Drug-safety alerts issued by regulatory authorities: usefulness of meta-analysis in predicting risks...

PHARMACOEPIDEMIOLOGYAND PRESCRIPTION

Drug-safety alerts issued by regulatory authorities: usefulnessof meta-analysis in predicting risks earlier

Carlos Alves & Francisco Batel Marques &Ana Filipa Macedo

Received: 14 November 2013 /Accepted: 18 March 2014 /Published online: 3 April 2014# Springer-Verlag Berlin Heidelberg 2014

AbstractPurpose The purpose of this study was to evaluate how riskestimates generated from cumulative meta-analysis performsover time for drugs having their benefit/risk ratio re-evaluateddue to safety issues and, additionally, assess whether resultsare consistent with regulatory authorities’ conclusions.Methods Four major regulatory authorities were searched fortheir issued safety alerts supported by longitudinal, compara-tive studies (experimentals and/or observationals). Therandom-effects model was used to pooled odds ratios (OR)over time by including studies according to the year they firstbecame available.Results Seventeen safety alerts were included in this study. In2008, proton-pump inhibitors (PPIs) were associated with anincreased risk for bone fractures [OR 1.25, 95 % confidenceinterval (CI) 1.00–1.55, P=0.049); the US Food and DrugAssociation (FDA) issued a safety alert in 2010 and added

warnings to the label. An increased risk for Clostridium-difficile-associated diarrhea was pooled for PPIs in 2004(OR 1.89, 1.19–3.02, P=0.007); US FDA issued a safety alertin 2012, adding warnings to the label. PPIs were associatedwith pneumonia in 2009 (OR 1.40, 1.06–1.85, P=0.017); USFDA issued an alert in 2012 but concluded that the benefit/risk(B/R) ratio remains positive. Statins were associated with anincreased risk for diabetes (OR 1.07, 1.01–1.15, P=0.033) in2008. The EuropeanMedicines Agency (EMA) issued an alertin 2012, including warnings to the label. The remainingcumulative meta-analyses did not estimate increased risks inadvance of regulatory decisions.Conclusion This study demonstrates that meta-analysis mayhelp predict iatrogenic risks. However, between-study hetero-geneity can considerably affect the estimated results, andtherefore, this technique should not replace further assess-ments during BR ratio re-evaluations.

Keywords Meta-analysis . Safety alerts . Data sources .

Regulatory agencies

Introduction

After a medicine is issued a market authorization and becomesavailable, unknown adverse reactions (ADRs) can arise fromits everyday application [1]. This additional knowledge of thesafety profile deserves to be carefully evaluated for patientprotection [1]. Some ADRs are serious enough to change thebenefit/risk (B/R) profile of a particular drug, leading torestriction on its use or even to withdrawal from the market[2]. In order to keep patients and healthcare professionalsupdated, authorities frequently issue drug-safety alerts regard-ing B/R ratio re-evaluations being conducted and subsequentregulatory decisions [3, 4]. Randomized controlled trials(RCTs) provide the best design for evaluating the efficacy of

Electronic supplementary material The online version of this article(doi:10.1007/s00228-014-1670-5) contains supplementary material,which is available to authorized users.

C. Alves (*) : F. B. MarquesCentral Portugal Regional Pharmacovigilance Unit (UFC), Centrefor Health Technology Assessment and Drug Research (CHAD),AIBILI – Association for Innovation and Biomedical Research onLight and Image Azinhaga de Santa Comba, Celas,3000-548 Coimbra, Portugale-mail: [email protected]

F. B. Marquese-mail: [email protected]

C. Alves : F. B. MarquesSchool of Pharmacy, University of Coimbra, Coimbra, Portugal

C. Alves :A. F. MacedoHealth Sciences Research Centre, University of Beira Interior,Covilhã, Portugal

A. F. Macedoe-mail: [email protected]

Eur J Clin Pharmacol (2014) 70:745–756DOI 10.1007/s00228-014-1670-5

http://dx.doi.org/10.1007/s00228-014-1670-5

a drug and its most common adverse effects [5, 6]. However,not all harmful effects can be easily anticipated in RCTs, andeven if identified, their reporting is usually inadequate.Observational studies usually support regulatory decisionson rare and/or long-latency adverse events (AEs), such asfractures, cardiovascular events, or malignancies [7].

Different types of epidemiological data supportpharmacovigilance activities, and its collection and evaluationare crucial steps for regulatory authorities in order to establishthe most accurate B/R ratio [8, 9]. Postmarketing drug-safetysurveillance can be considered a dynamic prospective processthat provides timely assessment of drug risks, as well as higher-quality and better-documented scientific evidence [1].Therefore, a considerable period of time can elapse betweenevidence supporting the association of drug exposure and anew AE that will eventually lead to a decision from regulatoryauthorities. Meta-analysis provides the conceptual and quanti-tative framework for rigorous literature review, combiningeffect measures when appropriate, and allowing an objectivepresentation and analysis of available data [10]. This techniquehas been commonly used to pool data from RCTs primarily toevaluate efficacy endpoints. [11]. Although not frequently usedto evaluate safety issues, analysis of evidence from cumulativemeta-analyses has demonstrated that more highly appropriateand timely decisions could have been made concerning car-diovascular events associated with rofecoxib [12].

This study is aimed to evaluate how risk estimates gener-ated from cumulative meta-analyses performs over time fordrugs for which the B/R ratio was re-evaluated due to safetyissues and, additionally, to assess whether the results areconsistent with regulatory authorities’ conclusions.

Methods

Selecting safety alerts

A previous study reviewed the type and publication status ofdata sources supporting B/R ratio re-evaluations conducted bythe US Food and Drug Administration (US FDA), HealthCanada, the European Medicines Agency (EMA), and theAustralian Therapeutics Goods Administration (TGA) [13].A total of 59 safety alerts were evaluated. Only safety alertsregarding the evaluation of a causal relation between exposureto a suspected drug and the occurrence of an AE, which wereissued for the first time between January 2010 and December2012, were considered for inclusion. Natural and healthcareproducts, medical devices, contrast agents, drug–drug interac-tions, drug–food interactions, medication errors, evaluationsof lack of efficacy, and AEs occurring during off-label usewere not considered for inclusion. Only safety alertsconcerning drugs with market authorization and simulta-neously included in one of the 30most prescribed drug classes

worldwide used in the ambulatory setting were considered forinclusion. The complete methodology by which safety alertswere selected is described in a previous study [13].

This study only included safety alerts in which regulatoryauthorities’ decisions were supported by longitudinal, com-parative studies [RCTs and/or observational studies (cohort orcase–control)]. Studies included in meta-analyses used tosupport regulatory decisions were retrieved and pooled inthe respective cumulative meta-analyses. No further biblio-graphic references were requested from regulatory authoritiesbeyond those published on their websites. Bibliographic elec-tronic searches were not conducted.

The following information from each safety alert wasextracted:

1. Date of first publication2. Regulatory authority issuing the safety alert3. Suspected drug(s)4. AE5. Type of studies supporting the evaluation6. Updated drug label; “Benefit/risk ratio unchanged” was

considered when no change was made.

Updates of the same safety alert were reviewed in order toretrieve further information. Two safety alerts were consideredto be evaluating the same clinical issue when they assessed thesame AE for the same suspected drug(s). Clinical questionrefers to the investigational hypothesis evaluated by a regula-tory authority. Regulatory authorities may have decided tokeep the suspected drug(s) under revision despite labelingchanges having been made.

Meta-analysis

For each clinical question, a cumulative meta-analysis wasperformed to determine pooled evidence over time. In thecumulative meta-analysis, studies were included accordingto the year they first became available: i.e., the earliest ofonline publication date (Epub ahead of print date) or thecorrespondent journal issue publication date. Studies musthave provided risk estimates [relative risk (RR), odds ratio(OR), or hazard ratio (HR)] for patients treated with thesuspected drug compared with a control group; or dataallowing calculation of such risk estimates. The most adjustedestimate was used for studies presenting more than one riskestimate. A minimum of three studies was needed in order tocarry out a cumulative meta-analysis.

Meta-analyses were conducted using the DerSimonian andLaird random-effects model in order to pool the OR with their95 % CIs [14]. It was assumed that the OR was an unbiasedestimate of the RR. This model was chosen as it is moreconservative than a fixed-effect model in the presence ofbetween-studies heterogeneity. Between-studies heterogeneity

746 Eur J Clin Pharmacol (2014) 70:745–756

was assessed using the chi-square test and the I2 measure ofinconsistency [15]. The influence of a study’s publication dateover the primary outcomes risk considered in each safety alertwas assessed using meta-regression according to the methodof moments. Publication bias was visually examined by afunnel plot and statistically evaluated by Egger’s regressionasymmetry test [16, 17]. A sensitivity analysis was performedto explore the influence of study design in the summaryestimates. All statistical analyses were performed using theComprehensiveMeta-analysis Version 2 (Biostat, Englewood,NJ, USA).

Results

Figure 1 summarizes the selection process of safetyalerts. Of the 59 safety alerts, 39 were excluded, asthey were not supported by longitudinal, comparativestudies. Twenty safety alerts were selected for furtherrevision, of which three were excluded: valproate andimpaired cognitive development, because the revisedstudies did not provide data to calculate RR estimates;lamotrigine and increased risk of sudden unexpecteddeath, as any type of death occurred in studies in which

patients were treated with lamotrigine; and antiepilepticsand bone disorders, as a considerable proportion ofstudies compared patients with epilepsy receiving treat-ment with healthy individuals.

Characteristics of the selected safety alerts are de-scribed in Table 1. The 17 safety alerts evaluated ninedifferent clinical questions. Two clinical questions(statins and increased blood sugar; statins and cognitiveside effects) were evaluated by the US FDA in the samesafety alert released on 28 February 2012. Four clinicalquestions were evaluated by only one regulatory author-ity; five clinical questions were evaluated by at leasttwo regulatory authorities.

Table 2 describes the results of cumulative meta-analysesover time according to the year of publication of each study,metaregression estimates, and publication bias assessment.

Fluoxetine and cardiovascular birth defects

Fluoxetine was not associated with a significant risk forcardiovascular birth defects (final result: OR 1.19; 95 % CI0.86–1.65, P=0.304; I2=28.3 %, P=0.21). Only two studiesreported an increased risk.w1, w4

20 Safety alerts fully reviewed

17 Safety alerts included2 issued by TGA2 issued by Health Canada6 issued by US FDA7 issued by EMA

59 Safety alerts reviewed in Alves et al. 20135 issued by TGA13 issued by Health Canada16 issued by US FDA25 issued by EMA

3 Excluded safety alertStudies did not provide risk estimates:

valproate and impaired cognitive development

Any event occurred in patients treated with the suspected drug: lamotrigine and increased risk of sudden unexpected death

Heterogeneity between study designs: antiepileptics and bone disorders

39 Initially excluded safety alerts:Supported by post -marketing spontaneous

reports and/or case reports and/or case series: 15

Supported by less than three longitudinal, comparative studies: 10

Supported by post -marketing spontaneous reports and/or case reports and/or case series and by less than threelongitudinal, comparative studies: 7

Supported by post -marketing spontaneous reports and/or case reports and/or case series and supported by non -comparative studies: 3

Supported by pregnancy register databasesstudies: 2

Supporting data sources not provided: 2Based upon other regulatory authority

evaluation: 1References not provided: 1

Fig. 1 Safety alerts selected forcumulative meta-analysis

Eur J Clin Pharmacol (2014) 70:745–756 747

Table 1 Safety alerts, data sources evaluated, studies included in cumulative meta-analyses, and decisions reached by regulatory authorities

Date Regulatory authority Evidence supportingregulatory decision

Studies included in the cumulativemeta-analysis

Label section update/regulatorydecision

Fluoxetine and cardiovascular birth defects

25 February 2010 EMA 1 cohort, 1 meta-analysis(9 observational studies)w1, w2

4 case–control, 2 cohort,1 retrospective cohort w1, w3–w8

Pregnancy section updated

Proton-pump inhibitors and bone fractures

25 May 2010 US FDA 5 case–control, 2 cohort, 1cross-sectionalw9–w16

6 case–control, 4 cohort,1 retrospective cohortw9–w15,w17, w18, w21, w22

Warnings and precautions;dosage recommendations

22 March 2012 EMA 5 case–control, 3 cohort, 2 meta-analysis (11 observational studies),1 retrospective cohortw9–w15, w17–w20

Warnings and precautions

Angiotensin receptor blockers and cancer

15 July 2010 US FDA 3 meta-analysis of randomizedclinical trials, 1 cohortw23–w26

23 RCTs, 1 cohortw26–w50 B/R ratio remains positive

20 October 2011 EMA 1 meta-analysis of randomizedclinical trialsw23

B/R ratio remains positive

Randomized clinical trials,observational studies (unspecified) a

Pioglitazone and bladder cancer

17 September 2010 US FDA 1 randomized clinical trial, 1 cohortw51, w52 2 cohort, 1 case–control,1 RCTw51–w54


16 March 2011 EMA Postmarketing spontaneous reports Warnings and precautions;contraindications2 cohort, 1 randomized clinical trial,

1 case–control, 1 meta-analysis ofrandomized clinical trialw51–w54,b

17 June 2011 Health Canada Postmarketing spontaneous reports Warnings and precautions;contraindications1 randomized clinical trial,

1 cohortw51, w52

18 June 2011 TGA 2 cohortw52, w53 Precautions

Combined hormonal contraceptives containing drospirenone and venous thromboembolism

27 May 2011 EMA 4 case–control, 2 cohort,1 retrospective cohortw55–w61

4 case–control, 2 cohort,1 retrospective cohortw55–w61

Warnings and precautions;contraindications

31 May 2011 US FDA 3 case–control, 2 cohort,2 retrospective cohortw55–w60, c

Warnings and precautions;remains under revision

07 June 2011 Health Canada 2 case-controlw55, w56,other studies (unspecified) a

Warnings and precautions;contraindications

06 July 2011 TGA 2 case-controlw55, w56,other studies (unspecified) a

Precautions; contraindications;remains under revision

Statins and increased blood sugar

10 January 2012 EMA 3 RCTs, 1 meta-analysis(13 RCTs)w62–w65

16 RCTs, 1 cohortw62, w63,w, 67, w73–w87


28 February 2012* US FDA Revieww66 Warnings and precautions3 RCTs, 3 meta-analysis

(16 RCTs), 1 cohort, 1 retrospectivecohortw62, w67–w73

Proton-pump inhibitors and Clostridium-difficile-associated diarrhea

08 February 2012 US FDA Postmarketing spontaneous reports,case reports a

Warnings and precautions;remainsunder revision17 case–control, 5 retrospective cohort,

3 cohort, 2 meta-analyses(30 observational studies)w88–w113

24 case–control, 5 retrospectivecohort, 3 cohortw88–w100, w102–104,w106–w120

Statins and cognitive side effects

28 February 2012* US FDA Postmarketing spontaneous reports Adverse drug reactionssection updated2 case reports 1 revision of database with

postmarketing spontaneous reports,1 survey (prospective)w121–w124

6 cohort, 2 case-controlw132–w139

4 randomized clinical trials, 2 cohort, 1case–control, 1 cross-sectional,1 meta-analysis (7 observationalstudies), 1 survey (retrospective)w125–w134

Proton-pump inhibitors and pneumonia

26 July 2012 EMA Postmarketing spontaneous reports B/R ratio remains positive;remains under reviewSystematic revieww140, randomized

clinical triala10 case–control, 5 retrospective

cohort, 4 cohortw146–w164


Proton-pump inhibitors and bone fractures

A statistically significant increased risk for bone fracturesassociated with proton-pump inhibitors (PPIs) was initiallyidentified in 2006 by pooled data from two studies (OR 1.31,95 % CI 1.08–1.59, P=0.007). In 2008, the risk was definedas not statistically significant (OR 1.18, 95 % CI 0.93–1.48,P=0.169) with the publication of Kaye et al.w15 In the sameyear, a statistically significant association was found frompooled studies by Targownik et al.w16 (OR 1.25, 955 CI1.00–1.55, P=0.049), as well as from the final result (OR1.27, 95 % CI 1.17–1.37, P<0.001; I2=77.0 %, P<0.001).

Angiotensin-receptor blockers and cancer

Angiotensin-receptor blockers are not associated with an in-creased risk for cancer (final result: OR 0.99, 95 % CI 0.95–1.05, P=0.674; I2=0 %, P=0.49). Metaregression showedthat results remained stable over time [metaregression esti-mate and standard error (SE) 0.001 (0.006); P=0.85].

Pioglitazone and bladder cancer

A statistically significant risk for bladder cancer associatedwith pioglitazone was identified after the publication of thefirst study in 2012 (OR 1.24, 95 % CI 1.04–1.48, P=0.012)and remained significant when the results of all studies werepooled (OR 1.32, 95 % CI 1.08–1.62, P=0.020; I2=37.4 %,P=0.19).

Combined hormonal contraceptives containing drospirenoneand venous thromboembolism

Two early studies published in 2007 did not report an in-creased risk between combined hormonal contraceptives con-taining drospirenone and venous thromboembolism. Laterstudies established an increased risk, which was confirmedin 2011 by meta-analysis [OR 1.70, 95 % CI 1.13–2.57, P=0.011; I2=81.0 %, P<0.001; metaregression estimate (SE)=0.22 (0.12); P=0.06].

Statins and increased blood sugar

The outcome of interest evaluated was “newly diagnoseddiabetes mellitus.” The cumulative meta-analysis of studiesin 2008 associated statins with an increased risk for diabetesmellitus (OR 1.07, 95 % CI 1.01–1.15, P=0.034). The resultwas determined as statistically nonsignificant after analysis ofpooled data from a cohort study (OR 1.11, 95% CI 0.99–1.23,P=0.055; I2=72.7 %, P<0.001) [metaregression estimate(SE)=0.02 (0.006); P<0.001].w73

Proton-pump inhibitors and Clostridium-difficile-associateddiarrhea

Cumulative meta-analysis showed that a statistically signifi-cant increased risk for C.-difficile-associated diarrhea withPPIs became evident when the fifth study was published in2005 (OR 1.89, 95 % CI 1.19–3.02, P=0.007). Final OR

Table 1 (continued)

Date Regulatory authority Evidence supportingregulatory decision

Studies included in the cumulativemeta-analysis

Label section update/regulatorydecision

10 case–control, 5 retrospective cohort,4 cohort, 3 meta-analyses (12 observationalstudies), 1 pooled analysis of RCTs, 1retrospective, noncomparativestudyw141–w164

EMA European Medicines Agency, TGATherapeutic Goods Administration, US FDA US Food and Drug Administration, RCT randomized controlledtrial, B/R benefit/risk ratio

*Issued in the same safety alerta Reference not providedb Meta-analysis of RCTs unpublishedc Retrospective cohort unpublishedw24 Published 29 November 2010w25 Published April 2011w26 Published 11 April 2011w52 Published 21 March 2011w53 Published 31 March 2012w54 Published 31 May 2012w89 Study designed as simultaneous case–control and cohortw67, w101 Published as short communication; reference list (represented by "w" numbers) is presented in electronic supplementary material (ESM) 2


Tab

le2

Cum

ulativeodds

ratio

s(O

Rs)and95

%confidence

intervals(CIs)

Safety

alerts

Studies

Design

Year

Cum

ulative

OR

Heterogeneity

Metaregression

PBa

OR(95%

CI)

PI2

PEstim

ate(SE)

PP

Fluoxetin

eandcardiovascular

birthdefectsEMA25

February

2010

Chambersetal.

Cohort

1996

4.19

(0.43–40.78)

0.217

Malm

etal.

Case–control

2005

1.78

(0.78–4.05)

0.172

Källénetal.

Case–control

2007

1.27

(0.79–2.03)

0.308

Alwan

etal.

Case–control

2007

1.24

(0.87–1.77)

0.231

Louik

etal.

Case–control

2007

1.10

(0.83–1.46)

0.503

Oberlanderetal.

R.cohort

2008

1.06

(0.82–1.39)

0.652

Diav-Citrin

etal.

Cohort

2008

1.19

(0.86–1.65)

0.304

28.3

%0.21

−0.102

(0.097)

0.29

0.06

Proton-pum

pinhibitorsandbone

fracturesUSFD

A25

May

2010

EMA22

March

2012

Vestergaard

etal.

Case–control

2006

1.18

(1.04–1.33)

0.008

Yangetal.

Case–control

2006

1.31

(1.08–1.59)

>0.007

Kayeetal.

Case–control

2008

1.18

(0.93–1.48)

0.169

Targow

niketal.

Cohort

2008

1.25

(1.00–1.55)

0.049

Yuetal.

Cohort

2008

1.25

(1.06–1.48)

0.007

Rouxetal.

Cohort

2008

1.29

(1.08–1.53)

0.004

deVries

etal.

R.cohort

2009

1.25

(1.09–1.41)

0.001

Grayetal.

Cohort

2010

1.24

(1.13–1.37)

<0.001

Corleyetal.

Case–control

2010

1.25

(1.15–1.36)

<0.001

Pouw

elsetal.

Case–control

2010

1.24

(1.15–1.34)

<0.001

Chiuetal.

Case–control

2010

1.27

(1.17–1.37)

<0.001

77.0

%<0.001

−0.0002(0.027)

0.99

0.12

Angiotensin-receptorblockersandcancer

USFD

A15

July

2010

EMA20

October

2011

IRMA2

RCT

2001

1.27

(0.26–6.13)

0.767

RENAAL

RCT

2001

1.31

(0.44–3.91)

0.627

IDNT

RCT

2001

0.86

(0.53–1.40)

0.556

Val-H

eFT

RCT

2001

0.89

(0.72–1.11)

0.313

LIFE

RCT

2002

1.04

(0.92–1.18)

0.556

ALPINE

RCT

2003

1.04

(0.92–1.18)

0.557

CHARM

Alternative

RCT

2003

1.04

(0.92–1.17)

0.578

VALIA

NT

RCT

2003

1.02

(0.91–1.14)

0.759

CHARM

RCT

2004

1.02

(0.93–1.13)

0.654

VALUE

RCT

2006

0.96

(0.88–1.03)

0.255

TROPH

YRCT

2006

0.95

(0.88–1.03)

0.221

SCOPE

RCT

2007

0.97

(0.90–1.04)

0.365

JIKEI

RCT

2007

0.97

(0.90–1.04)

0.366

ONTA

RGET(vsACEi)

RCT

2008

0.99

(0.93–1.05)

0.736

PROFE

SSRCT

2008

0.99

(0.93–1.04)

0.619

TRANSC

END

RCT

2008

1.00

(0.95–1.05)

0.986

DIRECT(O

verall)

RCT

2008

1.01

(0.95–1.08)

0.726

I-PR

ESE

RVE

RCT

2008

1.01

(0.95–1.07)

0.828


Tab

le2

(contin

ued)

Safety

alerts

Studies

Design

Year

Cum

ulative

OR

Heterogeneity

Metaregression

PBa

OR(95%

CI)

PI2

PEstim

ate(SE)

PP

GISSI-A

FRCT

2009

1.01

(0.95–1.07)

0,846

HIJ-CREATE

RCT

2009

1.00

(0.95–1.06)

0.900

KYOTO

RCT

2009

1.00

(0.95–1.06)

0.939

NAVIG

ATOR

RCT

2010

1.01

(0.96–1.06)

0.749

ACTIV

E-I

RCT

2011

0.99

(0.95–1.05)

0.943

Pasternaketal.

Cohort

2011

0.99

(0.96–1.02)

0.674

0%

0.49

0.001(0.006)

0.85

0.56

Pioglitazoneandbladdercancer

EMA17

Septem

ber2010

USFD

A16

March

2011

Health

Canada17

July

2011

TGA18

July

2011

Dormandy

etal.

RCT

2005

2.84

(1.02–7.89)

0.045

Lew

isetal.

Cohort

2011

1.59

(0.72–3.52)

0.162

Neumannetal.

Cohort

2012

1.24

(1.04–1.48)

0.012

Azoulay

etal.

Case–control

2012

1.32

(1.08–1.62)

0.020

37.4

%0.19

−0.09(0.08)

0.23

0.07

Com

binedhorm

onalcontraceptives

containing

drospirenone

andvenous

thromboem

bolism

EMA27

May

2011

USFD

A31

May

2011

Health

Canada07

July

2011

TGA06

July

2011

Dingeretal.

Cohort

2007

0.90

(0.57–1.42)

0.652

Seeger

etal.

R.cohort

2007

0.90

(0.63–1.29)

0.566

Lidegaard

etal.

Cohort

2009

1.15

(0.73–1.82)

0.544

vanHylckam

aViegetal.

Case–control

2009

1.61

(0.85–3.02)

0.143

Dingeretal.

Case–control

2010

1.45

(0.87–2.43)

0.158

Parkin

etal.

Case–control

2011

1.62

(0.99–2.63)

0.053

Jick

etal.

Case–control

2011

1.70

(1.13–2.57)

0.011

81.0

%<0.001

0.22

(0.12)

0.06

0.78

StatinsandincreasedbloodsugarEMA10

January

2012

USFD

A28

February2012*

Pravastatin

MSG

RCT

1993

3.02

(0.12–75.11)

0.500

4SRCT

1994

1.04

(0,84–1.28)

0.750

AFC

APS

/TEXCAPS

RCT

1998

1.02

(0,85–1.22)

0.834

GISSI

PREVENZIO

NE

RCT

2000

0,98

(0,84–1.14)

0.817

WOSC

OPS

RCT

2001

0.94

(0.82–1.08)

0.407

PROSP

ER

RCT

2002

1.01

(0.85–1.19)

0.914

ALLHAT

RCT

2002

1.04

(0.91–1.19)

0.551

ASC

OT-LLA

RCT

2003

1.06

(0.95–1.19)

0.325

HPS

RCT

2003

1.08

(0.98–1.19)

0.110

LIPID

RCT

2003

1.06

(0.97 –1.16)

0.216

PROVE-ITTIM

I22

RCT

2004

1.06

(0.97–1.15)

0.183

ATHEROMA

RCT

2005

1.06

(0.98–1.15)

0.154

MEGA

RCT

2006

1.07

(0.99–1.15)

0.091

CORONA

RCT

2007

1.07

(1.00–1.15)

0.051

GISSI

HF

RCT

2008

1.07

(1.01–1.15)

0.033

JUPITER

RCT

2008

1.09

(1.03–1.16)

0.006

Culveretal.

Cohort

2012

1.11

(0.99–1.23)

0.051

71.08%

<0.001

0.02

(0.006)

<0.001

0.003

Proton-pum

pinhibitorsandClostridium

-difficile-associateddiarrhea

USFD

A08

February

2012

Shah

etal.

Case–control

2000

0.86

(0.47–1.59)

0.625

Yip

etal.

Case–control

2001

1.61

(0.37–7.06)

0.530


Tab

le2

(contin

ued)

Safety

alerts

Studies

Design

Year

Cum

ulative

OR

Heterogeneity

Metaregression

PBa

OR(95%

CI)

PI2

PEstim

ate(SE)

PP

Kyneetal.

Cohort

2002

1.67

(0.69–4.04)

0.253

Cunningham

etal.

Case–control

2003

1.87

(0.97–3.60)

0.060

Dialetal.

Cohort

2004

1.89

(1.19–3.02)

0.007

Dialetal.

Case–control

2004

2.01

(1.36–2.99)

0.001

Al-Tureihi

etal.

Case–control

2005

2.08

(1.45–2.99)

<0.001

Dialetal.

Case–control

2005

2.26

(1.66–3.08)

<0.001

Loo

etal.

Case–control

2005

2.01

(1.37–2.95)

<0.001

Pepinetal.

R.cohort

2005

1.85

(1.23–2.78)

0.003

Modenaetal.

Case–control

2005

1.95

(1.33–2.87)

0.001

Mutoetal.

Case–control

2005

1.98

(1.38–2.84)

<0.001

Gillisetal.

Case–control

2006

1.92

(1.37–2.70)

<0.001

Kazakovaetal.

Case–control

2006

1.98

(1.43–2.75)

<0.001

Low

eetal.

Case–control

2006

1.86

(1.33–2.60)

<0.001

Dialetal.

Case–control

2006

1.82

(1.34–2.46)

<0.001

Yearsleyetal.

Case–control

2006

1.82

(1.37–2.43)

<0.001

Akhtaretal.

Case–control

2007

1.83

(1.40–2.39)

<0.001

Beaulieuetal.

R.cohort

2007

1.75

(1.35–2.27)

<0.001

Cadleetal.

Case–control

2007

1.81

(1.40–2.39)

<0.001

Dubberkeetal.

R.cohort

2007

1.90

(1.46–2.49)

<0.001

Jayatilakaetal.

Case–control

2007

1.94

(1.49–2.51)

<0.001

Aseerietal.

Case–control

2008

1.98

(1.54–2.56)

<0.001

Baxteretal.

Case–control

2008

1.92

(1.52 –2.44)

<0.001

Dialetal.

Case–control

2008

1.90

(1.52–2.37)

<0.001

Daltonetal.

R.cohort

2009

1.90

(1.54–2.35)

<0.001

Debastetal.

Case–control

2009

1.89

(1.53–2.32)

<0.001

Turco

etal.

Case–control

2010

1.92

(1.56–2.36)

<0.001

Bajajetal.

Case–control

2010

1.95

(1.59–2.39)

<0.001

How

elletal.

Cohort

2010

1.94

(1.60–2.34)

<0.001

Kim

etal.

Case–control

2010

1.96

(1.63–2.37)

<0.001

Linskyetal.

R.cohort

2010

1.94

(1.61–2.32)

<0.001

87.9

%<0.001

0.02

(0.04)

0.59

0.14

Statinsandcognitive

side

effectsUSFDA28

February

2012*

Jick

etal.

Case–control

2000

0.29

(0.13–0.64)

0.002

Rodriguez

etal.

Cohort

2002

0.41

(0.21–0.82)

0.011

Rockw

oodetal.

Case–control

2002

0.38

(0.23–0.63)

<0.001

Lietal.

Cohort

2004

0.52

(0.24–1.16)

0.111

Rea

etal.

Cohort

2005

0.67

(0.39–1.15)

0.144

Zandi

etal.

Cohort

2005

0.75

(0.47–1.19)

0.216


Tab

le2

(contin

ued)

Safety

alerts

Studies

Design

Year

Cum

ulative

OR

Heterogeneity

Metaregression

PBa

OR(95%

CI)

PI2

PEstim

ate(SE)

PP

Beydoun

etal.

Cohort

2011

0.67

(0.42–1.07)

0.091

Betterm

anetal.

Cohort

2012

0.65

(0.43–0.98)

0.039

75.9

%<0.001

−0.0008(0.06)

0.99

0.15

Proton-pumpinhibitorsandpneumoniaEMA26

July

2012

Mallowetal.

Cohort

2004

1.00

(0.38–2.60)

0.999

Laheijetal.

Case–control

2004

1.62

(0.95–2.77)

0.076

Gulmez

etal.

Case–control

2007

1.56

(1.31–1.87)

<0.001

Sarkar

etal.

Case–control

2008

1.74

(1.37–2.21)

<0.001

Beaulieuetal.

R.cohort

2008

1.45

(1.08–1.95)

0.015

Marciniak

etal.

Case–control

2009

1.46

(1.10–1.95)

0.009

Rougheadetal.

R.cohort

2009

1.37

(0.99–1.89)

0.055

Myles

etal.

Case–control

2009

1.40

(1.06–1.85)

0.017

Herzigetal.

Cohort

2009

1.39

(1.09–1.78)

0.008

Miano

etal.

R.cohort

2009

1.43

(1.13–1.82)

0.003

Myles

etal.(2)

Cohort

2009

1.36

(1.08–1.71)

0.008

Rodriguez

etal.

Case–control

2009

1.34

(1.08–1.66)

0.007

Gau

etal.

Case–control

2010

1.34

(1.08–1.63)

0.006

Eurichetal.

Case–control

2010

1.30

(1.07–1.59)

0.009

Dublin

etal.

Case–control

2010

1.29

(1.07–1.55)

0.009

Redelmeier

etal.

R.cohort

2010

1.26

(1.05–1.51)

0.015

Kasuyaetal.

R.cohort

2010

1.28

(1.07–1.54)

0.008

Meijvisetal.

Case–control

2011

1.32

(1.13–1.61)

0.003

Laheijetal.

Cohort

2011

1.35

(1.13–1.61)

0.001

95.7

%<0.001

−0.02(0.04)

0.62

0.47

EMAEuropeanMedicines

Agency;

TGATherapeuticGoods

Adm

inistration;

USFDAUSFoodandDrugAdm

inistration,

RCTrandom

ized

controlledtrial,SE

standard

error,R.cohortretrospective

cohort,P

Bpublicationbias

*Issuedin

thesamesafety

alert


estimate was 1.94 (95 % CI 1.61–2.37, P<0.001; I2=87.9 %,P<0.001).

Statins and cognitive side effects

A protective effect of statins on dementia and cognitive im-pairment was found (OR 0.65, 95 % CI 0.43–0.98, P=0.039;I2=75.9 %, P<0.001). Metaregression showed that resultswere stable over time [metaregression estimate (SE) −0.0008(0.06), P=0.99].

Proton-pump inhibitors and pneumonia

Cumulative meta-analysis showed a statistically significantincreased risk for pneumonia associated with PPIs becameevident when the third study was published in 2007 (OR 1.56,95 % CI 1.31–1.87, P<0.001). However, when the study byRoughead et al. was published in 2009, the result becamestatistically nonsignificant (OR 1.37, 95 % CI 0.99–1.89, P=0.055).w154 Following the publication by Myles et al., anincreased risk was observed again (OR 1.40, 95 % CI 1.06–1.85, P=0.017).w150 The final OR for cumulative meta-analysis was 1.35 (95 % CI 1.13–1.61, P=0.001; I2=95.7 %, P<0.001).

Publication bias

Egger’s asymmetry test was not statistically significant formost analyses but was significant for statins and increasedblood sugar (P=0.003) (Table 2). After exclusion of thecohort study, no statistically significant asymmetry was found(P=0.773).w73 Few studies were considered for pioglitazoneand bladder cancer analysis, which may not allow firm con-clusions despite the nonstatistically significant Egger’s asym-metry test (P=0.07). Subjective evaluation of publication biaswas based on visual inspection of the funnel plot.

Sensitivity analysis

Sensitivity analysis according to different study designs didnot significantly change results with respect to observedbetween-studies heterogeneity (ESM Appendix, Table 1).Regarding subgroup analysis according to study designs,timing changed following three pooled risk estimates. WhenRCTs only were considered to estimate the risk for increasedblood sugar (newly diagnosed diabetes mellitus) associatedwith statins, the estimate yielded a statistically significantincreased OR (1.07, 95 % CI 1.01–1.15, P=0.034). The sameincreased risk was observed in the cohort study, although thefinal estimate of different pooled study designs wasnonsignificant.w73 Considering cohort designs only, the ulti-mate increased risk estimate for diarrhea due to C. difficileassociated with PPIs was observed in 2009 only (OR 1.75,

1.00–3.07, P=0.05). The definitive increased risk for fracturesassociated with PPIs was observed in 2010 (OR 1.23, 1.07–1.40, P=0.003), when case–control studies only wereconsidered.

Discussion

Findings of this study show that for the majority of casescenarios (seven of nine), results yielded by meta-analysiswere in line with conclusions of regulatory authorities.Warnings could have been added to the label of PPIs in2004 for C.-difficile-associated diarrhea and in 2008 for bonefractures. The label of PPIs was first updates in 2012 and 2010regarding those AEs, respectively. These two decisions weresupported by observational data only. Increased blood sugarwas associated with statins in 2008 after pooling data fromRCTs. The inclusion of a cohort study in the estimate returneda final result that was statistically nonsignificant and associat-ed with considerable heterogeneity. The label on statins wasupdated to properly advise users of the risk of diabetes.

However, caution is needed when interpreting these riskestimates, as they could be biased by the inherent confoundingvariables of the included studies [10]. According to the sensi-tivity analysis, results of meta-analyses exclusively integratingdata from RCTs were characterized by low heterogeneity.Frequent and acute AEs are regularly identified from RCTs,particularly when they are pre-established endpoints of inter-est. When regulatory authorities and investigators are dealingwith rare AEs, which may be detected in RCTs, it is commonto pool data using a meta-analysis. This was the case whenAEs such as cancer and increased blood sugar were evaluatedusing meta-analysis [18–20].

All safety issues analyzed in this study were evaluated withat least one type of observational methodology by regulatoryauthorities that reviewed the drugs’ B/R ratio. This could bedue to the fact that most AEs were considered as rare and/orlong-latency events, such as malignancies, cardiovascularevents, and diabetes, which are typically better evaluated inpostauthorization safety studies. These studies offer the ad-vantage of a naturalistic observation, which may better repre-sent the incidence of iatrogenic events that occurred in clinicalpractice [8, 21].

The final conclusion of B/R re-evaluations conducted bythe regulatory authorities may contradict risk estimates usingmeta-analyses of pooled data. Additional data sourcessupporting a causal relation between an AE and a drug canbe used to substantiate regulatory decisions. In such cases,meta-analyses of the existing evidence can return inconclusiveresults, as occurred when authorities added warnings to thelabel of glucagon-like peptide-1 (GLP-1) receptor agonistsdue to the risk of acute pancreatitis [22]. In this study, anincreased risk of pneumonia associated with PPIs was


suspected in 2009. This is in line with results of previousmeta-analyses that yielded increased risk estimates and weresubsequently reviewed by EMA [23–25]. However, EMArecommended that no steps be taken to minimize risk at thattime, and kept this class of drugs under review. The authorityconsidered that evidence from observational studies of anassociation between PPIs as a class and pneumonia wasinconsistent and might be subject to residual confounding[26]. Methodological differences may also be responsible fordelays in meta-analysis to yield a statistical significant result,as occurred in relation to venous thromboembolism associatedwith orally administered contraceptives containingdrospirenone. Differences between venous thromboembolismdefinition, risk factors of included patients, and type of con-traceptives used as control group may explain why the firsttwo studies published in 2007 reported a null association [27].Later studies took these methodological and clinical issuesinto consideration and reported increased risks [27]. This mayhelp explain why only in 2011 was an increased risk identifiedby cumulative meta-analysis; the same year, all regulatoryauthorities suggested label updates.

According to metaregression results, most risk estimateswere stable over time. The exception was statins/increasedblood sugar, for which the risk progressively increased. Thissuggests that conducting cumulative meta-analysis of evi-dence may have prompted regulatory authorities to take moretimely action. Studies estimate that withdrawal of a drug fromthe market occurs in the first 2 years after its release and thatlabel changes take, on average, between 7 and 11 years [2, 28,29]. This study assessed the most commonly consumed drugsworldwide in the ambulatory care setting that were had beenapproved for several years and are thus used to treat millionsof people. Recently approved drugs may be more likely to belinked with unexpected serious AEs, therefore leading tomorerapid regulatory actions. In this study, none of the regulatorydecisions led to drug withdrawal from the market.Postmarketing drug safety requires careful evaluation ofexisting evidence by regulatory authorities. However, timelyascertainment of drug risks, with higher-quality and better-documented scientific evidence, seems to require improve-ment [2].

Taking into account the safety issues evaluated in this studyand the corresponding regulatory decisions, it is not possibleto draw definitive recommendations about the requirements ofconducting meta-analyses every time safety signals are deter-mined from longitudinal, comparative studies. Observationalstudies are more susceptible to bias and confounding, andintegrating data from such designs in meta-analyses mayreturn results with excessive heterogeneity, as observed formost cases evaluated. In an attempt to reduce such uncertainty,sensitivity analyses based on study design was conducted, butthe results did not differ significantly. When there is littleheterogeneity among studies, one may be willing to accept

evidence from a meta-analysis as helping to establish a B/Rratio [10]. In the presence of substantial heterogeneity, it isdifficult to draw conclusions, and acceptance of results may beslow. This might be one reason that regulatory authorities tookseveral years to make conclusions regarding an increased riskin some cases addressed in this study.

Some limitations to this study need to be noted. Safetyissues were evaluated by four major regulatory authorities;other authorities were not included. This could result in theexclusion of important information. The study also intended toanalyze safety alerts and communications that included earlynotices about such issues, some of which are still underrevision and the authorities’ final decisions have not yet beenmade. Only bibliographic references used by the regulatoryauthorities as sources of information to support safety alertswere considered for this study. Systematic reviews of biblio-graphic evidence for each clinical question were not conduct-ed; additionally, all data sources reviewed by regulatory au-thorities may not have been published on their websites.Therefore, some studies may be absent from our cumulativemeta-analyses. Despite the fact that Egger’s asymmetry testand visual inspection of funnel plots may not indicate publi-cation bias for most cases analyzed, no specific bibliographicresearches have been conducted. However, the extent towhich regulatory authorities have taken into account all pub-lished scientific evidence when a B/R ratio is evaluated for asafety issue was not within the scope of this study.

The role of meta-analysis in pharmacovigilance is amatter of ongoing debate, and efforts are being made todevelop guidelines on its use in drug-safety assessments[30]. However, a number of methodological consider-ations must be taken into account when designing andconducting meta-analyses, particularly when observa-tional studies are used, either exclusively or in combi-nation with RCTs [10, 30]. Assessing B/R ratios after asafety issue has been raised is a highly important sci-entific exercise that should be supported by differentsources of scientific evidence, which may sometimeshave conflicting results. Nonetheless, the quality of themeta-analysis is of extreme importance when safetypolicy measures may need to be taken [10]. Althoughregulatory authorities and independent investigators mayidentify increased iatrogenic risks for some drugs priorto official risk minimization strategies, uncertainties dueto heterogeneity or even the inclusion of different studydesigns may delay the decision-making process.

In conclusion, this study demonstrates that meta-analysis can be useful to assess causal relations of drugAEs and predict iatrogenic risks earlier. Although cu-mulative meta-analysis has been used with success toevaluate how risk estimates perform over time, as in thecase of rofecoxib, results can be affected by consider-able heterogeneity [12]. Therefore, this technique does


not replace further assessment methods during the B/Rratio evaluation process.

Conflict of interest Carlos Alves is supported by a research grant fromthe Foundation for Science and Technology, Portugal; reference:SFRH/BD/64957/2009.

Ana Filipa Macedo declares that she has no conflict of interest.Francisco Batel Marques declares that she has no conflict of interest.

Details of ethical approval No ethics approval was required for thisstudy

References

1. Madre LK, Califf RM, Reynolds RF et al (2006) Views fromAcademia, Industry, and Regulatory Agencies. In: Strom BL,Kimmel SE (eds) Textobook of pharmacoepidemiology. JohnWiley & Sons, Chichester, England, pp 63–88

2. Moore TJ, Singh S, Furberg CD (2012) The FDA and new safetywarnings. Arch Intern Med 172(1):78–80

3. US Food and Drug Administration (2013) Drug SafetyCommunications. Available at: http://www.fda.gov/Drugs/DrugSafety/ucm199082.htm. Accessed 24 September 2013

4. EuropeanMedicines Agency (2013) Safety monitoring of medicines.Available at: http://www.ema.europa.eu/ema/index.jsp?curl=pages/special_topics/general/general_content_000456.jsp&mid=WC0b01ac05801ae8fb. Accessed 24 September 2013

5. Wysowski DK, Swartz L (2005) Adverse drug event surveillance anddrug withdrawals in the United States, 1969–2002: the importance ofreported suspected reactions. Arch Intern Med 165(12):1363–1369

6. Medicines and Healthcare products Regulatory Agency. Safety in-formation. Drug safety update 2011;5(5). Citalopram andescitalopram: QT interval prolongation—new maximum daily doserestrictions (including in elderly patients), contraindications, andwarnings. Available at: http://www.mhra.gov.uk/Safetyinformation/DrugSafetyUpdate/CON137769. Accessed 24 September 2013

7. van Staa TP, Leufkens HG, Cooper C (2001) Use of inhaled cortico-steroids and risk of fractures. J Bone Miner Res 16:581–588

8. Papanikolaou PN, Christidi GD, Ioannidis JPA (2006) Comparisonof evidence on harms of medical interventions in randomized andnonrandomized studies. CMAJ 174(5):635–641

9. US Food and Drug Administration (2012) Centre for DrugEvaluation and Research. Guidance drug safety information –FDA’s Communication to Public. Available at: http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM295217.pdf. Accessed 22 May, 2013

10. Kim CJ, Berlin JA (2006) The use of meta-analysis inpharmacoepidemiology. In: Strom BL, Kimmel SE (eds)Textobook of pharmacoepidemiology. John Wiley & Sons,Chichester, England, pp 353–365

11. Alves C, Batel-Marques F, Macedo AF (2012) Data sources on drugsafety evaluation: a review of recent published meta-analyses.Pharmacoepidemiol Drug Saf 21:21–33

12. Jüni P, Nartey L, Reichenbach S et al (2004) Risk of cardiovascularevents and rofecoxib: cumulative meta-analysis. Lancet 364:2021–2029

13. Alves C, Macedo AF, Batel Marques F (2013) Sources of informa-tion used by regulatory agencies on the generation of drug safetyalerts. Eur J Clin Pharmacol. doi: 10.1007/s00228-013-1564-y [Epubahead of print]

14. DerSimonian R, Laird N (1986) Meta-analysis in clinical trials.Control Clin Trials 7:177–188

15. Higgins JP, Thompson SG, Deeks JJ, Altman DG (2003) Measuringinconsistency in meta-analysis. BMJ 327:557–560

16. Publication bias (2009) Borenstein M, hedges LV, Higgins JPT,Rothstein HR (eds). Introduction to meta-analysis. John Wiley &Sons, Chichester, England, pp 277–291

17. Egger M, Davey Smith G, Schneider M, Minder C (1997) Bias inmeta-analysis detected by a simple, graphical test. BMJ 315:629–634

18. Bangalore S, Kumar S, Kjeldsen SE et al (2011) Antihypertensivedrugs and risk of cancer: network meta-analyses and trial sequentialanalyses of 324,168 participants from randomized trials. LancetOncol 12(1):65–82

19. Sattar N, Preiss D, Murray HM et al (2010) Statins and risk ofincident diabetes: a collaborative meta-analysis of randomised statintrials. Lancet 375(9716):735–742

20. Sipahi I, Debanne SM, Rowland DY, Simon DI, Fang JC (2010)Angiotensin-receptor blockade and risk of cancer: meta-analysis ofrandomised controlled trials. Lancet Oncol 11(7):627–636

21. Vandenbroucke JP, Psaty BC (2008) Benefits and risks of drugtreatments: how combine the best evidence on benefits with the bestdata about adverse effects. JAMA 300(20):2417–2419

22. Alves C, Batel-Marques F, Macedo AF (2012) A meta-analysis ofserious adverse events reported with exenatide and liraglutide: acutepancreatitis and cancer. Diabetes Res Clin Pract 98(2):271–84

23. EomCS, Jeon CY, Lim JW, Cho EG, Park SM, Lee KS (2011) Use ofacidsuppressive drugs and risk of pneumonia: a systematic reviewand meta-analysis. Can Med Assoc J 183(3):310–319

24. Giulianio C, Wilhelm SM, Kale-Pradhan PB (2012) Are proton pumpinhibitors associated with the development of community-acquiredpneumonia?: a meta-analysis. Expert Rev Clin Pharmacol 5(3):337–344

25. Johnstone J, Nerenberg K, Loeb M (2010) Meta-analysis: protonpump inhibitor use and the risk of community-acquired pneumonia.Aliment Pharmacol Ther 31(11):1165–1177

26. European Medicines Agency (2012). Working parties and othergroups. Retired working parties. CHMP PharmacovigilanceWorking Party. Monthly reports. Monthly report on safety concerns,guidelines and general matters - July 2012. Available at: http://www.ema.europa.eu/docs/en_GB/document_library/Report/2012/07/WC500130391.pdf. Accessed 11 October 2013

27. Jick SS, Hernandez RK (2011) Risk of non-fatal venous thrombo-embolism in women using oral contraceptives containingdrospirenone compared with women using oral contraceptives con-taining levonorgestrel: case–control study using United States claimsdata. BMJ. doi:10.1136/bmj.d2151

28. Lester J, Neyarapaly GA, Lipowski E, Graham CF, Hall M, Dal PanG (2013) Evaluation of FDA safety-related drug label changes in2010. Pharmacoepidemiol Drug Saf 22(3):302–305

29. Lasser KE, Allen PD, Woolhandler SJ, Himmelstein DU, Wolfe SM,Bor DH (2002) Timing for new black box warnings and withdrawalsfor prescription medications. JAMA 287(17):2215–2220

30. Council for International Organizations ofMedical Sciences. CIOMSWorking Group X on “Considerations for applying good meta-analysis practices to clinical safety data within the biopharmaceuticalregulatory process”. http://www.cioms.ch/index.php/2012-06October 08-47-53/working-groups/working-group-x. Accessed 11November 2013

Contributions of authors

Carlos Alves conceived the study, collected and analyzed data, and wrotethe paper. Ana FilipaMacedo and Francisco Batel Marques conceived thestudy, analyzed data, and reviewed the paper. In case of disagreementbetween Carlos Alves and Ana Filipa Macedo, the opinion of FranciscoBatel Marques was sought.


http://www.fda.gov/Drugs/DrugSafety/ucm199082.htm

http://www.fda.gov/Drugs/DrugSafety/ucm199082.htm

http://www.ema.europa.eu/ema/index.jsp?curl=pages/special_topics/general/general_content_000456.jsp&mid=WC0b01ac05801ae8fb



http://www.mhra.gov.uk/Safetyinformation/DrugSafetyUpdate/CON137769

http://www.mhra.gov.uk/Safetyinformation/DrugSafetyUpdate/CON137769

http://www.fda.gov/downloads/Drugs/GuidanceComplianceRegulatoryInformation/Guidances/UCM295217.pdf



http://dx.doi.org/10.1007/s00228-013-1564-y

http://www.ema.europa.eu/docs/en_GB/document_library/Report/2012/07/WC500130391.pdf



http://dx.doi.org/10.1136/bmj.d2151

http://www.cioms.ch/index.php/2012-06-10-08-47-53/working-groups/working-group-x

http://www.cioms.ch/index.php/2012-06-10-08-47-53/working-groups/working-group-x

Drug-safety alerts issued by regulatory authorities: usefulness of meta-analysis in predicting risks...

Documents

Transcript of Drug-safety alerts issued by regulatory authorities: usefulness of meta-analysis in predicting risks...