Multi-country rapid adverse drug event assessment: the Asian Pharmacoepidemiology Network (AsPEN)...

ORIGINAL REPORT

Multi-country rapid adverse drug event assessment: the AsianPharmacoepidemiology Network (AsPEN) antipsychotic and acutehyperglycaemia study

Nicole Pratt1, Morten Andersen2, Ulf Bergman2, Nam-Kyong Choi3, Tobias Gerhard4,5, Cecilia Huang4,Michio Kimura6, Tomomi Kimura6, Kiyoshi Kubota7, Edward Chia-Cheng Lai8, Nobuhiro Ooba7, Urban Ösby9,Byung-Joo Park10,11, Tsugumichi Sato7, Ju-Young Shin10, Anders Sundström2, Yea-Huei Kao Yang8

and Elizabeth E Roughead1*

1Quality Use of Medicines and Pharmacy Research Centre, Sansom Institute for Health Research, University of South Australia, Adelaide,Australia2Centre for Pharmacoepidemiology, Karolinska Institutet, and Division of Clinical Pharmacology, Department of LaboratoryMedicineKarolinska Institutet, Karolinska University Hospital Stockholm, Sweden3Medical Research Collaborating Centre, Seoul National University College of Medicine/Seoul National University Hospital, Seoul, Korea4Institute for Health, Health Care Policy and Aging Research, Rutgers University, New Brunswick, NJ, USA5Department of Pharmacy Practice and Administration, Ernest Mario School of Pharmacy, Rutgers University, Piscataway, NJ, USA6Department of Medical Informatics, Hamamatsu University, School of Medicine, Shizuoka, Japan7Department of Pharmacoepidemiology, University of Tokyo Graduate School of Medicine, Tokyo, Japan8Institute of Clinical Pharmacy and Pharmaceutical Sciences, Health Outcome Research Centre, National Cheng KungUniversity, Tainan, Taiwan9Neurogenetics Unit, Department of Molecular Medicine and Surgery, Karolinska Institutet, and Centre for Molecular Medicine,Stockholm, Sweden10Department of Preventative Medicine, Seoul National University College of Medicine11Medical Research Collaborating Centre, Seoul National University College of Medicine, Seoul National University Hospital, Seoul, Korea

ABSTRACTPurpose To undertake a multi-country study to investigate the risk of acute hyperglycaemia with antipsychotic use.Methods Using a distributed network model with a common minimal data set, we performed a prescription sequence symmetry analysis(PSSA) to investigate the risk of acute hyperglycaemia associated with antipsychotic initiation. Incident insulin prescriptions were used as aproxy indicator of acute hyperglycaemia. Participating countries and population datasets included Australia (300,000 persons), Japan I(300,000 persons), Japan II (200,000 persons), Korea (53 million persons) Taiwan (1 million persons), Sweden (9 million persons), USA-Public(87 million persons) and USA-Private (47 million persons).Results Olanzapine showed a trend towards increased risk in most databases, with a significant association observed in the USA-Publicdatabase (Adjusted sequence ratio (ASR) = 1.14; 95% Confidence Interval (CI) 1.10–1.17) and Sweden (ASR= 1.53; 95% CI 1.13–2.06).Null or negative associations were observed for haloperidol, quetiapine and risperidone.Conclusion Acute hyperglycaemia appears to be associated with olanzapine use, however, this effect was only observed in two largedatabases. Despite different patterns of utilization of both antipsychotics and insulin, PSSA analysis results for individual antipsychoticmedicines were qualitatively similar across most countries. PSSA, used in conjunction with existing methods, may provide a simple andtimely method further supporting multi-national drug safety monitoring. Copyright © 2013 John Wiley & Sons, Ltd.

key words—Asian Pharmacoepidemiology Network (AsPEN); distributed network model; multi-national collaboration; antipsychotics;hyperglycaemial; prescription sequence symmetry analysis; pharmacoepidemiology

Received 15 May 2012; Revised 26 February 2013; Accepted 03 March 2013

INTRODUCTION

When medicines’ first reach the market many of themedicines adverse reactions are unknown. Safety

*Correspondence to: E. Roughead, Quality Use of Medicines and PharmacyResearchCentre; Sansom Institute, University of SouthAustralia, Adelaide, Australia.E-mail: [email protected]

Copyright © 2013 John Wiley & Sons, Ltd.

pharmacoepidemiology and drug safety 2013Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/pds.3440

issues can go undetected in clinical trials due tosmall study sizes and short-term follow-up. Activepost-marketing surveillance of medicines is critical toensuring that unrecognized harms are identified early.The safety of newly marketed medicines is tradition-ally monitored through spontaneous reports of adversedrug reactions to regulatory agencies. This system islimited due to under-reporting of possible adversereactions and the inability to determine incidence ofadverse events. The advent of large population basedcomputerized health databases provides the opportunityto complement the spontaneous reporting system bydeveloping alternative methods for rapid medicinesafety monitoring. While countries can undertake thiswork individually, cross national studies have theadvantage of large populations available for studywith resultant increased power to detect rare and oftenserious adverse events.The Asian Pharmacoepidemiology Network, or

AsPEN, is a multi-country pharmaco-epidemiologicresearch network which was formed to provide amechanism to support the conduct of pharmaco-epidemiological research and to facilitate the promptdetection and communication of emerging safetyissues between countries.1 AsPEN is similar to otherpharmacovigilance initiatives that have been developedin Europe and the United States (US). The EU-ADR2

project of the European Commission is a collaborationbetween four countries (Denmark, Italy, Netherlands,United Kingdom) using eight electronic health carerecord databases. The EU-ADR project employs acommon data frameworkwith distributed data processing.Common aggregated, de-identified input files arecreated locally and sent in encrypted format to acentral repository managed in the Netherlands.2 TheUS FDAMini-sentinel System3 also uses a distributeddata network for data analysis, but unlike the EU-ADRsystem, participating health plans and organizationsretain possession of their created data files. Standardizedanalysis programs are distributed by the coordinatingcentre to participants for use on their created data files.In this paper, we describe the first study of AsPEN

which used a distributive network model and acommon standardized analysis program. We employedthe prescription sequence symmetry analysis4 (PSSA)method as a tool for rapid adverse event signaldetection. PSSA examines asymmetry in the distribu-tion of an incident event (e.g. either prescription ofanother medicine or hospitalization) before and afterthe initiation of a specific medicine of interest. Asym-metry may indicate an effect of the specific medicineof interest on the risk of incident events.4 The PSSAmethod was chosen as a test case for these analyses

due to its computational speed and its minimaldata set requirement. Only three data elements arerequired, a patient identifier, medication code andmedication supply date. The method was initiallydeveloped in 19965 and has been further developedfor application to the identification of adverse medicineevents.4,6 Another advantage of PSSA in a distributednetwork model is that it effectively controls forconstant patient-specific unmeasured confoundingdue to its within-person design which means thatdetailed information on potential constant patient-specific confounders are not required.7 Given thehigher prevalence of diabetes in Asian populations,8,9

and the capacity for PSSA to detect acute events, thisstudy was undertaken to determine if antipsychoticuse was associated with an increased risk of acutehyperglycaemia.Case reports and observational studies have identified

an association between antipsychotics and hospitaliza-tion for acute hyperglycaemia10 in both patients withpre-existing diabetes11 and those without diabetes12

including life-threatening hyperglycaemia and acidosis.13

Analysis of the effect of administration of olanzapineand risperidone on fasting glucose and insulin-relatedmetabolic measures in the chronic schizophrenicpopulation found that olanzapine-treated patients hada greater increase in 1-h glucose levels, insulin resis-tance, and decreased insulin sensitivity compared torisperdione-treated patients.14 An observational studyfound that the risk of hyperglycaemia associatedwith antipsychotic use was exacerbated in patientswith pre-existing diabetes and the risk was highestfollowing antipsychotic treatment initiation.13 Animalstudies have also identified this risk, with acuteadministration of olanzapine and quetiapine in miceassociated with a 100% increase in plasma glucoselevels, while risperidone was associated with a 30%reduction in plasma glucose levels.15 A Cochranereview of 50 studies identified that olanzapine wasassociated with significantly larger weight gain andmore metabolic problems than other atypical anti-psychotics.16 Based on the published literature, wehypothesized that olanzapine would be associatedwith increased risk of acute hyperglycaemia and thatrisperidone would not be associated with increasedrisk of acute hyperglycaemia.

PURPOSE

To undertake a multi-country study to determinethe association between initiation of antipsychoticmedication and initiation of insulin, the latter a markerfor acute hyperglycaemia.

n. pratt et al.

Copyright © 2013 John Wiley & Sons, Ltd. Pharmacoepidemiology and Drug Safety, 2013DOI: 10.1002/pds

METHODS

Currently, AsPEN is a collaboration between six coun-tries using eight different datasets; The AustralianGovernment Department of Veterans’ Affairs healthcare claims Database (Australia),17 Japan Medical DataCentre insurance claims database (Japan I),18 hospitaldatabase in Hamamatsu University School of Medicine(Japan II),19,20 Korea Health Insurance Review andAssessment Service (Korea),21,22 the Swedish pre-scribed drug register (Sweden),23 National HealthInsurance Research Database (Taiwan),24 MedicaidAnalytic Extract (USA-Public),25,26 and MedstatMarketScan (USA-Private).27 Four datasets containhealth care claims for specifically insured populations(Australia, Japan I, USA-Public and USA-Private),and three are national datasets (Korean, Taiwan,Sweden), and one is a hospital-based dataset (Japan II).All countries had dispensing data sets at the patient

level including: a patient identifier, patient demo-graphics, date of medicine supply, medicine dis-pensed, quantity, and strength. All countries exceptfor Japan, Korea, and USA used WHO AnatomicalTherapeutic Chemical (ATC) classification codes toclassify medicines. For those countries without ATCcodes, specific country medicine codes were translatedto ATC codes for the purpose of analysis.There were 42 different antipsychotic medicines

available across the datasets. Only six medicinesclassified in the ATC antipsychotic class were commonto all countries; haloperidol, olanzapine, quetiapine,risperidone, lithium, and prochlorperazine. There were15 unique medicines available in one or two countriesonly. Lithium, which is primarily used for bipolar disor-der, and prochlorperazine, predominantly used fornausea and vomiting, were not included in our analysisas they are not used as antipsychotics. Our analysis,therefore, was limited to the four antipsychotics thatwere common to all countries; olanzapine, haloperidol,risperidone, and quetiapine.The AsPEN initiative uses a distributive network

model that required participants to create a commonminimum data set including (i) a unique patient identi-fier, (ii) a variable to identify the medicine dispensedbased on the WHO standard ATC code, and (iii) avariable to identify the date of medicine supply. Theco-ordinating centre for this study. the University ofSouth Australia, developed the statistical analysis codeas a stand-alone SAS program for execution by eachparticipant in their home institution. The SAS programemployed global macro variables which requiredparticipants to enter the variable names used in their datasets rather than forcing the creation of a data file with

common data variable names. This approach eliminatesa complex programming burden for participants andovercomes barriers due to language and disparate datastructures. Participants executed the SAS code and astandardized summary results file was returned to theco-ordinating centre for collation. These standardizedfiles included graphs of the number of people dispensedantipsychotics and diabetes medicines each month(prevalent population), number of people startingantipsychotics and diabetes medicines each month(incident population), and prescription symmetryanalyses including the PSSA distribution graphs.The PSSA method was used for this analysis. In

PSSA, the date of first insulin prescription and the dateof first antipsychotic prescription are determined for anindividual patient. If these two first dispensings arewithin one year of each other, then they are includedin the analysis. Patients were included as initiators ofan individual antipsychotic even if they had beenpreviously prescribed another antipsychotic in theclass. We excluded patients who initiated either medi-cine in the first year of data coverage in a country aswe would not be able to distinguish prevalent use fromincident use in this period. The ratio of the number ofpersons with insulin initiated after antipsychotic initia-tion is compared to the number of persons with insulininitiated before antipsychotics. This ratio is describedas the crude sequence ratio (SR). The SR estimatesthe incidence rate ratio of the event in exposed com-pared to non-exposed person time6. The PSSA methoduses a within person design, making it robust towardsconfounders that are stable over time;4 however, it issensitive to prescribing trends over time. For example,if a medicine is prescribed with increasing incidence, anonspecific excess of persons with that medicine pre-scribed second would be expected.4 In order to adjustfor such temporal trends, a null-effect SR is calculated.The null effect SR is the expected SR one wouldexpect in the absence of a causal association, given theincident medicine use and events in the backgroundpopulation. A description of the formula used for thisvalue is provided elsewhere.6 An adjusted SR isobtained by dividing the crude SR by the null-effectratio.6 The PSSA analyses were restricted to sequencesof incident antipsychotic and insulin initiations within12 months of each other to limit the effect of age andother potential time-varying covariates on the probabilityof exposure and outcome.

RESULTS

This analysis was performed in eight different datasetsranging in size from 175,000 in a Japanese hospitalization

rapid assessment of adverse drug events


dataset to over 87 million in the USA Medicaid dataset.Mean age of the population ranged from 27 years in theUSA Medicaid dataset up to 84 years in the Australianveterans dataset (Table 1).Olanzapine initiation was associated with a signifi-

cantly increased risk of insulin initiation in the USA-Public database (ASR= 1.14; 95% Confidence Interval(CI) 1.10–1.17) and the Swedish prescribed drugs reg-ister (ASR= 1.53; 95% CI 1.13–2.06). The trend in allother databases except Korea was towards increasedrisk (Figure 1). Risperidone was found to have a nullor negative association in seven of the eight databases;however, a significant positive result was observed inthe USA-Public database (Figure 2). Quetiapine andhaloperidol were found to have a null or negative asso-ciations with insulin initiation in all databases studied(Figure 3 and Figure 4, respectively).Temporal analysis of the sequence of events for

olanzapine and insulin revealed a high rate of insulininitiation in the first two weeks after olanzapineinitiation in the USA-Public database (Figure 5).Similar patterns were observed in other countriesexcept Korea which had high numbers of insulininitiators in the weeks before antipsychotic initiation(Figure 6).

DISCUSSION

This multi-country study assessing the associationbetween antipsychotic use and acute hyperglycaemiafound a trend towards increased insulin initiationfollowing olanzapine initiation. Null or negative asso-ciations were found for other antipsychotic medicinesand insulin initiation. In this study, the PSSA methodwas used which relies solely on its self-controlleddesign rather than numerical covariate adjustment tocontrol potential confounding. Using insulin initiationas a proxy for acute hyperglycaemia, we have foundsimilar results to other observational studies,28,29 witholanzapine associated with increased risk, and risperi-done predominantly not associated with increasedrisk. Randomized controlled trials found minimalincreases in fasting glucose levels and insulin resis-tance associated with risperidone, in contrast tosignificant increases in these measurements foundwith olanzapine.30

While many countries found a positive associationbetween olanzapine and insulin initiation, a significanteffect was observed only in two of the larger data-bases, the USA-Public population and the Swedishpopulation. Differences across countries may be dueto chance, to differences in the populations studied,or differences in indication and dose of the prescribed

antipsychotic. The USA-Public population had a highproportion of persons with mental illness31 and isrelatively younger than the other countries (Table 1).Various indications for antipsychotic use may alsocontribute to some of the differences observedbetween countries in this study; for example, lowerdose antipsychotics may be more widely used fordementia-related symptoms in the datasets with elderlypopulations (Australian DVA population) and higherdose antipsychotic for the treatment of schizophreniain the younger populations (USA-Public dataset).Additionally, a review of studies found that diabetesonset effects of antipsychotics may be attenuated inthe elderly.32 The Korean dataset was the only data-base in which a negative association was observedfor olanzapine. Temporal analysis of the sequenceof events (Figure 6) revealed that the antipsychoticinitiation was predominantly occurring in the fourthand fifth week after insulin initiation. Greater stigmafor mental health issues may have led to people notseeking treatment in Korea33; however, once intreatment for an unrelated condition, such as diabetes,treatment for mental health issues may havecommenced.In the majority of countries, risperidone use was not

associated with increased risk, and significant negativeassociations were seen in five countries and a significantincreased risk was observed in the USA-Public popula-tion only. Similar to the olanzapine analysis, the higherrisk in the US-Public population may be due to the useof higher doses of risperidone in younger patients.Quetiapine analyses predominantly showed null associ-ations, while haloperidol analyses predominantlyshowed significant negative associations. Many of theassociations appear to suggest a protective associationbetween antipsychotic and insulin initiation; however,this may be due to the increased numbers of patientsinitiating antipsychotics in the first few weeks afterinsulin is initiated. This may indicate that patients arebeing treated with antipsychotics, potentially for delir-ium, due to hypoglycaemia following insulin initia-tion.34 Alternatively, the protective association may beexplained by protopathic bias. In this example, proto-pathic bias could result from an increased likelihoodof initiating antipsychotic treatment after diabetes treat-ment initiation due to symptomatic adverse events ofthe underlying diabetes disease, such as treatment ofdelirium from infection associated with underlyingundiagnosed diabetes.Due to the potential variation in availability of infor-

mation on potential confounding variables acrosscountries, we employed the PSSA technique which,due to the within person study design, eliminates the

n. pratt et al.


Table

1.Characteristicsof

theAsPEN

datasets

Australia

Japan(I)

Japan(II)

Korea

Sweden

Taiwan

bUSA

c(I)

USA

d(II)

Populationcharacteristics

Population

(N)(‘000)

248

330

175

50,291

9256

1000

87,404

51,422

Years

covered

2001–2

010

2005–2

009

1999–2

010

2001–2

010

July

2005–June

2010

1997–2008

2001–2005

2001–2007

Age

(Mean

(SD))

84.0

(13.4)

36.8

(13.6)

43.3

(24.4)

37.5

(20.4)

41.0

(NAa )

39.5

(20.6)

26.9

(23.6)

32.0

(18.3)

Gender

(Male(%

))59.0

56.6

48.2

50.2

49.7

49.5

38.7

48.8

Group

covered

Veteransandtheir

dependants

Workers

andfamily

mem

bers

ofsix

health

insurance

unions

operated

bylargefirm

s

Patientswho

visited

Ham

amatsu

University

School

ofMedicinehospital

Nationalinsurance

claimsdatabase

National

One

millionrandom

lysampled

individuals

from

original

NHIRD

Medicaideligible

individuals

Privately

insured

individualsfrom

>150

contributin

gem

ployers

andhealth

plans

Typeof

Database

Adm

inistrative

claims

Adm

inistrative

claims

Hospitalrecords

Adm

inistrative

claims

Alldispensings

ofprescribed

drugs

Adm

inistrative

claims

Adm

inistrative

claims

Adm

inistrative

claims

Medicineinform

ation

Coding

system

ATC

Dom

estic

code

Dom

estic

code

Dom

estic

code

ATC

ATC

NDC

NDC

Dateof

prescriptio

nNo

Yes

Yes

Yes

Yes

Yes

No

No

Dateof

dispensing

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

a Not

available

bTaiwan

launched

asingle-payer

NationalHealth

Insuranceprogram

onMarch

1,1995.Asof

2007,2

2.60

millionof

Taiwan’s

22.96millionpopulatio

nswereenrolledin

thisprogram.

c MedicaidAnalytic

Extract

(USA).

dMedstat

MarketScan(U

SA).



need to numerically control for time-stable con-founders. While our study found that the adjustedPSSA analysis results for individual medicines weresimilar in the direction of the associations across coun-tries, we did find that some countries had apparentlyoutlying results. For example, Korea found a negativeassociation between olanzapine and insulin while allother studies had results suggestive of a positive asso-ciation. Future work will develop methods to calculatepooled effect estimates using meta-analysis techniquesand tests for heterogeneity across countries. Such

pooled PSSA analyses will be implemented using thesummary ASRs and will not require access to individualpatient-level data. This will mitigate data-privacy issuesand protect patient confidentiality. Like many otherobservational study designs, the PSSAmethod has somelimitations. First, the method is sensitive to bias due toconfounding by indication. Confounding by indicationwould occur if the underlying disease, for which theantipsychotic is prescribed, schizophrenia or dementia,is associated with an increased risk of diabetes develop-ment. This bias would result in a false positive association

Figure 2. Hyperglycaema (insulin initiation) in relation to risperidone use, by country. Adjusted sequence ratio from Prescription Sequence SymmetryAnalyses

Figure 1. Hyperglycaema (insulin initiation) in relation to olanzapine use, by country. Adjusted sequence ratio from Prescription Sequence SymmetryAnalyses Olanzapine and Insulin

n. pratt et al.


between antipsychotic and insulin initiation. Becauseonly olanzapine was found to be significantly associ-ated with insulin use confounding by indication isunlikely to be the explanation. Additionally, a signifi-cantly increased association may be explained eitherby olanzapine inducing hyperglycaemia or that diabetesprotects against doctors prescribing olanzapine(reverse causation4). The PSSA method is also sensi-tive to changes in the underlying distribution of treat-ment initiators over time. For example if insulininitiation was increasing over the study period by

more than antipsychotic initiation, then we wouldexpect more patients to be prescribed insulin second.The PSSA analysis uses an adjustment process toeliminate the effect of prescribing trends in theestimation of the SR.5 Given the wide variability inincidence use of antipsychotics and insulin betweenthe countries, the similarity between the ASRssuggests that this process appropriately accounts fordifferences in utilization patterns within countrieshighlighting the usefulness of this approach forcross-country safety signal detection.

Figure 3. Hyperglycaema (insulin initiation) in relation to quetiapine use, by country. Adjusted sequence ratio from Prescription Sequence Symmetry Analyses

Figure 4. Hyperglycaema (insulin initiation) in relation to haloperidol use, by country. Adjusted sequence ratio from Prescription Sequence SymmetryAnalyses



A 12 month period between incident prescriptionswas chosen for our prescription symmetry analysis aswe were interested in acute hyperglycaemic events.This period was chosen to ensure sufficient cases inthe smaller datasets, and to limit the effects of potentialtime varying confounders in the analysis. It would bepossible in future work to limit the follow-up periodto 3 or 6 month intervals in the larger datasets. Visualinspection of the PSSA weekly distribution graphs(Figures 5 and 6) indicate that much of the increasedincidence of insulin initiation after olanzapine initia-tion occurs in the first 3 months. The shape of the

symmetry graphs (Figures 5 and 6) show increasingnumbers of people using both treatments on either sideof time zero, that is time of antipsychotic initiation.A potential reason for this is that a patient initiatingon a medicine may be more likely to be assessed bythe physician and consequently initiated on treatmentfor other health issues. This more intense treatmentbehavior is just as likely after initiating antipsychoticsas after initiating insulin and therefore unlikely to biasthe SR. For the purpose of this analysis, we selectedincident insulin prescriptions as a proxy indicator ofacute hyperglycaemia. Insulin was selected as a proxy

Adj RR (95% CI) = 0.598 ( 0.533 - 0.671)

Num

ber

of p

atie

nts

0

10

20

30

40

50

60

Weeks since first Olanzapine prescription

Weeks from olanzapine initiation to insulin initiationWeeks from insulin initiation to olanzapine initiation

Figure 6. Frequency distribution of insulin initiation by number of weeks before or after olanzapine initiation for patients started both medicines within oneyear of each other in the Korean population

Weeks since first Olanzapine prescription

Adj RR (95% CI) = 1.136 ( 1.105 - 1.168)N

umbe

r of

pat

ient

s

0

100

200

300

400

500

600

700

Weeks from olanzapine initiation to insulin initiationWeeks from insulin initiation to olanzapine initiation

Figure 5. Frequency distribution of insulin initiation by number of weeks before or after olanzapine initiation for patients started both medicines within oneyear of each other in the USA-Public population

n. pratt et al.


indicator to investigate the relatively severe/acutehyperglycaemic events induced by antipsychoticsrather than oral antihyperglycaemic medicines suchas metformin, which would be commonly used forchronic management of diabetes. This method wouldnot be suitable for investigating the occurrence ofmetabolic syndrome associated with antipsychoticuse which would require a longer time period tobecome clinically evident. Patients were included asinitiators of an individual antipsychotic even if theyhad been previously prescribed another antipsychoticin the class. We excluded patients who initiated eithermedicine in the first year of data coverage in a countryas we would not be able to distinguish prevalent usefrom incident use in this period. In most countries,individuals do not enter and leave the coveredpopulations so our assumption that the first prescrip-tion observed in the dataset is likely to be an indicatorof incident prescription rather than prevalent use islikely to be valid. For countries with more dynamicpopulation coverage, however, such as the US andJapan, there is a possibility that some patients whowere in fact prevalent users of the medicines may havebeen included. Future work will explore the possibilityof including enrolment date in the inclusion criteria forthe prescription symmetry analysis.This study has demonstrated that a distributive

network model using a common minimum datasetand common analytic code is both feasible and practicalfor real-time studies of medicine safety. Utilizingexisting analytic code, tailored to the specific needs ofeach country, minimized personnel time, and maxi-mized efficiency with study results usually returned tothe co-ordinating centre within one month after supplyof the analytic code. Continued efforts to harmonizeterminology and classification systems will furtherenhance the ability of the AsPEN network to embarkon more detailed pharmacoepidemiological studies.Specific medication availability and the use ofcountry-specific medicine codes were challenges inthe implementation of this project. Of 42 individualantipsychotics available across countries, only fourwere common to all countries, and mapping ofcountry-specific medicine codes to ATC codes wasnecessary for some countries. The use of globalSAS macro variables in the standardized analyticcode means that the PSSA methodology is easilyextended to other medicines and outcome events.Demonstration of the feasibility of implementing the

method cross-nationally using global SAS macros thatdo not require programming expertise raises the possi-bility of the method to be used routinely by regulatoryagencies to identify potential signals of adverse events.

The method would be suitable for suspected cases wherethe adverse event was treated with specific treatmentsavailable in the dataset and the outcome did not influencesubsequent use of the treatment. Potential signals wouldneed to be confirmed by triangulation with other data,such as a signal from self-report systems, as well as bio-logical plausibility. The method may assist as a screeningtool to help prioritize more intensive research studies thatwould provide confirmatory evidence.In summary, we have found that despite different

patterns of utilization of both antipsychotics and insulin,PSSA analysis was easy to implement using adistributed network model across multiple countries andthat results for individual antipsychotic medicines werequalitatively similar across most countries. PSSA, usedin conjunction with existing methods, may provide asimple and timely method further supporting multi-country drug safety monitoring.

CONFLICT OF INTEREST

MA participates in studies funded by AstraZeneca,Merck Sharp & Dohme and Pfizer with grants paidto Karolinska Institutet. TK is an employee of JanssenPharmeceutical K.K. All other authors declare nocompeting interests including specific financial inter-ests and relationships and affiliations relevant to thesubject of this manuscript.

KEY POINTS• Acute hyperglycaemia, as measured by insulin initia-tion, was associated with olanzapine use, but this as-sociation was only significant in two large databases

• Risperidone, quetiapine, and haloperiodol werenot associated with increased risk.

• The Asian PharmacoEpidemiology Network(AsPEN) provides a mechanism to support multi-country rapid assessment of adverse drug events

• Prescription Sequence Symmetry Analysis is afeasible tool for distributed safety surveillanceprograms due to its minimal dataset requirements

ACKNOWLEDGEMENT

Drs. Gerhard and Huang were supported by the Agency ofHealthcare Research and Quality grant U18 HS016097.

REFERENCES

1. AsPENCollaborators. The Asian Pharmacoepidemiology Network (AsPEN): promot-ing multi-national collaboration for pharmacoepidemiologic research in Asia.Pharmacoepidemiol Drug Saf 2013; doi: 10.1002/pds.3439.



2. Coloma PM, Schuemie MJ, Trifiro G, et al. Combining electronic healthcaredatabases in europe to allow for large-scale drug safety monitoring: TheEU-ADR project. Pharmacoepidemiol Drug Saf 2011; 20: 1–11.

3. Behrman RE, Benner JS, Brown JS, McClellan M, Woodcock J, Platt R.Developing the sentinel system--a national resource for evidence development.N Engl J Med2011; 364: 498–499

4. Hallas J. Evidence of depression provoked by cardiovascular medication:a prescription sequence symmetry analysis. Epidemiology 1996; 7(5):478–484.

5. Petri H, de Vet HC, Naus J, Urquhart J. Prescription sequence analysis: a newand fast method for assessing certain adverse reactions of prescription drugs inlarge populations. Stat Med 1988; 7: 1171–1175.

6. Tsiropoulos I, Andersen M, Hallas J. Adverse events with use of antiepilepticdrugs: a prescription and event symmetry analysis. Pharmacoepidemiol DrugSaf 2009; 18: 483–491.

7. Maclure M, Fireman B, Nelson JC, et al. When should case-only designs be usedfor safety monitoring of medical products? Pharmacoepidemiol Drug Saf 2012;21(Suppl 1): 50–61.

8. Kim SM, Lee JS, Lee J, et al. Prevalence of Diabetes and Impaired FastingGlucose in Korea. Diabetes Care 2006; 29(2): 226–231.

9. The DECODA Study Group. Age- and Sex-Specific Prevalence of Diabetes andImpaired Glucose Regulation in 11 Asian Cohorts. Diabetes Care 2003; 26(6):1770–1780.

10. Kohen I, Gampel M, Reddy L, Manu P. Rapidly developing hyperglyce-mia during treatment with olanzapine. Ann Pharmacother 2008; 42(4):588–591.

11. Lipscombe LL, Levesque L, Gruneir A, et al. Antipsychotic drugs andhyperglycemia in older patients with diabetes. Arch Intern Med 2009;169(14): 1282–1289.

12. Lipscombe LL, Levesque LE, Gruneir A, et al. Antipsychotic drugs and the riskof hyperglycemia in older adults without diabetes: a population-based observa-tional study. Am J Geriatr Psychiatry 2011; 19(12): 1026–1033.

13. Varma MK, Connolly K, Fulton B. Life-threatening hyperglycemia and acidosisrelated to olanzapine: a case report and review of the literature. J Intensive CareMed 2007; 22(1): 52–55.

14. Smith M, Hopkins D, Peveler RC, Holt RI, Woodward M, Ismail K. First- v.second-generation antipsychotics and risk for diabetes in schizophrenia:systematic review and meta-analysis. Br J Psychiatry 2008; 192(6):406–411.

15. Savoy YE, Ashton MA, Miller MW, et al. Differential effects of various typicaland atypical antipsychotics on plasma glucose and insulin levels in the mouse:evidence for the involvement of sympathetic regulation. Schizophr Bull 2010;36(2): 410–418.

16. Komossa K, Rummel-Kluge C, Hunger H, et al. Olanzapine versus other atypicalantipsychotics for schizophrenia. Cochrane Database Syst Rev 2010(3):CD006654.

17. Data Management and Information Systems, Department of Veterans’ Affairs,Canberra, Australia.

18. Japan Medical Data Center database: Claims data assembled from a couple ofprivate health insurance unions, Japan. http://www.jmdc.co.jp/en/mission/index.html (last accessed 22 October 2012).

19. Watanabe H, Kimura T, Hori K, Kawakami J, Kimura M. [D?D: Analytical clinicalinformation retrieval system based on hospital information system – overview anduse cases.] Jpn J Pharmacoepidemiol. 2010; 15(2): 97–106. (In Japanese)

20. Kimura M, et.al. High speed clinical data retrieval system with event timesequence feature - with 10 years of clinical data of Hamamatsu UniversityHospital CPOE. Methods Inf Med 2008; 47(6): 560–568.

21. Korea Health Insurance Review and Assessment Service (HIRA). Major Activitiesof HIRA. HIRA, Seoul; March 2007. http://hira.or.kr/cms/rb/rbb_english/13/13_01/01/review.html (last accessed 22 October 2012).

22. Korea National Health Insurance Corporation (NHIC). Statistical Information.NHIC, Seoul; March 2010. http://stat.kosis.kr/nsieu/view/tree.do?task=branchView&id=350_35001_B*MT_OTITLE&hOrg=350 (last accessed 22 Oc-tober 2012).

23. Wettermark B, Hammar N, Fored CM, et al. The new Swedish Prescribed DrugRegister--opportunities for pharmacoepidemiological research and experiencefrom the first six months. Pharmacoepidemiol Drug Saf 2007; 16(7): 726–735.

24. National Health Insurance Research Database (NHIRD). Taiwan. http://www.nhri.org.tw/nhird/en/index.htm (last accessed 22 October 2012)

25. Crystal S, Akincigil A, Bilder S, Walkup JT. Studying prescription drug use andoutcomes with medicaid claims data: strengths, limitations, and strategies. MedCare 2007; 45(10 Supl 2): S58–65.

26. Centers for Medicare and Medicaid Services. Medicaid Analytic eXtract (MAX)General Information. http://www.cms.gov/MedicaidDataSourcesGenInfo/07_MAXGeneralInformation.asp. (last accessed 22 October 2012).

27. Adamson DM, Chang S, Hansen LG. White Paper: Health Research Data for theReal 1World: the MarketScan Databases. Atlanta, GA: Thomson Medstat; 2006.

28. Koro CE, Fedder DO, L’Italien GJ, et al. Assessment of independent effect ofolanzapine and risperidone on risk of diabetes among patients with schizophre-nia: population based nested case control study. BMJ 2002; 325(7358): 243–247.

29. Lambert BL, Chou CH, Chang KY, Tafesse E, Carson W. Antipsychoticexposure and type 2 diabetes among patients with schizophrenia: a matchedcase–control study of California Medicaid claims. Pharmacoepidemiol DrugSaf 2005; 14(6): 417–425.

30. Newcomer JW. Second-generation (atypical) antipsychotics andmetabolic effects: acomprehensive literature review. CNS Drugs 2005; 19(Suppl 1): 1–93.

31. Center for Mental Health Services. Mental Health, United States, 2004.Manderscheid, R.W., and Berry, J.T., eds. DHHS Pub no. (SMA)-06-4195. Rock-ville, MD: Substance Abuse and Mental Health Services Administration, 2006.

32. Trifiro G, Spina E, Gambassi G. Use of antipsychotics in elderly patients withdementia: do atypical and conventional agents have a similar safety profile?Pharmacol Res 2009; 59(1): 1–12.

33. Kim YJ, Kim G, Hansen L, Chiriboga D. Attitudes of Older Korean AmericansToward Mental Health Services. J Am Geriatr Soc 2007; 55(4); 616–620.

34. Miller MO. Evaluation and management of delirium in hospitalized olderpatients. Am Fam Physician 2008; 78(11): 1265–1270.

n. pratt et al.


http://www.jmdc.co.jp/en/mission/index.html

http://www.jmdc.co.jp/en/mission/index.html

http://hira.or.kr/cms/rb/rbb_english/13/13_01/01/review.html

http://hira.or.kr/cms/rb/rbb_english/13/13_01/01/review.html

http://stat.kosis.kr/nsieu/view/tree.do?task=branchView&id=350_35001_B*MT_OTITLE&hOrg=350

http://stat.kosis.kr/nsieu/view/tree.do?task=branchView&id=350_35001_B*MT_OTITLE&hOrg=350

http://www.cms.gov/MedicaidDataSourcesGenInfo/07_MAXGeneralInformation.asp

http://www.cms.gov/MedicaidDataSourcesGenInfo/07_MAXGeneralInformation.asp

Multi-country rapid adverse drug event assessment: the Asian Pharmacoepidemiology Network (AsPEN)...

Documents

Transcript of Multi-country rapid adverse drug event assessment: the Asian Pharmacoepidemiology Network (AsPEN)...