Applying quantitative methods for detecting new drug safety signals in pharmacovigilance national database

  • Published on

  • View

  • Download

Embed Size (px)



    Applying quantitative methodsn


    School of Pharmacy, Tehran University of Medical Sciences, Tehran, Islamic Republic of Iran

    Received 10 September 2006; Accepted 28 June 2007

    reactions (ADRs). Discovery of previously undetected and a drug, the relationship being unknown or,

    different in quantity (more frequency) or quality (more

    pharmacoepidemiology and drug safety 2007; 16: 11361140Published online 17 August 2007 in Wiley InterScience ( DOI: 10.1002/pds.1459

    Islamic Republic of Iran. E-mail: shalviri_g@yahoo.comCopyright # 2007 John Wiley & Sons, Ltd.adverse effects of medicines has been defined as one of incompletely documented previously.2 Howeverthe concept of signal is also included in any newinformation from various sources, indicating apossible risk associated with a medicine, which is

    * Correspondence to: G. Shalviri, Iranian ADR Monitoring Center,Ministry of Health, Fakhre Razi Ave, Enghlab Ave, Tehran 13145,INTRODUCTION

    The principle of pharmacovigilance science is timelydetection of new and previously unknown adverse drug

    the potential contributions of pharmacoepidemiol-ogy.1 The World Health Organization (WHO) definesa drug safety signal as Reported information on apossible causal relationship between an adverse eventcovigilance database over the period of 19982005.Methods All adverse drug reactions (ADRs) reported to Iranian Pharmacovigilance Center (IPC) from March 1998through January 2005, were used for the analysis. The data were analysed based on three different signal detection methodsincluding reporting odds ratios (RORs), information component (IC) and proportional reporting ratios (PRRs). The signalsdetected were categorised based on the number of reports per drug-adverse event combination, severity of the event andlabelled or unlabelled ADRs.Results During the study period, 6353 cases of ADR reports describing 11 130 reactions were received by IPC. The datasetinvolved 4975 drug-adverse event combinations. The count of drug-event combinations was 1, 2 and 3 or more for 3470, 726and 779 combinations, respectively. According to PRRs, there were 2838, 872 and 488 drug-event combinations known as asignal for the pairs with the reporting frequency of 1, 2 and 3 reports, respectively. The results of estimating RORs showedthat 2722, 862 and 481 drug-adverse event combinations were detected to be signal for the pairs with the reporting frequencyof 1, 2 and 3 reports, respectively, while measuring IC and IC-2SD detected 1120, 378 and 235 for the same reportingfrequencies. Diclofenac-induced paralysis and tramadol-induced severe reactions were the most important signals.Conclusion Applying quantitative signal detection methods to the database of national pharmacovigilance centres isnecessary to early detection of drug safety alerts. Copyright # 2007 John Wiley & Sons, Ltd.

    key words signal; adverse drug reaction; pharmacoepidemiologyObjectives To applies three different methods of signal dSUMMARY

    etection to the registered adverse events in Iranian Pharma-Present Address: Iranian ADR Monitoring center, Ministry of Health, Fakhre Razi Ave, Enghlab Ave, Tehran 13145,Islamic Republic of Iran3signals in pharmacovigilance

    Gloria Shalviri1,2*, Kazem Mohammad1, Reza

    1School of Public Health and Institute of Public Health ReseRepublic of Iran2for detecting new drug safetyational database

    Majdzadeh1 and Kheirollah Gholami3

    rch, Tehran University of Medical Sciences, Tehran, Islamic

  • seriousness) from available information.3 Duringthe last four decades, national pharmacovigilancecentres in the world have tried tremendously to find away to better detect unknown signals of drug-relatedproblems. Creation of different quantitative methods for

    these meetings, the causality and seriousness ofreports are determined based on WHO definitions.16

    Furthermore, the urgent preliminary meeting would beset up within 3 hours of receiving reports of death orlife-threatening reactions, permanent damage ordisability and/or birth defect. Although, individualcase assessment is well performed in IPC, there islack of computerised aggregation case assessment.Signals are detected based on traditional methods byclinical observation and IPC has not defined compu-terised methods to automatically detect signals. In thisstudy we tried to examine three quantitative methods


    detecting new drug safety signals 1137data mining, especially in large databases of spon-taneous reporting, has been the result of global effortsin this area. Recently, calculating proportional report-ing ratios (PRRs) based on 2 2 contingency tableshas been adopted for use by UK Medicines Agency.Evans et al.4 revealed the results of utilisation of thePRR method for 15 newly marketed drugs with thehighest levels of ADR reporting during 19961998.The authors concluded that PRRs were valuable aid tosignal generation from spontaneous reporting data.Rothman et al.5 offered some advantages for reportingodds ratios (RORs) over PRRs for signal detection inpharmacovigilance; however, the subject is still amatter of controversy.6,7 Calculating odds ratios fordetecting signals has been undertaken by somenational ADR centres such as the Netherlands LarebFoundation.8 The Uppsala Monitoring Center, coor-dinating department of WHO for International DrugMonitoring, has developed a new model for generatingsignals from international database of adverse events,since 1998.9,10 The WHO model for signal detection,known as Bayesian Confidence Propagation NeuralNetwork (BCPNN), was designed based on Bayestheory which was able to recognise early alerts ofsome new adverse reactions such as topiramate-inducedglaucoma, olanzepin-induced granolocytopenia andcaptopril-induced cough.11 Another model of Baye-sian data mining has been invented by DuMouchel forapplication in the database of the US Food and DrugAdministration (FDA) since 1999.12 The model calledEmpirical Bayes Screening (EBS) stratifies the ADRreports by (at least) gender and year of report andmathematically is based on a two gamma mixturedistribution and ranking of product-adverse effectpairs by degree of interest.13,14

    IPC, as a full member of WHO International DrugMonitoring Program, is the national centre for con-ducting Pharmacovigilance activities in Iran. IPCstarted to collect and assess ADR reports from all overthe country since 1998. The first ADR reportssubmitted to WHO were the results of a researchproject conducted in a teaching hospital for monitor-ing ADRs.15 Based on the results of that study, themethods for evaluating ADR reports were defined atthe centre. All ADR reports received by IPC areassessed in weekly case assessment meeting by anadvisory committee consisted of physicians, clinicalpharmacists, epidemiologist and ADR specialists. In

    Copyright # 2007 John Wiley & Sons, Ltd. Phof signal detection in IPC. This is the first timethat IPC database has been assessed systematicallyfor signals.


    All ADRs reported to IPC from March 1998 to January2005 were analysed. The WHO Terminology is used toharmonise reported ADRs in IPC database.17 Report-ed ADRs to IPC are routinely registered in MicrosoftExcel file. All registered ADRs were ranked based onthe frequency of drug-event combinations registeredin the database. In order to count the reportedfrequency of each recorded combination, MicrosoftAccess was used. The contingency table (Table 1) wascalculated for each combination as the basic table forusing quantitative methods.

    The data were analysed based on three differentsignal detection methods including PRRs, informationcomponent (IC) and 95% confidence interval (CI)(IC-2SD) and RORs.

    In order to use PRRs for detecting signals, twodifferent criteria were considered. The first criterionwas defined as the combinations with the reportingfrequency of 3 reports, PRR 2 and x2 4. Thesecond one was the signals detected using PRR and95%CI.

    The second method used for signal detection in thisstudy was performed by estimating the measure ofdisproportionality known as IC and the IC-2SD.Whenever IC-2SD was greater than zero, the combina-tion was considered a possible signal for furtherevaluation.

    Table 1. A 2 2 contingency table for calculating measure ofdisproportionality

    Suspected drug All other drugs

    Suspected reaction a bAll other reactions c d

    rmacoepidemiology and Drug Safety, 2007; 16: 11361140DOI: 10.1002/pds

  • Estimating RORs and 95%CI for each drug-eventcombination was the third method used in this study.

    The signals detected by each method were cate-gorised based on the number of reports per drug-adverse event combination, severity of the event andlabelled or unlabelled ones. In this respect, measuresof disproportionality were calculated for the combi-nations with number of reports 1, 2 or 3separately. Every drug-event combination which wasnot mentioned in the 2004 version of The MartindaleExtra Pharmacopea was classified as an unlabelledreaction. To determine severe reactions, the WHOdefinition for serious ADR was used.18 Accordingly, aserious reaction is one which resulted in one or moreof the following:

    1. Death or life-threatening situation2. Hospitalisation or prolonged hospital stay3. Persistent organ damage or disability4. Congenital birth defect.

    signal for the pairs with the reporting frequency of1,2 and 3 reports, respectively. The signals detectedwith the criteria of PRR2, x2 4 were 2930, 872 and480 for the pairs with the same reporting frequencies.Among signals detected by PRR and 95%CI, 7.75%were serious-labelled events and 12.16% wereserious-unlabelled ones for the combinations withnumber of reports 1.

    The results of estimating RORs and 95%CI showedthat 2722, 862 and 481 drug-event combinations weredetected to be signal for the pairs with the reportingfrequency of 1, 2 and 3 reports, respectively,while measuring IC and IC-2SD detected 1120, 378and 235 for the same reporting frequencies. Assessingseriousness and being labelled ADRs revealed that3.04% of signals detected by ROR belonged toserious-labelled reactions versus 5.51% seriou-s-unlabelled ones. Evaluation of new adverse eventsand their seriousness by IC-2SD expressed 11.43% ofthem as serious-labelled and 17.32% as serious-unlabelled in the total reported combinations.





    1138 g. shalviri ET AL.1 PRR 2, x2 4ROR, CI> 1PRR, CI> 1

    IC, IC-2SD> 02 PRR 2, x2 4

    ROR, CI> 1PRR, CI> 1

    IC, IC-2SD> 03 PRR 2, x2 4

    ROR, CI> 1PRR, CI> 1

    IC, IC-2SD> 0

    Copyright # 2007 John Wiley & Sons, Ltd. Phthe combination d

    Frequency of Method No.Table 2. Comparing signal detection methods in detecting new anThe methods applied to signal detection were thencompared in recognising different types of adverseevents.


    There were 6353 cases of ADR reports describing11 130 reactions, reported to IPC during the studyperiod. The dataset involved 4975 drug-event com-binations, which were assessed for detecting signals.

    The count of drug-event combinations was 1, 2 and3 or more for 3470, 726 and 779 combinations,respectively. There were 500 drug items responsiblefor 468 reaction terms in the database.

    According to PRR and 95%CI, there were 2838,872 and 488 drug-event combinations known as aTable 2 shows the result of comparing threemethods taken in detecting new and serious signals.There were 234 signals commonly detected by allthree methods of signal detection used in this study.There was no signal detected by IC-2SD which wasnot detected by other two methods.

    Of the 234 common signals detected by the threemethods, signals related to paralysis induced byintramascular injection of diclofenac sodium andsevere reactions induced by tramadol injection(dyspnoea, hypotension, bronchospasm, respiratorydepression, nausea and vomiting) were previously ledto a request for the manufacturer to change the productlabelling. These signals were detected traditionallybased on clinical presentations. Following spon-taneous reports of serious adverse effects, parenteral

    serious signals


    Labelled andserious (%)

    Unlabelled andserious (%)

    30 7.9 11.822 3.04 5.5138 7.75 12.1620 11.43 17.3272 5.62 11.0762 2.32 4.8772 6.08 12.6178 7.41 18.8080 3.33 11.6781 4.57 12.4788 5.12 11.0735 6.38 21.27

    rmacoepidemiology and Drug Safety, 2007; 16: 11361140DOI: 10.1002/pds

  • bronchospasm, respiratory depression, nausea andvomiting). Diclofenac injection induced paralysis wasa previously unknown reaction recognised by IPC,while reported ADRs with tramadol generated a signalindicating more frequency and seriousness of these

    Table 3. Top 15 detected signals with the highest number ofreports

    Rank Reaction Drug Total count

    1 Rigors Streptokinase 240

    detecting new drug safety signals 1139use of diclofenac sodium and tramadol was limited tohospital at 2001 and 2003 respectively.19 Table 3shows top 15 drug-event combinations with thehighest number of reports, meeting minimum criteriafor potential signals. As shown in the table, thesesignals are related to eight medicines in whichstreptokinase, tramadol, diclofenac and co-trimoxazolrelated adverse reactions need further evaluations.


    Early attempts for computerising signal detection goback to 1970s.20 Most activities currently centre on:RORs, Bayesian data mining and PRRs.21

    Applying the PRR method with the cut-off asPRR 2, x2 4 for the combinations with3 reports,identified 480 (9.65% of reported preferred terms)signals in this study in which 390 (81.4%) were

    2 Vomiting Tramadol 1423 Nausea Tramadol 1394 Paralysis Diclofenac 1345 Skin discolouration Flourescine 1196 Skin disorder Lipiodol 1177 Vertigo Tramadol 1168 Pruritus Co-trimaxazole 749 Injection site mass Deferoxamine 66

    10 Pruritus Deferoxamine 6411 Asthenia Tramadol 6112 Rash Co-trimaxazole 6113 Headache Trinitroglycerin 5614 Injection site necrosis Diclofenac 5415 Urticaria Co-trimaxazole 52recognised as known ADRs and 89 (18.6%) neededmore evaluation. In a similar study conducted byEvans et al.,4 the method identified 487 potentialsignals (10% of reported preferred terms), 339 (70%)were recognised adverse reactions, 80 (17%) requiringfurther evaluation, three cases led to labelling change.Similarly, there were three signals in our study relatedto ADRs which previously led to labelling changes forsafety reasons. Also, two detected signals were relatedto ADRs which previously led to regulatory decisionmaking to shift the availability of the products fromcommunity to hospital use only. These reactionsinvolved paralysis induced by IM injection ofdiclofenac sodium and severe reactions induced bytramadol IM injection (dyspnoea, hypotension,

    Copyright # 2007 John Wiley & Sons, Ltd. PhaADRs which led to considerable number of hospital-isation and three cases of death. These cases of ADRsled to regulatory decision resulting in limitations ofthe injection pharmaceutical forms of diclofenacsodium and tramadol to hospital use in 2001 and 2003,respectively. Although these signals were detectedwith traditional methods, applying quantitativemethods sys...


View more >