinuation of the PID↑.

Fifteen patients (17%) developed hypokalemia (potassium2 <3.5 mEq/L).

The results were not influenced by length of stay, age, sex, renal function and type

of medication discontinued.

Conclusions

The effects of serum potassium-influencing drugs need to be monitored not only

after starting the medication but also after stopping the medication. The same may

hold true for the effects of other drugs. Clinical risk management should therefore

focus on the risks not only when a new medication is prescribed, but also when

medication is stopped.

93

3.3

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Introduction

Medications can be an important source of unintended patient harm, which may be

caused by non-preventable adverse effects or medication errors.1 When considering

initiation of a new medication, it is important to determine the appropriateness

and fit of the patient’s drug therapy, including checking for potential drug-

related problems such as correct dosing, monitoring necessities and drug-drug

interactions. To support this evaluation, some hospitals have incorporated a

basic clinical decision support system in the computerized physician order entry

(CPOE) system. The discontinuation of a medication, however, may have important

consequences as well. A tapering scheme may be necessary, the loss of enzyme

induction or inhibition may require dosage adjustment of other medications, or

electrolyte disturbances may occur. Ignoring these consequences may lead to an

adverse drug withdrawal event.2 To our knowledge, there is hardly any literature

about medication surveillance when a prescription is stopped, even when the

prescription is involved in a drug-drug interaction.

Literature has now shown that the risk on hyperkalemia is one of the most

frequently occurring problems with drug therapy.3,4 After publication of the RALES

(Randomized Aldactone Evaluation Study) results in 1999, it has been increasingly

recognized that the use of serum potassium increasing drugs (PIDs↑) such as renin-

angiotensin system (RAS) inhibitors, potassium-sparing diuretics, or potassium

supplements may heighten a patient’s risk for developing hyperkalemia.5-7

Recommendations from the RALES report not only resulted in an increased

prescription rate of spironolactone, from 34 per 1000 patients to 149 per 1000

patients, but also led to an increased rate of hospitalizations for hyperkalemia from

2.4 to 11 per 1000 patients, and an associated rise in mortality from 0.3 to 2.0 per

1000 patients.6

On the other hand, the use of serum potassium lowering drugs (PLDs↓), such

as loop diuretics or thiazide diuretics, is associated with an increased risk for

developing hypokalemia.8 As with hyperkalemia, hypokalemia is a risk factor for

developing cardiac arrhythmias, a potentially lethal adverse effect.9 Monitoring of

serum potassium levels after the start of a PID↑ or PLD↓ is recommended in drug

information leaflets and guidelines; however, recommendations to monitor serum

potassium levels after discontinuation of these agents are lacking.10

Many patients use both a PID↑ and PLD↓, (eg, an RAS-inhibitor and a diuretic),

which is often seen in patients with heart failure or hypertension and diabetes or

renal failure. Although not recognized by all drug-drug interactions databases, the

combination of a PID↑ with a PLD↓ is an example of a pharmacologic interaction. As

such, extra attention is warranted when one of the drugs is discontinued. Therefore,


94

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

the purpose of this study was to investigate the consequences of discontinuation of

one of these agents on serum potassium levels in patients using both a PID↑ and

a PLD↓.

Methods

Setting This retrospective follow-up study was performed in the University Medical Centre

Utrecht (UMCU), a 1042-bed academic medical centre located in the center of the

Netherlands. In 2004-2009, approximately 30,000 clinical hospitalizations took

place annually. All medications for hospitalized patients were prescribed using

a CPOE system with a basic clinical decision support system incorporated. For

research purposes all prescriptions are routinely exported in the Utrecht Patient

Oriented Database (UPOD). UPOD is an infrastructure of relational databases



UMCU since 2004, and has been described in detail elsewhere.11 The utilization of

UPOD is in accordance with guidance of the Institutional Review Board.

Study design and populationPatients were eligible for inclusion if they had been hospitalized in the UMCU in

2004-2009, used both a PID↑ and a PLD↓, and had one of these drugs discontinued

during the hospitalization period. Serum potassium levels before (potassium1)

and after (potassium2) discontinuation were compared in patients who stopped

the PLD↓, (i.e. continued the PID↑), and in patients who stopped the PID↑, (i.e.

continued the PLD↓) (Figure 1). Patients were included in our study if both the PID↑

and the PLD↓ were prescribed within 12 hours after hospital admission, assuming

that this represents continuation of medication in use before admission. If a patient

was admitted to the hospital more than once during the study period, only the first

admission was included in the analysis.

PIDs↑ were defined as RAS-inhibitors, potassium-sparing diuretics or potassium

supplements. PLDs↓ were defined as loop diuretics or thiazide diuretics. Potassium1

was defined as the serum potassium level measured closest to the moment

of discontinuation, within a window of 48 hours before and 12 hours after

discontinuation. The 12 hour period after discontinuation was chosen because the

potassium influencing drug is still present in the body for several hours, depending

on the elimination half-life after the last dosage. Potassium2 was defined as the

first serum potassium level measured within 48 -144 hours after the moment of

discontinuation (Figure 1).

95

3.3

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Patients were excluded if they were younger than 18 years, on hemodialysis,

using a potassium or phosphate binding drug or if the potassium-influencing

medication was changed (another PID↑/PLD↓ stopped or added) between the time

of discontinuation and the measurement of potassium2. Patients were also excluded

if potassium1 or

potassium

2 were not available.

Figure 1: Schematic representation of the inclusion criteria

Figure 1: Schematic representation of the inclusion criteria

OutcomeTwo primary outcome variables were studied: the change in serum potassium

levels after the moment of discontinuation (change in serum potassium levels =

potassium2

– potassium1) and the serum potassium level after discontinuation

(potassium2). Also, the percentage of patients developing either hypokalemia or

hyperkalemia was calculated. Hypokalemia was defined as potassium2 < 3.5 mEq/L

and hyperkalemia as potassium2 > 5.5 mEq/L.

Data analysisBoth outcome measures were computed separately for patients who discontinued

the use of the PLD↓ (i.e. continued the PID↑) and for patients who discontinued

the use of the PID↑ (i.e. continued the PLD↓). Univariate linear regression analysis

(continuous variables) and 1-way analysis of variance (categorical variables) were

used to study the influence of length of stay, age, sex, renal function and type of

medication discontinued. Estimated glomerular filtration rate was estimated using

the Modification of Diet in Renal Disease equation12 and the serum creatinine

level measured closest to time of discontinuation during the hospitalization. SPSS

Package for Social Sciences software version 17.0 for windows (SPSS Inc., Chicago,

IL) was used to analyze the collected data.


96

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Results

The selection profile of the study is depicted in Figure 2. Of the 420 patients who

met the inclusion and exclusion criteria, 156 (37%) were subsequently excluded

as no serum potassium was measured 48-144 hours after discontinuation of 1 of

both drugs. In the end, a total of 241 patients were included in the study, 154 of

whom discontinued the PLD↓ and 87 of whom discontinued the PID↑. Typical for

both groups was the relatively long length of stay (Table 1). Kidney function was

comparable in both groups.

Table 1: Baseline characteristics

Characteristic

PLD↓ Discontinued,

PID↑ Continued (n=154)

PID↑ Discontinued,

PLD↓ Continued (n=87)

Length of stay, days, median

(interquartile range)

13.2 (8.6-21.0) 16.7 (10.8-28.9)

Age, y, mean (95% CI) 70.4 (68.1-72.8) 67.8 (64.7-71.0)Male, n (%) 73 (47.4%) 48 (55.2%)eGFR, ml/min, a n (%)

≥80 77 (50.0%) 47 (51.5%)50-80 50 (32.5%) 24 (27.6%) ≤50 27 (17.5%) 16 (18.4%)

Type of drug stopped, n (%)potassium supplementb 31 (36%)potassium sparing diureticc 14 (16%)RAS-inhibitord 42 (48%)loop diuretice 108 (70%)thiazide diureticf 46 (30%)

Type of drug continued, n (%)potassium supplementb 16 (10%)potassium sparing diureticc 25 (16%)RAS-inhibitord 113 (73%)loop diuretice 75 (86%)thiazide diureticf 12 (14%)

eGFR: estimated glomerular filtration rate;

PID↑= potassium-increasing drug; PLD↓= potassium-lowering drug aGlomerular filtration rate estimated by the Modification of Diet in Renal Disease equation.b Oral en intravenous potassium chloride, potassium phosphate.c Spironolactone, triamterened Candesartan, captopril, enalapril, fosinopril, irbesartan, lisinopril, losartan, perindopril, quinapril,

ramipril, valsartan,e Bumetanide; furosemidef hydrochlorothiazide, chlorthalidone, indapamide

97

3.3

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Figure 2: In- and exclusion profile

292,594 admissions 126,444 patients



exclusion criteria (total): 640 admissions 586 patients

no follow up for more than 48hours after discontinuation 289 admissions 263 patientsdialysis or potassium/phospate resorber 80 admissions 74 patientschanges in potassium influencing drug use after discontinuation 271 admissions 249 patients

432 admissions 420 patients

potassium1or potassium2 not available a 187 admissions 179 patientspotassium1 not available 83 admissions 79 patientspotassium2 not available 164 admissions 156 patients

245 admissions 241 patientsa numbers do not add up as patients may have both serum potassium levels not available

both potassium1 and potassium2 available

admissions UMCU >18 year

using 1 PID and 1PLD upon admission

PID or PLD stops

evaluable admissions/patients

Figure 2: In- and exclusion profile

Patients who discontinued the PLD↓, and continued the PID↑Overall, in the group of patients who discontinued the PLD↓, mean change in

serum potassium levels was 0.19 mEq/L (range -0.9–1.8 mEq/L). After the time of

discontinuation, serum potassium levels increased in 91 (59%) of the patients

(Figure 3). Surprisingly, 51 (33%) patients who discontinued the PLD↓ showed a

decrease in serum potassium levels.

Five patients (3.2%) developed hyperkalemia after discontinuation. Five patients

(3.2%) developed hypokalemia after discontinuation, and another 11 (7.1%) patients

already had hypokalemia before discontinuation (Figure 4).

Patients who discontinued the PID↑, and continued the PLD↓Overall, in the group of patients who discontinued the PID↑, mean change in

serum potassium levels was -0.40 mEq/L (range -2.6-0.7 mEq/L). After the time of

discontinuation, serum potassium levels decreased in 61 (70%) patients (Figure 3).

Surprisingly, 20 (23%) patients who discontinued the PID↑ showed an increase in

serum potassium levels.

Fifteen (17%) patients developed hypokalemia after discontinuation. Two patients

(2.3%) developed hyperkalemia after discontinuation (Figure 4).


98

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

In both groups of patients, results were not influenced by length of stay, age, sex,

renal function or type of medication discontinued.

Figure 3: Changes in serum potassium levels after discontinuation (potassium2-potassium1),

Figure 3: Changes in serum potassium levels after discontinuation (potassium

2-potassium

1),

Figure 4: Serum potassium levels after discontinuation (potassium2)

Figure 4: Serum potassium levels after discontinuation (potassium2)

99

3.3

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Discussion

Serum potassium levels decreased in 70% of the patients who discontinued the PID↑

but continued the PLD↓, and increased in 59% of the patients who discontinued

the PLD↓ but continued the PID↑. Of the patients who discontinued the PID↑, 17%

developed hypokalemia and 3.2% of those who discontinued the PLD↓ developed

hyperkalemia. The strongest effect was found in the group of patients who

discontinued the PID↑ and continued to take a PLD↓.

Notably, in 23-33% of patients, serum potassium levels shifted in the opposite

direction from the one expected. This may have been caused by minor changes

in serum potassium levels that fell within the range of normal physiological

fluctuations and the variation of the assay method (2%). The magnitude of the

shift in the expected direction was larger than the shift in the opposite direction.

In the group of patients who discontinued the PID↑, the mean change in serum

potassium levels in the patients whose serum potassium levels decreased was

-0.66 mEq/L, while the mean change in serum potassium levels in patients whose

serum potassium levels increased was only 0.30 mEq/L. In the group of patients

who discontinued the PLD↓, the mean change in serum potassium levels in

patients whose levels increased was 0.51 mEq/L, while the mean change in serum

potassium levels in patients whose levels decreased was only -0.33 mEq/L. However,

as the variation of the assay was only 2%, this is not a complete explanation for

the data. It is likely that pathophysiologic factors such as morbidity (eg, nausea,

diarrhea, organ failure), nutritional status and concomitant (eg, laxatives) may

have influenced the results. Lack of data on these issues is a limitation of this study.

Another limitation is the fact that strict inclusion and exclusion criteria have been

applied. This resulted in a selected group of patients with an average length of

stay within the hospital of 19.3 days (median 14.9 days). Between 2004 and 2009,

the average length of stay within the UMCU was only 7.8 days, meaning that the

population in our study probably was more severely ill than the average hospital

population. Furthermore, this study included patients using a combination of a

PID↑ and PLD↓. Results may therefore not be extrapolated to patients only one of

these drugs. This may compromise the generalizability of the results.

Finally, a limitation of this study is that the doses of the PIDs↑ and PLDs↓ were not

accounted for, while it is known that these drugs have a dose-dependent effect on

the serum potassium levels. However, it was not our objective to determine to what

extent the discontinuation of drugs that interfere with potassium homeostasis

contributes to changes in serum potassium levels. Rather, it was our intention

to create awareness that attention is needed not only when a new medication

is started, but also when a medication is discontinued. We have demonstrated


100

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

this need with regard to discontinuation of drugs that interfere with potassium

homeostasis, but the need may hold true for many others drugs, such as those

involved in pharmacokinetic interactions.13,14 From the in- and exclusion profile of

our study it can be deducted that serum potassium levels were not monitored after

discontinuation of one of the 2 types of serum potassium-influencing drugs in 37%

of the patients.

In conclusion, this study demonstrated that the effects of serum potassium-

influencing drugs need to be monitored, not only after starting the medication,

but also after stopping it. The same may hold true for the effects of other drugs.

Clinical risk management should therefore be focused not only on risks when a

new medication is prescribed, but also on those that may occur when medication

is stopped.

Acknowledgment




101

3.3

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

References

1. Leendertse AJ, Egberts AC, Stoker LJ, et al. Frequency of and risk factors for preventable medication-

related hospital admissions in the Netherlands. Arch Intern Med 2008;168:1890-6.

2. Bain KT, Holmes HM, Beers MH, et al. Discontinuing medications: a novel approach for revising the

prescribing stage of the medication-use process. J Am Geriatr Soc 2008;56:1946-1952

3. Zwart-van Rijkom JE, Uijtendaal EV, ten Berg MJ, et al. Frequency and nature of drug-drug interactions

in a Dutch university hospital. Br J Clin Pharmacol 2009;68:187-193

4. Buurma H, De Smet PA, Egberts AC. Clinical risk management in Dutch community pharmacies: the

case of drug-drug interactions. Drug Saf 2006;29:723-732

5. Henz S, Maeder MT, Huber S, et al. Influence of drugs and comorbidity on serum potassium in 15 000

consecutive hospital admissions. Nephrol Dial Transplant 2008;23:3939-45.

6. Juurlink DN, Mamdani MM, Lee DS, et al. Rates of hyperkalemia after publication of the Randomized

Aldactone Evaluation Study. N Engl J Med 2004;351:543-551

7. Pitt B, Zannad F, Remme WJ, et al. The effect of spironolactone on morbidity and mortality in patients

with severe heart failure. Randomized Aldactone Evaluation Study Investigators. N Engl J Med

1999;341:709-717

8. Liamis G, Milionis H, Elisaf M. Blood pressure drug therapy and electrolyte disturbances. Int J Clin

Pract 2008;62:1572-1580

9. Cohen HW, Madhavan S, Alderman MH. High and low serum potassium associated with cardiovascular

events in diuretic-treated patients. J Hypertens 2001;19:1315-1323

10. Bootsma JE, Warle-van Herwaarden MF, Verbeek AL, et al. Adherence to biochemical monitoring

recommendations in patients starting with Renin Angiotensin system inhibitors: a retrospective

cohort study in the Netherlands. Drug Saf 2011;34:605-614

11. ten Berg MJ, Huisman A, van den Bemt PM, et al. Linking laboratory and medication data: new

opportunities for pharmacoepidemiological research. Clin Chem Lab Med 2007;45:13-19

12. Levey AS, Coresh J, Greene T, et al. Using standardized serum creatinine values in the modification

of diet in renal disease study equation for estimating glomerular filtration rate. Ann Intern Med

2006;145:247-254

13. Shullo MA, Schonder K, Teuteberg JJ. Elevated tacrolimus levels associated with intravenous

azithromycin and ceftriaxone: a case report. Transplant Proc 2010;42:1870-1872

14. Castberg I, Helle J, Aamo TO. Prolonged pharmacokinetic drug interaction between terbinafine and

amitriptyline. Ther Drug Monit 2005;27:680-682

3.4Influence of a strict glucose protocol

on serum potassium levels in intensive

care patients



Dylan W. de Lange

Arief Lalmohamed


Toine C.G. Egberts

Chapter 3.4 | Influence of a strict glucose protocol on serum potassium levels in intensive care patients

104

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Abstract

Objective

Tight glucose control therapy (TGC) has been implemented to control hyperglycemia

in intensive care (ICU) patients. TGC, however, may also influence serum potassium

levels. We therefore investigated the influence of TGC on both serum glucose and

serum potassium levels and associated mortality.

Design

Retrospective analysis

Setting

Mixed ICU of a tertiary hospital

Patients

Patients admitted to the ICU for 24 hours or more and with at least three serum

glucose and serum potassium levels between 1999-2001 (conventional period),

2002-2006 (implementation period) or 2007-2009 (TGC period).

Measurements and main results

Means and standard deviations (SDs) of serum glucose and serum potassium

levels, and rate of hypoglycemia (≤2.2mmol/L) and hypokalemia (≤3mmol/L) levels,

were compared between the TGC and conventional period using time series

analysis. Although mean serum glucose levels dropped 2.1 mmol/L (95%CI =-1.8

to -2.3mmol/L, p<0.002), mean serum potassium levels did not change (absolute

increase 0.02mmol/L; 95%CI = -0.06 to 0.09mmol/L, p=0.64). The rate of hypoglycemia

increased with 5.9% (95%CI=-3.0 to -8.9, p<0.002), but the rate of hypokalemia stayed

equal (absolute reduction 4.8%; 95%CI = -11.1% to 1.5%, p=0.13). The SD of serum

glucose levels within a patient did not change, while the SD of serum potassium

levels even decreased 0.04 mmol/L (95%CI = -0.01 to -0.07, p=0.01).

Mean serum glucose levels, mean serum potassium levels and SDs of both serum

glucose and serum potassium levels were all independently associated with ICU

mortality. Highest mortality rates were seen at both the lowest and highest mean

values (U/J-shaped association) and mortality rates increased with increasing

variability (SDs) for both serum glucose and serum potassium levels.

Conclusion

Our study shows that a TGC was not associated with an increased risk of serum

potassium related events. Low and high mean values and high variability of both

serum glucose and serum potassium levels should be prevented.

105

3.4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Introduction

Hyperglycemia in response to critical illness has been associated with increased

morbidity and mortality1. The mechanism suggested for this increased risk is that

elevated glucose concentrations increase the level of several toxic intracellular

derivatives that are generated as by-products of the glycolytic pathway2,3. Especially

during severe illness, the expression of insulin-independent glucose transporters

on the membranes of several cell types is up-regulated, which may allow high

circulating glucose levels to overload and damage these cells4-7. Based upon this line

of reasoning, vd Berge et al investigated in surgical intensive care unit (ICU) patients

whether a tight glycemic control (TGC) protocol (target serum glucose level 4.4-6.1

mmol/L (80-110 mg/dL)) would reduce mortality8. In this landmark clinical trial, an

absolute mortality reduction of 3.4% was found. This led to the implementation of

TGC protocols on many ICUs worldwide.

Subsequent studies performed in medical ICU patients, however, failed to

reproduce the reduction in mortality9-11. The NICE-sugar investigators even reported

an increased mortality when TGC was compared to conventional treatment12. Since

then, TGC has become a major area of debate among medical specialties involved in

the care of acutely ill patients. Several hypotheses have been postulated to explain

the contradictory results.

First, the characteristics of study populations differed between the different clinical

trials that were carried out. TGC seems to benefit surgical ICU patients more than

medical ICU patients13. In the setting where hyperglycemia is triggered by surgery,

the delay between onset of hyperglycemia and the start of glycemic control is short.

Medical ICU patients may have suffered from chronic diseases and hyperglycemia

before ICU admission and time from the onset of symptoms to the start of TGC

may be longer. As such the TGC protocol may be more beneficial in surgical

ICU patients. Second, a TGC protocol can be expected to increase the incidence

of hypoglycemia, which in itself gives a higher risk on mortality. An increased

incidence of hypoglycemia (serum glucose levels ≤ 2.2 mmol/L (40 mg/dL)) was

seen in all studies that applied a TGC protocol14. The highest rates of hypoglycemia

were generally found in studies that applied the lowest serum glucose targets,

varying, however, with the complexity of the TGC protocols15. An increased rate of

hypoglycemia may also reflect a large fluctuation of serum glucose levels implying

that these vulnerable patients may also be exposed to high glucose levels, although

mean serum glucose levels are relatively low16. In vitro studies showed that a large

variability in glucose levels may enhance cell apoptosis17,18 and some clinical studies

confirmed an independent association between large variations in serum glucose

levels and mortality 19-22. Third, in addition to serum glucose, serum potassium


106

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

may also be a significant factor influencing patients’ outcomes. It is well known

that insulin induces a shift of potassium from the extracellular to the intracellular

compartment. As a result, the implementation of a TGC protocol may also induce

larger variations in serum potassium levels and more frequently hypokalemia

and associated complications such as arrhythmia. In the NICE-sugar study there

was an increased risk of cardiovascular deaths in the TGC-group, but its cause

remained unclear12. Some studies reported an increased incidence of hypokalemia,

and emphasized the necessity of combined glucose and potassium monitoring to

prevent hypokalemia-induced arrhythmia23,24. Other studies did not mention the

measurement of serum potassium levels as a part of the TGC protocol.

Because of the ongoing discussion, we investigated within our own academic ICU

what the influence was of the implementation of a TGC protocol on the rate of

hypoglycemia and hypokalemia and the variation in serum glucose and potassium

levels. Furthermore, the association between mortality and both serum glucose

and potassium levels and variability was studied during the TGC period.

Methods

Setting and study populationThis retrospective observational study was carried out at the 32-bed Intensive Care

Unit (ICU) of the University Medical Centre Utrecht (UMCU), which is a tertiary care

teaching hospital in the Netherlands. Surgical, internal, neurological and cardio-

thoracic ICU patients are treated on this ICU.

A TGC policy was consecutively implemented on all specialties of the ICU during

the years 2002-2006. These years were defined as the implementation period. The

years before implementation, 1999-2001, were defined as the conventional period

and the years 2007-2009 were defined as the TGC period.

During the conventional period, conventional serum glucose control was applied

to all patients, which implied that insulin therapy was started if the serum glucose

levels were above 12 mmol/L (215 mg/dL) and that serum glucose levels were

maintained between 10-11 mmol/L (180 and 200 mg/dL). During the TGC period,

a TGC protocol was followed, aiming to maintain serum glucose levels between

4.4-6.1 mmol/L (80-110 mg/dL). In the conventional period, the policy was to start

potassium supplementation if serum potassium dropped below 4 mmol/L; this

policy did not change in the TGC period.

Adult patients (≥18 years) who were admitted to the ICU for more than 24 hours in

the years 1999-2009 were eligible for inclusion. To be able to calculate serum glucose

and potassium variability, patients with less than three valid serum glucose or

less than three valid serum potassium measurements during ICU admission were

107

3.4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

excluded. Moreover, if a patient was admitted to the ICU more than once, only the

first admission was included (Figure 1).

For research purposes all laboratory values are routinely exported to the Utrecht

Patient Oriented Database (UPOD). UPOD is an infrastructure of relational databases



UMCU since 2004, and has been described in detail elsewhere25. As only routinely

documented patient data were used in this study, the ethics board waived the need

for informed consent.

Figure 1: Inclusion of study patients

Figure 1: Inclusion of study patients

OutcomesFour outcome parameters were determined for each individual patient:

1) mean serum glucose and potassium levels: means of serum glucose and serum

potassium levels were calculated for each patient during ICU stay

2) hypoglycemia and hypokalemia: proportion of patients with at least one

hypoglycemic episode and proportion of patients with at least one hypokalemic

episode. Hypoglycemia was defined as a serum glucose measurement ≤ 2.2 mmol/L

and hypokalemia was defined as a serum potassium measurement ≤ 3.0 mmol/L

3) variability of serum glucose and serum potassium levels: mean standard

deviation (SD) of individually calculated standard deviations of serum glucose and

serum potassium.

4) in the TGC period: ICU mortality.


108

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Data analysisFirst, interrupted time series analysis was performed to compare the TGC period

with the conventional period. Patient characteristics were compared using t-test

(mean values), the chi-square test (proportions) and Mann-Whitney U test (median

values).

Segmented regression analysis was used to estimate changes in outcomes that

occurred after the intervention controlling for pre-intervention trends26. The

intervention was the implementation of a TGC protocol between January 2002 and

January 2007. Outcome values during the implementation period were modelled

as a separate segment. Each time point in the time series data represents a three

month period resulting in the recommended minimum of 12 time points before

and 12 time points after the intervention26. The following linear regression model

was used:

Yt = β

0 + β

1* time

1 + β

2 * time

2 + β

3 * time

3 (1)

(Yt = mean serum glucose level at time = t; β

0= intercept = mean serum glucose

level at time = 0; β1= baseline trend = change in serum glucose level during the

conventional period; time1 = time after start of the conventional period(Jan 1, 1999);

β2= trend change during the implementation period, adjusted for the baseline

trend; time2 = time after start of the implementation period (Jan 1, 2002); β

3 = trend

change during the TGC Period, adjusted for the baseline trend and trend change

during the implementation period; time3 = time after start of the TGC Period (Jan

1, 2007). The change in mean serum glucose values between the end of the control

period and the start of the TGC period are calculated from:

Change = (β1 +

β2) * duration of implementation (2)

The same regression model was used for the proportion of patients with a

hypoglycemia, the mean SDs for serum glucose levels, mean serum potassium

levels, proportion of patients with hypokalemia and mean SDs for serum potassium

levels.

Second, the relationship between mean serum glucose levels, mean serum

potassium levels and ICU mortality was studied. This relationship was also studied

for the SD of serum glucose, SD of serum potassium levels and mortality. For this

analysis, only patients who either survived or who died on the ICU during the TGC

period were included. Logistic regression analysis was performed to determine the

association between mean glucose levels and risk on ICU mortality. Adjustments

109

3.4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

were made for mean serum potassium levels, age, gender and length of ICU stay.

A similar procedure was performed to determine the association between mean

serum potassium levels (with adjustments for mean serum glucose levels, age,

gender and length of stay), SD of serum glucose levels and SD of serum potassium

levels and ICU mortality. Moreover, to study the relationship between the combined

mean serum glucose and mean serum potassium levels on ICU mortality, patients

were divided in eight equal subgroups for their mean serum glucose and eight

equal subgroups for their mean serum potassium values. This resulted in 64

possible mean glucose-mean potassium subgroup-combinations. ICU mortality was

calculated for each combination.

Time series analysis was performed using SAS 9.2, SAS Institute Inc, Cary, USA using

PROC REG procedure. All other data analyses were performed using SPSS release 20

(SPSS, Inc., Chicago, Illinois).

Results

During the study period, a total of 21,662 admissions were available for selection.

After application of the inclusion criteria, 2538 patients could be included in the

conventional period, 4925 in the implementation period and 2730 in the TGC

Period (Figure 1). Baseline characteristics of the patients are shown per period in

Table 1. Compared to patients included in the conventional period, patients in the

TGC period were older (59.6 vs 57.6 years), mean serum creatinine levels were lower

(112 vs 121 mmol/L), the length ICU and hospital stay was shorter (3.1 vs 4.5 and 16

vs 18 days respectively) and the number of patients admitted due to an emergency

was lower (45 vs 54%). ICU mortality was also lower in the TGC period (14.4 vs 18.6%).

The median number of serum glucose measurements increased approximately 3.4

times (p<0.002) after implementation of the TGC protocol. The median number of

serum potassium measurements per patient, decreased from 10 measurements to

9 (p<0.002).


110

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40 Tab

le 1

: B

asel

ine

char

acte

rist

ics

Con

ven

tion

al

per

iod

1999

-200

0-20

01(n

=253

8)

Imp

lem

enta

tion

Pe

riod

20

02-2

003-

2004

-200

5-20

06

(n=4

925)

Tigh

t G

luco

se

Con

trol

Per

iod

20

07-2

008-

2009

(n=2

730)

p

Mal

e ge

nd

er (%

)60

.363

.160

.70.

782a

Age

(yrs

) mea

n (9

5% C

I)57

.6 (5

7.0-

58.3

)59

.9 (5

9.4-

60.3

)59

.6 (5

9.0-

60.2

)<0

.002

b

Len

gth

of

stay

IC

U (d

ays)

med

ian

(IQ

R)

4.5

(2.0

-11.

3)2.

8 (1

.1-8

.2)

3.1

(1.8

-8.3

)<0

.002

c

Len

gth

of

stay

hos

pit

al (d

ays)

med

ian

(IQ

R)

18.0

(9.0

-38.

7)14

.0 (7

.9-3

0.5)

16.0

(8.1

-32.

8)<0

.002

c

Hos

pit

alis

atio

n d

ue

to e

mer

gen

cy (%

)54

.041

.845

.1<0

.002

a

Die

d o

n I

CU

n (%

)47

2 (1

8.6)

659

(13.

4)39

3 (1

4.4)

<0.0

02a

Die

d i

n h

osp

ital

(%)

135

(5.3

)27

5 (5

.6)

147

(5.4

)0.

741a

Nu

mbe

r of

glu

cose

mea

sure

men

ts, t

otal

4354

218

3393

1846

59n

.a.

Nu

mbe

r of

glu

cose

mea

sure

men

ts p

er p

atie

nt,

med

ian

(IQ

R)

8 (5

-18)

12 (6

-37)

27 (1

3-76

)<0

.002

a

Nu

mbe

r of

pot

assi

um

mea

sure

men

ts, t

otal

4495

477

407

4895

9n

.a.

Nu

mbe

r of

pot

assi

um

mea

sure

men

ts p

er p

atie

nt,

med

ian

(IQ

R)

10 (6

-21)

9 (6

-17)

9 (6

-20)

<0.0

02a

pH

, mea

n (9

5% C

I)7.

42 (7

.38-

7.45

)†7.

39 (7

.39-

7.39

)†††

††7.

39 (7

.39-

7.39

)††

0.12

5b

Seru

m c

reat

inin

e (m

mol

/L) m

ean

(95%

CI)

120.

6 (1

15.7

-125

.4)†

††10

8.5

(105

.6-1

11.4

) †††

†††

111.

5 (1

07.5

-115

.5)†

†††

<0.0

02b

p-v

alu

es r

epre

sen

t d

iffe

ren

ces

betw

een

Con

ven

tion

al p

erio

d a

nd

TG

C p

erio

da =

Ch

i sq

uar

e te

stb

= in

dep

end

ent

sam

ple

s t-

test

c = M

ann

Wh

itn

ey U

tes

t95

% C

I =

95%

con

fid

ence

in

terv

al; I

QR

= i

nte

rqu

arti

le r

ange

n.a

. = n

ot a

pp

lica

ble

† 0.

6% m

issi

ng

†† 8

.6%

mis

sin

g††

† 0.

1% m

issi

ng

††††

0.7

% m

issi

ng

††††

† 0.

6% m

issi

ng

††††

†† 0

.5%

mis

sin

g

111

3.4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Comparison of the TGC period with the conventional periodThe mean serum glucose levels, proportion of patients with a hypoglycemia and SD

values for serum glucose levels over time are displayed in the figures 2a, 2b and 2c.

After implementation of the TGC protocol, mean serum glucose levels dropped from

8.7 to 6.6 mmol/L, an absolute reduction of 2.1 mmol/L (95%CI = -1.8 to -2.3, p<0.002,

Figure 2a). The percentage of patients with a hypoglycemia increased from 1.7% to

7.6%, an absolute increase of 5.9% (95%CI = 3.0% to 8.9%, p<0.002). The variability

of serum glucose levels within a patient did not change after implementation of

the TGC protocol (absolute difference -0.11 mmol/L; 95%CI = -0.26 to -0.03, p=0.13).

Similarly to serum glucose, values over time for serum potassium are displayed

in figure 2d, e, and f. Mean serum potassium levels stayed the same (absolute

increase 0.02 mmol/L, 95%CI = -0.06 to 0.09 mmol/L, p=0.64). The percentage of

patients with a hypokalemia dropped from 23.8 to 19.0%, an absolute reduction of

4.8% (95%CI = -11.1% to 1.5%, p=0.13), however this reduction was not statistically

significant. The variability of serum potassium levels within a patient decreased

after implementation of the TGC; SD values decreased by 0.04 mmol/L (95%CI =

-0.01 to -0.07, p=0.01).

Relationship between serum glucose and potassium levels and ICU mortality during TGCDuring TGC, overall ICU mortality was 15.2%. Multivariate logistic regression

analysis revealed a U-shaped relationship between mean serum glucose levels and

mortality (figure 3a). A J-shaped relationship was found between serum potassium

level and mortality (figure 3b). Both trends were statistically significant and

independent of age, gender, length of ICU stay, and mean serum potassium or

mean serum glucose levels respectively.

Mortality increased with increasing variability (SDs) of both serum glucose and

serum potassium levels (figure 3b and 3d). These trends were also independent of

age, gender, length of stay and SD for serum potassium or SD for serum glucose

levels respectively.

The percentages of patients who died at the ICU for each combined mean serum

glucose and mean serum potassium level subgroup is displayed in Table 2b. When

stratified for both mean serum glucose and mean serum potassium levels, extremes

in mortalities were seen in both the combined highest mean serum potassium

and lowest serum glucose subgroup (45.2%) and the combined highest mean

serum glucose and lowest mean serum potassium subgroup (40.5%). ICU mortality

increased with increasing variability of both serum glucose and serum potassium

levels with extremes at the highest SD’s of the combination (table 2b).


112

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Figure 2a: Mean serum glucose levels during the study period y=8.720875 - 0.00002*time

1 - 0.0011*time

2 + 0.001605*time

3

Figure 2b: Percentage of patients with a period of hypoglycemia (≤2.2 mmol/L) during the study period

y=2.268667 - 0.00056*time1 + 0.003807*time

2 - 0.00659*time

3

Figure 2c: Means of individual standard deviations for serum glucose levels during the study period

y=2.085461 - 0.00019*time1 + 0.000133*time

2 + 1.342228*time

3

113

3.4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Figure 2d: Mean serum potassium levels during the study period y=4.117654 - 0.0000025*time

1 + 0.0000114*time

2 + 0.0000374*time

3

Figure 2e: Percentage of patients with a period of hypokalemia (≤3 mmol/L) during the study period

y = 24.14247 - 0.00029 * time1 - 0.00236 * time

2 - 0.004526 * time

3

Figure 2f: Means of individual standard deviations for serum potassium levels during the study period

y=0.493879 - 0.000021*time1 - 0.000033*time

2 - 0.0000311*time

3


114

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Figure 3a: Relationship between mean serum glucose levels and adjusted* ODDs ratio on ICU mortality.

* adjusted for age, gender, length of ICU stay and mean serum potassium levels

Figure 3b: Relationship between mean serum potassium levels and adjusted* ODDs ratio on ICU mortality.

* adjusted for age, gender, length of ICU stay and mean serum glucose levels.

115

3.4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Figure 3c: Relationship between SD for serum glucose levels and adjusted* ODDs ratio on ICU mortality.

* adjusted for age, gender, length of ICU stay and SD for serum potassium levels

Figure 3d: Relationship between SD for serum potassium levels and adjusted* ODDs ratio on ICU mortality.

* adjusted for age, gender, length of ICU stay and SD for serum glucose levels

116

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Tab

le 2

a:

ICU

mor

tali

ty (%

an

d n

) per

str

atu

m o

f m

ean

ser

um

glu

cose

an

d m

ean

ser

um

pot

assi

um

lev

els

du

rin

g th

e TG

C p

erio

d

aver

age

1 (2

.78-

3.77

)2

(3.7

8-3.

91)

3 (3

.92-

4.00

)4

(4.0

1-4.

10)

5 (4

.11-

4.21

)6

(4.2

2-4.

34)

7 (4

.35-

4.57

)8

(4.5

8-6.

04)

pot

assi

um

n=3

17n

=339

n=3

07n

=334

n=3

20n

=311

n=3

36n

=319

tota

l

aver

age

glu

cose

1 (3

.18-

6.00

)n

=319

4/41

(9.8

%)

6/40

(15.

0%)

8/40

(20.

0%)

5/51

(9.8

%)

9/41

(22.

0%)

9/36

(25.

0%)

5/28

(17.

9%)

19/4

2 (4

5.2%

)65

/319

(20.

4%)

2 (6

.01-

6.25

)n

=322

2/35

(5.7

%)

2/43

(4.7

%)

4/43

(9.3

%)

1/40

(2.5

%)

6/48

(12.

5%)

6/34

(17.

6%)

8/45

(17.

8%)

15/3

4 (4

4.1%

)44

/322

(13.

7%)

3 (6

.26-

6.46

)n

=338

1/44

(2.3

%)

2/49

(4.1

%)

3/47

(6.4

%)

4/41

(9.8

%)

8/41

(19.

5%)

6/37

(16.

2%)

7/35

(20.

0%)

14/4

4 (3

1.8%

)45

/338

(13.

3%)

4 (6

.47-

6.67

)n

=304

2/26

(7.7

%)

3/37

(8.1

%)

4/40

(10.

0%)

5/51

(9.8

%)

0/34

(0%

)4/

50 (8

.0%

)11

/47

(23.

4%)

7/19

(36.

8%)

36/3

04 (1

1.8%

)

5 (6

.68-

6.93

)n

=331

3/43

(7.0

%)

2/54

(3.7

%)

4/35

(11.

4%)

4/49

(8.2

%)

3/37

(8.1

%)

11/3

9 (2

8.2%

)9/

38 (2

3.7%

)9/

36 (2

5.0%

)45

/331

(13.

6%)

6 (6

.94-

7.29

)n

=332

9/42

(21.

4%)

3/47

(6.4

%)

1/35

(2.9

%)

2/32

(6.3

%)

7/41

(17.

1%)

5/44

(11.

4%)

6/44

(13.

6%)

12/3

7 (3

2.4%

)45

/322

(14.

0%)

7 (7

.30-

7.85

)n

=325

6/44

(13.

6%)

5/37

(13.

5%)

7/40

(17.

5%)

1/38

(2.6

%)

3/36

(8.3

%)

4/31

(12.

9%)

5/53

(9.4

%)

10/4

6 (2

1.7%

)41

/325

(12.

6%)

8 (7

.86-

29.9

8)n

=322

17/4

2 (4

0.5%

)6/

32 (1

8.8%

)3/

27 (1

1.1%

)4/

32 (1

2.5%

)8/

42 (1

9.0%

)3/

40 (7

.5%

)10

/46

(21.

7%)

21/6

1 (3

4.4%

)72

/322

(22.

4%)

tota

ln

=258

344

/317

(13.

9%)

29/3

39 (8

.6%

)34

/307

(11.

1%)

26/3

34 (7

.8%

)44

/320

(13.

8%)

48/3

11 (1

5.4%

)61

/336

(18.

2%)

107/

319

(33.

5%)

393/

2583

(15.

2%)

Tab

le 2

b:

ICU

mor

tali

ty (%

an

d n

) per

str

atu

m o

f st

and

ard

dev

iati

on o

f se

rum

glu

cose

an

d s

eru

m p

otas

siu

m l

evel

s d

uri

ng

the

TGC

per

iod

SD p

otas

siu

m1

(0.0

0-0.

22)

2 (0

.23-

0.28

)3

(0.2

9-0.

34)

4 (0

.35-

0.39

)5

(0.4

0-0.

45)

6 (0

.46-

0.53

)7

(0.5

4-0.

66)

8 (0

.67-

2.11

)

n=3

23n

=291

n=3

70n

=298

n=3

30n

=312

n=3

37n

=322

tota

l

SD g

luco

se1

(0.0

0-0.

99)

n=3

183/

68 (4

.4%

)3/

67 (4

.5%

)0/

59 (0

.0%

)1/

47 (2

.1%

)0/

28 (0

.0%

)1/

16 (6

.3%

)2/

15 (1

3.3%

)2/

18 (1

1.1%

)12

/318

(3.8

%)

2 (1

.00-

1.23

)n

=330

0/58

(0.0

%)

6/41

(14.

6%)

1/47

(2.1

%)

2/49

(4.1

%)

5/49

(10.

2%)

2/39

(5.1

%)

4/24

(16.

7%)

14/2

3 (6

0.9%

)34

/330

(10.

3%)

3 (1

.24-

1.42

)n

=318

0/39

(0.0

%)

3/44

(6.8

%)

3/54

(5.6

%)

1/39

(2.6

%)

6/35

(17.

1%)

7/41

(17.

1%)

8/42

(19.

0%)

5/24

(20.

8%)

33/3

18 (1

0.4%

)

4 (1

.43-

1.61

)n

=336

1/41

(2.4

%)

1/30

(3.3

%)

4/53

(7.5

%)

1/53

(1.9

%)

5/44

(11.

4%)

4/35

(11.

4%)

8/46

(17.

4%)

8/34

(23.

5%)

32/3

36 (9

.5%

)

5 (1

.62-

1.81

)n

=311

1/32

(3.1

%)

1/26

(3.8

%)

3/36

(8.3

%)

5/32

(15.

5%)

5/48

(10.

4%)

13/5

1 (2

5.5%

)10

/42

(23.

8%)

19/4

4 (4

3.2%

)57

/311

(18.

3%)

6 (1

.82-

2.12

)n

=322

5/29

(17.

2%)

3/22

(13.

6%)

5/43

(11.

6%)

1/28

(3.6

%)

5/43

(11.

6%)

10/4

4 (2

2.7%

)20

/67

(29.

9%)

19/4

6 (4

1.3%

)68

/322

(21.

1%)

7 (2

.13-

2.60

)n

=323

0/27

(0.0

%)

1/37

(2.7

%)

2/42

(4.8

%)

2/30

(6.7

%)

4/38

(10.

5%)

15/4

7 (3

1.9%

)20

/46

(43.

5%)

22/5

6 (3

9.3%

)66

/323

(20.

4%)

8 (2

.61-

24.4

5)n

=325

2/29

(6.9

%)

2/24

(8.3

%)

2/36

(5.6

%)

6/20

(30.

0%)

3/45

(6.7

%)

15/3

9 (3

8.5%

)17

/55

(30.

9%)

44/7

7 (5

7.1%

)91

/325

(28.

0%)

tota

ln

=258

312

/323

(3.7

%)

20/2

91 (6

.9%

)20

/370

(5.4

%)

19/2

98 (6

.4%

)33

/330

(10.

0%)

67/3

12 (2

1.5%

)89

/337

(26.

4%)

133/

322

(41.

3%)

393/

2583

(15.

2%)

117

3.4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Discussion

Although implementation of TGC resulted in a significant absolute reduction of

mean serum glucose levels of 2.1 mmol/L, mean serum potassium levels did not

change significantly. The proportion of patients with a hypoglycemia increased

from 1.7 to 7.6% while the proportion of patients with a hypokalemia did not

change. The variability of serum glucose levels within patients did not change

while the variability of serum potassium values decreased. The highest mortality

was seen in patients with low serum glucose levels combined with high serum

potassium levels and in patients with high serum glucose levels combined with

low serum potassium levels. ICU mortality increased with increasing variability for

both serum glucose and serum potassium levels.

Implementation of a TGC led to 3.4 times more frequent measurement of serum

glucose levels (27 vs 8), lower mean glucose levels and a higher incidence of

hypoglycemia. As both moderate and severe hypoglycemia are associated with an

increased risk of death, an elevated rate of hypoglycemia should be prevented27,28.

In our setting, after implementation of the TGC protocol, the incidence of

hypoglycemia was 7.6%. This percentage compares favorably to the percentages

5.1 – 28.6% found in other studies8, 11, 14. It is hard, however, to unravel whether the

increased incidence of hypoglycemia is caused by the lower mean glucose levels or

whether it is caused by the more frequent measuring of serum glucose levels, as

more frequent monitoring may also increase the chance to detect a hypoglycemia.

As insulin therapy induces a shift of potassium from the extracellular to the

intracellular space, we expected that the proportion of patients with a hypokalemia

would increase. However, in our study mean serum potassium levels and the

proportion of patients with a hypokalemia did not change significantly after

implementation of the TGC protocol. One study did find an increased incidence

of hypokalemia after implementation of a TGC, however, this study used a higher

threshold to define hypokalemia12, 29. Another study reported an increase of patient

days with hypokalemia just after application of TGC, but this increase did not

persist throughout the study period23. We believe that a serum potassium level ≤

3 mmol/L represents a more appropriate threshold as this comprises a serious risk

on arrhythmia30. Apparently, the unchanged policy in our setting to keep serum

potassium levels above 4 mmol/L was successful in preventing severe hypokalemia.

Some studies showed that TGC increased the variability of serum glucose levels.

This is of clinical importance as an increased variability is associated with a rise

in mortality. In our study, after implementation of TGC, serum glucose SD values

stayed about equal. This means that TGC can be implemented in the general practice


118

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

without an increase of serum glucose SD values. SD of serum potassium levels even

decreased, while the target levels and monitoring frequency remained unchanged.

Different parameters have been reported to express glucose variability, among them

the Mean Absolute Glucose per hour (MAG) and SD. The MAG sums the absolute

differences between two consecutive serum glucose levels and averages these over

the time. Consequently, the MAG is influenced by the number of measurement31.

As in our study the number of measurements was 3.4 times higher in the TGC

period than in the conventional period, the MAG could not be used to compare the

variability. As the SD is the most frequently used parameter in literature22, the SD

was chosen to report the variability.

During TGC, the highest ICU mortality was seen at both low and high mean serum

glucose levels. A similar relationship between mean glucose levels and mortality

was recently described for cardiac patients admitted to the ICU32. Mortality, however,

was also high at both low and high serum potassium levels. This means that both

mean serum glucose and serum potassium levels were independently associated

with ICU mortality. Moreover, mortality for low and high mean serum potassium

levels was higher than for low and high mean serum glucose levels. Apparently, the

association between mean serum potassium levels and ICU mortality was stronger

than the association between serum glucose levels and ICU mortality.

The combined mean serum glucose – mean serum potassium levels table reveals

highest mortality rates in both the combined lowest serum glucose and highest

serum potassium subgroup (45.2%) and the combined highest serum glucose and

lowest serum potassium subgroup (40.5%). Patients in the first subgroup, may have

been severely ill, which is often accompanied with high serum potassium levels.

When these high serum potassium levels were treated with insulin, hypoglycemia

may have resulted. In the latter subgroup, the TGC protocol might have attributed

to the high mortality as correction of the high serum glucose level in combination

with the relatively low serum potassium level may have resulted in an unintended

hypokalemia.

ICU mortality also increased with increasing SD values for both serum glucose and

serum potassium levels and highest mortality rates were seen in the combined

highest serum glucose SD and serum potassium SD subgroups. Sakr et al recently

also showed a relationship between variability of serum sodium levels and

mortality, suggesting that high variability is unfavorable for a patient’s outcome33.

High variability may have been caused by the TGC protocol but may also be the

result of the severity of disease. As no information was available about the severity

of illness, it was not possible to unravel the association between variability, severity

of illness and mortality, which limits the causal interpretation.

119

3.4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Another limitation is that no information was available about the individual use of

insulin, parenteral feeding or co-medication. The use of co-medication might have

been relevant as it is known that many medications may influence both serum

glucose and/or serum potassium levels34-37. Beta2-sympathicomimetics, for example,

may induce hypokalemia and beta2-sympathicomimetics are widely used in the

ICU38. The use of co-medications, however, is not supposed to vary between the

conventional and TGC period to an extent that would have influenced the outcome

measures.

In conclusion, our study shows that TGC can be implemented was not associated

with an increased risk of serum potassium related events. During TGC, both low

and high serum glucose and low and high mean serum potassium levels should be

prevented, as these laboratory findings were associated with an increased mortality

rate. The same applies for a high variability of both serum glucose and serum

potassium levels.

Acknowledgement





120

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

References

1. Lepper PM, Ott S, Nüesch E, et al: Serum glucose levels for predicting death in patients admitted to

hospital for community acquired pneumonia: prospective cohort study. BMJ 2012;344:e3397

2. Brownlee M. Biochemistry and molecular cell biology of diabetic complications. Nature 2001;414:813–

820

3. Weekers F, Giulietti A, Michalaki M, et al: Metabolic, endocrine and immune effects of stress

hyperglycemia in a rabbit model of prolonged critical illness. Endocrinology 2003;144:5329–5338

4. Mizock BA: Alterations in carbohydrate metabolism during stress: A review of the literature. Am J Med

1995;98:75-84

5. Cromphaut van SJ: Hyperglycaemia as part of the stress response: the underlying mechanisms. Best

Pract Res Clin Anaesthesiol 2009;23:375–386

6. Dungan KM, Braithwaite SS, Preiser JC: Stress hyperglycaemia. Lancet 2009; 373: 1798–1807

7. Berghe van den G: How does blood glucose control with insulin save lives in intensive care? J Clin

Invest 2004;114:1187-1195

8. Berghe van den G, Wouters P, Weekers F, et al: Intensive insulin therapy in critically ill patients. N

Engl J Med 2001;345:1359-67

9. Berghe van den G, Wilmer A, Hermans G, et al: Intensive insulin therapy in the medical ICU. N Engl J

Med 2006;354:449-61.

10. De La Rosa GDC, Donado JH, Restrepo AH, et al: Strict glycaemic control in patients hospitalised in a

mixed medical and surgical intensive care unit: a randomised clinical trial. Crit Care 2008;12:R120

11. Arabi YM, Dabbagh OC, Tamim HM, et al: Intensive versus conventional insulin therapy: a randomized

controlled trial in medical and surgical critically ill patients. Crit Care Med 2008;36:3190-3197

12. The NICE-SUGAR Study Investigators: Intensive versus conventional glucose control in critically ill

patients. N Engl J Med 2009;360:1283-1297

13. Berghe van den G. Mesotten D, Vanhorebeek I: Intensive insulin therapy in the intensive care unit.

CMJA 2009;180:799-800

14. Wiener SR, Wiener D, Larson R: Benefits and Risks of Tight Glucose Control in Critically Ill Adults.

JAMA 2008;300:933-944

15. Kansagara D, Fu R, Freeman M, Wolf F, Helfand M: Intensive Insulin Therapy in Hospitalized Patients:

A Systematic Review. Ann Intern Med 2011;154:268-282

16. Bagshaw SM, Bellomo R, Jacka1 MJ et al: The impact of early hypoglycemia and blood glucose

variability on outcome in critical illness. Crit Care 2009;13:R91

17. Quagliaro L, Piconi L, Assaloni R et al: Intermittent high glucose enhances apoptosis related to

oxidative stress in human umbilical vein endothelial cells: the role of protein kinase C and NAD(P)

H-oxidase activation. Diabetes 2003;52:2795-2804

18. Monnier L, Mas E, Ginet C, et al: Activation of Oxidative Stress by Acute Glucose Fluctuations Compared

With Sustained Chronic Hyperglycemia in Patients With Type 2 Diabetes. JAMA. 2006;295:1681-1687

19. Egi M, Bellomo R, Stachowski E et al: Variability of Blood Glucose Concentration and Short-term

Mortality in Critically Ill Patients. Anesthesiology 2006;105:244–252

20. Krinsley JS: Glycemic variability: A strong independent predictor of mortality in critically ill patients.

Crit Care Med 2008;36:3008-3013

21. Ali NA, O’Brien JM, Dungan K, et al: Glucose variability and mortality in patients with sepsis. Crit Care

Med 2008;36:2316-21

22. Eslami S, Taherzadeh Z, Schultz M, Abu-Hanna A. Glucose variability measures and their effect on

mortality: a systematic review. Int Care Med 2011;37:583-593

23. Onyenweni AJ, Winterstein AG. Rates of hypokalaemia after implementation of aggressive insulin

dosing in critical care patients. Pharmacoepidemiol Drug Saf 2007;16:S011

24. Berghe van den G, Schetz M, Vlasselaers D, et al: Intensive insulin therapy in critically ill patients:

NICE-SUGAR or Leuven blood glucose target? J Clin Endocrin Metab 2009;94:3163-3170

25. Berg ten MJ, Huisman A, van den Bemt PM, et al: Linking laboratory and medication data: new

opportunities for pharmacoepidemiological research. Clin Chem Lab Med 2007;45:13-19

121

3.4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

26. Wagner AK, Soumerai SB, Zhang F, et al: Segmented regression analysis of interrupted time series

studies in medical use research. J Clin Pharm Therapeut 2002;27:299-309

27. Hermanides J, Bosman RJ, Vriesendorp TM, et al: Hypoglycemia is associated with intensive care unit

mortality. Crit Care Med 2010;38:1430-1434

28. The NICE-SUGAR Study Investigators: Hypoglycemia and risk of death in critically ill patients. N Engl

J Med 2012;367:1108-1118

29. Vlasselaers D, Milants I, Desmet L, et al: Intensive insulin therapy for patients in paediatric intensive

care: a prospective, randomised controlled study. Lancet 2009;373:547–556

30. Paltiel O, Salakhov E, Ronen I, et al: Management of severe hypokalaemia in hospitalized patients.

Arch Int Med 2001;161:1089-1095

31. Harmsen RE, Spronk PE, Schultz MJ, Abu-Hanna A: May frequency of blood glucose measurement be

blurring the association between MAG and mortality? Crit Care Med 2011;39:224

32. Lipton JA, Barendse RJ, Van Domburg RT et al: Hyperglycemia at admission and during hospital stay

are independent risk factors for mortality in high risk cardiac patients admitted to an intensive

cardiac care unit. Eur Heart J Acute Cardiovasc Care 2013:2:306-313

33. Sakr Y, Rother S, Ferreira AM, Ewald C, et al: Fluctuations in serum sodium level are associated with

an increased risk of death in surgical ICU patients. Crit Care Med 2013;41:133-142

34. Rimmer JM, Horn JF, Gennari J: Hyperkalaemia as a complication of drug therapy. Arch Intern Med

1987;147:867-869

35. Paice BJ, Paterson KR, Onyanga-Omara F et al: Record linkage study of hypokalaemia in hospitalized

patients. Postgrad Med J 1986;62:187-191

36. Salem CB, Fathallah N, Hmouda H and Bouraoui K: Drug-Induced Hypoglycaemia, an update. Drug

Saf 2011;34 :21-45

37. Luna B, MN Feinglos: Drug-induced hyperglycemia. JAMA 2001; 286: 1945-1948

38. Zanen P: Dalingen van het kaliumgehalte in het serum ten gevolge van ß2-sympathicomimetica. Ned

Tijdschr Geneeskd. 1990;134:688-689

4General discussion and summary

Chapter 4 | General discussion and summary

124

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

125

4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

General discussion

Pharmaceutical care aims to optimize the benefit/harm balance of drug treatment

for the individual patient. Approved medicines have, on a population level, a by

regulatory authorities judged positive benefit / harm balance given appropriate use

by health care professionals and patients. On the level of the individual patient

there is in clinical practice, however, a large variability in treatment response:

for the individual patient the benefit/harm balance may substantially differ from

the population average. In a way, each time a medicine is prescribed, it can be

considered an n=1 experiment and adverse effects may occur. Medication errors are

an important cause of, often preventable, adverse drug events (ADEs)1. Medication

errors most frequently result from flaws in the complex medication management

iterative process consisting of prescribing, transcription and verification, pharmacy

dispensing, nurse administration to the patient or use by the patient himself,

monitoring the effects of, and evaluation of the pharmacotherapy2. Especially the

prescribing and monitoring stages are prone for errors3. Pharmaceutical care aims

to prevent such errors and should therefore especially focus on these two stages.

Drug-drug interactions (DDIs) have shown to contribute significantly to the

negative consequences of drug treatment4-6. There are many guidelines to support

healthcare professionals to mitigate the potential clinical consequences of DDIs

(pDDIs) for the individual patient7-10. These guidelines advise physicians and

pharmacists how to handle the pDDI for the individual patient at the moment

of either prescribing, for example by suggesting alternative medication, and/

or during the course of drug use by adequate monitoring. This monitoring may

consist of clinical monitoring, such as the clinical observation of desired and

unintended effects (e.g. skin reactions), but also by measurement of biomarkers,

which may consist of physical parameters like blood pressure and ECG or of

laboratory markers such as serum drug concentrations, serum creatinine or serum

electrolytes. Analysis of the Dutch clinical guidelines on pDDI management showed

that 33% of the risk mitigation strategies advise to monitor one or more laboratory

values11. As a limited set of laboratory markers is particularly involved in the risk

mitigation strategies of ADEs12,13, exchange between health care professionals of

information on drug serum levels, INR, serum creatinine, potassium, sodium levels

and pharmacogenetic markers became obliged by the Dutch law in 201214. Studies

on risk mitigation strategies should therefore especially focus on these markers.

The medication management process is in most hospitals nowadays supported by

information technology applications such as computerized physician order entry

systems (CPOE)15,16. A CPOE usually includes a basic form of clinical decision support

(CDS) for example for to alert health care professionals for the risk of pDDIs at


126

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

the moment of drug prescribing: each new prescription is checked for drug-drug

interactions given the other active prescriptions for that patient. The moment

of prescribing is indeed an important moment to reflect on the appropriateness

and fit of the patient’s drug therapy, including the risk of the potential negative

effects of a pDDI. At this stage the physician can only decide to either continue or

discontinue the medication while many combinations of drugs are not necessarily

harmful when these are adequately monitored during the course of drug taking.

This means that at the moment of prescribing, the physician has to decide which

monitoring, when and how often has to be performed in order to minimize the risk

of actual occurrence of an ADE. Clinical guidelines of pDDI management, however,

often lack this clarity17,18.

It has been shown that in ambulatory care the risk of pDDIs is often not

monitored19-21. It may be expected that within the hospital setting patients are

monitored more adequately, because of the easy access to laboratory facilities and

the availability of other monitoring facilities in hospitals as well as the exacerbated

disease state of the patient. However, hospital stay is only short and ADEs that result

from pDDIs may therefore also become manifest after discharge22. This means that

monitoring should continue after discharge and that information on both drug use

and monitoring should accompany the patient when he moves from hospital care

to primary care. Additional to differences related to setting (hospital or at home),

risk mitigation strategies may also depend on patient characteristics. Geriatric

patients differ from children and ICU patients are different again. ICU patients are

in general more severely ill, which may interfere with their drug metabolism and

their capacity to counteract and deal with physiological disturbances. Although

these patients are monitored intensively, treatment is complex and risk mitigation

strategies therefore deserve extra attention.

The objective of this thesis was to describe the frequency and potential clinical

relevance of drug therapy monitoring with laboratory markers in hospitalized

patients with a special focus on pDDIs, potassium and ICU patients.

127

4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Summary

A high burden of nonspecific computer alerts is known to undermine the safety

purposes of an alerting system. As basic pDDI alerting systems do not take relevant

patient characteristics into account, current systems have a high sensitivity but

a low specificity. Chapter 2 showed to what extent this problem may apply for

hospitalized patients. Indeed pDDIs occurred frequently in hospitalized patients.

Of all hospitalized patients, 25% encountered at least one pDDI, and mainly due to

the higher number of medication orders per patient, this percentage was twice as

high for the ICU population (54%). pDDIs occurring in ICU patients were of special

interest as medications used at the ICU are often categorized as “high risk” and

patients may be, due to their severity of illness, more vulnerable to ADEs resulting

from these pDDIs23,24. Chapter 2.1 and chapter 2.2 showed that both in the general

hospital setting and the ICU, a limited number of drug pairs were responsible for

the majority of the pDDI alerts. The top ten was responsible for 53% of the pDDIs at

general hospital departments and for 79% of the pDDIs occurring at the ICU. The

most frequently occurring pDDIs on the ICU was a reflection of the specific high

risk medications frequently used at the ICU and the top-10 therefore differed from

the top-10 pDDIs observed in patients hospitalized at general hospital departments.

The most frequently occurring possible outcome of the pDDIs was an increased

risk of side effects. Laboratory monitoring was the most frequently advised risk

mitigation strategy, namely in approximately 50% of all pDDIs (48.6% for patients

hospitalized at general hospital departments and 51.6% for patients hospitalized at

the ICU). This means that an advanced clinical decision support system which links

laboratory data to prescription data may be an important tool for a more efficient

and more effective management of risks associated with these pDDIs.

Chapter 3 showed that measurement of serum potassium, sodium and/or creatinine

was performed in about 50% of the patients somewhere during hospitalization.

These three markers were frequently involved in clinical risk management of drug

therapy12. If one of these three biomarkers was measured, the other two were

measured as well in over 90% of the patients, suggesting that these three markers

are often monitored simultaneously. Overall, patient- and admission related factors

such as age and medical specialism appeared to be the stronger predictors of

monitoring than the use of specific medications. The use of specific medication

requiring the monitoring of one of these three markers did not, or only modestly,

increase the frequency of monitoring. For example, the monitoring of serum

creatinine, was not more frequent in patients using medication that was renally

cleared and required dose modification in case of decreased renal function (DRF-

drugs). The computerized physician order entry system (CPOE), however, did not


128

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

signal when a DRF drug was prescribed. As 77% of the patients were using such

a DRF drug, signaling would have led to a high burden of alerts. An exception of

medication leading to an increased percentage of patients with a measurement

is the use of potassium supplements, for which serum potassium measurement

increased from 41 to 99% of the patients.

To assess whether the laboratory values measured during hospitalization are

suitable for medication verification at discharge, the time of the last measurement

was also determined for each of these markers. The percentage of patients with

a measurement within the last 48 hours before discharge was less than 25%,

suggesting that there is still room for improvement to communicate a relevant test

result at transition of care.

As serum potassium is involved in the risk mitigation strategy of one of the most

frequently occurring pDDIs, e.g. those between serum potassium increasing drugs

(PIDs) and this pDDI has potentially serious consequences25, monitoring of this

electrolyte was studied in more detail. Chapter 3.2 showed that in the presence of

a CPOE with pDDI alerting, serum potassium levels were measured slightly more

frequently in patients with a pDDI between two potassium increasing drugs than

when the patients were using monotherapy of potassium increasing drugs (67% vs

58%). Whether this difference can be attributed to pDDI alerting, however, is hard

to say as baseline characteristics of patients in the pDDI group were different from

those in the monotherapy group. Even though physicians received a direct pop-up

to monitor serum potassium levels in case of prescribing two PIDs concomitantly,

serum potassium levels were not measured in 33% of patients and 10% of the

patients developed hyperkalemia

Although monitoring is often advised when medication is started17, there is less

attention for the role of monitoring when medication is discontinued, even when

this medication is involved in a pDDI. Inhibitors of the renin-angiotensin-aldosteron

system (RAS-i) and spironolactone are often combined with diuretics in patients

with cardiac failure. As these drugs have an opposite effect on serum potassium

levels, stopping one of these drugs may have consequences on serum potassium

levels and when these levels fall outside of normal ranges, even lead to a potassium

related cardiac event in these vulnerable patients. After discontinuation of the

potassium lowering drug, serum potassium levels increased in 59% of the patients

and 3.2% of the patients developed hyperkalemia (potassium>5.5 mmol/L) (chapter

3.3). When the potassium increasing drug was stopped, serum potassium levels

decreased in 70% of the patients and 17% developed hypokalemia (potassium <3.5

mmol/L). These results show that monitoring is not only important after starting

therapy, but also after discontinuation. As this need may also be true for patients

using other drugs, clinical risk management should not only be focused on risks

129

4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

that occur when a new medication is prescribed, but also on those that may occur

when the medication is stopped.

Serum potassium levels may also be influenced by insulin which is involved in

tight glucose protocols (TGC). TGC aims to lower glucose levels and the effect on

serum glucose levels has therefore been extensively studied. To what extent TGC

also influences serum potassium levels and possibly may result in potassium

related adverse events, is barely studied. TGC protocols therefore often lack clarity

on when and how often to monitor serum potassium levels as well. Results of

our study showed that TGC was not associated with an increased risk of serum

hypokalemia, although both low and high mean values and high variability of both

serum glucose and serum potassium levels were associated with an increased ICU

mortality and should therefore be prevented (chapter 3.4).

All studies that are presented in this thesis show the importance of monitoring

laboratory markers as a risk mitigation strategy in drug therapy for hospitalized

patients. Individual studies focused on the role of monitoring for patients with

pDDIs and vulnerable patients admitted to the ICU. In this chapter, the studies in

this thesis will be discussed in a broader perspective of three themes. In chapter

2 it is shown that a CDS based on basic pDDI alerting is leading to a high burden

of non-specific alerts (pDDI alerts generated for 8-10% of the prescriptions). The

result that only a limited number of drug-drug combinations was responsible for

the high burden of alerts and that the most frequently advised risk mitigation

strategy consisted of monitoring, offer possibilities to enhance these CDS systems

with monitoring data and improve the specificity for pDDI alerting. Moreover,

as in chapter 3 is shown that adherence to monitoring guidelines is hampered,

advanced CDS should also support adherence to monitoring guidelines. The next

part therefore discusses the restrictions of current CDS systems on mitigating

risks of pDDIs and options to improve CDS systems with a focus on improving the

specificity of alerting and adherence to monitoring guidelines. The second part

addresses the role of advanced clinical decision support on the ICU and in the final

part, the role of the pharmacist or clinical pharmacologist at the ICU, additional

to CDS is discussed.

Clinical decision supportGuidelines on medical and pharmaceutical care are continuously developing given

the dynamics in research and evidence, and health care professionals are expected

to practice in accordance with these guidelines. Health care professionals, however,

not always adhere to clinical guidelines including drug therapy guidelines. For

example, adherence of pharmacists in ambulatory care to pDDI guidelines was

overall only 69%, but varied largely depending on gender, age and number of drugs


130

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

used26. The degree of adherence also varied with the nature of the recommended

risk mitigation strategy. Low adherence was found for alerts where the only

proposed mitigation strategy was drug substitution. Adherence was high when

the pharmacist could communicate the risk mitigation strategy directly with

the patient. Physicians may be hindered to adhere to guidelines for a variety of

reasons27,28. Lack of awareness and familiarity to therapy guidelines contribute to

decision making based on clinical experience rather than evidence or guidelines

which may in turn create risks for adverse events. Other frequently mentioned

reasons for a hampered adherence are lack on agreement on the content of the

guideline, lack of believe that the guideline recommendation can be performed,

lack of outcome expectancy, lack of ability to overcome the inertia of previous

practice, unclear guideline recommendations, and external barriers such as time

constraints or lack of a reminder system to perform recommendations.

The use of information technology in health care, especially the introduction

of (advanced) clinical decision support (CDS) which links data from different

systems and integrates clinical guidelines, may result in substantial improvement

in patient safety29. A broadly accepted definition of CDS is: “Providing clinicians

or patients with computer-generated clinical knowledge and patient-related

information, intelligently filtered or presented at appropriate times, to enhance

patient care”30. At present, CDSs have been developed for a myriad of clinical

issues, including systems for facilitating diagnosis31, reminders for prevention and

disease management 32 and medication management33,34. Clinical trials with CDS

as intervention have shown positive results, but evidence that CDS systems can

improve long term patient outcomes is still inconclusive35-37.

Current situation: basic CDS

Basic clinical decision support for medication management is in place in most

currently used computerized order entry systems (CPOEs) and includes drug-

allergy checking, dosing guidance, formulary decision support, duplicate therapy

checking, contra-indication checking and drug–drug interaction management34.

Basic CDS (bCDS) systems usually only check whether the newly prescribed drug is

in accordance with general guidelines on a population level, for example, whether

the drug potentially interacts with another drug that is already in use by the patient.

Additional clinical data specific for the patient at hand such as the patients’ actual

laboratory test results are not taken into account, and the alert is not tailored to a

specific setting given its monitoring facilities (e.g. hospital, ICU) nor to the expertise

/ specialty of the individual physician. For example, when two potassium drugs

are prescribed concomitantly, the bCDS on pDDI alerting will always generate an

alert for the risk of hyperkalemia, even when baseline serum potassium levels are

131

4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

low and the two potassium increasing drugs are prescribed with the intention to

increase serum potassium concentration. Moreover, bCDS also lacks the ability

to deal with different problems simultaneously, i.e. it assesses the clinical risk of

a pDDI and that of renal insufficiency separately from each other. As such, most

bCDS within computerized order entry systems (CPOEs) have a high sensitivity but

a low specificity, i.e. many false positive alerts appear in the prescribing stage. As

a consequence, to assess whether the patients is really at risk for a hyperkalemia,

prescribers have to interrupt their prescribing process to check in another system

or another module of the EPR for additional data on (co)morbidity, kidney function

and serum potassium levels. When they have to consult other information systems,

this may be a time consuming process. Although physicians generally believe that

a bCDS on pDDIs improves their prescribing performance, they also fear the high

burden of nonspecific alerts38.

Chapter 2 of this thesis confirms that basic pDDI alerting indeed generates a large

number of nonspecific warnings for patients both hospitalized at general hospital

department and at the ICU. This may contribute to alert fatique and non-adherence

to the risk mitigation advises given in the alerts39,40.

From bCDS to more advanced CDS

Several studies have investigated methods to increase the specificity of the alerts

but only some have shown to be partially effective (table 1) 41-44.


132

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Table 1: Methods to improve specificity of DDI alerting and to prevent alert fatigue

Reference Method to enhance DDI Outcome Remarks

vd Sijs 200841 turning off frequently overriden drug alerts

no consensus between physicians

turning off alerts per physician is still option

Paterno 200942 divide pDDIs into severity level indicatorsfor life-threatening DDIs: hard stop for one of the interacting drugs for less serious pDDIs: alert + select override reasonfor least serious pDDIs: alert providing information only

Acceptance rates for tiered versus non-tiered sites:100% vs 34% (p<0.001) for life-threatening DDIs;29% vs 11% (p<0.001) for less severe DDIs

only 0.2% of the alerts is life-threatening

Shah 200643 designate only critical to high-severity alerts to be interruptive to clinician workflow

acceptance rates: 100% for the life-threatening DDIs41% of the severe DDIs

acceptance rates are not compared to general DDI alerting

Duke 201344 integration of relevant patient specific laboratory data for serum potassium or creatinine into pDDI alerts for two potassium increasing drugshigh risk patients baseline potassium >5.0 mEq/l and/or creatinine ≥1.5 mg/dl (132 μmol/l)

no significant difference in alert-adherence in high-risk patients between the intervention group (15.3%) and the control group (16.8%) ( p=0.71)

adherence was defined as discontinuation of one of the interacting drugs; other risk mitigation strategies such as monitoring frequency were not taken into account

One study investigated the possibility to turn off alerts, but this option was not

successful as physicians could not agree on which alerts could be turned off41.

Another trial integrated relevant patient specific test results for serum potassium

and serum creatinine levels into the pDDI alert when two potassium increasing

drugs were prescribed concomitantly, but this option did not improve adherence

in terms of stopping one of the interacting drugs, even when baseline serum

potassium levels were high (> 5.5 mmol/L)44. Only when pDDIs were accompanied

with information on severity and combined with a preprogrammed order to

discontinue one of the interacting drugs, adherence improved42. These studies,

however, always measured adherence in terms of stopping one of the interacting

drugs, while often multiple options for risk mitigation strategies are advised.

Chapter 2, however, shows that the most frequently advised risk mitigation

strategy consists of monitoring and that 48% of this monitoring even constituted

of measurement of laboratory markers. Likewise, Seidling et al reported that the

possible outcome of 43% of the pDDI pairs might be modulated by patient-related

133

4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

factors which mainly consisted of measurement of three laboratory values, i.e.

renal function, potassium, and therapeutic drug monitoring results45. This means

that physicians, who did not stop one of the interacting drugs, still might have

adhered to the pDDI guidelines by performing monitoring. Because it is known

that “willing to monitor” is the most frequently mentioned reason for “not willing

to stop”42,46, appropriate monitoring would have been a better outcome measure

than discontinuation of one of the two interacting drugs. Linking laboratory values

to pharmacy data is therefore still considered as a promising option to enhance

pDDI alerting.

Although physicians have good intentions on monitoring, literature indicates that

adherence to monitoring protocols may be poor47. Poor adherence to monitoring

was indeed confirmed in chapter 3.1 in which is shown that the use of specific

medications requiring the monitoring of serum potassium, sodium or creatinine

did not, or only modestly, increase the percentage of monitoring. Strategies to

improve adherence to pDDI guidelines, should therefore also focus on monitoring.

This monitoring may consists of both measurement of baseline values and follow-

up values. Electronic reminders have shown to improve the adherence to baseline

monitoring48, but as pDDI alerts are only generated at moment of prescribing in

bCDS, improvement of follow-up measurement is not likely. The proposal of a

second “corollary” order for monitoring that directly follows the first order, may

overcome this mismatch in timing49. For example, when an alert is generated for

a pDDI between two potassium increasing drugs, the alert may automatically be

followed by a corollary order for serum potassium and creatinine measurement.

The corollary order may even consist of a series of pending orders for serum

potassium and creatinine measurements and propose to measure these values

according to the guideline as long as the pDDI is active. When also other orders

for measurements are pending, the system may propose to combine these. As

monitoring is important not only when a new drug is started, but sometimes also

when a drug is discontinued (chapter 3.3) or when the dose is changed, corollary

orders may also be useful in these situations. Subsequently, when new information

on laboratory values becomes available, these values might show that risk-benefit

balance for a drug therapy has been changed for a specific patient. The physician

has than again to decide whether to change the drug therapy regimen. The CDS

should therefore also be triggered when new information on laboratory values is

entered into the system. The same may apply for other relevant information such

as for example on (co)morbidity or patient data.

A model for an advanced CDS that may account for changes in a multifactorial

approach is proposed in Figure 1. The model is adapted from Berner et al. who

described an advanced CDS on diagnosing50. The scheme starts from a database


134

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

including all kind of structured patient related data and may for example consist

of basic patient characteristics, medications, biomarkers, and (co)morbidity (1). The

use of “smart forms” may, for example, help to enter these data in a structured

way51. When new information on one of the patient related data is entered into the

database (2) the reasoning engine is triggered (3) to activate relevant clinical rules

(4). Based on information in the knowledge base (5), which includes information on

guidelines, such as threshold values for laboratory markers and risk estimates on

adverse events, the reasoning engine estimates whether there is an increased risk

on ADEs in the new situation or not. When the knowledge base indicates that also

data is needed on other patient related parameters, the reasoning engine extracts

this information from the patient related data. When the engine determines

an increased risk that requires action, the action is communicated through the

output (6). As a rule-based method requires explicit knowledge of the details of each

medical domain, this may result in many complex rules in the reasoning engine.

Rules also use knowledge from the knowledge base, which is based on clinical

guidelines that are developed for populations and not for the individual patients.

An alternative for rule based reasoning may be case based reasoning. Cased based

reasoning uses knowledge in the form of specific cases to solve a new problem,

and the solution is based on the similarities between the new problem and the

available cases. Experiments have shown that addition of a case based engine to a

rule based engine may add details that are necessary in the knowledge base and

that the output is more appropriate to the individual patient. Despite, case based

reasoning also faces a number of problems. First, case based reasoning is based

on historic cases, which means that comparable cases need to be available in a

case database. Secondly, information which has been derived from historic cases

may also lack reliability. And last, but not least, case based reasoning is based on

complex algorithms that need to be programmed. Despite these limitations case

based reasoning seems promising to overcome the limitations of the rule based

engine when it is added to rule based reasoning52.

Figure 1: a schematic representation of advanced CDS on the monitoring stage.

For example, when a new test result for a serum potassium is entered into the patient related database

(=input), the reasoning engine is triggered to check whether this new information on serum potassium

alters the risk on an ADE. The knowledge base may provide risk estimates on serum potassium levels

that may increase the risk of an ADE, in the presence or absence of certain potassium increasing or

decreasing medications or combination of such medications, kidney function and (co)morbidity. When

the reasoning engine estimates an increased risk and action is required, the engine generates an output.

The output can consist of a preprogrammed/corollary order or an advice to the doctor, the pharmacist

or even to the patient48. The model is now elaborated on pDDIs, but when rules and knowledge are

introduced on for example monitoring of efficacy, this concept may also support monitoring of

efficacy.

Some remarks have to be made for this concept. First, for the rule based engine, the knowledge base

should be expanded with relevant data on risks. Guidelines on risk mitigation strategies, for pDDIs for

example, only contain free text fields with information on monitoring data, while these guidelines can

only work when this information is available in separate data fields on the identity, normal ranges, and

monitoring frequency of the relevant biomarkers. The databases on risk management of pDDIs should

therefore be expanded with these fields. Second, as every change in the steady state of the database

will trigger the reasoning engine, it may be necessary to expand the knowledge base with a set of

biomarkers, medications and comorbidities which may canalize the set of rules that are activated.

Finally, as input may consist of multiple kinds of data, it should be determined to whom, when, and by

which means the output is communicated. As the treating physician is primarily responsible for the

Figure 1: a schematic representation of advanced CDS on the monitoring stage.

135

4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

For example, when a new test result for a serum potassium is entered into the

patient related database (=input), the reasoning engine is triggered to check

whether this new information on serum potassium alters the risk on an ADE.

The knowledge base may provide risk estimates on serum potassium levels that

may increase the risk of an ADE, in the presence or absence of certain potassium

increasing or decreasing medications or combination of such medications, kidney

function and (co)morbidity. When the reasoning engine estimates an increased risk

and action is required, the engine generates an output. The output can consist of

a preprogrammed/corollary order or an advice to the doctor, the pharmacist or

even to the patient48. The model is now elaborated on pDDIs, but when rules and

knowledge are introduced on for example monitoring of efficacy, this concept may

also support monitoring of efficacy.

Some remarks have to be made for this concept. First, for the rule based engine,

the knowledge base should be expanded with relevant data on risks. Guidelines

on risk mitigation strategies, for pDDIs for example, only contain free text fields

with information on monitoring data, while these guidelines can only work

when this information is available in separate data fields on the identity, normal

ranges, and monitoring frequency of the relevant biomarkers. The databases on

risk management of pDDIs should therefore be expanded with these fields. Second,

as every change in the steady state of the database will trigger the reasoning

engine, it may be necessary to expand the knowledge base with a set of biomarkers,

medications and comorbidities which may canalize the set of rules that are

activated. Finally, as input may consist of multiple kinds of data, it should be

determined to whom, when, and by which means the output is communicated.

As the treating physician is primarily responsible for the patient, it is reasonable

that risks and required actions are communicated with the treating physician.

Although, there are also situations in which action is required from another health

care professional, such as clinical observation by the nurse. Information may then

also be communicated to another health care professional. It can even be imagined

that action of a health care professional is not always necessary, for example when

measurement of serum potassium levels is advised, the system may check whether

the test can be performed in material which is still available from the patient in

the laboratory. When material is still available, the order can automatically be

generated without an active action of the health care professional. However, this

may only work as long as all information on required and the performed actions

are recorded in the central patient information system.

Current basic CDS systems are developed to provide guidance on drug treatment

but often lack specificity. ICT builders should therefore work on a multifactorial

approach of CDS systems. Only when the reasoning engine is able to use


136

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

multifactorial data, i.e. not only medication data but also data on laboratory values

and morbidity, specificity of the output can be improved. ICT builders should also

work on the possibility to trigger the CDS by multifactorial changes. When changes

on laboratory data or comorbidity may also trigger the system, the monitoring

stage of the medication process is also supported. Moreover, it should be possible

that the reasoning engine works with both a static knowledge base and a flexible

knowledge based that is filled with information that is obtained from cases. Finally,

the CDS should be able to generate several types of output to several kind of health

care providers, i.e. output may also consist of (pending) corollary orders and a

change of a laboratory value may also alert the pharmacist.

Clinical decision support on the ICUBecause patients as well as the setting are different, possibilities to increase

the specificity of pDDI alerts may differ for the ICU compared to the general

hospital. The ICU differs in many aspects from general hospital departments. ICU

patients are more severely ill compared to patients admitted at general hospital

departments and often require support of their vital physiological functions like

mechanical ventilation, renal replacement therapy and intravenous support of

their blood pressure. Given the urgent need for life supporting therapy, higher

risks may be also accepted for medical treatment. Higher risks are also accepted

for treatment with medications, despite it is known that these patients, due to

their altered drug disposition and metabolism and their decreased capacity to

deal with physiological disturbances, may be more vulnerable to the side effects

of drugs. The life supporting therapy is often provided with only a limited set of

highly pharmacologic active medications. Although these medications are often

considered as “high-risk” medications53, ICU physicians are very familiar with the

use of these medications. Parenteral administration and the short half-life enable to

titrate the dose to a desired effect which is needed in acute illness while the effect

may also be easily reversed by discontinuation of the medication. Due to the focus

on life supporting therapies, ICU physicians may be less familiar with medications

the patients use at home for chronic diseases. Although these medications are often

(temporarily) discontinued when a patient is admitted at the ICU, these medications

may possibly create a higher risk especially when the patient is transferred to a

next health care provider again. To mitigate risks, ICU patients are intensively

monitored, both clinically, physiologically and biochemically. Chapter 2 shows that

the most frequently advised risk mitigation strategy is monitoring. The high rate of

monitoring at the ICU therefore offers possibilities to enhance a CDS for pDDIs at

the ICU. As suggested in the previous part, these laboratory data could be integrated

in the CDS. Whether laboratory enhancement of pDDI alerts at the moment of

137

4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

prescribing will actually change drug therapy, however, may be argued, also when

at the moment of prescribing it is known that laboratory values deviate from

reference values. As laboratory marker values may vary highly within the same ICU

patient over time, the single previous value at moment of prescribing does usually

not adequately represent the patient’s condition over time. Moreover, medical need

of drug treatment will often prevail, despite laboratory values outside normal

limits at moment of prescribing. Risks, however, can also be accepted more easily

as patients are continuously monitored and deviations may, due to the technical

facilities and intravenous access of ICU patients, be corrected quickly. Due to the

intensive monitoring, some pDDIs may even lack clinical relevance on the ICU, such

as for example the combination of insulin with a beta-blocker. In general practice

this combination is discouraged because beta-blockers may mask the symptoms of

a hypoglycemia, resulting in a possible delay of its detection and correction. ICU

patients, in contrast, usually do not notice hypoglycemia themselves anyway, and

due to the tight glucose protocol, glucose levels are frequently monitored. As such,

alerting for this pDDI is not clinically relevant in an ICU setting.

Because patients on the ICU are usually intensively monitored, and physicians

rather act on deviating monitoring values than on alerts of possible risks on pDDIs,

significant deviation of biomarkers might be a better moment to trigger the CDS on

the ICU and reflect whether there is a possible relation with the use of medications54.

The CDS may then not only be triggered when laboratory values fall outside normal

limits, but also when they are highly variable, for example when drug levels deviate

more than 50% compared to the previous level. Large within patient variability

has for example been associated with negative outcomes for lithium55. Likewise,

chapter 3.4 shows that highly variable serum potassium levels are also associated

with increased ICU mortality. As many medications may influence potassium levels,

physicians should always have to consider the role of medication in this. However,

due to their severity and complexity of disease, deviations in biomarkers are very

common in ICU patients. Often, it will not be clear which part of the deviation

can be attributed to the use of a certain drug or combination of drugs (DDI) and

which part is attributable to the myriad of other potential conditions, such as renal

insufficiency and infection. To overcome these problems, it has been suggested that

case based reasoning would be a valuable additional strategy especially at the ICU.

A conceptual version of a combined rule based and case based reasoning engine has

indeed shown promising results52,56, but as far as known, case based reasoning has

still not been implemented on ICUs.

Possibly, ICU patients are too complex to capture all covariates in CDS. This means

that human interpretation of data always stays necessary to make decisions in

complex situations. ICU physicians are therefore highly trained professionals who


138

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

are able to make decisions for complex patients. In contrast to the computer, they

can make decisions on the information which is available, also when part of the

information is lacking. Patient information should therefore be easily available.

Over the past ten years, patient data monitoring systems (PDMS) have been developed

in a way that all relevant patient-specific information can be presented in a logical

way, such as in dashboards on topics like fluid balance, circulation, respiration and

medications57. Although it can be argued whether “organization of information into

a PDMS” fits into the definition of a CDS30, organization of information may be very

helpful in human decision making. CDS may also be integrated when information

on for example the dashboard on current drug use is enhanced with information

on for example normal dosage ranges for the prescribed drugs, normal ranges of

therapeutic drug levels, information on current pDDIs and relevant monitoring

data. Logical organization of information in dashboards, may offer the possibility

to evaluate the patient at moments that may optimally fit into the physicians work

flow and it is known that CDS that fits into the work process of the physicians, are

more likely to be successful46.

The myriad of information which is presented in a PDMS, however, also generates

the risk that important information on relevant problems is easily overlooked.

Therefore there is still a need to be alerted for relevant problems. It has been

shown that clinical rules may improve prescribing when a problem is demarcated

such as dosing of antimicrobials for patients with renal failure58. As the twenty

most frequently occurring pDDIs at the ICU are responsible for more than 90%

of the alerts, this may also apply for pDDIs on the ICU. First, these pDDIs should

be judged on clinical relevance, given the monitoring facilities on the ICU. When

clinical relevance is lacking, these pDDIs may be turned off. Secondly, when the

risk of pDDIs may be relevant, and risks may be mitigated by the measurement of

laboratory values, a rule may link these laboratory data to prescription data and

signal when laboratory data deviate from normal limits which are determined

for ICU patients. The alert should than not only be generated when laboratory

values fall outside normal limits at the moment of prescribing, but also in the

monitoring stage of the pDDI. Thirdly, when a combination of drugs is potentially

life-threatening, and the clinician is required either to cancel the order he or she

is writing or discontinue the pre-existing drug order, pDDIs may be signaled in a

basic way. As this method may still generate a high burden of alerts, and physicians

do not want to be disturbed for a high burden of low specific alerts, the alerts may

also be presented on for example the medication dashboard of the PDMS. As direct

action is often not needed, the clinical pharmacist may than judge on the clinical

relevance for the specific patient and use this information in the multidisciplinary

consultation. When direct action is needed, such as for example the pDDI between

139

4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

valproate and meropenem, the pharmacist may also decide to directly consult the

physician59. Or, when medication with a narrow therapeutic index is involved in a

pDDI, such as for ciclosporine and tacrolimus, the pharmacist may even decide on

therapeutic drug monitoring.

The role of the clinical pharmacist or pharmacologist at the ICUThe previous parts of this general discussion described that it is important to

support the complex processes on the ICU with a PDMS, CPOE and CDS. CPOEs are

increasingly implemented into these PDMS and have shown to reduce the number

of prescribing errors compared to the hand-written medication order process but

the support of complex medication management processes with CDS systems is

still lagging behind60,61. Basic CDS systems lack specificity and although specificity

may be increased with rules that integrate multifactorial patient data, it will often

be too complex to integrate all these data in clinical rules for the ICU. CDS may

also not detect all drug related problems on the ICU. Recent research showed that a

CPOE/CDS generated an alert for only 8% of the drug related problems which were

identified by the clinical pharmacist62. CDS on medication is therefore often not

implemented into the CPOE of a PDMS on the ICU and evaluation of the risk/benefit

ratio of treatment with medications therefore still relies on the judgment of highly

trained ICU physicians. It was reported, however, that ADEs frequently occur at

ICUs of hospitals in the United States of America (US), and that approximately two-

third of the actually ADEs were judged preventable63. Pharmacists therefore still

search for methods to support the highly trained ICU physicians in the prevention

of these ADEs. Since several trials that were performed at ICUs of hospitals in

the US showed that participation of a clinical pharmacist on the ICU, especially

those that are trained in critical care, may reduce the frequency and costs of drug-

related problems, these clinical pharmacists are increasingly involved in ICU64-68.

The activities that were performed by these clinical pharmacists are summarized

in table 2. It was estimated that in 2006 clinical pharmacists were involved in

approximately two-third of the ICUs69.


140

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Table 2: Activities performed by clinical pharmacists on the ICU64-68

participation multidisciplinary medicine roundson ward stay for consultationperforming chart reviews on all ICU patientperforming medication / allergy interviews - drug histories and medication verificationproviding drug informationproviding pharmacokinetic monitoring / drug therapy monitoringmaking and documenting recommendationsidentifying, reporting, and assisting in the management and/or prevention of ADEs and medication errorsserving as a liaison between the pharmacy, nursing and medical/surgical servicescounseling patients, family members, and/or caretakersin-service educationnutrition team participation

The setting of a US hospital pharmacy, however, cannot be compared to setting

of a Dutch hospital pharmacy. In the US much more pharmacists are employed

in hospital pharmacies, which makes it easier to allocate time to clinical services.

Moreover, in the US, the order-entry and verification process is still performed by

pharmacists, while in the Netherlands, this task is mainly performed by pharmacy

technicians supervised by pharmacists. Dutch pharmacists therefore traditionally

focus on improving the medication management process (system approach) rather

than on individual patient care. This systems pharmacy approach may consist

of several preventive risk minimization strategies, such as optimization of the

logistic chain, formulary management, participation in guideline development

and implementation and maintenance of a CPOE/CDS70. The risks associated

with the drugs logistic chain, for example, can be mitigated by implementation

of automated dispensing cabinets and bar code verification at moment of drug

dispensing or drug administration71,72. A specific ICU formulary defined by a

multidisciplinary ICU drugs and therapeutic committee may lead to selection of

medications that have less interactions than other medications within the same

therapeutic group. A limited formulary will increase familiarity and knowledge of

all professionals involved and thereby decrease the likelihood of medication errors.

Furthermore, pharmacists increasingly offer services consisting of preparation of

ready-to-use parenteral medication in a decentralized pharmacy satellite located

in and dedicated solely to the care of patients. Pharmacists usually also participate

in guideline development for the ICU. Guidelines that are tailored to the needs

of ICU patients, may also prevent ADEs. The guideline on gastric protection, for

example, is often not applied as the use of proton-pump inhibitors is associated

with an increased risk of pneumonia in patients that are mechanically ventilated.

Because these guidelines may also be necessary to develop a CDS based on rules,

pharmacists may also play a role in improving CDS on medications.

141

4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

In 2010, however, Klopowotska et al. showed that on-ward participation of a

specially trained hospital pharmacist also reduced the number of ADEs occurring

at a Dutch ICU73. The on-ward participation consisted of chart review of the patients’

medication and participation into the multidisciplinary patient review meeting

(MDR). Errors that were prevented mainly consisted of drug or dose omissions,

monitoring errors (TDM and renal function), and improper dosage errors. Presence

of the pharmacist on wards and discussion of the issues and recommendations

with the physicians in the MDR apparently led to better acceptance of the advices.

Better acceptance was also reported by Schepers et al, who found that clinicians

most prefer advises that were performed by the pharmacy74. When a potential risk

is established by a pharmacist, the pharmacist can provide additional information

to the physician and motivate why the risk may be relevant for the specific patient.

Pharmacist participation at the ICU may also lead to a better accessibility, enabling

physicians and nurses to approach the pharmacist for questions they otherwise

would not have called for.

The activities performed by clinical pharmacists in the US, however, have shown

to be highly variable and direct patient care often only including providing

distribution functions in a satellite ICU pharmacy69. Chart review, consisting of

evaluation of new medication orders for its appropriateness for given indication,

duration of therapy, drug dosage and frequency, risk of drug-drug and drug-disease

interactions, pharmacological duplications and drug omissions is not always

performed on daily basis. Despite that elements from the chart review might be

performed standardly, a complete chart review is also often not performed by Dutch

pharmacists. The potential problems found by Klopowotoska et al. such as errors of

omission and monitoring errors, however, can only be found with a complete chart

review. When medication safety has to be improved, pharmacists should therefore

first agree on which service is most relevant and effective69,75,76. When is agreed that

chart review is necessary, and this fundamental service is standardly provided at

ICUs, pharmacies may upgrade their services to a “desirable” and “optimal” level

as described by Dager et al. and attend a well- trained pharmacist to the MDR77.

Because the content of desirable and optimal services may depend on the setting of

the ICU such as the number of beds, the level of the ICU and the presence of a CDS

on medications, pharmacists should discuss with the medical staff which level is

desired for the care of their ICU patients.

In all trials that showed a reduction of ADEs and costs, the attending pharmacist was

experienced or did have a special training on critical care. Because the basic training

programs of clinical pharmacists in the US mainly consists of pharmacotherapy

for a broad range of diseases, special training programs on critical care have been

developed by the American organization of hospital pharmacists78-80. This training


142

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

program consists of training on ICU-patient specific pharmacotherapy and solving

drug therapy problems in the context of the pathophysiological changes during

critical illness77. The training program of Dutch hospital pharmacists differs

from the basic training of the clinical pharmacists in the US which consists of

training on a broad range of skills such as drug preparation, therapeutic drug

monitoring, and clinical activities. To perform chart review and participate at MDR

activities, a special training program might also be necessary for Dutch hospital

pharmacists. The clinical pharmacologist in the Netherlands, however, may already

have the knowledge to participate into the MDR of at a Dutch ICU. In contrast

to many other countries where only medical doctors can become certified as a

clinical pharmacologists, hospital pharmacists may also be certified as a clinical

pharmacologist in the Netherlands. The certification program consists of achieving

knowledge on drug-metabolism, pharmacokinetics, pharmacodynamics and

exploration of mechanism of action and effects also in specific patient populations.

Because this specific patient population may consist of the vulnerable intensive care

patient, the knowledge of a clinical pharmacologist may contribute very well to the

safe use of medications on the intensive care81. Because clinical pharmacologists

are experienced to teach pharmacology to medical students, they may also provide

an educational role in pharmacotherapy on the ICU. The attendance of a clinical

pharmacist/pharmacologist to the ICU, however, does not mean that health care

professionals should not invest in the development of CDS systems on medications.

Although CDS systems on medications lack specificity on the ICU, CDS can still

be used to make the medication review process more efficient82. Probably the

combination of improvements in the ICU medication management process (systems

pharmacy) and the attendance of a pharmacist / clinical pharmacologist that is

specialized in critical care, is the most effective strategy to reduce the frequency of

adverse drug events.

Recommendations for clinical practiceThis thesis leads to the following recommendations:

- pDDI guidelines should include which patients are at risk provide

clear instructions on when and how monitoring should be performed,

especially for medications which may influence serum potassium levels.

- CDS systems should integrate pharmacy data with laboratory data and

other clinical patient data with the aim to increase the specificity of pDDI

alerts.

- CDS systems should be developed that support adherence to monitoring

guidelines. Corollary orders and automated reminders on monitoring are

suggested as options to improve adherence, but further research is needed

to investigate whether these options actually prevent ADEs.

143

4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

- To prevent ADEs, CDS systems should not only enable triggering when

a new medication is prescribed, but also when new information on

monitoring values or other patient specific data such as morbidity

becomes available or medication is discontinued.

- When measurement of laboratory values is relevant to prevent ADEs

which may occur after discharge of hospitalized patients, measurement

of laboratory values should also be performed before discharge. These

laboratory data should accompany pharmacy data which are provided to

the next health care provider.

- Developers should invest in CDS that is tailored to the needs of the ICU.

Options to improve such a CDS are:

o alerting when significant deviations of biomarkers are detected

in the presence of a pDDI

o alerting to another health care provider, such as a clinical

pharmacist who can judge on the clinical relevance

- The clinical setting and expertise of the professional should be taken into

account to increase the specificity of a CDS on DDI alerting.

- A specific G-standard should be developed for the ICU that includes

medications that are frequently used on the ICU, dosages that are accepted

at the ICU and pDDIs that are relevant for the ICU.

ConclusionsThe findings from the studies presented in this thesis show that only few drug-

drug combinations are responsible for the majority of the pDDI alerts and that

the risk of the majority of these alerts may be mitigated by monitoring. It was

also shown, however, that adherence to these guidelines varies largely, depending

on both patient characteristics and type of medication used. Physicians seem to

adhere most to monitoring guidelines when the effect on laboratory markers is

intended. CDS should therefore better support the monitoring stage of drug

therapy. Special attention for monitoring is needed when risk factors change such

as when medication which is involved in a pDDI is discontinued or at moments

that patients are transferred in care.

A limited number of drug-drug pairs were also responsible for the majority of

the alerts at the ICU, but as monitoring of general laboratory markers is already

frequently performed at the ICU, not all pDDIs are relevant. For the majority of the

pDDIs, however, linking laboratory data to prescription data may be an important

tool for effective management of risks associated with these pDDIs. For other

relevant pDDIs, it may still be worthwhile to signal, despite higher risks may be

accepted for medical treatment at the ICU.


144

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

References

1. Bates DW Cullen DJ, Laird N et al. Incidence of adverse drug events and potential adverse drug events.

Implications for prevention. ADE Prevention Study Group. JAMA 1995;274:29-34

2. Leape LL, Bates DW, Cullen DJ, Cooper J, Demonaco HJ, Gallivan T, Hallisey R, Ives J, Laird N, Laffel G,

et al. Systems analysis of adverse drug events. ADE Prevention Study Group. JAMA 1995;274:35-43

3. Gurwitz JH, Field TS, Harrold LR, Rothschild J Debellis K, Seger AC, Cadoret C, Fish LS, Garber L,

Kelleher M, Bates DW. Incidence and preventability of adverse drug events among older persons in the


4. Juurlink DN, Mamdani M, Kopp A, Laupacis A, Redelmeier DA. Drug-Drug Interactions among elderly

patients hospitalized for drug toxicity JAMA 2003;289:1652-1658

5. Pirmohamed M, James S, Meaking S, Adverse drug reactions as cause of admission to hospital:

prospective analysis of 18 820 patients. Drug Saf 2006;29:723-732

6. Becker ML, Kallewaard M, Caspers PWJ, Visser LE, Leufkens HGM, Stricker BH. Hospitalisations and

emergency department visits due to drug–drug interactions: a literature review. Pharmacoepidem

Drug Saf 2007;16:641–651

7. Up-to-Date, Inc.: Lexi-Interact Online. Available from: www.uptodate.com/crlsql/interact/frameset.jsp

(Accessed at July 1th, 2014)

8. Thomson Reuters (Healthcare) Inc.: Micromedex healthcare series. Interactions. Updated periodically.

Available from: www.thomsonhc.com/hcs/Librarian?submit = I+have+read+the+above+notice+-

+Start+MICROMEDEX (Accessed at July, 2014)

9. Pharmaceutical Press: Stockley’s Drug Interactions. Available from: http://www.medicinescomplete.

com/ (Accessed at July 1th, 2014)

10. Z-index. Available from: Z-index .http://www.z-index.nl/ (Accessed at July 1th, 2014)

11. Geerts AFJ, de Koning GHP, de Smet PGAM, van Solinge WW, Egberts ACG. Laboratory Tests in the

Clinical Risk Management of Potential Drug-Drug Interactions. Drug Saf 2009;32:1189-1197

12. Van den Bemt PMLA, Egberts ACG, Leendertse A. Hospital Admissions Related to Medication (HARM).

Een prospectief, multicenter onderzoek naar geneesmiddel gerelateerde ziekenhuisopnames. Utrecht:

Division of Pharmacoepidemiology & Pharmacotherapy, Utrecht Institute for Pharmaceutical

Sciences, 2006

13. Warlé-van Herwaarden MF, Kramers C, Sturkenboom MC, van den Bemt PMLA, De Smet PAGM.

Targeting outpatient drug safety: recommendations of the Dutch HARM-Wrestling task force. Drug

Saf 2012;35:245-259

14. Kingdom of the Netherlands. Regulation of the Ministry of Health, Welfare and Sport of 9

December 2011, amending the regulation on the Dutch Medicines Act (in Dutch). https://zoek.

officielebekendmakingen.nl/stcrt-2011-22913.html (Accessed at March 1th, 2014)

15. Medicatieveiligheid: Elektronisch voorschrijven (EVS). http://www.igz.nl/onderwerpen/curatieve-

gezondheidszorg/medicatieveiligheid/medicatieoverdracht/elektronisch_voorschrijven evs/.)

(Accessed at July 1th, 2014)

16. Richtlijn electronisch voorschrijven: Koninklijke Nederlandsche Maatschappij tot bevordering der

Geneeskunst. 2013. www.knmg.nl/elektronisch-voorschrijven (Accessed at July 1th, 2014)

17. Geerts AFJ, de Koning GHP, van Solinge WW, de Smet PGAM, Egberts ACG. Instructions on laboratory

monitoring in 200 drug labels. Clin Chem Lab Med 2012;50:1351-1358

18. Floor-Schreudering A, De Smet PA, Buurma H, Amini S, Bouvy ML. Clarity and applicability of

drug-drug interaction management guidelines: a systematic appraisal by general practitioners and

community pharmacists in the Netherlands. Drug Saf 2011;34:683-690

19. Raebel MA, Lyons EE, Andrade SE, Chan KA, Chester EA, Davis RL, Ellis JL, Feldstein A, Gunter MJ,

Lafata JE, Long CL, Magid DJ, Selby JV, Simon SR, Platt R. Laboratory Monitoring of Drugs at Initiation

of Therapy in Ambulatory Care. J Gen Intern Med 2005;20:1120-1126

20. Coleman JJ, McDowell SE, Evans SJW, Gill PS, Ferner RE. Oversight: a retrospective study of biochemical

monitoring in patients beginning antihypertensive drug treatment in primary care. Br J Clin Pharm

2010;70:109-117

145

4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

21. Van Blijderveen JC, Straus SM, de Ridder MA, Stricker BH, Sturkenboom MC, Verhamme KM. Adherence

to renal function monitoring guidelines in patients starting antihypertensive therapy with diuretics

and RAAS inhibitors: A Retrospective cohort study. Drug Saf 2014;37:369–377

22. Forster AJ, Murff HJ, Peterson JF, Gandhi TK, Bates DW. The incidence and severity of adverse events

affecting patients after discharge from the hospital. Ann Intern Med 2003;138:161-167

23. Spriet I, Meersseman W, de Hoon J et al. Mini series II-clinical aspects: clinically relevant CYPP450-

mediated drug interactions in the ICU. Intensive Care Med 2009;35:603-612

24. Krishnan V, Murray P. Pharmacologic issues in the critically ill. Clin Chest Med 2003;24:671–688

25. Juurlink DN, Mamdani MM, Lee DS, Kopp A, Austin PC, Laupacis A, Redelmeier DA. Rates of

hyperkalemia after publication of the Randomized Aldactone Evaluation Study. N Engl J Med

2004;351:543–551

26. Buurma H, Schalekamp T, Egberts AC, De Smet PA. Compliance with national guidelines for the

management of drug-drug interactions in Dutch community pharmacies. Ann Pharmacother

2007;41:2024-2031

27. Cabana MD, Rand CS, Powe NR, Wu AW, MH Wilson, Abboud PAC, Rubin HR. Why don’t physicians

follow clinical practice guidelines?: A framework for improvement. JAMA 1999;282:1458-1465

28. Lugtenberg M, Zegers-van Schaick JM, Westert GP Burgers JS. Why don’t physicians adhere to

guideline recommendations in practice? An analysis of barriers among Dutch general practitioners.

Implementation Science 2009;4:54

29. Bates D, Cohen M, Leape L, Overhage JM, Shabot MM, Sheridan T. Reducing the frequency of errors in

medicine using information technology. J Am Med Inform Assoc 2001;8:299-308

30. Osheroff JA, Pifer EA, Teich JM, Sittig DF, Jenders RA. Improving outcomes with clinical decision

support: an implementer’s guide. Health Information Management and Systems Society (2005). http://

www.providersedge.com/ehdocs/ehr_articles/Clinical_Decision_Support_Implementers_Workbook.

pdf (Accessed at July 1th, 2014).

31. Berner ES, Maisiak RS, Cobbs CG, Taunton OD. Effects of a decision support system on physicians’

diagnostic performance. J Am Med Inform Assoc 1999;6:420-427

32. Sequist TD, Gandhi TJ, Karson AS, Fiskio JM, Bugbee D, Sperling M, Cook EF, Orav EJ, Fairchild DG,

Bates DW. A randomized trial of electronic clinical reminders to improve quality of care for diabetes

and coronary artery disease. J Am Med Inform Assoc 2005;12:431–437

33. Kaushal R, Shojania KG, Bates DW. Effects of computerized physician order entry and clinical decision

support systems on medication safety: a systematic review. Arch Intern Med 2003;163:1409-1416

34. Kuperman GJ, Bobb A, Payne TH, MD, Avery AJ, Gandhi TK, Burns G, Classen DC, Bates DW. Medication-

related Clinical Decision Support in Computerized Provider Order Entry Systems: A Review. J Am Med

Inform Assoc 2007;14:29-40

35. Johnston ME, Langton KB, Haynes RB, Mathieu A. Effects of Computer-based Clinical Decision Support

Systems on Clinician Performance and Patient Outcome: A Critical Appraisal of Research. Ann Intern

Med 1994;120:135-142

36. Garg AX, Adhikari NKJ, McDonald H, Rosas-Arellano MP, Devereaux PJ, Beyene J, Sam J,Haynes RB.

Effects of computerized clinical decision support systems on practitioner performance and patient

outcomes: a systematic review. JAMA 2005;293:1223-1238

37. Bright TJ, Wong A, Dhurjati R, Bristow E. Effect of clinical decision-support systems; a systematic

review. Ann Intern Med 2012;157:29-43

38. Glassman PA, Simon B, Belperio P, Lanto A. Improving recognition of drug interactions: benefits and

barriers to using automated drug alerts. Med Care 2002;40:1161–1171

39. Weingart SN, Toth M, Sands DZ, Aronson MD, Davis RB, Phillips RS. Physicians’ decisions to override

computerized drug alerts in primary care. Arch Intern Med 2003;163:2625-2631

40. Sijs H, Mulder A, van Gelder T, Aarts J, Berg M, Vulto A. Drug safety alert generation and overriding in


41. Sijs H, Aarts J, van Gelder T, Berg M, Vulto A. Turning off frequently overridden drug alerts: limited

opportunities for doing it safely. J Am Med Inform Assoc 2008;15:439–448

42. Paterno MD, Maviglia SM, Gorman PN, Seger DL, Yoshida E, Seger AC, Bates DW, Gandhi TK. Tiering

drug–drug interaction alerts by severity increases compliance rates. J Am Med Inform Assoc

2009;16:40–46


146

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

43. Shah NR, Seger AC, Seger DL, Fiskio JM, Kuperman GJ, Blumenfeld B, Recklet EG, Bates DW, Gandhi TK.

Improving override rates for computerized prescribing alerts in ambulatory care. AMIA Annu Symp

Proc 2005; 2005:1110

44. Duke JD, Li X, Dexter P. Adherence to drug–drug interaction alerts in high risk patients: a trial of

context-enhanced alerting. J Am Med Inform Assoc 2013;20:494–498

45. Seidling HM, Klein U, Schaierd M, Czocka D, Theilea D, Pruszydloa MG, Kaltschmidta J, Mikusa G,

Haefelia WE. What, if all alerts were specific – Estimating the potentialimpact on drug interaction

alert burden. Int J Med Inform 2014;83:285–291

46. Bates DW, Kuperman GJ, Wang S, Gandhi T, Kittler A, Volk L, Spurr C, Khorasani R, Tanasijevic M,

Middleton B . Ten commandments for effective clinical decision support: making the practice of

evidence-based medicine a reality. J Am Med Inform Assoc 2003;10:523–530

47. Goldman RE, Soran CS, Hayward GL, Simon SR. Doctors’ perceptions of laboratory monitoring in

office practice. J Eval Clin Pract 2010;16:1136-1141

48. Feldstein AC, Smith DH, Perrin N, Yang X, Rix M, Raebel MA, Magid DJ, Simon SR, Soumerai SB.

Improved therapeutic monitoring with several interventions: a randomized trial. Arch Intern Med

2006;166:1848-1854

49. Overhage JM, Tierney WM, Zhou XH, McDonald CJ. A randomized trial of “corollary orders” to prevent

errors of omission. J Am Med Inform Assoc 1997;4:364-375

50. Berner ES. Clinical Decision Support systems –theory and practice. 2th edition Springer Science and

Business Media; New York 2007

51. Schnipper JL, Linder JA, Palchuk MB, Einbinder JS, Li Q, Postilnik A, Middleton B. “Smart Forms” in

an Electronic Medical Record: Documentation-based Clinical Decision Support to Improve Disease

Management. J Am Med Inform Assoc 2008;15:513–523

52. Kumar KA, Singh Y, Sanyal S. Hybrid approach using case-based reasoning and rule-based reasoning

for domain independent clinical decision support in ICU. Expert Syst Appl 2009;36:65–71

53. Institute of Safe Medication Practices. ISMP’s list of high alert medications 2012. http://www.ismp.org/

tools/highalertmedications.pdf (Accessed at July 1th, 2014)

54. Tegeder I, Levy M, Muth-Selbach U, Oelkers R, Neumann F, Dormann H, Azaz-Livshits T, Criegee-

Rieck M, Schneider T, Hahn EG, Brune K, Geisslinger G. Retrospective analysis of the frequency and

recognition of adverse drug reactions by means of automatically recorded laboratory signals. Br J Clin

Pharmacol 1999;47:557-564

55. Melik EJM, Souverein PC, Breeijen JH, Tusveld CE, Egberts TCG, Wilting I. Age as a ceterminant of

instability of serum lithium concentrations. Ther Drug Monit 2013;35:643–648

56. Nillson M, Sollenborn M. Advancements and trends in medical case-based reasoning: An overview of

systems and system development. American Association for Artificial Intelligence 2004

57. Musen MA, Shahar Y. Clinical Decision-Support Systems . In Shortliffe EH, Cimino JJ. Biomedical

Informatics. Springer-Verlag; London 2014

58. Helmons PJ, Grouls RJ, Roos AN, Bindels AJ, Wessels-Basten SJ, Ackerman EW, Korsten EH. Using a

clinical decision support system to determine the quality of antimicrobial dosing in intensive care

patients with renal insufficiency. Qual Saf Health Care 2010;19:22-26

59. Spriet I, Goyens J, Meersseman W, Wilmer A, Willems L, Van Paesschen W. Interaction Between

Valproate and Meropenem: A Retrospective Study Ann Pharmacother 2007;41:1130-1136

60. Shamliyan TA, Duval S, Du J, Kane RL: Just what the doctor ordered. Review of the evidence of the

impact of computerized physician order entry system on medication errors. Health Serv Res 2008;

43:32-53

61. Shulman R, Singer M, Goldstone J, Bellingan G. Medication errors: a prospective cohort study of

hand-written and computerised physician order entry in the intensive care unit. Crit Care 2005;

9:R516-R521

62. Zaal RJ, Jansen MMPM, Duisenberg-van Essenberg M, Tijssen CC, Roukema JA, van den Bemt PMLA.

Identification of drug-related problems by a clinical pharmacist in addition to computerized alerts.

Int J Clin Pharm 2013;35:753–762

63. Kopp BJ, Erstad BL, Allen ME, Theodorou AA, Priestley G. Medication errors and adverse drug events

in an intensive care unit: Direct observation approach for detection. Crit Care Med 2006;34:415–425

147

4

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

64. Leape LL, Cullen DJ, Clapp MD, et al. Pharmacist participation on physician rounds and adverse drug

events in the intensive care unit. JAMA 1999;282:267–70

65. Papadopoulos, Rebuck JA, Lober C, Pass SE, Seidl EC, Shah RA, Sherman DS. The Critical Care

Pharmacist: An Essential Intensive Care Practitioner. Pharmacother 2002;22:1484–1488

66. Kane SL, Weber RJ, Dasta JF. The impact of critical care pharmacists on enhancing patient outcomes.

Intensive Care Med 2003;29:691–698

67. Kopp BJ, Mrsan M, Erstad BL, Duby JJ. Cost implications of and potential adverse events prevented by

interventions of a critical care pharmacist. Am J Health-Syst Pharm 2007;64:2483-2487

68. Lee AJ, Chiao TB, Lam JT, Khan S, Boro MS. Improving medication safety in the ICU: the pharmacist’s

role. Hosp Pharm 2007;42:337–344

69. McLaren R, Devlin JW, Martin SJ, Dasta JF, Rudis MI, Bond CA. Critical Care Pharmacy Services in

United States Hospitals Ann Pharmacother 2006;40:612-618

70. Moyen E, Camiré E, Stelfox HT. Clinical review: Medication errors in critical care. Critical Care

2008;12:208

71. Helmons PJ, Wargel LN, Daniels CE. Impact of bar-coded medication administration (BCMA) on

medication administration errors and accuracy in multiple patient care areas: a before-and-after

study. Am J Health Syst Pharm 2009;66:1202-1210

72. Hassink JJM, Jansen MMPM, Helmons PJ. Effects of bar-code assisted medication administration

(BCMA) on frequency, type and severity of medication administration errors: a review of the literature

Eur J Hosp Pharm Sci Pract 2012;19:489–494

73. Klopotowska JE, Kuiper R, van Kan HJ, de Pont AC, Dijkgraaf MG, Lie-A-Huen L, Vroom MB, Smorenburg

SM. On-ward participation of a hospital pharmacist in a Dutch intensive care unit reduces prescribing

errors and related patient harm: an intervention study. Critical Care 2010; 14:R174

74. Scheepers-Hoeks AMJW, Rene JE Grouls RJE, Neef C, Ackerman EW, Korsten EHM. Physicians´ response

to clinical decision support on an intensive care unit - comparison of four different alerting methods.

Artif Intell Med 2013;59:33-38

75. Rudis MI, Brandl KM, for the Society of Critical Care Medicine and American College of Clinical

Pharmacy Task Force on Critical Care Pharmacy Services. Position paper on critical care pharmacy

services. Crit Care Med 2000;28:3746–3750

76. Hassan E. Critical care pharmacotherapy: issues and approaches. Curr Opin Crit Care 2000;6:299–303

77. Dager W, Bolesta S, Brophy G, Dell K, Gerlach A, Kristeller J, Micek S, Hess M, Bentley M, Juang P,

MacLaren R, Kane-Gill S, Haase K, Mann H, Miller C, Pepe D, Southwood R, Weant K, Rebuck J, Wood

C. An Opinion Paper Outlining Recommendations for Training, Credentialing, and Documenting and

Justifying Critical Care Pharmacy Services. Pharmacotherapy 2011;31:135e–175e

78. ASHP Critical care traineeship. http://www.ashpfoundation.org/MainMenuCategories/Education/

Traineeships/CriticalCareTraineeship (Accessed at July 1th, 2014).

79. American Society of Health-System Pharmacists: ASHP supplemental standard and learning objectives

for residency training in critical-care pharmacy practice. American Society of Hospital Pharmacists.

Am J Hosp Pharm 1990;47:609–612

80. Expert group department of health, NHS and UKCPA. Adult critical care specialist pharmacy practice:

new says of working. http://www.codeg.org/fileadmin/codeg/pdf/Critical_care_specialist_guidance.

pdf (Accessed at July 1th, 2014)

81. Moore N. The role of the clinical pharmacologist in the management of adverse drug reactions. Drug

Saf 2001;24:1-7

82. Helmons PJ, Kosterink JGW. Drug-drug interaction checking assisted by clinical decision support: a

return on investment analysis. Thesis Rijksuniversiteit Groningen 2014

Samenvatting

List of co-authors

List of publications

Dankwoord

Curriculum vitae

Samenvatting

150

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

151

S

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Samenvatting

Farmaceutische zorg heeft tot doel de balans tussen risico en baten van behandeling

met geneesmiddelen te optimaliseren. Registratie-autoriteiten beoordelen deze

risico-baten balans voor de populatie waarvoor het geneesmiddel is onderzocht.

Als deze positief uitvalt worden de geneesmiddelen geregistreerd. In de klinische

praktijk bestaan echter vaak grote verschillen tussen patiënten. Dit betekent dat

de risico-baten balans voor de individuele patiënt kan verschillen van die van het

groepsgemiddelde en dat iedere patiënt die een geneesmiddel gebruikt kans heeft

op onverwachte bijwerkingen. Een groot deel van de bijwerkingen kan worden

vermeden door het toepassen van risicomitigerende strategieën. De samenvatting

van productkenmerken van geregistreerde geneesmiddelen en (behandel)richtlijnen

adviseren vaak welke risicomitigerende strategieën kunnen worden toegepast. Als

deze richtlijnen niet goed worden nageleefd, kunnen vermijdbare bijwerkingen

ontstaan. Deze bijwerkingen worden gezien als medicatiefout. Medicatiefouten

ontstaan vaak in het complexe medicatieproces dat bestaat uit het voorschrijven van

geneesmiddelen, het interpreteren en verifiëren van de voorschriften, het afleveren

van geneesmiddelen, de toediening door de zorgverlener of patiënt zelf, het

monitoren van het effect en de evaluatie van de farmacotherapie. Medicatiefouten

blijken vooral te worden gemaakt in het stadium van voorschrijven en stadium

van monitoren. Omdat de farmaceutische zorg mede tot doel heeft deze fouten te

voorkomen, is het van belang dat aandacht wordt besteed aan de medicatiefouten

die optreden in deze twee stadia.

Geneesmiddeleninteracties blijken in belangrijke mate bij te dragen aan de

negatieve effecten van behandeling met geneesmiddelen. Er bestaan daarom veel

richtlijnen die de zorgprofessional ondersteunen om de mogelijke gevolgen van deze

geneesmiddeleninteracties te voorkomen. Voorschrijfsystemen met een geïntegreerd

beslissingsondersteunend systeem voor geneesmiddeleninteracties kunnen op basis

van deze richtlijnen advies geven over de risicomitigerende strategieën die kunnen

worden toegepast op de individuele patiënt. Dergelijke systemen signaleren een

mogelijke geneesmiddeleninteractie vaak op het moment van voorschrijven en

geven dan bijvoorbeeld een voorstel om te kiezen voor een alternatief geneesmiddel

of om tijdens gebruik van deze geneesmiddelen het effect van de interactie goed

te monitoren. Dit monitoren kan bestaan uit klinische monitoring (observatie

van ongewenste effecten zoals huidreacties), het meten van fysische parameters

(bijvoorbeeld het opnemen van de bloeddruk of van een elektrocardiogram) of het

meten van laboratorium markers (geneesmiddelenconcentraties, serum creatinine

waarden, serum elektrolyt waarden en dergelijke). Uit analyse van de Nederlandse

Samenvatting

152

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

richtlijn voor geneesmiddeleninteracties is gebleken dat maar liefst 33% van alle

risicomitigerende strategieën adviseert om een of meer laboratoriummarkers

te meten. Ook is gebleken dat hierbij slechts een beperkte set van laboratorium

markers betrokken is. In 2012 heeft de Nederlandse wet zorgverleners daarom

verplicht gesteld om informatie over deze beperkte set van markers, bestaande

uit geneesmiddelenconcentraties in bloed, international normalized ratio, serum

creatinine, kalium, natrium en farmacogenetische markers uit te wisselen.

Onderzoek naar risicomitigerende strategieën zou zich daarom op deze markers

moeten concentreren.

Het voorschrijven van medicatie wordt in de meeste ziekenhuizen tegenwoordig

informatietechnologisch ondersteund via een geautomatiseerde voorschrijfsysteem

(computerized order entry system, CPOE). In het voorschrijfsysteem is vaak

een basale vorm van klinische beslissingsondersteuning geïntegreerd waarbij

zorgverleners tijdens het voorschrijven gewaarschuwd worden voor het risico

op mogelijke geneesmiddeleninteracties. Daarbij wordt elk nieuw voorschrift

gecontroleerd op een mogelijke interactie met de actuele medicatie die de patiënt

voorgeschreven heeft gekregen en wordt de voorschrijver gewaarschuwd voor het

risico van de mogelijke geneesmiddeleninteractie. Het moment van voorschrijven

is immers een belangrijk moment om na te denken over de geschiktheid van

het geneesmiddel voor de individuele patiënt en hierbij wordt ook het risico op

bijwerkingen door een mogelijke interactie tussen geneesmiddelen in overweging

genomen. Op het moment van voorschrijven kan de arts echter alleen besluiten

om het gebruik van geneesmiddelen te stoppen dan wel voort te zetten, terwijl

de combinatie van twee interacterende geneesmiddelen niet schadelijk hoeft te

zijn als klinische parameters en biomakers van de patiënt adequaat gemonitord

worden tijdens het gebruik van de geneesmiddelen. Dit betekent dat de arts op het

moment van voorschrijven moet beslissen over wat te monitoren, wanneer en hoe

vaak. Het is gebleken dat richtlijnen voor geneesmiddeleninteracties hier vaak geen

duidelijkheid over geven.

Ook is gebleken dat het risico van een mogelijke geneesmiddeleninteractie vaak

niet gemonitord wordt bij patiënten in de ambulante zorg. Het is de verwachting

dat het risico van een mogelijke geneesmiddeleninteractie beter gemonitord

wordt bij patiënten die in het ziekenhuis opgenomen zijn, enerzijds vanwege de

beschikbaarheid van laboratorium- en andere monitoringsfaciliteiten en anderzijds

omdat de ziekte van de opgenomen patiënt dit noodzakelijk maakt. De duur van

de gemiddelde ziekenhuisopname is echter kort en bijwerkingen ten gevolge van

mogelijke geneesmiddeleninteracties kunnen daarom ook pas na ontslag ontstaan.

153

S

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Dat betekent dat de monitoring zou moeten doorgaan nadat de patiënt ontslagen

is en dat relevante informatie over zowel het geneesmiddelengebruik als de

monitoring tijdens het ontslag moet worden overgedragen aan de eerstelijns zorg.

Naast de verschillen in setting (ziekenhuis of thuis), kan de risicomitigerende

strategie ook afhangen van de karakteristieken van de individuele patiënt.

Eigenschappen van geriatrische patiënten verschillen van die van kinderen, en

beide verschillen weer van die van Intensive Care (IC) patiënten. IC patiënten zijn in

het algemeen ernstiger ziek waardoor hun capaciteit om fysiologische verstoringen

te kunnen compenseren verstoord kan zijn. Ondanks de intensieve monitoring

van deze patiënten, is de behandeling complex en risicomitigerende strategieën

verdienen daarom extra aandacht.

Hoofdstuk 2 en 3 beschrijven de frequentie en de klinische relevantie van

monitoring van geneesmiddelentherapie met laboratorium markers van in het

ziekenhuis opgenomen patiënten. Speciale aandacht wordt besteed aan mogelijke

geneesmiddeleninteracties, kalium en patiënten die opgenomen zijn op de

Intensive Care.

Het is bekend dat een computersysteem dat veel niet-specifieke waarschuwingen

genereert het veiligheidsdoel van deze waarschuwingen ondermijnt.

Computersystemen met een basale vorm van interactiebewaking houden geen

rekening met individuele patiëntkarakteristieken. Ze signaleren mogelijke

geneesmiddeleninteracties wel nauwkeurig (hoge sensitiviteit) maar de

waarschuwingen zijn vaak niet relevant voor de individuele patiënt (lage

specificiteit). Hoofdstuk 2 laat zien in welke omvang dit probleem heeft binnen

de ziekenhuispopulatie. Potentiële geneesmiddeleninteracties bleken veel voor te

komen bij ziekenhuispatiënten. Van alle patiënten die in het ziekenhuis opgenomen

waren, had 25% minimaal 1 potentiële geneesmiddeleninteractie (hoofdstuk 2.1).

Dit percentage was ruim twee maal zo hoog (54%) bij patiënten die opgenomen

waren op de IC (hoofdstuk 2.2). Het verschil met de algemene ziekenhuispopulatie

werd vooral veroorzaakt door het grotere aantal geneesmiddelen dat de IC

patiënten gebruikten. Hoofdstuk 2.1 en 2.2 laten zien dat slechts een beperkt aantal

geneesmiddelencombinaties verantwoordelijk was voor het grootste deel van de

potentiële geneesmiddeleninteracties. Dit betrof zowel algemene verpleegafdelingen

als IC. Op de algemene verpleegafdelingen waren de 10 meest voorkomende

interacties verantwoordelijk voor 53% van de potentiële geneesmiddeleninteracties

en op de IC voor 79% van de potentiële geneesmiddeleninteracties. In hoofdstuk 2.2

is nader aandacht besteed aan de geneesmiddeleninteracties die optraden bij IC

Samenvatting

154

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

patiënten omdat veel geneesmiddelen die op de IC worden gebruikt een hoog

intrinsiek risico hebben en IC patiënten, door de ernst van hun ziekte, kwetsbaarder

zijn voor de negatieve effecten van potentiële geneesmiddeleninteracties. De

potentiële geneesmiddeleninteracties op de IC bleken inderdaad vooral betrekking

hebben op de specifieke hoog-risico medicamenten die op de IC gebruikt worden.

De top-10 van meest voorkomende potentiële geneesmiddeleninteracties op

de IC verschilt daarom van die op algemene verpleegafdelingen. De meeste

geneesmiddeleninteracties gaven een verhoogde kans op bijwerkingen. Een kans

op verminderde effectiviteit kwam minder vaak voor. De nationale richtlijn voor

het managen van potentiële geneesmiddeleninteracties adviseerde meestal om te

monitoren. Deze monitoring bestond uit het meten van laboratorium waarden,

klinische monitoring op toxiciteit of verminderde effectiviteit en het monitoren

van fysische parameters zoals het opnemen van een elektrocardiogram en het

meten van de bloeddruk. Monitoring van laboratorium markers werd het meest

frequent geadviseerd, namelijk bij 48,6% van de patiënten met een interactie op een

algemene verpleegafdeling en bij 51,6% van de patiënten met een interactie op de IC.

Deze resultaten laten zien dat een geavanceerd klinisch beslissingsondersteunend

systeem waarbij laboratorium waarden gekoppeld worden aan voorschrijfdata

een belangrijk hulpmiddel kan zijn om het geneesmiddeleninteractiesysteem

specifieker te maken.

Hoofdstuk 3.1 laat zien dat bij 50% van de patiënten tijdens opname een meting

gedaan is voor serumkalium, -natrium en/of –creatinine spiegels. Deze drie

laboratoriummarkers zijn vaak betrokken bij de strategieën om bijwerkingen

van geneesmiddelentherapie te voorkomen. Het is echter niet bekend wat de

determinanten zijn om deze laboratoriummarkers te meten. Als één van de drie

laboratoriummarkers gemeten werd, bleek 90% van de patiënten ook een meting

te hebben voor de andere twee laboratoriummarkers. Dit suggereert dat de drie

laboratoriummarkers vaak tegelijkertijd gemeten worden. In het algemeen waren

patiënt- en opname gerelateerde factoren zoals leeftijd en behandelspecialisme

belangrijkere voorspellers voor het meten van deze laboratoriummarkers dan

het gebruik van specifieke geneesmiddelen. Het gebruik van medicatie waarvoor

controle van een van deze laboratoriummarkers geadviseerd wordt, had nauwelijks

invloed op het aantal patiënten waarbij een laboratoriumparameter gemeten werd.

Bij patiënten die een geneesmiddel gebruikten dat door de nier geklaard wordt

en waarvoor de nationale standaard een doseringsaanpassing adviseert als de

nierfunctie verminderd is (VNF geneesmiddelen), is niet vaker een serumcreatinine

meting gedaan dan bij patiënten die geen VNF geneesmiddel gebruikten. Het

voorschrijfsysteem signaleerde echter niet als een VNF geneesmiddel werd

155

S

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

voorgeschreven en adviseerde dus ook niet om een serum creatinine spiegel

te meten. Signalering zou tot een hoog aantal meldingen hebben geleid, 77%

van de patiënten gebruikte tijdens opname in het ziekenhuis namelijk een

VNF geneesmiddel. Medicatie waarvoor wel vaker gemeten werd, waren de

kaliumsupplementen. Wanneer patiënten kaliumsupplementen gebruikten, nam

het percentage patiënten met een kaliummeting toe van 41 naar 99%.

Om vast te stellen of de uitslag van de metingen tijdens ziekenhuisopname ook

geschikt zijn voor de verificatie van medicatie bij ontslag, is het tijdstip van de

laatste meting van deze laboratorium markers bepaald. Het percentage patiënten

waarbij binnen 48 uur vóór ontslag nog een meting is gedaan bleek minder dan 25%

te zijn. Dit suggereert dat verbetering mogelijk is in de aanvraag en communicatie

van relevante testuitslagen aan de zorgverlener voorafgaand aan een overgang van

zorglijn.

Kalium is betrokken bij één van de meest frequent voorkomende interacties

tussen geneesmiddelen, namelijk die tussen twee kaliumverhogende

geneesmiddelen. Omdat deze interactie mogelijk tot ernstige gevolgen kan leiden,

is in hoofdstuk 3.2 de monitoring van deze elektrolyt nader bestudeerd. In

aanwezigheid van een voorschrijfsysteem met signaleringsfunctie voor potentiële

geneesmiddeleninteracties, bleken kaliumspiegels iets vaker te worden gemonitord

als patiënten een interactie hadden tussen twee of meer kaliumverhogende

geneesmiddelen ten opzichte van patiënten die maar één kaliumverhogend

geneesmiddel gebruikten (67% versus 58% van de patiënten). Of dit verschil kan

worden toegeschreven aan de signaleringsfunctie van het voorschrijfsysteem was

niet vast te stellen. De basiskenmerken van patiënten in de interactie-groep verschilde

namelijk van die van de monotherapie-groep. Ondanks dat de arts tijdens het

voorschrijven een melding kreeg met het advies om kaliumspiegels te monitoren,

bleek bij 33% van de patiënten met een interactie tussen kaliumverhogende

geneesmiddelen geen kalium meting gedaan en 10% van de patiënten ontwikkelde

een hyperkaliëmie.

Bij het starten van geneesmiddelen wordt vaak geadviseerd om te monitoren

maar er is weinig aandacht voor de rol van monitoring als geneesmiddelen

worden gestopt, zelfs als deze geneesmiddelen onderdeel zijn van een potentiële

geneesmiddeleninteractie. Patiënten met hartfalen krijgen vaak een combinatie van

remmers van het renine-antiotensine-aldosteron systeem (RAS-i), spironolactone en

diuretica. Omdat deze geneesmiddelen een tegengesteld effect hebben op de serum

kaliumspiegels, kan het stoppen van een van deze geneesmiddelen gevolgen hebben

Samenvatting

156

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

op de kalium spiegels. Als de kaliumspiegels hierdoor buiten de normaalwaarden

vallen, kan dit bij deze kwetsbare patiëntgroep leiden tot een kalium gerelateerde

hartritmestoornis. In hoofdstuk 3.3 wordt beschreven wat er gebeurt als één van

deze middelen wordt gestopt. Bij 59% van patiënten die zowel een kalium-verhogend

als een kalium-verlagend geneesmiddel gebruikten en stopten met het gebruik

van de kalium-verlager, nam de kaliumspiegel toe. Bij 3,2% van deze patiënten

ontwikkelde zich een hyperkaliëmie (serum kaliumspiegels > 5,5 mmol/L). Als

het kalium-verhogende middel werd gestopt, nam de kaliumspiegel bij 70% van

de patiënten af en ontwikkelde 17% een hypokaliëmie (serum kaliumspiegels

< 3,5 mmol/L). Deze resultaten laten zien dat monitoring belangrijk is bij zowel

starten als stoppen van therapie met geneesmiddelen.

Kaliumspiegels kunnen ook worden beïnvloed door het gebruik van insuline.

Insuline wordt toegepast om glucosespiegels te verlagen en is een belangrijk

onderdeel van de strikte glucose controle (SGC) protocollen op de Intensive Care.

De invloed van insuline op glucosespiegels is ook uitgebreid in de literatuur

beschreven. In welke mate de SGC protocollen invloed hebben op de kaliumspiegels

en mogelijk kunnen leiden tot een kalium gerelateerde bijwerking, is nauwelijks

onderzocht. SGC protocollen zijn daarom niet duidelijk over wanneer en hoe vaak

kaliumspiegels moeten worden gecontroleerd. De resultaten van hoofdstuk 3.4 laten

zien dat SGC niet geassocieerd was met een toegenomen risico op hypokaliëmie.

Dit ondanks de bevinding dat zowel hoge en lage gemiddelde glucose en kalium

waarden en hoge variabiliteit van glucose en kaliumwaarden geassocieerd waren

met een hoge mortaliteit. Het is daarom wel van belang zowel hoge als lage glucose

en kaliumspiegels evenals een hoge variabiliteit van glucose en kaliumspiegels te

voorkomen.

Alle onderzoeken die in dit proefschrift beschreven zijn, tonen aan dat het belangrijk

is om laboratoriummarkers te meten als onderdeel van de risicomitigerende

strategie van geneesmiddelengebruik door in het ziekenhuis opgenomen patiënten.

De afzonderlijke onderzoeken concentreren zich op de rol van monitoring

bij patiënten met een geneesmiddeleninteractie en kwetsbare patiënten die

opgenomen zijn op de Intensive Care. In het laatste hoofdstuk van dit proefschrift,

hoofdstuk 4, worden de resultaten van deze onderzoeken in een breder perspectief

geplaatst aan de hand van drie thema’s. In het eerste thema wordt besproken welke

rol een geavanceerd klinisch beslissingsondersteunend systeem (aCDS) kan spelen in

het verhogen van de specificiteit van een klinisch beslissingsondersteunend systeem

voor geneesmiddeleninteracties. Een basaal klinisch beslissingsondersteunend

systeem (CDS) op basis van potentiële geneesmiddeleninteracties leidt immers tot

157

S

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

een hoog aantal niet-specifieke waarschuwingen (waarschuwingen werden gegeven

bij 8-10% van de voorschriften) terwijl slechts een beperkt aantal combinaties

van geneesmiddelen verantwoordelijk was voor een groot aantal meldingen. Het

koppelen van patiënt-specifieke monitoringsgegevens aan de algemene gegevens

over geneesmiddelen speelt hierin een belangrijke rol. Ook wordt aandacht

besteed aan de wijze waarop een geavanceerd CDS systeem de naleving van

monitoringsrichtlijnen zou kunnen verbeteren. De risicomitigerende strategie

die in de voorgaande hoofdstukken het meest geadviseerd werd bestond immers

uit monitoring, maar de monitoringsrichtlijnen werden niet altijd nageleefd.

Op de intensive care worden patiënten echter al uitgebreid gemonitord. In het

tweede thema worden daarom de mogelijkheden besproken van een klinisch

beslissingsondersteunend systeem ten behoeve van interactiebewaking binnen de

setting van de intensive care. Tot slot wordt in dit laatste hoofdstuk ingegaan op de

rol van de (ziekenhuis)apotheker en/of klinisch farmacoloog indien gebruik wordt

gemaakt van een geavanceerd klinisch beslissingsondersteunend systeem op de IC.

Naar aanleiding van de onderzoeken in het proefschrift worden de volgende

aanbevelingen gedaan:

− Neem duidelijke instructies op in richtlijnen over wat te monitoren, bij wie

en wanneer, vooral bij geneesmiddelen die de kaliumspiegel beïnvloeden;

− Integreer zowel voorschrijfdata als laboratorium data en andere klinische

patiënt gegevens in een klinisch beslissingsondersteunend systeem om de

specificiteit van interactiesignalen te verhogen;

− Ontwikkel klinische beslissingsondersteunende systemen verder zodat ze

ook de naleving van monitoringsrichtlijnen ondersteunen;

− Ontwikkel klinische beslissingsondersteunende systemen verder zodat

zij niet alleen waarschuwen voor risico’s als geneesmiddelen worden

voorgeschreven, maar ook als gebruik van geneesmiddelen stopt.

− Neem in de richtlijnen op dat laboratoriummarkers kort voor ontslag

worden gemeten als deze betrokken kunnen zijn bij het ontstaan van

bijwerkingen na ontslag;

− Ontwikkel klinische beslissingsondersteunende systemen die specifiek

gericht zijn op de behoefte op de IC. Opties hiervoor zijn om het systeem

Samenvatting

158

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

te laten signaleren als biomarkers afwijken in aanwezigheid van een

geneesmiddeleninteractie en/of de alert door een andere zorgprofessional

te laten beoordelen op klinische relevantie;

− Ontwikkel klinische beslissingsondersteunende systemen die rekening

houden met de klinische setting en expertise van de professional zodat de

specificiteit van geneesmiddeleninteractiebewaking verbetert

− Ontwikkel een specifieke interactiestandaard voor de IC

Concluderend is uit het onderzoek van dit proefschrift gebleken dat slechts

een beperkt aantal geneesmiddelencombinaties verantwoordelijk is voor het

grootste deel van de potentiële geneesmiddeleninteracties en dat het risico van

het grootste deel van deze potentiële geneesmiddeleninteracties kan worden

gemitigeerd door de patiënt goed te monitoren, met name door het meten van

laboratoriummarkers. Ook is gebleken dat de mate waarin monitoringsrichtlijnen

worden nageleefd variëren met de eigenschappen van de patiënt, de setting en

het soort medicatie dat wordt gebruikt. Op de Intensive Care afdeling worden

patiënten uitgebreid gemonitord. Patiëntveiligheid kan worden vergroot door

klinische beslissingsondersteunende systemen te ontwikkelen die beter gebruik

kunnen maken van laboratoriummarkers en die het naleven van richtlijnen om

laboratoriummarkers te meten kunnen verbeteren.

159

L

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

List of co-authors

Affiliations at the time the research was conducted

Presented in alphabetical order

Dr. M. (Maarten) J. ten Berg

Department of Clinical Chemistry and Haematology, University Medical Centre

Utrecht, Utrecht, the Netherlands

Dr. O. (Olaf) L. Cremer

Department of Intensive Care, University Medical Centre Utrecht, Utrecht, the

Netherlands

Prof. dr. A. (Toine) C.G. Egberts

Department of Clinical Pharmacy, University Medical Center Utrecht, the

Netherlands & Department of Pharmacoepidemiology and Clinical Pharmacology,

Utrecht Institute for Pharmaceutical Sciences, Utrecht University, The Netherlands

Drs. L. (Lieke) L.M. van Harssel

Department of Hospital Pharmacy, Diakonessenhuis, Utrecht, the Netherlands

Dr. G. (Gerard) W.K. Hugenholtz


Drs. E. (Emile) M. Kuck


Dr. D. (Dylan) W. de Lange

Department of Intensive Care, University Medical Centre Utrecht, Utrecht, the

Netherlands

Prof. dr. W. (Wouter) W. van Solinge

Department of Clinical Chemistry and Haematology, University Medical Centre

Utrecht, Utrecht, the Netherlands & Department of Pharmacoepidemiology and

Clinical Pharmacology, Utrecht Institute for Pharmaceutical Sciences, Utrecht

University, The Netherlands

List of co-authors

160

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Dr. J. (Jeannette) E.F. Zwart- van Rijkom

Department of Clinical Pharmacy, University Medical Center Utrecht, the

Netherlands & Department of Pharmacoepidemiology and Clinical Pharmacology,

Utrecht Institute for Pharmaceutical Sciences, Utrecht University, The Netherlands

161

L

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

List of publications related to this thesis

Uijtendaal EV, Harssel LL, Hugenholtz GW, Kuck EM, Zwart-van Rijkom JE, Cremer

OL, Egberts TC. Analysis of Potential Drug-Drug Interactions in Medical Intensive

Care Unit Patients. Pharmacotherapy 2014; 34, 213-219.

Uijtendaal EV, Zwart-van Rijkom JE, van Solinge WW, Egberts TC. Serum

Potassium Influencing Interacting Drugs: Risk-modifying strategies also needed at

discontinuation. Ann Pharmacother 2012; 46: 176-182.

Uijtendaal EV, Zwart-van Rijkom JE, van Solinge WW, Egberts TC. Frequency of

laboratory measurement and hyperkalaemia in hospitalised patients using serum

potassium concentration increasing drugs. Eur J Clin Pharmacol. 2011; 67: 933-940.

Zwart-van Rijkom JE, Uijtendaal EV, ten Berg MJ, van Solinge WW, Egberts AC.

Frequency and nature of drug-drug interactions in a Dutch university hospital. Br J

Clin Pharmacol 2009; 68: 187-193.

Other publicationsvan Harssel LM, Uijtendaal EV, Hugenholtz GWK, Kuck EM, Zwart-van Rijkom JEF,

Cremer OL, Egberts TCG. Frequentie en aard van interacties tussen geneesmiddelen

bij Intensive care patiënten. Pharm Weekbl 2012; 6: a1218.

Verstraete E, Veldink JH, Huisman MH, Draak T, Uijtendaal EV, van der Kooi AJ,

Schelhaas J, de Visser M, van der Tweel I, van den Berg LH. Lithium lacks effect on

survival in amyotrophic lateral sclerosis: a phase IIb randomised sequential trial. J

Neurol Neurosurg Psychiatry 2012; 83: 557-564.

Zwart-van Rijkom J, Uijtendaal E, Ten Berg M, Van Solinge W, Egberts A. Frequency

and nature of drug-drug interactions in a Dutch university hospital. Br J Clin

Pharmacol 2010; 70 :618.

van de Vlekkert J, Hoogendijk JE, de Haan RJ, Algra A, van der Tweel I, van der

Pol WL, Uijtendaal EV, de Visser M; Dexa Myositis Trial. Oral dexamethasone pulse

therapy versus daily prednisolone in sub-acute onset myositis, a randomised clinical

trial. Neuromuscul Disord 2010; 20: 382-389.

List of publications

162

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Piepers S, Veldink JH, de Jong SW, van der Tweel I, van der Pol WL, Uijtendaal EV,

Schelhaas HJ, Scheffer H, de Visser M, de Jong JM, Wokke JH, Groeneveld GJ, van

den Berg LH. Randomized sequential trial of valproic acid in amyotrophic lateral

sclerosis. Ann Neurol 2009; 66: 227-234.

de Smet AM, Kluytmans JA, Cooper BS, Mascini EM, Benus RF, van der Werf TS, van

der Hoeven JG, Pickkers P, Bogaers-Hofman D, van der Meer NJ, Bernards AT, Kuijper

EJ, Joore JC, Leverstein-van Hall MA, Bindels AJ, Jansz AR, Wesselink RM, de Jongh

BM, Dennesen PJ, van Asselt GJ, te Velde LF, Frenay IH, Kaasjager K, Bosch FH, van

Iterson M, Thijsen SF, Kluge GH, Pauw W, de Vries JW, Kaan JA, Arends JP, Aarts LP,

Sturm PD, Harinck HI, Voss A, Uijtendaal EV, Blok HE, Thieme Groen ES, Pouw ME,

Kalkman CJ, Bonten MJ. Decontamination of the digestive tract and oropharynx in

ICU patients. N Engl J Med 2009; 360: 20-31.

de Visser M, van de Vlekkert J, Hoogendijk JE, de Haan RJ, Algra A, van der Tweel I,

van der Pol WL, Uijtendaal EV. TO5 Oral dexamethasone pulse therapy versus daily

prednisolone in subacute inflammatory myopathies: A randomised clinical trial.

Neuromuscul Disord 2008; 18: 833.

Kneyber MC, van Woensel JB, Uijtendaal E, Uiterwaal CS, Kimpen JL; Dutch

Antibiotics in RSV Trial (DART) Research Group. Azithromycin does not improve

disease course in hospitalized infants with respiratory syncytial virus (RSV) lower

respiratory tract disease: a randomized equivalence trial. Pediatr Pulmonol 2008;

43: 142-149.

Uijtendaal EV, Vingerhoets RW, van Orshoven NP, Schobben AFAM, Jansen PAF.

Postprandial hypotension in elderly: effectiveness of midodrine and desmopressin

vs placebo assessed with the portapress; a pilot study. Br J Clin Pharmacol 2006; 62:

732-733.

Uijtendaal EV, Rademaker CM, Schobben AF, Fleer A, Kramer WL, van Vught AJ, van

Dijk A. Once-daily versus multiple-daily gentamicin in infants and children. Ther

Drug Monit 2001; 23: 506-513.

Uijtendaal EV, Rademaker CMA, Schobben AFAM, Fleer A. Aminoglycosides once

daily in children: the need for more evidence. Rev Med Microbiol 2000; 11, 87-100.

Bekers O, Uijtendaal EV, Beijnen JH, Bult A, Underberg WJM. Cyclodextrins in the

pharmaceutical field. Drug Dev Ind Pharm 1991; 17: 1503-1549.

163

D

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Dankwoord

Mama, kom je vandaag ook helpen bij de crea-middag op school? Nee, nog even

niet, mama moet nog even wat afmaken. Wat dan? Een boekje. Wat voor een boekje?

Een boekje over mama’s werk. En als het boekje klaar is? Dan gaan hooggeleerde

heren en dames hierover nog vragen stellen. Hooggeleerde heren? Zoals bij Harry

Potter? Nou, daar hadden ze wel oren naar! Zo is het wel een tijdje gegaan, vooral

in het laatste jaar, tijdens de eindsprint. Weekendjes zeilen met het gezin werden

spaarzaam en kindervakantieweken werden gevuld met sportkampen. Gelukkig

vonden ze het altijd leuk om te gaan en hebben hierdoor geleerd dat je met de

juiste sportmentaliteit de finish wel haalt! Zo geldt dat nu ook voor mij: de finish

is in zicht.

Net als bij topsport, wordt promoveren in een bepaald stadium een onderdeel van

het leven, en daarin spelen veel mensen een rol.

Toine, jij kent het traject als geen ander. Je hebt me in de gelegenheid gesteld om

verlof op te nemen voor de eindsprint en ik vond het een bijzondere ervaring dat je

(zonder verlof) in hetzelfde tempo mee sprintte naar de finish. Voor de zesde keer

naar een stuk kijken dat weer gereviseerd was, vaak op tijden dat normale mensen

op één oor liggen, ik heb er respect voor!

Wouter, heerlijk zoals je kan kijken naar de “grote lijn”. Met een grote grijns op je

gezicht kon je vragen stellen als : “Waar doen we dit allemaal voor?” of “Hoe denk je

dat die andere zorgverlener dit leest?”. Het ging om de grote lijn, maar het zijn de

kleine dingen die het doen. Dank hiervoor!

Jeanette, copromotor. Ik bewonder je bijzondere vaardigheid om “concise” te

kunnen schrijven. Daarnaast kan je ook efficiënt werken, dat is ook nodig want

alleen met deze eigenschap red je het om een drukke baan, copromotorschap en

gezin te combineren. Fijn dat we met de warme band die we als collega’s hebben

opgebouwd verder mogen werken aan de toekomst van de apotheek.

Hanneke, wat vond ik het heerlijk dat je wilde meedenken over het onderzoek. En

dat meedenken, dat doe je graag! Je liet zien dat het belangrijk is om de vraagstelling

vooraf helder te hebben, want dan pas kon je de juiste data selecteren. Bovendien

bekeek je de vraag kritisch: Zou je niet beter zus of zo? Je hebt een belangrijke

inbreng gehad. Ik waardeer het daarom enorm dat je ook mijn paranimf wil zijn!

Dankwoord

164

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Marcel, collega van het eerste uur, oud-kamergenoot, vriend. Op jou kan ik bouwen.

Veel woorden zijn hiervoor niet nodig. Fijn dat je mijn andere paranimf wil zijn.

Monique, Mo, fijn dat je de honneurs voor de IC (-apotheek) hebt waargenomen

tijdens mijn verlof. Je denkt IC, je ademt IC en bent eigenlijk gewoon IC. Ik kon het

werk dus met een gerust hart aan je overlaten.

Collega’s, ook dank aan jullie. Het werk dat bleef liggen moest toch verdeeld

worden en jullie hebben dat toch gefixt! Barbara, Erik, met z’n drieën deden we de

eindsprint. Ik heb nooit het gevoel gehad dat we een wedstrijd deden om de eerste

plaats. Integendeel: waar nodig hielpen we elkaar als echte teamgenoten zodat we

alle drie de finish konden halen.

Vrienden, zonder jullie een voor een te noemen, dank voor jullie geduld. Afspraakjes

zijn nog wel eens uitgesteld geweest, maar ze komen er weer aan hoor! Lot, dank

dat de kids altijd bij jou terecht konden. Caby, gelukkig waren er nog momenten

tijdens de hockeytraining van de jongens om bij te praten. Nu wordt het tijd voor

onze eigen sportmoment!

Lieve Heinz, ondanks jouw drukke baan, heb je een niet te onderschatten rol

gespeeld in de ondersteuning thuis. Waar nodig was je mijn luisterend oor,

meelezer en brede schouder. Ons bootje heeft tot jouw leedwezen erg veel aan de

kant gelegen. Gelukkig is Fred er nog die door weer en wind met je mee wil. Na

het behalen van een mijlpaal bepalen we echter steeds weer een nieuwe koers en

daardoor is het nooit saai. Ik ben benieuwd wat de nieuwe koers gaat brengen!

Pap, mam, jullie hebben gezorgd voor de juiste basis. Jullie hebben jezelf altijd

weggecijferd zodat wij, Raymond en ik onszelf konden ontplooien. Jullie zijn altijd

mijn trouwste supporters geweest en hebben er altijd in geloofd dat ik de finish zou

halen. Ik ben dan ook heel blij dat jullie hiervan getuige mogen zijn. Ik draag dit

boekje dan ook op aan jullie!

Tot slot, Nienke en Tjebbe, eindelijk is het dan zover. De finish is in zicht: op naar

Zweinstein!

165

CV

R1

R2

R3

R4

R5

R6

R7

R8

R9

R10

R11

R12

R13

R14

R15

R16

R17

R18

R19

R20

R21

R22

R23

R24

R25

R26

R27

R28

R29

R30

R31

R32

R33

R34

R35

R36

R37

R38

R39

R40

Curriculum vitae

Esther Véronique Uijtendaal was born in ’s-Hertogenbosch, the Netherlands on

April 1, 1967. She graduated Atheneum at the “Sint Janslyceum” in ‘s-Hertogenbosch

in 1985, and subsequently started her studies Pharmaceutical Sciences at Utrecht

University. She obtained her Master of Science degree in pharmacy in 1991 and in

1993 she completed an additional-master program in policy and management at

the Centre of Policy and Management of the University Utrecht (currently Utrecht

University School of Governance (USG)). In 1993 she also completed her PharmD

exam.

She started her professional career as a pharmacist at “Apotheek ‘t Eiland”

in Leerdam, pharmaceutical wholesaler “OPG Medico” in Alkmaar and the

“Waterland Ziekenhuis” in Purmerend. In 1995 she started working as a pharmacist

at the University Medical Centre in Utrecht (UMCU) to implement the electronic

prescribing system “Medicator” and in 1996 she started her training to become

a hospital pharmacist. After finishing her training in 2000, she continued her

professional career as a hospital pharmacist in the UMCU with special interest in

clinical trials. She registered as a clinical pharmacologist in 2007 and is a member

of the medical ethical review board of the UMCU since 2007. In 2009 she started her

PhD research. She combined this PhD thesis with her work as a hospital pharmacist

with special interest in intensive care medicine. Currently she still works as a

hospital pharmacist and clinical pharmacologist in the UMCU with special interest

in vital functions.

Esther lives together with Heinz. They are the proud parents of their daughter

Nienke (2003) and son Tjebbe (2006).

M

onitorin

g drug th

erapy in

hosp

italized patien

ts Esth

er Uijten

daal 2014


Esther Uijtendaal

UITNODIGING



patients





Academiegebouw



Paranimfen:


06-52803009


06-52097419

M

onitorin

g drug th

erapy in

hosp

italized patien

ts Esth

er Uijten

daal 2014


Esther Uijtendaal

UITNODIGING



patients





Academiegebouw



Paranimfen:


06-52803009


06-52097419

Proefschrift Uijtendaal

Documents

Transcript of Proefschrift Uijtendaal