Data management and data analysis techniques in pharmacoepidemiological studies using a pre-planned multi-database approach: a systematic literature review

REVIEW Da a managemen and da a analysis echniques in pha macoepidemiological s udies using a p e-planned mul i-da abase app oach: a sys ema ic li e a u e e iew † Ma loes T. Bazelie 1 *, I ene E iksson 2 , F ank de V ies 1,3 , Ma janka K. Schmid 4 , Jani Rai anen 5,6 , Ja i Haukka 7 , Jakob S a up-Linde 8,9 , Ma ie L. De B uin 1 and Mo en Ande sen 2 1 Di ision o Pha macoepidemiology and Clinical Pha macology, U ech Ins i u e o Pha maceu ical Sciences, U ech Uni e si y, Ne he lands 2 Cen e o Pha macoepidemiology, Ka olinska Ins i u e , S ockholm, Sweden 3 Depa men o Clinical Pha macy and Toxicology, Maas ich Uni e si y Medical Cen e, Maas ich , Ne he lands 4 Di ision o Molecula Pa hology, Ne he lands Cance Ins i u e, Ne he lands 5 School o Heal h Sciences, Uni e si y o Tampe e, Finland 6 UKK Ins i u e o Heal h P omo ion, Tampe e, Finland 7 Uni e si y o Helsinki, Helsinki, Finland 8 Aalbo g Uni e si y, Aalbo g, Denma k 9 Depa men o Endoc inology and In e nal Medicine, Aa hus Uni e si y Hospi al, Aa hus, Denma k ABSTRACT Pu pose To iden i y pha macoepidemiological mul i-da abase s udies and o desc ibe da a managemen and da a analysis echniques used o combining da a. Me hods Sys ema ic li e a u e sea ches we e conduc ed in PubMed and Embase complemen ed by a manual li e a u e sea ch. We included pha macoepidemiological mul i-da abase s udies published om 2007 onwa ds ha combined da a o a p e-planned common analysis o quan i a i e syn hesis. In o ma ion was e ie ed abou s udy cha ac e is ics, me hods used o indi idual-le el analyses and me a- analyses, da a managemen and mo i a ions o pe o ming he s udy. Resul s We ound 3083 a icles by he sys ema ic sea ches and an addi ional 176 by he manual sea ch. A e ull- ex sc eening o 75 a icles, 22 we e selec ed o final inclusion. The numbe o da abases used pe s udy anged om 2 o 17 (median = 4.0). Mos s udies used a coho design (82%) ins ead o a case–con ol design (18%). Logis ic eg ession was mos o en used o indi idual-le el analyses (41%), ollowed by Cox eg es- sion (23%) and Poisson eg ession (14%). As me a-analysis me hod, a majo i y o he s udies combined indi idual pa ien da a (73%). Six s udies pe o med an agg ega e me a-analysis (27%), while a semi-agg ega e app oach was applied in h ee s udies (14%). In o ma ion on cen al p og am- ming o he e ogenei y assessmen was missing in app oxima ely hal o he publica ions. Mos s udies we e mo i a ed by imp o ing powe (86%). Conclusions Pha macoepidemiological mul i-da abase s udies a e a well-powe ed s a egy o add ess sa e y issues and ha e inc eased in popula i y. To be able o co ec ly in e p e he esul s o hese s udies, i is impo an o sys ema ically epo on da abase managemen and analysis echniques, including cen al p og amming and he e ogenei y es ing. © 2015 The Au ho s. Pha macoepidemiology and D ug Sa e y published by John Wiley & Sons, L d. key wo ds—pha macoepidemiology; mul i-da abase; sys ema ic e iew; da a managemen ; analysis echniques Recei ed 18 Decembe 2014; Re ised 29 May 2015; Accep ed 08 June 2015 INTRODUCTION The need o pos -app o al su eillance o d ug sa e y has been widely ecognized o mo e han ou decades. F om he ea ly days o pha macoepidemiology, ini ia- i es ha e been unde aken o s udy sa e y and, mos e- cen ly, e ec i eness o medica ions using ou inely collec ed heal hca e da a. O e he pas decade, an inc easing numbe o s udies ha e been pe o med using heal hca e da abases om mul iple coun ies, egions o *Co espondence o: M. T. Bazelie , Depa men o Pha macoepidemiology and Clinical Pha macology, U ech Ins i u e o Pha maceu ical Sciences, U ech Uni e si y, Uni e si ei sweg 99, 3584 CG U ech , Ne he lands. E-mail: [email p o ec ed] † P io pos ings and p esen a ions: Some findings om his s udy we e p esen ed a he ICPE con e ence 2014 du ing he wo kshop ‘The Common Da a Model: Lessons om Pas P ojec s and Ongoing Ini ia i es’(Pha macoepidemiol D ug Sa . 2014 Oc ; 23 Suppl 1: 361). © 2015 The Au ho s. Pha macoepidemiology and D ug Sa e y published by John Wiley & Sons, L d. pha macoepidemiology and d ug sa e y 2015; 24: 897–905 Published online 14 July 2015 in Wiley Online Lib a y (wileyonlinelib a y.com) DOI: 10.1002/pds.3828 This is an open access a icle unde he e ms o he C ea i e Commons A ibu ion-NonComme cial-NoDe i s License, which pe mi s use and dis ibu ion in any medium, p o ided he o iginal wo k is p ope ly ci ed, he use is non-comme cial and no modifica ions o adap a ions a e made. heal hca e o ganiza ions. Using da a om mul iple da a- bases o e s a numbe o po en ial ad an ages such as in- c eased sample size (da ase s become la ge enough o gi e p ecise es ima es o medica ion isks and benefi s e en o a e ou comes and exposu es) and gene aliz- abili y (when simila esul s a e ound in s udies u ilizing he same me hodology in he e ogeneous popula ions). Fu he mo e, i o e s a po en ial o using a s anda d- ized me hodological app oach ac oss da a sou ces. A numbe o la ge ini ia i es ha e been launched o de elop me hods o combining da a om mul iple da abases and egis e s. In he Uni ed S a es (U.S.), he Obse a ional Medical Ou comes Pa ne ship (OMOP) 1 and he Mini-Sen inel p og am 2 ha e been un since 2007 and 2008, espec i ely. Da a a e o en combined using he heal h main enance o ganiza ion (HMO) Resea ch Ne wo k, a conso ium o 19 la ge heal hca e deli e y o ganiza ions in he U.S. 3 The Ca- nadian Ne wo k o Obse a ional D ug E ec S udies (CNODES), a dis ibu ed ne wo k o Canadian e- sea che s and da a cen e s, was launched in 2011. 4,5 Eu opean ini ia i es ha add ess logis ical and me h- odological p oblems o conduc ing mul i-da abase s udies include EU-ADR 6 and IMI-PROTECT. 7 Recen ly, he Asian Pha macoepidemiology Ne wo k (AsPEN), 8 a collabo a ion including Asian coun ies, was also s a ed. Besides hese la ge p og ams, a hand- ul o smalle esea ch p ojec s ha e also combined da a om se e al heal hca e da abases. 9–13 The e a e a ious ways o combine da a om se - e al independen da abases, which all ha e di e en ad an ages and disad an ages. A combina ion o ag- g ega e esul s does no equi e sha ing o indi idual pa ien da a and makes op imal use o locally a ailable da a (e.g. in o ma ion on con ounde s). Howe e , co ec ing o he e ogenei y be ween he da abases may no be ully possible when combining summa y es ima es. A combina ion o indi idual pa ien da a opens mo e space o explo ing and co ec ing o he - e ogenei y; i o e s an oppo uni y o use exac ly he same defini ions o exposu es, ou comes, co a ia es, and ime windows. This app oach may, howe e , e- sul in a comp omise when ele an in o ma ion is los i no a ailable in all da abases (possibly leading o la ge esidual con ounding). To ou knowledge, no sys ema ic e iew has been pe o med ye o selec mul i-da abase s udies and o illumina e which me hods ha e mos equen ly been used o combine da a. Ou aim was he e o e o iden- i y pha macoepidemiological s udies using a p e- planned mul i-da abase app oach and o desc ibe da a managemen and da a analysis echniques used o combining da a. METHODS Li e a u e sea ch We conduc ed sys ema ic li e a u e sea ches in PubMed and Embase, as well as a manual li e a u e sea ch, o iden i y ele an mul i-da abase obse a- ional s udies. We ollowed he PRISMA guideline (www.p isma-s a emen .o g). The ollowing inclusion c i e ia we e applied: (i) pee - e iewed pha macoepidemiological s udy (de- fined as an obse a ional s udy abou he sa e y o e - ec i eness o medica ion); (ii) published om 2007 onwa ds; (iii) s udy subjec s we e selec ed om wo o mo e independen heal hca e da abases (i.e. da a- bases co e ing a di e en s udy popula ion); (i ) all da abases wi hin he s udy we e analyzed o answe he same esea ch ques ion; and ( ) da a we e com- bined o a p e-planned common analysis o quan i a- i e syn hesis (i.e. da a we e combined ei he a indi idual pa ien -le el o he es ima es ob ained om indi idual da abases we e combined in analyses). We excluded d ug u iliza ion s udies, a icles ha ocused on he de ec ion o ad e se d ug eac ions (ADRs) and o he pha maco igilance s udies, s udies ha only epo ed esul s om sepa a e da abases wi hou p o- iding any combined es ima es, pu ely me hodological pape s (e.g. desc ibing me hods o combine da a, bu no ac ually doing so in he pape ), as well as s udies published in languages o he han English (exclusion c i e ia we e applied sequen ially). Sys ema ic sea ch s a egies o PubMed and Embase we e de eloped unde he supe ision o a e- sea ch lib a ian and included ex wo ds and ele an indexing o cap u e pha macoepidemiological s udies sa is ying he inclusion c i e ia. The s a egies we e adap ed o ma ch he s uc u e o each da abase and we e based on sea ch e ms ha included ‘d ugs’,‘da- abases’, and ‘epidemiology’/‘obse a ional s udies’ (using MeSH, Em ee, and ee ex e ms); bo h da a- bases we e sea ched om 2007 o Oc obe 2013 (see e- ables 1 and 2 o he ull sea ch s a egy in Pubmed and Embase, espec i ely). The da e fil e om 2007 onwa ds was applied as ou pilo sea ches did no iden i y any ele an publica ions be o e 2007. We he e o e belie ed ha i would help imp o e he p e- cision o he sea ch wi hou nega i e impac on i s sen- si i i y. One au ho (IE) sc eened he i les and abs ac s o he iden ified a icles using ou inclusion and exclusion c i e ia o selec a icles eligible o ull- ex sc eening. The elec onic li e a u e sea ch was supplemen ed wi h a manual sea ch (2007 o Decembe 2013). We iden ified publica ions lis ed on he ollowing esea ch m. . bazelie e al. 898 © 2015 The Au ho s. Pha macoepidemiology and D ug Sa e y published by John Wiley & Sons, L d. Pha macoepidemiology and D ug Sa e y, 2015; 24: 897–905 DOI: 10.1002/pds p ojec s’websi es: (i) OMOP (h p://omop. nih.o g and www.omop.o g); (ii) Mini-Sen inel (www.mini- sen inel.o g); (iii) he HMO Resea ch Ne wo k (www.hmo esea chne wo k.o g); (i ) EU-ADR (www. euad -p ojec .o g); ( ) IMI-PROTECT (www.imi-p o- ec .eu); ( i) AsPEN (www.aspenne .asia); and ( ii) CNODES (www.cnodes.ca). In addi ion, esea che s wi hin ou eam we e consul ed o add mul i- da abase p ojec s ha hey we e awa e o (and ha we e no iden ified in he au oma ic and websi e sea ches). The i les and abs ac s we e sc eened o de e mine which a icles we e eligible o ull- ex sc eening. Two au ho s (IE and MB) e iewed he ull- ex o all po en ially ele an s udies o de e mine final inclusion. Da a ex ac ion p ocess A da a ex ac ion o m was designed o ex ac ele- an da a om he selec ed s udies. Da a ex ac ion was fi s pe o med by one e iewe (MB) wi h subse- quen quali y assu ance pe o med by he second e- iewe (IE). Disag eemen s we e esol ed h ough discussion wi h a hi d e iewe (MA). Analysis o s udy cha ac e is ics We e ie ed in o ma ion abou he objec i e o he s udies, he exposu e, he ou come, he s udy design, and he numbe o di e en da abases and coun ies. Fu he , we classified s udies acco ding o he me hods ha we e used o indi idual-le el analyses and me a- analyses. Rega ding he me a-analyses, we defined h ee di e en le els o combining da a: (1) An agg ega e le el app oach, in which sepa a e analyses a e pe o med on da ase s om each da abase and o e all esul s (adjus ed e ec es ima es wi h confidence in e als) a e collec ed o me a-analysis. The analyses a e usually ‘da abase-op imized’in he sense ha he bes a ailable da a o each da abase a e being used. This app oach allows using he no mal s a is ical echniques o me a-analysis, including andom- e ec s models, o accoun o he e ogenei y o s udy esul s. Fu he , me a- eg ession can be used o assessing a ia ion in e ec s ela ed o co a i- a es, which may explain some o he o e all he e ogenei y. (2) A semi-agg ega e le el app oach, in which s a i- fied da ase s wi h e en coun s (and o coho s udies pe son ime) a e collec ed om all da a- bases o one common analysis (e.g. a dis ibu ed da a ne wo k). Da ase s can be s a ified on ou - come, exposu e, and co a ia e pa e ns (age, sex, ime since ini ia ion, and selec ed con ounde s). Fo a coho s udy, his app oach employs a Poisson eg ession model on ables o e en num- be s and pe son ime s a ified by exposu e and co a ia e pa e ns. Fo a case–con ol s udy, a lo- gis ic eg ession model can simila ly be used o analyze equency ables o cases and con ols s a ified by di e en co a ia es. (3) An indi idual le el app oach, in which indi idual pa ien da a a e collec ed om all da abases o one common analysis. In his scena io, he in o - ma ion om di e en da abases has o be made compa ible wi h ega d o defini ions o expo- su es, ou comes, co a ia es, and ime windows. He e ogenei y o s udy popula ions and a ia ion in he e ec s bo h o exposu e and co a ia es be- ween da abases may a ec he esul s. This may be accoun ed o using s a is ical echniques co ec ing o o e all a ia ion wi hin and be- ween da abases. Wi h ega d o da a managemen , we we e in e es ed in which da a we e collec ed cen ally and whe he cen al p og amming o he use o dis ibu ed common p og ams was men ioned in he a icle. Mo eo e , mo i a ions o conduc ing he s udy (as s a ed in he a icle) we e classified. Fo all ques ions he e was a ca ego y ‘no specified’ ha could be icked. This ca ego y is only shown in he esul ables i he e was a leas one s udy wi h missing da a. F equen- cies we e calcula ed o desc ibe he s udy cha ac e - is ics, da a analysis echniques, da a managemen , and mo i a ions. RESULTS The PubMed and Embase sea ches iden ified 3083 publica ions (see Figu e 1). A e he sc eening o i- les and abs ac s, 44 a icles we e selec ed o a ull- ex sc eening. The manual sea ch iden ified an addi- ional 176 publica ions; 22 om OMOP, 64 om Mini-Sen inel, 34 om he HMO Resea ch Ne wo k (based on a Pubmed sea ch, because no a icles we e ound on he p ojec ’s websi e), 36 om EU-ADR, 11 om IMI-PROTECT, 2 om AsPEN, 2 om CNODES, and 5 om indi idual p ojec s (da a no shown in he flowcha ). A e e iew o i les and ab- s ac s we selec ed 31 a icles o a ull- ex sc eening. Figu e 1 shows ha om he 75 a icles selec ed o ull- ex sc eening, 22 we e finally selec ed o inclusion. mul i-da abase s udies: a sys ema ic li e a u e e iew 899 © 2015 The Au ho s. Pha macoepidemiology and D ug Sa e y published by John Wiley & Sons, L d. Pha macoepidemiology and D ug Sa e y, 2015; 24: 897–905 DOI: 10.1002/pds E- able 3 gi es an o e iew o he 22 s udies in- cluded in ou e iew. 9–30 Six s udies we e published be ween 2007 and 2010, while a majo i y o 16 s udies was published be ween 2011 and 2013. Fo se en s udies, he e was a me hod objec i e in addi ion o an o e all (mos ly clinical) objec i e. 10,13,15,17,18,23,27 Figu e 2 shows which coun ies con ibu ed o he mul i-da abase p ojec s. F om he 22 s udies, he e we e 14 (64%) ha used a leas one da abase om he U.S. o Canada. Hal o he s udies (50%) used a leas one Eu opean da abase. Wi hin Eu ope, da a om G ea B i ain we e mos o en used: he e we e nine s udies (41%) ha used a B i ish da abase (wi h he Clinical P ac ice Resea ch Da alink (CPRD) being he mos equen ly used one, i.e. fi e imes). Fu he , he e we e fi e s udies ha used a leas one Scandina ian da abase (23%), fi e s udies ha used a da abase om he Ne he lands (23%), and fi e s udies ha used an I alian da abase (23%). Table 1 shows ha mos s udies add essed sa e y (82%) a he han e ec i eness issues (23%). The de- sign ha was mos o en used was he coho s udy de- sign (82%) as opposed o he case–con ol design (18%). The numbe o da abases used pe s udy anged om 2 o 17 (median = 4.0), while he numbe o coun- ies in ol ed anged om 1 o 6 (median = 2.5). The ype o exposu e ha was mos equen ly s udied was medica ion ela ed o he ne ous sys em (41%), o example an idep essan s and dopamine agonis s. In e ms o ou comes, ca diac diso de s we e mos e- quen ly ep esen ed (36%), ollowed by all-cause mo - ali y (23%) and ne ous sys em diso de s (18%). The Figu e 1. Flowcha o he selec ion o a icles. * No pape s we e excluded because hey we e non-English ( he e we e 122 non-English pape s bu hey we e excluded because o o he exclusion c i e ia) m. . bazelie e al. 900 © 2015 The Au ho s. Pha macoepidemiology and D ug Sa e y published by John Wiley & Sons, L d. Pha macoepidemiology and D ug Sa e y, 2015; 24: 897–905 DOI: 10.1002/pds code sys em used o iden i y exposu e was o en no specified (82%), while o he ou come his was less o en he case (18%). Table 2 shows ha logis ic eg ession was mos e- quen ly used o indi idual-le el analyses (41%), ollowed by Cox eg ession (23%) and Poisson eg es- sion (14%). As me a-analysis me hod, a majo i y o he s udies used an indi idual le el app oach (73%) (see e- able 4 o he exac ca ego iza ion). The e we e h ee s udies wi h a semi-agg ega e app oach (14%), and six s udies pe o med an agg ega e me a-analysis (27%). Fo wo s udies, he agg ega e me a-analysis was he only me a-analysis ha was conduc ed (be- cause da a we e collec ed on an agg ega e le el). The o he ou agg ega e me a-analyses we e conduc ed in addi ion o an indi idual pa ien da a me a-analysis ( h ee s udies) o a semi-agg ega e me a-analysis (one s udy). The e we e ou s udies ha collec ed semi-agg ega e da ase s (18%), bu one o hem did no use his in o ma ion in a me a-analysis ( eflec edin heca ego y ‘me a-analysis: none’). A quan i a i e he e ogenei y as- sessmen was conduc ed in almos hal o he s udies (45%) wi hou one specific es being mos popula . Fo 12 s udies, cen al p og amming was men ioned in he publica ion. The use o dis ibu ed common p og ams was men ioned o fi e s udies; in h ee cases hese common p og ams we e used in a semi- agg ega e le el app oach, and he o he wo cases we e ela ed o an agg ega e le el app oach. A majo i y o he a icles mo i a ed hei mul i-da abase s udy by powe o men ioned powe as a s eng h o hei s udy (86%) (da a no shown in a able). Only one a icle explici ly men ioned a e exposu e as an a gumen , and h ee a icles made men ion o a a e ou come. Ex e nal alidi y was s essed in six publica ions (27%), while only h ee a icles men ioned compa ing di e en popula ions o da abases as an a gumen o pe o m a mul i-da abase p ojec (14%). Figu e 2. Map o coun ies in ol ed in mul i-da abase s udies. Legend: - he numbe s eflec how many imes he coun y (a leas one da abase) was in- ol ed in a mul i-da abase s udy ( he da ke he colo , he highe he numbe o s udies) - s ipes: coun y no in ol ed in any mul i-da abase s udy mul i-da abase s udies: a sys ema ic li e a u e e iew 901 © 2015 The Au ho s. Pha macoepidemiology and D ug Sa e y published by John Wiley & Sons, L d. Pha macoepidemiology and D ug Sa e y, 2015; 24: 897–905 DOI: 10.1002/pds DISCUSSION Ou sys ema ic li e a u e sea ch iden ified 22 pha macoepidemiological mul i-da abase s udies, in which da a we e combined o a p e-planned common analysis o quan i a i e syn hesis. Fo indi idual-le el analyses, logis ic eg ession was mos equen ly used, ollowed by Cox and Poisson eg ession. Fo me a- analyses, 16 s udies combined indi idual pa ien da a, while a semi-agg ega e le el analysis was conduc ed in h ee s udies and an agg ega e le el analysis in six. I was a challenge o cap u e mul i-da abase obse - a ional s udies in a sys ema ic li e a u e sea ch. Usu- ally, he clinical opic was well ep esen ed in sea ch e ms o keywo ds, bu he use o mo e han one inde- penden da abase was much ha de o cap u e. The e is a Mesh e m called ‘Mul icen e s udy’, bu his was on he one hand oo b oad (e.g. also in ol ing s udies ha selec ed hei pa ien s om wo di e en hospi als, which was no he defini ion o mul i-da abase s udy we we e looking o ) and on he o he hand missing ou on ele an a icles. Among s udies ha used mo e han one da abase, i was almos impossible o dis in- guish s udies wi h pooled es ima es om s udies ha only showed esul s om he sepa a e da abases, using sea ch e ms. Because o hese issues, we we e qui e libe al in defining he sea ch s a egy and manually sc eened o e 3000 a icles, which only esul ed in 44 a icles ha we e selec ed o a ull- ex sc eening. S ill, we missed ou on 31 a icles ha we e ound in he addi ional manual sea ch. Among he s udies finally selec ed o inclusion, he ange o di e en exposu es and ou comes was qui e b oad. This indica es ha he upcoming end o pe o ming mul i-da abase obse a ional s udies s e ches ou o e he en i e field o pha macoepidemiology. App oxima ely hal o ou selec ed s udies used a leas one da abase om he U.S. o Canada, and his pe cen age was simila o Eu opean da abases. This indica es ha No he n Ame ica and Eu ope cu en ly ake an almos equal pa in con ibu ing o his ela- i ely new field o mul i-da abase esea ch. Rega ding he me hods o pe o ming such a s udy, we ound ha he combina ion o indi idual pa ien da a was he mos equen ly used echnique. I should howe e be no ed ha one o ou inclusion c i e ia was ha da a we e combined o a p e-planned common analysis. The eby we excluded all ‘s anda d’me a- analyses ha pool es ima es om di e en s udies o- ge he on a pos -hoc basis, i.e. he app oach o com- bining esul s om published li e a u e. This la e ype o me a-analysis is equen ly used in he field o clinical ials and is p obably also qui e common Table 1. Objec i e, design, exposu e, and ou come Numbe o s udies ( o al: n = 22) % Objec i e ca ego y E idence gene a ion 20 91% Me hod / easibili y s udy 15% Dual pu pose (combina ion o he abo e)15% S udy ype (a) Sa e y 18 82% E ec i eness 5 23% Design ca ego y Coho s udy 18 82% Case–con ol s udy 4 18% Numbe o da abases Range [2,17] Mean 5.9 Median 4.0 Numbe o coun ies Range [1,6] Mean 2.4 Median 2.5 D ug exposu e: ATC ca ego y N—Ne ous sys em 9 41% A—Alimen a y ac and me abolism 4 18% C—Ca dio ascula sys em 3 14% M—Musculo-skele al sys em 3 14% B—Blood and blood o ming o gans 15% R—Respi a o y sys em 15% S—Senso y o gans 15% Ou come: All-cause mo ali y/MedDRA SOC (a) Ca diac diso de s 8 36% All-cause mo ali y 5 23% Ne ous sys em diso de s 4 18% Congeni al, amilial and gene ic diso de s 3 14% Gas oin es inal diso de s 29% Musculoskele al and connec i e issue diso de s 29% Blood and lympha ic sys em diso de s 15% Eye diso de s 15% In ec ions and in es a ions 15% P egnancy,pue pe ium,and pe ina al condi ions 15% Renal and u ina y diso de s 15% Respi a o y, ho acic and medias inal diso de s 15% Skin and subcu aneous issue diso de s 15% D ug code sys ems (a) ATC 4 18% BNF 29% No specified 18 82% Ou come code sys ems (a) ICD-9 13 59% ICD-10 8 36% ICPC 29% RCD 29% CCP 15% CPT 15% mRS 15% No specified 4 18% (a) One s udy could con ibu e o mo e han one ca ego y. Abb e ia ions: ATC, Ana omical The apeu ic Chemical; BNF, B i ish Na- ional Fo mula y; CCP, Canadian Classifica ion o Diagnos ic, The apeu ic and Su gical P ocedu es; CPT, Cu en P ocedu al Te minology; ICD-9, In e na ional Classifica ion o Diseases—9 h e ision; ICD-10, In e na- ional Classifica ion o Diseases—10 h e ision; ICPC, In e na ional Clas- sifica ion o P ima y Ca e; mRS, modified Rankin Scale; RCD, READ CODE Classifica ion; SOC, sys em o gan class. m. . bazelie e al. 902 © 2015 The Au ho s. Pha macoepidemiology and D ug Sa e y published by John Wiley & Sons, L d. Pha macoepidemiology and D ug Sa e y, 2015; 24: 897–905 DOI: 10.1002/pds in he field o obse a ional s udies, as i can be done ela i ely quickly wi hou he need o ge ing access o pa ien -le el heal hca e da a. I his ype o me a- analysis would ha e been included in ou e iew, he p opo ion o agg ega e me a-analyses would he e o e ha e been much highe . The e we e h ee s udies ha compa ed an indi id- ual le el me a-analysis o an agg ega e le el (fixed o andom e ec ) me a-analysis, using he same da a. 10,13,18 In all h ee cases, he esul s we e simila be ween he wo app oaches. Impo an ly, hese agg ega e le el analyses we e designed wi h a p e- specified common analysis plan, i.e. he same plan as was used o he indi idual le el analyses, bu now wi hou combining he indi idual pa ien da a bu pooling da abase-specific es ima es oge he . F om a logis ic poin o iew, his ype o agg ega e me a- analysis could be an in e es ing al e na i e o an indi- idual le el app oach, as no indi idual pa ien da a ha e o be ans e ed. O he me hods ha do no e- qui e sha ing o indi idual pa ien da a in ol e a dis- ibu ed ne wo k app oach such as he EU-ADR 15 o a case-cen e ed logis ic eg ession app oach as de- sc ibed by Toh and cowo ke s. 27,31 I was no always clea ly s a ed in he a icles how he da a we e combined. Especially he echnique o combining indi idual pa ien da a (ins ead o combin- ing semi-agg ega e o agg ega e da a) was no always explici ly men ioned. Mos o he a icles clea ly had a clinical ocus, and desc ip ions o da a managemen we e o en e y sho o comple ely missing. Cen al p og amming was men ioned o 12 s udies, bu he ac ual numbe o s udies ha did so is p obably highe , because his echnique is e y likely o he s udies ha combined indi idual pa ien da a (16 in o al). Howe e , he use o dis ibu ed common p og ams was equen ly men ioned o semi-agg ega e and ag- g ega e le el app oaches. This is a good p ac ice, be- cause di e en ways o s a is ical p og amming may lead o he e ogenei y be ween esul s om di e en da abases. In he indi idual le el app oach, i was o - en no clea ly desc ibed i and how defini ions we e kep simila be ween da a om di e en sou ces. Powe was he mos s a ed eason o pe o ming a mul i-da abase s udy. Compa ing di e en popula ions o da abases was seldom men ioned. Many pape s, es- pecially he indi idual pa ien da a s udies, did no show any cha ac e is ics o pa ien s om he sepa a e da abases. O e all, only hal o he s udies pe o med a quan i a i e he e ogenei y assessmen . The e a e impo an s eng hs and limi a ions o his e iew. To ou knowledge, we we e he fi s o pe - o m a sys ema ic li e a u e sea ch abou me hods used in mul i-da abase s udies. Full- ex sc eening was pe - o med by wo independen e iewe s, and da a ex ac- ion was quali y checked by a second e iewe as well. E en hough we pe o med a sys ema ic li e a u e sea ch, we p obably missed ele an a icles, because he o mula ion o a sui able sea ch s a egy was no s aigh o wa d. Ou sea ches we e limi ed om 2007 onwa ds; howe e , in ou pilo sea ches wi h no da e fil e applied we we e no able o iden i y any s udies published p io o 2007 ha would po en- ially fi ou inclusion and exclusion c i e ia. Table 2. Da a analysis echniques and da a managemen Numbe o s udies ( o al: n = 22) % Indi idual-le el analyses Logis ic eg ession 9 41% Cox p opo ional haza ds model 5 23% Poisson eg ession 3 14% Incidence a e /incidence a e a io 29% P e alence /p e alence a io 15% Rela i e isk 15% Gene alized linea model eg ession 15% Exposu e– ime ela ion Time-dependen exposu e 14 64% In en ion o ea (e e /ne e ) 7 32% Cumula i e exposu e (dose o ime)15% Con ounde con ol Con en ional 11 50% P opensi y sco e 7 32% Disease isk sco e 15% None 3 14% Me a-analysis me hod (a) Indi idual 16 73% Semi-agg ega e 3 14% Agg ega e 6 27% Fixed e ec 4 18% Fixed e ec / andom e ec 29% None 15% He e ogenei y assessmen Quan i a i e es 10 45% I-squa ed 29% Chi-squa ed 15% Coch an’s Q s a is ic 15% In e ac ion by da a sou ce 15% Kaplan–Meie s a ified by da abase 15% No specified 4 18% Quali a i e s a emen s only 15% No specified 11 50% P og amming Cen al (leading cen e ) 12 55% Decen al 00% No specified 10 45% Da a collec ed cen ally Indi idual-based egis e da a 16 73% Semi-agg ega e da ase s 4 18% Agg ega e esul s 29% Dis ibu ed common p og ams Yes 5 23% No specified 17 77% (a) One s udy could con ibu e o mo e han one ca ego y. mul i-da abase s udies: a sys ema ic li e a u e e iew 903 © 2015 The Au ho s. Pha macoepidemiology and D ug Sa e y published by John Wiley & Sons, L d. Pha macoepidemiology and D ug Sa e y, 2015; 24: 897–905 DOI: 10.1002/pds Classifica ion in o he di e en le els o da a combin- ing was done o he e iewe s’bes e o , bu was some imes based on e y li le in o ma ion. Classifica- ion o mo i a ions o pe o m he s udy may ha e been suscep ible o subjec i e in e p e a ion. In conclusion, mul i-da abase s udies a e becoming mo e popula in obse a ional esea ch. We eel ha he e is oom o imp o emen in making clea o he eade how da a om di e en da abases we e com- bined; on an indi idual, semi-agg ega e, o agg ega e le el. Fo all scena ios, i is use ul o know how defi- ni ions o exposu es, ou comes, con ounde s, and ime-windows we e kep consis en ac oss he da a- bases. Fu he , i should be explained how da a man- agemen was o ganized, which da a we e collec ed cen ally and whe he cen al p og amming o dis ib- u ed common p og ams we e used. I is use ul o show cha ac e is ics o pa ien s om he sepa a e da abases o enable he eade o e alua e whe he he e we e impo an di e ences. When combining da a om di e en da abases, he pe o mance o he e ogenei y assessmen s should become common p ac ice. E en i he objec i e o a s udy is clinical a he han me h- odological, all his in o ma ion enables a be e in e - p e a ion o he esul s o a mul i-da abase s udy. CONFLICT OF INTEREST Ma loes T. Bazelie ’s employmen a U ech Uni e - si y is unded by he CARING p ojec (Eu opean Communi y’s Se en h F amewo k P og am g an ag eemen numbe 282526). F ank de V ies is employed by U ech Uni e si y as a senio esea che , conduc ing esea ch coo dina ed by The Cen e o Resea ch Me hods. The Cen e o Resea ch Me hods has ecei ed un es ic ed unding om he Ne he lands O ganiza ion o Heal h Re- sea ch and De elopmen (ZonMW), he Du ch Heal h Ca e Insu ance Boa d (CVZ), he Royal Du ch Pha - macis s Associa ion (KNMP), he p i a e–public unded Top Ins i u e Pha ma (www. ipha ma.nl, in- cludes co- unding om uni e si ies, go e nmen , and indus y), he EU Inno a i e Medicines Ini ia i e (IMI), he EU 7 h F amewo k P og am (FP7), and he Du ch Minis y o Heal h and indus y (including GlaxoSmi hKline, Pfize , and o he s). Ma ie L. De B uin is employed by U ech Uni e - si y as a senio esea che conduc ing esea ch in col- labo a ion wi h he WHO Collabo a ing Cen e o pha maceu ical policy and egula ion. This Cen e e- cei es no di ec unding o dona ions om p i a e pa ies, including pha ma indus y. Resea ch unding om public–p i a e pa ne ships, e.g. IMI, TI Pha ma (www. ipha ma.nl) is accep ed unde he condi ion ha no company-specific p oduc o company ela ed s udy is conduc ed. The Cen e has ecei ed un e- s ic ed esea ch unding om public sou ces, e.g. Ne he lands O ganiza ion o Heal h Resea ch and De- elopmen (ZonMW), he Du ch Heal h Ca e Insu - ance Boa d (CVZ), EU 7 h F amewo k P og am (FP7), Du ch Medicines E alua ion Boa d (MEB), and Du ch Minis y o Heal h. Mo en Ande sen pa icipa es/has pa icipa ed in e- sea ch p ojec s unded by As aZeneca, Lundbeck, Me ck Sha p & Dohme, No a is, Nycomed, and Pfize wi h g an s ecei ed by he ins i u ions whe e he has been employed. He has pe sonally ecei ed ees o leading and eaching pha macoepidemiology cou ses a anged by Medicademy, he Danish Associ- a ion o he Pha maceu ical Indus y. KEY POINTS •The upcoming end o pe o ming mul i- da abase obse a ional s udies s e ches ou o e he en i e field o pha macoepidemiology. •Rega ding he me hods o pe o ming such a s udy, we ound ha he combina ion o indi idual pa ien da a was he mos equen ly used echnique. •The e is oom o imp o emen in making clea o eade s how da a om di e en da abases we e combined (on an indi idual le el o agg ega e le el), how da a managemen was o ganized, and whe he cen al p og amming was used. •I is use ul o show cha ac e is ics o pa ien s om he sepa a e da abases and he pe o mance o he e ogenei y assessmen s should become common p ac ice. ETHICS STATEMENT The au ho s confi m o ha e adhe ed o E hics p inci- ples du ing all phases o he s udy. ACKNOWLEDGEMENTS The esea ch leading o he esul s o his s udy has ecei ed unding om he Eu opean Communi y’s Se en h F amewo k P og amme (FP-7) unde g an ag eemen numbe 282526, he CARING p ojec . The unding sou ce had no ole in s udy design, da a collec ion, da a analysis, da a in e p e a ion, o w i ing o he epo . m. . bazelie e al. 904 © 2015 The Au ho s. Pha macoepidemiology and D ug Sa e y published by John Wiley & Sons, L d. Pha macoepidemiology and D ug Sa e y, 2015; 24: 897–905 DOI: 10.1002/pds REFERENCES 1. The obse a ional medical ou comes pa ne ship. 2009; A ailable a : h p:// www.omop. nih.o g [14 Ap il 2015]. 2. The mini-sen inel p og am. 2010; A ailable a : h p://www.mini-sen inel.o g [14 Ap il 2015]. 3. The HMO esea ch ne wo k. 2013; A ailable a : h p://www.hmo esea ch- ne wo k.o g [14 Ap il 2015]. 4. Suissa S, Hen y D, Cae ano P, e al. Canadian Ne wo k o Obse a ional D ug E ec S udies (CNODES). CNODES: he Canadian Ne wo k o Obse a ional D ug E ec S udies. Open Med 2012; 6(4): e134–e140. 5. The Canadian Ne wo k o Obse a ional D ug E ec S udies. 2013; A ailable a : h p://www.cnodes.ca [14 Ap il 2015]. 6. The EU-ADR p ojec . (2008) A ailable a : h p://euad -p ojec .o g [14 Ap il 2015]. 7. The Pha macoepidemiological Resea ch on Ou comes o The apeu ics by a Eu opean Conso ium (PROTECT). 2009; A ailable a : h p://www.imi-p o- ec .eu [14 Ap il 2015]. 8. AsPEN collabo a o s, Ande sen M, Be gman U, e al. The Asian Pha macoepidemiology Ne wo k (AsPEN): p omo ing mul i-na ional collabo a- ion o pha macoepidemiologic esea ch in Asia. Pha macoepidemiol D ug Sa 2013; 22(7): 700–704. 9. T ifi ò G, Mokhles MM, Dieleman JP, e al. Risk o ca diac al e egu gi a ion wi h dopamine agonis use in Pa kinson’s disease and hype p olac inaemia: a mul i-coun y, nes ed case–con ol s udy. D ug Sa 2012; 35(2): 159–171. 10. Mokhles MM, T ifi ò G, Dieleman JP, e al. The isk o new onse hea ailu e associa ed wi h dopamine agonis use in Pa kinson’s disease. Pha macol Res 2012; 65(3): 358–364. 11. Kiele H, A ama M, Engeland A, e al. Selec i e se o onin eup ake inhibi o s du ing p egnancy and isk o pe sis en pulmona y hype ension in he newbo n: popula ion based coho s udy om he fi e No dic coun ies. BMJ 2012; 344: d8012. 12. S ephansson O, Kiele H, Haglund B, e al. Selec i e se o onin eup ake inhibi- o s du ing p egnancy and isk o s illbi h and in an mo ali y. JAMA 2013; 309(1): 48–54. 13. Bazelie MT, de V ies F, Ves e gaa d P, e al. Risk o ac u e wi h hiazolidinediones: an indi idual pa ien da a me a-analysis. F on Endoc inol (Lausanne) 2013; 4(11): 1–9. 14. And ade SE, McPhillips H, Lo en D, e al. An idep essan medica ion use and isk o pe sis en pulmona y hype ension o he newbo n. Pha macoepidemiol D ug Sa 2009; 18(3): 246–252. 15. Coloma PM, Schuemie MJ, T ifi ò G, e al. Combining elec onic heal hca e da- abases in Eu ope o allow o la ge-scale d ug sa e y moni o ing: he EU-ADR P ojec . Pha macoepidemiol D ug Sa 2011; 20(1): 1–11. 16. Da is RL, Rubanowice D, McPhillips H, e al. Resea ch in The apeu ics. Risks o congeni al mal o ma ions and pe ina al e en s among in an s exposed o an i- dep essan medica ions du ing p egnancy. Pha macoepidemiol D ug Sa 2007; 16(10): 1086–1094. 17. Do mu h CR, Hemmelga n BR, Pa e son JM, e al. Use o high po ency s a ins and a es o admission o acu e kidney inju y: mul icen e , e ospec i e obse - a ional analysis o adminis a i e da abases. BMJ 2013; 346: 880. 18. Engel e ST, Soinne L, Ringleb P, e al. IV h ombolysis and s a ins. Neu ology 2011; 77(9): 888–895. 19. Filion KB, Cha eau D, Ta gownik LE, e al. P o on pump inhibi o s and he isk o hospi alisa ion o communi y-acqui ed pneumonia: eplica ed coho s udies wi h me a-analysis. Gu 2014; 63(4): 552–558. 20. Habel LA, Coope WO, Sox CM, e al. ADHD medica ions and isk o se ious ca dio ascula e en s in young and middle-aged adul s. JAMA 2011; 306(24): 2673–2683. 21. Hagiwa a M, Delea TE, S an o d RH, S empel DA. S epping down o flu icasone p opiona e o a lowe dose o flu icasone p opiona e/salme e ol combina ion in as hma pa ien s ecen ly ini ia ing combina ion he apy. Alle gy As hma P oc 2010; 31(3): 203–210. 22. Henk HJ, Tei elbaum A, Pe ez JR, Kau a S. Pe sis ency wi h zoled onic acid is associa ed wi h clinical benefi in pa ien s wi h mul iple myeloma. Am J Hema ol 2012; 87(5): 490–495. 23. Rassen JA, Choudh y NK, A o n J, Schneeweiss S. Ca dio ascula ou comes and mo ali y in pa ien s using clopidog el wi h p o on pump inhibi o s a e pe - cu aneous co ona y in e en ion o acu e co ona y synd ome. Ci cula ion 2009; 120(23): 2322–2329. 24. Ray WA, Va as-Lo enzo C, Chung CP, e al. Ca dio ascula isks o nons e oi- dal an iinflamma o y d ugs in pa ien s a e hospi aliza ion o se ious co ona y hea disease. Ci c Ca dio asc Qual Ou comes 2009; 2(3): 155–163. 25. Schelleman H, Bilke WB, Kimmel SE, e al. Me hylphenida e and isk o se ious ca dio ascula e en s in adul s. Am J Psychia y 2012; 169(2): 178–185. 26. Schwa z GF, Ko ak S, Ma dekian J, Fain JM. Incidence o new coding o d y eye and ocula in ec ion in open-angle glaucoma and ocula hype ension pa- ien s ea ed wi h p os aglandin analogs: e ospec i e analysis o h ee medical/pha macy claims da abases. BMC Oph halmol 2011; 11: 14. 27. Toh S, Reichman ME, Hous oun M, e al. Compa a i e isk o angioedema as- socia ed wi h he use o d ugs ha a ge he enin–angio ensin–aldos e one sys- em. A ch In e n Med 2012; 172(20): 1582–1589. 28. Toh S, Bake MA, B own JS, Ko negay C, Pla R, Mini-Sen inel In es iga o s. Rapid assessmen o ca dio ascula isk among use s o smoking cessa ion d ugs wi hin he US Food and D ug Adminis a ion’s Mini-Sen inel p og am. JAMA In e n Med 2013; 173(9): 817–819. 29. Tsai TT, Ho PM, Xu S, e al. Inc eased isk o bleeding in pa ien s on clopidog el he apy a e d ug-elu ing s en s implan a ion: insigh s om he HMO Resea ch Ne wo k-S en Regis y (HMORN-s en ). Ci c Ca dio asc In e 2010; 3(3): 230–235. 30. an Soes EM, Valkho VE, Mazzaglia G, e al. Subop imal gas op o ec i e co e age o NSAID use and he isk o uppe gas oin es inal bleeding and ul- ce s: an obse a ional s udy using h ee Eu opean da abases. Gu 2011; 60(12): 1650–1659. 31. Toh S, Reichman ME, Hous oun M, e al. Mul i a iable con ounding adjus men in dis ibu ed da a ne wo ks wi hou sha ing o pa ien -le el da a. Pha macoepidemiol D ug Sa 2013; 22(11): 1171–1177. SUPPORTING INFORMATION Addi ional suppo ing in o ma ion may be ound in he online e sion o his a icle a he publishe ’s web si e. mul i-da abase s udies: a sys ema ic li e a u e e iew 905 © 2015 The Au ho s. Pha macoepidemiology and D ug Sa e y published by John Wiley & Sons, L d. Pha macoepidemiology and D ug Sa e y, 2015; 24: 897–905 DOI: 10.1002/pds