thiazolidinediones and risk of heart failure in patients with or at high risk of type 2 diabetes...

Thiazolidinediones and Risk of Heart Failurein Patients with or at High Risk of Type 2Diabetes MellitusA Meta-Analysis and Meta-Regression Analysis of Placebo-Controlled

Randomized Clinical Trials

Adrian V. Hernandez,1 Ali Usmani,2 Anitha Rajamanickam2 and A. Moheet3

1 Health Outcomes and Clinical Epidemiology Section, Department of Quantitative Health Sciences, Lerner Research Institute,

Cleveland Clinic, Cleveland, Ohio, USA

2 Department of Hospital Medicine, Institute of Medicine, Cleveland Clinic, Cleveland, Ohio, USA

3 General Medicine Unit, Rochester General Hospital, Rochester, New York, USA

Abstract Background:Recent meta-analyses of randomized clinical trials (RCTs) demonstrated a higher risk of heart

failure (HF) with the use of thiazolidinediones (TZDs). However, this effect may have been diluted by including

active controls. Also, it is uncertain whether the risk of HF is similar with rosiglitazone and pioglitazone.

Objectives: This study quantified the risks of HF with the use of TZDs in patients with or at high risk of

developing type 2 diabetes mellitus (DM), and evaluated differential effects by type of TZD. Secondarily, we

evaluated risks of peripheral edema.

Methods: We performed a systematic review and meta-analysis of placebo-controlled RCTs evaluating the

effect of rosiglitazone or pioglitazone on investigator-reported HF and edema. Articles published before

31 December 2009 were searched in MEDLINE, TheWeb of Science, and Scopus, and the data were extracted

by three investigators. RCTs with ‡100 patients and ‡3 months of follow-up were included. We quantified

the effect of TZDs as odds ratios (ORs) by using the Mantel-Haenzel and alternative models. We further

evaluated the risk of serious/severe HF, and the effect of several trial characteristics on HF risk by subgroup

analysis and meta-regression analysis.

Results: 29 trials (n = 20 254) were evaluated. TZDs were significantly associated with HF (TZD 360/6807[5.3%] vs placebo 234/6328 [3.7%], OR 1.59; 95%CI 1.34, 1.89; p < 0.00001). The risk of HF was higher with

rosiglitazone than with pioglitazone (2.73 [95% CI 1.46, 5.10] vs 1.51 [1.26, 1.81]; p = 0.06). TZDs were

associated with a similar risk of serious/severe HF (OR 1.47; 95%CI 1.16, 1.87; p = 0.002). Use of TZDs was

also associated with edema (OR 2.04; 95% CI 1.85, 2.26; p < 0.00001). HF and edema risks were consistent

using Peto and random effects models. Risks of HF were significantly high for the subgroups of trials

including patients with or at high risk for type 2 DM, and for the subgroup of trials with ‡12 months of

follow-up. Meta-regression analysis showed that trials with lower overall baseline risk had higher HF risks.

Conclusion: In placebo-controlled trials of adult patients with or at high risk for type 2 DM, TZD therapy is sig-

nificantly and consistently associatedwith a higher risk ofHF. The risk of serious/severeHF is also increasedwith

the use of TZDs. HF risks are similar to those of meta-analyses combining active- and placebo-controlled trials.

The benefit/risk profile of TZDs should be considered when treating diabetic patients with or without prior HF.

Introduction

Harmful effects such as myocardial infarction (MI), stroke,

heart failure (HF), and peripheral edema associated with the

use of thiazolidinediones (TZDs) in patients with type 2 diabetes

mellitus (DM) have been reported.[1-6] HF and edema may be

explained by renal sodium reabsorption and fluid retention,[7,8]

given that TZDs do not directly affect left ventricular systolic

or diastolic function.[9] Type 2 DM itself is associated with

a higher risk of HF[10] by promoting atherosclerosis and

ORIGINAL RESEARCH ARTICLEAm J Cardiovasc Drugs 2011; 11 (2): 115-128

1175-3277/11/0002-0115/$49.95/0

ª 2011 Adis Data Information BV. All rights reserved.

coronary artery disease (CAD) or diabetic cardiomyopathy.[11]

Thus, when plasma volume is expanded beyond a certain thresh-

old due to the intake of TZDs, cardiac function may worsen in

diabetic patients.

A recent meta-analysis of seven randomized clinical trials

(RCTs) involving >12 months’ use of TZDs in 20 191 patients

with or at high risk of developing type 2 DM found a 72%higher relative risk of HF in patients treated with TZDs.[5] This

risk was largely confirmed in other meta-analyses: the use of

rosiglitazone for >12months in four RCTs of 14 291 type 2DM

patients found a 109% higher relative risk of HF;[4] the use of

rosiglitazone for >4 weeks in 132 trials involving 41 743 patientswith or without type 2 DM was associated with a 69% higher

relative risk of serious HF;[12] and the combined short- and

long-term use of pioglitazone in 19 RCTs involving 16 390 pa-

tients with type 2 DM found a 41% higher relative risk of se-

rious HF.[3] Another recent meta-analysis of 26 RCTs found a

126% higher odds of peripheral edema in 15332 diabetics with

short- and long-term use of TZDs.[6]

Published meta-analyses have combined active drugs and

placebo as comparators. Therefore the risk of HF may have

been diluted. It is known that metformin has been associated

with reductions in all-cause and cardiovascular mortality when

compared with sulfonylurea therapy in type 2 DM patients.[13]

However, metformin may be associated with a higher risk of

HF events, as shown in a recent cohort study.[14] Moreover,

these meta-analyses did not explore alternative methods and

models to evaluate the consistency of the risks. This is im-

portant when events are rare.[15-17] It is also important to

translate these risks intomore clinically oriented numbers, such

as the number needed to harm (NNH).[18] Finally, it is im-

portant to explore the effects of several trial characteristics on

HF risks using subgroup analysis andmeta-regression analysis.

We performed a systematic review and meta-analysis of

placebo-controlled RCTs to primarily evaluate the risk of HF

and secondarily the risk of peripheral edema in patients with or

at high risk for type 2 DM and receiving TZDs. Differential

effects between rosiglitazone and pioglitazone were also as-

sessed. Finally, we evaluated the consistency of HF risks in

subgroups of patients and the effect of trial characteristics on

HF risks in a meta-regression analysis.

Methods

Search Strategy and Trial Selection

We screened abstracts of published RCTs evaluating the

effects of rosiglitazone or pioglitazone onHF/peripheral edema

in patients with type 2 DM or at high risk for type 2 DM

(impaired glucose tolerance and/or fasting glucose, insulin

resistance syndrome, and metabolic syndrome). Literature

searches were independently conducted by three authors (AR,

AU, AM) in PubMed-MEDLINE (1960–31 December 2009)

and by one author (AVH) in EMBASE (1980–31 December

2009). The searches used the key words ‘thiazolidinedione,’

‘rosiglitazone’ or ‘pioglitazone’ and ‘type 2 diabetes’ or ‘im-

paired glucose tolerance’ or ‘impaired fasting glucose’ or ‘insulin

resistance syndrome’ or ‘metabolic syndrome’ and ‘adverse

events,’ ‘harmful events,’ ‘heart failure,’ or ‘peripheral edema’

and ‘randomized controlled trial’ or ‘randomized trial’.We also

evaluated unpublished studies by reviewing the conference

abstracts of the last 6 years from the American College of

Cardiology, American Heart Association, American College of

Chest Physicians, and European Society of Cardiology.

Our searches were restricted to placebo-controlled RCTs

conducted in humans, with available information on the effect

of TZD treatment on HF and/or peripheral edema, with more

than 100 adult patients (as RCTs with <50 patients per arm are

small andmay be of low quality), published in English, andwith

‡3 months of follow-up. We excluded RCTs with active com-

parators. Abstracts were reviewed independently by inves-

tigators, and full text articles were retrieved for abstracts that

fulfilled inclusion criteria. A manual review of references from

primary or review articles was also performed. Only RCTs that

clearly identified HF or peripheral edema events per treatment

arm were included.

Data Extraction

Three investigators (AU, AR, AM) extracted data, and the

results were compiled. Disagreement was resolved by con-

sensus. We collected for each trial: treatment arms; TZD dose;

patients per arm; trial duration; disease type; age and body

mass index (BMI); proportion male; proportionWhite; history

of HF, hypertension, CAD, and hyperlipidemia; HF type (new-

onset or exacerbation); peripheral edema severity; and number

of HF and peripheral edema events per treatment arm.

Primary Outcomes

Our primary outcome was investigator-reported incident

HF. Definition of HF varied between RCTs, and diagnosis

included typical signs/symptoms, radiologic findings, and/oruse of drugs (diuretics, vasodilators, or inotropes). HF events

had or did not have independent clinical event committee eval-

uation, and were or were not associated with hospital admission.

116 Hernandez et al.

ª 2011 Adis Data Information BV. All rights reserved. Am J Cardiovasc Drugs 2011; 11 (2)

HF events may have been either new or exacerbations. We also

analyzed HF events defined as serious, severe, or congestive by

independent committees or by the study investigators. Our

secondary outcome was peripheral edema. Peripheral edema

may have been mild, moderate, or severe.

Trial Quality Assessment

Two investigators (AU, AR) used the Jadad score, a vali-

dated measure to assess the quality of RCTs.[19] This score

ranges from 0 to 5, and evaluates randomization, blinding, and

reporting of withdrawals and dropouts. A trial with a score of

‡3 was considered to be of high quality.

Statistical Analysis

This study followed the reporting guidelines of the Preferred

Reporting Items for Systematic Reviews and Meta-Analyses

(PRISMA) statement.[20] We analyzed outcomes using the

Mantel-Haenzel (M-H) fixed effects model with treatment arm

continuity correction[16] to obtain pooled odds ratios (ORs) and

95% confidence intervals (CIs). Thismodel is themost appropriate

model when outcomes are rare (<10%) and when imbalance

between treatment arms is important. Statistical heterogeneity

was evaluated with the Cochran chi-squared (w2) and the

I2 statistical tests.[21] Funnel plots andEgger’s tests[22] were used

to evaluate publication bias. Results were also stratified by the

type of TZD – rosiglitazone or pioglitazone.

To evaluate the consistency of the TZD risks, we also cal-

culated ORs and 95% CIs from Peto fixed effects models,[23]

and the DerSimonian and Laird (DL) random effects mod-

els.[24] We also calculated risk ratios (RRs) using both fixed and

randommodels.We evaluated subgroups ofRCTs according to

the type of patients (type 2 DM vs non-diabetic), and follow-up

time (‡12 vs <12 months). The w2 test was used to evaluate

subgroup differences. To explore the effects of several trial

characteristics on the risk of HF, we performed a random ef-

fects meta-regression-analysis[25] with the use of the PROC

MIXED procedure of SAS. The risk of HFwith the use of TZD

was modelled as log OR. Trial characteristics included the

baseline risk for HF (risk for HF in the placebo arm, and a

measure of the risk profile of the patient population), type of

TZD, sample size, and average follow-up time.

Finally, we calculated the number needed to harm

(NNH).[18] The NNH is the inverse of the absolute risk increase

(risk in TZD group – risk in placebo group) and shows the

number of patients that have to be treated with TZD to produce

one harmful event. We reported NNH values per trial based on

their actual follow-up times, and did not attempt to combine

NNH across trials. The calculation of NNH per trial is also

closely related to the risk in the placebo arm and the absolute

risk increase. A p-value of <0.05 was considered significant.

We used the Review Manager (RevMan, version 5.0, Oxford,

England, 2008), S-plus 7 (Insightful Corporation, Seattle, WA,

USA), and SAS 9.2 (SAS Institute Inc., Cary, NC, USA).

Results

Eligible Studies

We identified 415 abstracts, and 386 were excluded for

various reasons (figure 1). Twenty-nine RCTs (18 with rosi-

glitazone[26-43] and 11 with pioglitazone[44-54]) were included for

the final analyses (see table I). The total number of participants

was 20 254 (TZD 11 270, placebo 8984). Only one study[26]

scored 2 in the Jadad Score, and the rest were of high quality.

Records identified throughdatabase searching

(n = 445)

Records after duplicates removed(n = 415)

227 RSG/188 PIO

Records excluded(n = 334)

177 RSG/157 PIO• 235 (129/106) n < 100• 90 (44/46) not placebo controlled• 5 (1/4) not RCTs• 3 (3/0) other diseases• 1 (0/1) not in humans

Full-text articles excluded,with reasons

(n = 52)32 RSG/19 PIO

• 51 (32/19) lackeddetailed informationon HF and edema

• 1 (0/1) withduplicated results

Records screened(n = 415)

Full-text articles of RCTsassessed for eligibility

(n = 81)50 RSG/31 PIO

RCTs included inqualitative synthesis

(n = 29)18 RSG/11 PIO

RCTs included inquantitative synthesis

[meta-analysis](n = 29)

Additional records identifiedthrough other sources

(n = 16)

Fig. 1. Search strategy profile of the meta-analysis according to the PRISMA

guidelines. HF= heart failure; PIO= pioglitazone; RCT= randomized clinical

trial; RSG= rosiglitazone.

Thiazolidinediones and Heart Failure Risk in Diabetics 117


Table

I.Generalcharacteristicsofrandomizedtrialsincludedin

themeta-analysis

Author,year

Sample

size

TZDdose

(mg/day)

Trial

duratio

n(y)

Typeofpatient

Mean

age

Male

(%)

White

(%)

Mean

HbA1c

Mean

BMI

Typeof

HF

HF

definition

Typeof

edema

Rosiglitazone

Agrawale

tal.,[2

6]2003

824

40.5

Type2DM

61.6

62

NA

9.2

29

--

Mild-M

od

Albertinie

tal.,[2

7]2007

135

80.25

Type2DM

56

64

95

NA

31

--

Mild

Barnettetal.,[2

8]2003

171

80.5

Type2DM

54

77

09.1

27

--

Mild-M

od

Bhattetal.,[2

9]2007

200

81

MS

59.4

80

98

5.8

33

New

Onset

IR-

Caoetal.,[3

0]2006

297

40.75

MS

60

58

0NA

29

New

Onset

IR-

Daileyetal.,[3

1]2004

365

4–8

0.5

Type2DM

57

59

74

8.1

32

--

Mild-M

od

Dargie

etal.,[3

2]2007

224

4–8

1Type2DM

64

79

96

7.8

29

Exa

cer

IR/IC

EC

Mild-M

od

Davidsonetal.,[3

3]2007

233

80.5

Type2DM

52.5

47

09.3

32

New

Onset

IRMild-M

od

Fonsecaetal.,[3

4]2000

339

4–8

0.5

Type2DM

58.3

68

80

8.8

30

--

Mild-M

od

Gerstein

etal.,[3

5]2006

5269

83

IGT/IF

G54.7

38

NA

NA

31

New

Onset

IR/IC

EC

Mild-M

od

Gomez-Perezetal.,[3

6]2002

105

2–4

0.5

Type2DM

53

26

510.0

28

--

Mild-M

od

Hedbladetal.,[3

7]2007

555

4–8

1Type2DM/IR

S67.5

46

NA

6.5

30

--

Mild-M

od

Hollanderetal.,[3

8]2007

568

2–4

0.5

Type2DM

53

51

57

9.0

33

New

Onset

IR/IC

EC

Mild-M

od

Lebovitz

etal.,[3

9]2001

493

2–4

0.5

Type2DM

60

66

74

8.9

30

--

Mild-M

od

Phillipsetal.,[4

0]2001

908

80.5

Type2DM

57.5

63

77

8.9

30

--

Mild

Raskin

etal.,[4

1]2001

313

4–8

0.5

Type2DM

56.5

56

70

9.0

32

Both

IRMild-M

od

Rosenstocketal.,[4

2]2006

227

4–8

2Type2DM

68.5

75

NA

7.7

30

Unclear

IRMild-M

od

Zhuetal.,[4

3]2003

530

4–8

0.5

Type2DM

58.9

45

09.8

25

--

Mild-M

od

Overallrosiglitazone

11756

1.69

57.2

50

55

8.6

30

Pioglitazone

Aronoffetal.,[4

4]2000

408

7.5–45

0.5

Type2DM

53.7

58

78

10.2

NA

--

Mild-M

od

Berhanuetal.,[4

5]2007

222

45

0.42

Type2DM

52.7

42

31

8.5

31

--

Mild-M

od

Continuednextpage



Study Characteristics

Table I summarizes characteristics of the included RCTs.

The included trials are heterogeneous with respect to these char-

acteristics. Studies had a follow-up of between 3 and 39months,

a weighted mean of 1.9 years (rosiglitazone: 1.7 years, piogli-

tazone: 2.2 years), and a median of 313 participants (rosiglita-

zone: 326, and pioglitazone: 297). Patients had a mean age of

58 years, mean hemoglobin A1c (HbA1c) of 8.5%, mean BMI of

30 kg/m2, 78% were White, and half were men. Ten RCTs re-

ported information aboutmedical history (hypertension, CAD,

hypercholesterolemia, and HF), and most RCTs excluded pa-

tients with New York Heart Association (NYHA) HF classes

III or IV. Five studies reported new-onset HF, and three studies

reported both new-onset and exacerbated HF (table I).

Ten RCTs had information on HF (eight rosiglitazone

[n = 7331], two pioglitazone [n = 5804]) [table I]. Investigator-

reported HF events were evaluated post hoc by independent

committees in four RCTs,[32,35,38,47] and these events were con-

firmed. The number of HF events was different between the

DREAM (Diabetes REduction Approaches with ramipril and

rosiglitazone Medications)[35] and PROactive (PROspective

pioglitAzone Clinical Trial In macroVascular Events)[47] trials.

These RCTs used different TZDs and patient populations

(younger patients at risk of diabetes and no evidence of CV

disease [DREAM] vs older diabetic patients with pre-existing

CV disease [PROactive]). The PROactive trial was the only one

defining serious HF events (hospitalization or prolongation of

hospitalization, fatal or life-threatening, persistent significant

disability or incapacity, and any of the following: acute left

ventricular failure, cardiac asthma, cardiac failure [acute or

chronic], cardiac failure congestive, cardiopulmonary failure,

congestive cardiomyopathy, left or right ventricular failure,

low cardiac output syndrome, pulmonary edema, or ventricular

dysfunction).

Other trials gave information on definitely worsening HF as

defined by hospitalization, worsening symptoms and signs,

suggestive findings on imaging and new medication;[32] on

confirmed HF based on typical signs and symptoms, typical

radiological evidence, and use of diuretics, vasolidators or ino-

tropes, without hospitalization;[35] on congestive HF without

any specific definition;[38,41,53] on moderate HF without a spe-

cific definition;[33] or on HF without a specific definition of

severity.[29,30]

In trials reporting HF events, insulin was the background

medication for all patients in three trials[38,41,53] and in 30% of

patients in another,[47] totaling 3018 patients (23% of the total

sample). Background use of ACE inhibitors or angiotensin IITable

I.Contd

Author,year

Sample

size

TZDdose

(mg/day)

Trial

duratio

n(y)

Typeofpatient

Mean

age

Male

(%)

White

(%)

Mean

HbA1c

Mean

BMI

Typeof

HF

HF

definition

Typeof

edema

Einhorn

etal.,[4

6]2000

328

30

0.33

Type2DM

55.6

57

84

9.8

32

--

Mild-M

od

Erdmannetal.,[4

7]2007

5238

15–45

3.25

Type2DM

61.8

66

99

7.9

31

Both

IR/IC

EC

Mild-Sev

Goldstein

etal.,[4

8]2006

142

45

0.25

Type2DM

57.5

63

87

7.2

30

--

Unclear

Herz

etal.,[4

9]2003

297

30–45

0.33

Type2DM

58.4

54

97

7.5

31

--

Mild-M

od

Kipnesetal.,[5

0]2001

560

15–30

0.33

Type2DM

56.6

89

79

9.9

32

--

Mild-M

od

Mattooetal.,[5

1]2005

289

30

0.5

Type2DM

58.9

43

97

8.8

32

--

Mild-M

od

Rosenblattetal.,[5

2]2001

197

30

0.44

Type2DM

54.5

53

66

10.5

31

--

Mild-M

od

Rosenstocketal.,[5

3]2002

566

15–30

0.33

Type2DM

57

43

65

9.8

34

Both

IRMild-M

od

Scherbaum

etal.,[5

4]2002

251

15–30

0.5

Type2DM

58.5

54

NA

9.1

29

--

Mild-M

od

Overallpioglitazone

8498

2.15

59.7

63

91

8.4

31

OverallTZD

20254

1.89

58.2

55

78

8.5

30

BMI=

bodymass

index;CAD=coronary

artery

disease;DM=diabetesmellitus;Exacer=exacerbationofHF;HbA1c=hemoglobin

A1C;HTN=hypertension;HC=hypercholesterolemia;

HF=heartfailure;IRS=insulin

resistancesyndrome;IG

T=im

pairedglucose

tolerance;IFG=im

pairedfastingglucose

;IR

=investig

ator-reportedHF;ICEC=independentclinicalevent

committee;Mod=moderate;MS=metabolic

syndrome;NA=notavailable;Sev=severe;TZD=thiazolidinedione.



type 1 receptor antagonists (angiotensin receptor blockers) was

reported in five studies[29,32,35,38,47] ranging between 49%[29]

and 100%[32] of patients and totaling 7222 patients (54% of

total sample). Nitrates were used as background medication by

2110 patients (16% of total sample); nitrate use was only re-

ported in two trials.[32,47] No differences were observed in the

use of these background medications between the TZD and

placebo arms.

Twenty-seven trials had information on peripheral edema

(16 rosiglitazone [n = 11 259], 11 pioglitazone [n = 8498]), whichwas mainly mild or moderate. 2058 edema events were report-

ed in 19 430 patients (TZD: 1393/11 016 [12.6%]; placebo:

665/8741 [7.6%]).

Meta-Analysis

Compared with placebo, patients who received TZDs had an

increased risk of HF (TZD 360/6807 [5.3%] vs placebo 234/6328[3.7%]; OR 1.59; 95% CI 1.34, 1.89; p< 0.00001; figure 2). TheHF risk was higher for rosiglitazone than for pioglitazone

(OR 2.73; 95% CI 1.46, 5.10 and 1.51; 1.26, 1.81, respectively;

w2 test = 3.5 [degrees of freedom = 1]; p = 0.06). There was no

heterogeneity of effectmeasures forHF across trials overall and

by drug (I2 = 0% for all). There was evidence of publication bias

in HF studies in the funnel plot (Egger’s test p < 0.0001).Patients who received TZDs also had an increased risk of

serious/severe HF (TZD 175/6807 [2.6%] vs placebo 118/6328[1.9%]; OR 1.47; 95% CI 1.16, 1.87; p= 0.002; figure 3).

Patients who received TZDs had an increased risk of peri-

pheral edema (1393/11 016 [12.6%] vs 665/8741 [7.6%]; OR 2.04;

95% CI 1.85, 2.26; p< 0.00001; figure 4). This risk was similar

for both rosiglitazone and pioglitazone. There was important

heterogeneity of effect measures for edema across trials overall

(I2= 48%) and within the rosiglitazone trials (I2 = 62%). There

was also evidence of publication bias in edema studies in the

funnel plot (Egger’s test p < 0.0001).

Numbers Needed to Harm

The overallNNH forHF ranged between 27 and 220 (table II).

The NNH for rosiglitazone ranged between 35 and 220, and

that for pioglitazone ranged between 27 and 95. The NNH for

serious/severe HF ranged between 62 and 134 overall, between

80 and 134 for rosiglitazone, and between 62 and 95 for pio-

glitazone. The overall NNH for peripheral edema ranged be-

tween 6 and 84, without differences between rosiglitazone and

pioglitazone.

Sensitivity Analysis of Models

The ORs for HF and peripheral edema using the Peto and

the random effects models were consistent with the ORs from

Odds ratioM-H, fixed [95% CI]

Study or subgroup

Rosiglitazone

PioglitazoneErdmann et al.[47] 2007Rosenstock et al.[53] 2002

Bhatt et al.[29] 2007 1 102 0 98 0.2% 2.91 [0.12, 72.33]1.93 [0.35, 10.72]1.87 [0.53, 6.57]3.00 [0.12, 74.40]7.03 [1.60, 30.96]4.43 [0.24, 82.80]2.01 [0.22, 18.21]1.29 [0.28, 5.88]2.73 [1.46, 5.10]

1.50 [1.26, 1.80]4.49 [0.24, 83.91]1.51 [1.26, 1.81]

1.59 [1.34, 1.89]

0.01 0.1

FavorsTZD

Favorsplacebo

1 10 100

1.0%1.8%0.2%1.0%0.3%0.6%1.4%6.5%

93.2%

93.5%

100.0%

0.3%

145114116

2634186104111

3508

2633187

2820

6328

2402013

12

222

222

234

0

4 1527 1101 117

14 26354 3824 2094 116

3823

2605379

2984

6807

39

3174

321

360

Cao et al.[30] 2006Dargie et al.[32] 2007Davidson et al.[33] 2007Gerstein et al.[35] 2006Hollander et al.[38] 2007Raskin et al.[41] 2001Rosenstock et al.[42] 2005Subtotal [95% CI]

Total [95% CI]

Subtotal [95% CI]

Total events

Total events

Total events

Heterogeneity: Chi2 = 3.19, df = 7 (p = 0.87); I2 = 0%Test for overall effect: Z = 3.13 (p = 0.002)

Heterogeneity: Chi2 = 0.54, df = 1 (p = 0.46); I2 = 0%Test for overall effect: Z = 4.51 (p < 0.00001)


TZD Placebo Odds ratioM-H, fixed, 95% CIevents total events total weight

Fig. 2. Risks of heart failure with the use of rosiglitazone and pioglitazone. CI= confidence interval; df= degrees of freedom; M-H=Mantel-Haenzel;

TZD= thiazolidinedione.



the M-H model (table III). Fixed and random RRs for HF and

edema were similar to the ORs.

Subgroup Analysis

We evaluated RCTs involving only type 2 DM patients

(n = 13 933, 69% of total sample).[26-28,31-34,36,38-54] In this sub-

group, there were 571 (TZD: 341/placebo: 230) HF events (96%of total HF events) and 1720 (1189/531) peripheral edema

events (84% of total edema events). The risks of both HF and

edema were similar to the overall analyses (table III). RCTs

involving patients at high risk for DM included 23 (19/4) HF

events and 338 (204/134) edema events; HF risks were non-

significantly higher and edema risks were non-significantly

lower than the overall risks. HF and edema risks did not differ

between the type 2 DM subgroup and the subgroup at high risk

for type 2 DM.

We also evaluated RCTs with ‡12 months of follow-up

(n = 11 713, 58% of total sample).[29,32,35,37,42,47] In this sub-

group, there were 574 (343/231) HF events (97% of total events)

and 1545 (972/573) peripheral edema events (75% of total

events). The risk of both HF and edema remained similar to

the overall analyses (table III). RCTs with shorter follow-

ups (<12 months) only included 20 (17/3) HF events and

513 (421/92) edema events; HF and edema risks were non-

significantly higher than the overall risks, and the HF risk was

not significant (p = 0.2).

Meta-Regression Analysis

In the univariable analysis, the risk of HF with the use of

TZDs decreases as the baseline risk for HF increases (p = 0.003)[figure 5]. Also, shorter average time of follow-up (p = 0.04) anduse of rosiglitazone versus pioglitazone (p = 0.01) were asso-

ciated with a higher risk of HF (table IV). There was no asso-

ciation between the risk of HF and trial sample size (p = 0.2). Inthe multivariable analysis, after adjustment for time of follow-

up and type of TZD, the association between risk of HF and

baseline risk for HF remained significant and in the same

direction (p = 0.01).

Discussion

Main Findings

This meta-analysis demonstrates that in 13 135 patients with

or at high risk for developing type 2 DM from ten placebo-

controlled RCTs, the odds of HF were 60% higher in patients

receiving TZDs. The risk of serious/severe HF was also in-

creased with the use of TZDs. The risks of HF were consistent

when using different methods and association measures. Trials

including patients with or at high risk for type 2DM, trials with

a follow-up of ‡12 months, and trials with lower overall base-

line risk for HF had significantly higher HF risks. In 29 RCTs

including 20 254 patients, the odds of peripheral edema were

Study or subgroup

Rosiglitazone

PioglitazoneErdmann et al.[47] 2007Rosenstock et al.[53] 2002

Bhatt et al.[29] 2007 1 102 0 98 0.4% 2.91 [0.12, 72.33]1.93 [0.35, 10.72]1.31 [0.34, 5.01]

4.43 [0.24, 82.80]2.01 [0.22, 18.21]1.29 [0.28, 5.88]1.75 [0.84, 3.68]

1.42 [1.10, 1.83]4.49 [0.24, 83.91]1.44 [1.12, 1.85]

1.47 [1.16, 1.87]

Not estimableNot estimable

1.8%3.3%

0.6%1.2%2.6%9.9%

89.5%0.6%

90.1%

100.0%

0.01 0.1

FavorsTZD

Favorsplacebo

1 10 100

145114116

2634186104111

3508

2633187

2820

6328

2400013

10

1080

108

118

152110117

2635382209116

3823

2605379

2984

6807

4500444

22

1494

153

175

Cao et al.[30] 2006Dargie et al.[32] 2007

TZD Placebo/Control Odds ratioM-H, fixed [95% CI]

Odds ratioM-H, fixed, 95% CIevents total events total weight

Davidson et al.[33] 2007Gerstein et al.[35] 2006Hollander et al.[38] 2007Raskin et al.[41] 2001Rosenstock et al.[42] 2005Subtotal [95% CI]

Total [95% CI]

Subtotal [95% CI]

Total events

Total events

Total events

Heterogeneity: Chi2 = 0.85, df = 5 (p = 0.97); I2 = 0%

Heterogeneity: Chi2 = 0.59, df = 1 (p = 0.44); I2 = 0%

Test for overall effect: Z = 1.48 (p = 0.14)

Test for overall effect: Z = 2.82 (p = 0.005)

Heterogeneity: Chi2 = 1.59, df = 7 (p = 0.98); I2 = 0%Test for overall effect: Z = 3.16 (p = 0.002)

Fig. 3. Risks of serious/severe heart failure with the use of rosiglitazone and pioglitazone. CI= confidence interval; df =degrees of freedom; M-H=Mantel-

Haenzel; TZD = thiazolidinedione.



100% higher in patients receiving TZDs, and the risk was

similar between rosiglitazone and pioglitazone.

Relationship to Prior Studies

We focused on placebo-controlled trials, as this design

provides reliable evidence of the effectiveness of new drugs, and

effects are not shadowed by active comparators.[55] Our results

revealed similar risks and directions to previous meta-analyses

of TZDs that reported results on placebo-controlled trials,

as shown in table V. Thus, we did not find a clear higher risk

for HF in our meta-analysis of placebo-controlled trials in

comparison to meta-analyses combining active- and placebo-

controlled trials. The risk for HF is therefore increased irre-

spective of the comparators used.

Singh et al.[56] showed the highest risk of HF based on only

three trials. However, recent extensive systematic reviews and

meta-analyses by Mannucci et al. of rosiglitazone[12] and pio-

glitazone[57] trials have demonstrated a lower risk of HF with

rosiglitazone and a similar risk of HF with pioglitazone both

compared with our findings and previous findings. These meta-

analyses incorporated several unpublished studies, and this is in

line with our finding of publication bias for trials reporting HF

events. Peripheral edema risks are also consistent with a pre-

vious meta-analysis by Berlie et al.[6] (table V). However, we are

not able to compare our findings with those from two meta-

analyses by Richter et al.[58,59] that did not report effects in

placebo-controlled trials.

Although it was an indirect comparison and a non-signif-

icant result, the risk of HF with rosiglitazone was higher than

Study or subgroup

Rosiglitazone

Pioglitazone

Agrawal et al.[26] 2003 17 405 2 419 0.3% 9.14 [2.10, 39.80]

14.49 [0.80, 261.39]3.77 [1.22, 11.69]3.55 [1.63, 7.73]3.20 [1.00, 10.23]3.58 [0.44, 29.46]1.43 [1.13, 1.81]4.60 [0.24, 87.92]3.26 [1.56, 6.80]0.76 [0.42, 1.37]4.71 [1.41, 15.74]3.52 [1.08, 11.49]3.45 [1.30, 9.15]3.06 [1.41, 6.68]

27.11 [1.66, 443.31]2.06 [1.73, 2.45]

6.26 [0.37, 106.86]2.14 [0.71, 6.48]2.47 [0.76, 8.04]1.99 [1.74, 2.28]1.48 [0.24, 9.13]0.93 [0.48, 1.79]3.71 [1.28, 10.75]4.66 [1.70, 12.78]4.95 [0.57, 43.15]2.42 [1.29, 4.54]2.55 [0.12, 53.78]2.03 [1.79, 2.30]

2.04 [1.85, 2.26]

3.38 [0.14, 84.41]0.1%0.1%0.7%1.3%0.7%0.2%

21.1%0.1%1.6%4.5%0.7%0.8%1.0%1.4%0.1%

34.9%

0.1%0.8%0.7%

55.2%0.4%3.3%0.9%0.8%0.2%2.7%0.1%

65.1%

100.0%

7187

184114116113

263434

278186158173104111105

4887

79112160

26337099

18714796

18784

3854

8741

004

1041

1240

1020335

100

196

054

4192

16451

130

469

665

6484

181110117226

263571

277382335735209116425

6372

329110168

260572

198373142101379167

4644

11016

16

1428127

1744

30322843312748

502

121010

7133

3028205

582

891

1393

Albertini et al.[27] 2007Barnett et al.[28] 2003Dailey et al.[31] 2004Dargie et al.[32] 2007Davidson et al.[33] 2007Fonseca et al.[34] 2000Gerstein et al.[35] 2006Gómez Perez et al.[36] 2002Hedblad et al.[37] 2007Hollander et al.[38] 2007Lebovitz et al.[39] 2001Phillips et al.[40] 2001Raskin et al.[41] 2001Rosenstock et al.[42] 2005Zhu et al.[43] 2003

Aronoff et al.[44] 2000Berhanu et al.[45] 2006Einhorn et al.[46] 2000Erdmann et al.[47] 2007Goldstein et al.[48] 2006Herz et al.[49] 2003Kipnes et al.[50] 2001Mattoo et al.[51] 2005Rosenblatt et al.[52] 2001Rosenstock et al.[53] 2002Scherbaum et al.[54] 2002

Subtotal [95% CI]

Subtotal [95% CI]

Total [95% CI]

Total events

Total events

Total events

TZD Placeboweightevents total events total

Odds ratioM-H, fixed [95% CI]

Odds ratioM-H, fixed, 95% CI



Heterogeneity: Chi2 = 50.38, df = 26 (p = 0.003); I2 = 48%0.01 0.1

FavorsTZD

Favorsplacebo

1 10 100Test for overall effect: Z = 13.80 (p < 0.00001)

Fig. 4. Risks of peripheral edema with the use of rosiglitazone and pioglitazone. CI = confidence interval; df= degrees of freedom; M-H=Mantel-Haenzel;

TZD= thiazolidinedione.



the risk with pioglitazone in our meta-analysis involving a

broad range of ages. This finding is consistent with those of

several recent large retrospective cohort studies in diabetic

patients older than 65 years where rosiglitazone and pioglita-

zone were directly compared.[60-62] Winkelmayer et al.[60]

studied 28 361 new TZDusers and found a significant increased

relative risk for hospitalizations due to HF of 13% (range

1–26%) with rosiglitazone versus pioglitazone. Juurlink et al.[61]

studied 39 736 Canadian patients and found a decreased risk of

acute HF with pioglitazone versus rosiglitazone over a period

of 3 years (hazard ratio [HR] 0.77; 95% CI 0.69, 0.87). Finally,

Graham et al.[62] studied 227 571 Medicare beneficiaries and

showed a sustained higher risk of hospitalizations due to HF in

patients receiving rosiglitazone in comparison to those receiv-

ing pioglitazone over a period of 18 months (HR 1.25; 95% CI

1.16, 1.34). Rosiglitazone is a more potent agonist of the per-

oxisome proliferator-activated receptor-g (PPARg) than pio-

glitazone, and activation of PPARg in the kidney seems to be

an important mechanism of TZD-induced salt and water re-

tention.[63] However, in a recent cohort study byWertz et al.[64]

in diabetic patients older than 18 years, rosiglitazone was not

associated with a higher risk of acute HF than pioglitazone in

the overall matched population of 28 938 patients (event rate

per 1000 person-years: rosiglitazone 13.2 vs pioglitazone 11.9).

Furthermore, when analyzing the subpopulation of patients

older than 65 years (n= 5377), the event rate was no higher (rosi-

glitazone 43.9 vs pioglitazone 42.4). The younger and healthier

overall population, the use of propensity score matching to

ensure similarity between rosiglitazone and pioglitazone groups,

and the limited number ofHF events in the >65-year populationmay explain differences with prior studies.

No previous meta-analysis has described or evaluated the

effect of trial characteristics on the risk of HF with the use

of TZDs. Patients at high risk of type 2 DM (vs those with

type 2 DM) and those who received TZDs for <12 months

(vs ‡12 months) were at higher risk of developing HF. However,

these findings carry a lot of uncertainty as the numberofHFevents

in those subgroups was low, and need to be confirmed. We also

found in meta-regression analysis that lower baseline risk for

HF was associated with a higher risk of developing HF. This find-

ingwasnotdrivenbyanyparticular trial.AsHFrisksdecreasewith

longer time of use of TZDs, we should be careful with extrapolat-

ing the occurrence of events in the short term to longer times of

follow-up. This is especially important for NNH calculations.

Clinical Relevance

Patients with type 2DM and those at high risk of developing

type 2 DM share a high risk of cardiovascular disease (CVD).

Table II. Number needed to harm (NNH) for overall and serious/severe heart failure (HF) events

Author, year Sample Follow-up

years

Overall HF

events

Overall event

rates

NNH for

overall HF

Serious HF

events

Serious event

rates

NNH for

serious HF

TZD placebo TZD placebo TZD placebo TZD placebo

Rosiglitazone

Bhatt et al.,[29] 2007 200 1 1 0 0.98 0 102 1 0 0.98 0 102

Cao et al.,[30] 2006 297 0.75 4 2 2.63 1.38 80 4 2 2.63 1.38 80

Dargie et al.,[32] 2007 224 1 7 4 6.36 3.51 35 5 4 4.55 3.51 96

Davidson et al.,[33] 2007 233 0.5 1 0 0.85 0 117 0 0 0 0 NA

Gerstein et al.,[35] 2006 5 269 3 14 2 0.53 0.08 220 0 0 0 0 NA

Hollander et al.,[38] 2007 568 0.5 4 0 1.05 0 96 4 0 1.05 0 96

Raskin et al.,[41] 2001 313 0.5 4 1 1.91 0.96 105 4 1 1.91 0.96 105

Rosenstock et al.,[42] 2006 227 2 4 3 3.45 2.70 134 4 3 3.45 2.70 134

Total rosiglitazone 7 331

Pioglitazone

Erdmann et al.,[47] 2007 5 238 3.25 317 222 12.17 8.43 27 149 108 5.72 4.10 62

Rosenstock et al.,[53] 2002 566 0.33 4 0 1.06 0 95 4 0 1/06 0 95

Total pioglitazone 5 804

Total TZDs 13135

NA= not applicable; TZD= thiazolidinedione.



These diseases may be associated with diastolic dysfunc-

tion.[11,65] In our meta-analysis, half of the patients with im-

paired glucose tolerance, impaired fasting glucose, or insulin

resistance syndrome[35-37] had hypertension, and a third had

hypercholesterolemia. The cardiovascular risk in patients with

metabolic syndrome[28,29] seemed higher: more than two-thirds

had hypertension, CAD, and hypercholesterolemia. The risks

of HF and edema were similar between overall and type 2 DM,

as most of these events occurred in diabetic patients.

Published meta-analyses on TZDs and HF or edema risk

have included either a broad range of follow-up times[3,6,58,59] or

only ‡12 months.[4,5,57] Although it may be unlikely that short-

term (<12 months) studies could demonstrate an impact on

harmful outcomes,[59] we found such a high risk forHF. Lincoff

et al.[3] did not find differences in HF risk according to time of

follow-up (p = 0.2). Ideally, an individual patient data meta-

analysis will adjust for different follow-up times among

patients.

Several trials removed patients with pre-existing HFNYHA

classes III and IV. Moreover, it is not clear whether HF events

were exacerbations or new events. Lago et al.[5] reported that six

of their seven trials analyzed only new-onset HF. We have in-

cluded three of those trials in our meta-analysis,[29,35,37] but one

of them[47] also included exacerbations of HF. The four trials

whose HF events were adjudicated by a committee confirmed

events reported by investigators. HF definitions were probably

different between trials, but we did not have information

available to confirm this. We also combined HF events that did

or did not require hospitalization, as we did not have details

about them. It also remains to be investigated whether or not

TZD-associated HF carries a more favorable prognosis than

HF induced by systolic or diastolic dysfunction.

Only two meta-analyses of trials have reported NNH cal-

culations. Singh et al.[4] found a broad 1-yearNNH forHFwith

the use of rosiglitazone, ranging from 30 to 383, and Lago

et al.[5] obtainedNNHvalues ranging from 35 to 491 (107 for an

average of 30 months of follow-up). These authors incorrectly

either adjusted NNH calculations to 1 year assuming equal risk

of HF over time, or calculated NNH for an average time of

follow-up. The NNH is a simple and valuable measure for

clinical practice and decision making.[15,18] However, it has

been criticized for its poor statistical properties.[66] We strongly

suggest calculating NNH values per trial, and reporting rele-

vant information to completely understand each NNH in

context, such as the baseline risk of the study population, the

absolute risk increase, and the specific time of follow-up.[66-68]

TZDs have been associated with MI,[2,4] stroke,[3] and frac-

tures.[69] Also, a non-significant increased risk of cardiovascular

Table III. Sensitivity and subgroup analyses for risks associated with

thiazolidinediones

Association measure

OR (95% CI) RR (95% CI)

Heart failure

Model (all patients)

M-H 1.59 (1.34, 1.89) 1.52 (1.30, 1.78)

Peto 1.59 (1.34, 1.88) N/A

Random 1.56 (1.32, 1.86) 1.49 (1.28, 1.75)

Type of patients

Type 2 DM

M-H 1.53 (1.29, 1.83) 1.47 (1.25, 1.72)

random 1.53 (1.28, 1.82) 1.46 (1.25, 1.71)

Othera

M-H 4.30 (1.54, 12.03) 4.25 (1.53, 11.77)

random 3.90 (1.35, 11.24) 3.83 (1.34, 10.95)

Follow-up

‡12 months

M-H 1.57 (1.31, 1.87) 1.50 (1.28, 1.76)

random 1.67 (1.16, 2.40) 1.66 (1.10, 2.50)

<12 months

M-H 2.71 (0.94, 7.79) 2.68 (0.93, 7.67)

random 2.56 (0.88, 7.44) 2.52 (0.88, 7.25)

Peripheral edema

Model (all patients)

M-H 2.04 (1.85, 2.26) 1.83 (1.68, 2.00)

Peto 1.99 (1.80, 2.18) N/A

Random 2.41 (1.91, 3.04) 2.18 (1.76, 2.70)

Type of patients

Type 2 DM

M-H 2.18 (1.95, 2.45) 1.90 (1.73, 2.09)

random 2.57 (1.97, 3.35) 2.35 (1.82, 3.04)

Othera

M-H 1.56 (1.25, 1.95) 1.52 (1.23, 1.88)

random 2.00 (0.91, 4.41) 1.91 (0.92, 3.98)

Follow-up

‡12 months

M-H 1.91 (1.71, 2.14) 1.71 (1.56, 1.88)

random 2.12 (1.57, 2.85) 1.85 (1.47, 2.34)

<12 months

M-H 2.58 (2.04, 3.25) 2.41 (1.94, 2.99)

random 2.80 (1.94, 4.05) 2.65 (1.84, 3.80)

a Includes: impaired glucose tolerance, impaired fasting glucose, insulin

resistance syndrome, and metabolic syndrome.

DM= diabetes mellitus; M-H=Mantel-Haenzel; NA= not applicable; OR= oddsratio; RR= risk ratio.



death has been found.[2-4] Thus, the benefit/harm balance of

TZDs has shifted in the direction of harm. Some factors can

predispose to TZD-associated fluid retention, such as older age,

female sex, left ventricular hypertrophy, elevated BNP, history

of CAD, renal insufficiency, and insulin therapy.[7] TZDs

should not be used in patients withHF, and should be used with

caution in patients with CVDwithout HF. In patients with type

2 DM without CVD and a low risk of HF, the use of TZDs

should be compared with the benefit/harm profile of other oral

antidiabetic drugs.[5] Recently, the Consensus of American and

European Diabetes associations relegated the use of pioglita-

zone to a third-line treatment, additionally stating that rosi-

glitazone was not recommended at any point.[70] The final

results of the long-term RECORD (Rosiglitazone Evaluated

for Cardiac Outcomes and Regulation of Glycaemia in Dia-

betes) trial[71] confirm the harmful effect of rosiglitazone on

fatal and non-fatal HF when compared with metformin and

sulfonylurea. In September 2010 the US FDA decided to re-

strict access to rosiglitazone: current users will only be able to

continue using the medication if they acknowledge and docu-

ment that they understand the risks associated with the drug,

and the drug will be available to patients not already taking it

only if they are unable to achieve glycemic control on other

medications and decide not to take pioglitazone for medical

reasons.[72] Simultaneously, the European Medicines Agency

recommended the suspension of the marketing authorizations

of rosiglitazone-containing antidiabetic medicines, which will

stop being available in Europe in the next few months.[73] Fi-

nally, the harmful profile of TZDs may also be associated with

increased costs related to hospitalization, use of imaging tech-

niques, and use of other drugs.

Study Limitations

Our meta-analysis has several limitations. First, we only

used data reported by authors. We did not contact primary

investigators for clarification of data or for additional data.

Indeed, availability of individual-patient data would have al-

lowed us to perform time-to-event analyses.[3,15] Second, HF

and edema events were not the primary outcomes of included

trials, and potential information bias may be present. Third, we

combined HF events of different definitions and variable reli-

ability. When we analyzed only serious/severe HF events, the

risks were consistent with the risk of overall HF events. Four of

the RCTs had HF events adjudicated by independent com-

mittees[32,35,38,47] comprising 86% (11 299/13 135) of the total

sample used to investigate HF risks; therefore the risk of

overestimation of this harmful effect is limited. Also, our HF

risks are consistent with previous meta-analyses. Fourth, there

is evidence of publication bias for both outcomes in our study.

The inclusion of several unpublished trials, frequently small

and with only a slight effect on HF, probably explains that

the effect of rosiglitazone on HF has been lower in a re-

cently published meta-analysis.[57] Finally, we included placebo-

controlled trials only and trials with a broad range of follow-up.

HF risks in placebo-controlled trials were similar to those from

published meta-analyses of trials controlling for both active

drugs and placebo. Therefore the HF risks of active drugs

2.00

1.75

1.50

1.25

Log

OR

of H

F

1.00

0.75

0.50

0.25

00 1 2 3 4

Baseline risk for HF (%)

5 6 7 8 9

p = 0.003

Fig. 5. Relationship between the risk of heart failure (HF) with the use of

thiazolidinediones and thebaseline risk forHF in the placeboarm.The size of the

circle corresponds to the inverse variance of the log odds ratio (OR) in that trial.

Table IV. Association between trial characteristics and the risk of heart failure in the meta-regression analysis

Trial variable Univariable Multivariable

coefficient (SE) p-value coefficient (SE) p-value

Baseline risk of the population (per 1% increase) -0.09 0.003 -0.24 0.01

Follow-up (per 1 year increase) -0.17 0.04 0.28 0.1

Rosiglitazone (vs pioglitazone) 0.50 0.01 0.62 0.2

Sample size (per 100 patients increase) -0.005 0.2



are probably small and our provided NNH values may be of

clinical utility. Some could argue that long-term trials are

necessary to observe harmful events. Indeed, we found that the

subgroup of trials with ‡12 months of follow-up had signifi-

cantly high HF risks, which is consistent with meta-analyses

that combinedplaceboandactive controls and included>12-month

trials only.

Conclusion

The use of TZDs is associated with a higher risk of incident

HF and peripheral edema in patients with or at high risk of

developing type 2 DM. The risk of serious/severe HF is also

increased with the use of TZDs.HighHF risks were observed in

the subgroup of trials with ‡12 months of follow-up. These HF

risks are similar to those from meta-analyses combining active-

and placebo-controlled trials. The growing evidence regarding

these harmful effects suggest that TZDs should not be used in

patients with HF, and in patients without HF TZD use should

be cautious or avoided in favor of other antidiabetic drugs that

have a better benefit/risk profile.

Acknowledgments

No external funding was used to support this work. The authors are

solely responsible for the design and conduct of this study, all study

analyses, the drafting and editing of the manuscript, and its final contents.

The authors have no relevant conflicts of interest.

Table V. Meta-analyses evaluating the effect of thiazolidinediones on heart failure and peripheral edema in overall and placebo-controlled (PC) randomized

clinical trials (RCTs)

Study TZD Population No. of

RCTs

Total

sample

No. of PC

RCTs

Sample in

PC trials

Follow-up Method/model

Relative risk metric (95% CI)

Heart failure

Lincoff et al.[3] Pioglitazone Type 2 DM 19 16390 9 8067 >4mo FE All: HR 1.41 (1.14, 1.76)

PC: HR 1.39 (1.09, 1.77)

Singh et al.[4] Rosiglitazone High risk of or

with type 2 DM

4 14291 2 5493 >12mo FE All: RR 2.09 (1.52, 2.88)

PC: Unknown

Lago et al.[5] Both High risk of or

with type 2 DM

7 20191 4 10931 >12mo FE All: RR 1.72 (1.21, 2.42)

PC: RR 1.97 (0.94, 4.13)

Mannucci et al.[12] Rosiglitazone With and without

type 2 DM

132 41743 103 25973 >1mo RE All: OR 1.69 (1.20, 2.36)

PC: OR 1.45 (0.90, 2.35)

Singh et al.[56] Both High risk of or

with type 2 DM

3 10731 3 10731 >12mo RE OR 2.1 (1.08, 4.08)

Mannucci et al.[57] Pioglitazone With and without

type 2 DM

94 21180 27 4202 >1mo RE All: RR 1.32 (0.88, 1.98)

PC: RR 2.0 (1.1, 5.0)a

Hernandez et al.

(current)

Both High risk of or

with type 2 DM

10 13135 10 13135 >3mo FE All: OR 1.59 (1.34, 1.89)

rosiglitazone:

OR 2.73 (1.46, 5.10)

pioglitazone:

OR 1.51 (1.26, 1.81)

Peripheral edema

Berlie et al.[6] Both Type 2 DM 26 16679 10 3079 >4mo RE All: OR 2.67 (2.22, 3.22)

PC: OR 2.35 (1.40, 3.93)

Richter et al.[58] Rosiglitazone Type 2 DM 9 4739 Unclear Unclear >6mo FE

RE

All: OR 2.27 (1.83, 2.81)

All: OR 4.62 (2.28, 9.38)

Richter et al.[59] Pioglitazone Type 2 DM 18 11565 Unclear Unclear >6mo FE

RE

All: OR: 2.22 (1.96, 2.52)

All: OR: 3.15 (2.34, 4.23)

Hernandez et al.

(current)

Both High risk of or

with type 2 DM

27 19757 27 19757 >3mo FE OR: 2.04 (1.85, 2.26)

rosiglitazone: 2.06 (1.73, 2.45)

pioglitazone: 2.03 (1.79, 2.30)

a An approximation from figure 2 of the manuscript. This association was regarded as significant only, but no actual association and 95% CI were given.

DM= diabetes mellitus; FE= fixed effects; HR= hazard ratio; OR= odds ratio; RE = random effects; RR= risk ratio.



References1. Selvin E, Bolen S, Yeh HC, et al. Cardiovascular outcomes in trials of oral

diabetes medications: a systematic review. Arch Intern Med 2008; 168: 2070-80

2. Nissen SE, Wolski K. Effect of rosiglitazone on the risk of myocardial in-

farction and death from cardiovascular causes.NEngl JMed 2007; 356: 2457-71

3. Lincoff AM, Wolski K, Nicholls SJ, et al. Pioglitazone and risk of cardio-

vascular events in patients with type 2 diabetes mellitus: a meta-analysis of

randomized trials. JAMA 2007; 298: 1180-8

4. Singh S, Loke TK, Furberg CD. Long-term risk of cardiovascular events with

rosiglitazone: a meta-analysis. JAMA 2007; 298: 1189-95

5. Lago RM, Singh PP, Nesto RW. Congestive heart failure and cardiovascular

death in patients with pre-diabetes and type 2 diabetes given thiazolidinedi-

ones: a meta-analysis of randomized clinical trials. Lancet 2007; 370: 1129-36

6. Berlie HD, Kalus JS, Jaber LA. Thiazolidinediones and the risk of edema:

A meta-analysis. Diabetes Res Clin Pract 2007; 76: 279-89

7. Karalliedde J, Buckingham RE. Thiazolidinediones and their fluid-related

adverse effects: facts, fiction and putative management strategies. Drug Saf

2007; 30: 741-53

8. BuckinghamRE,HannaA. Thiazolidinedione insulin sensitizers and the heart:

a tale of two organs? Diabetes Obes Metab 2008; 10: 312-28

9. St John SuttonM, Rendell M, Dandona P, et al. A comparison of the effects of

rosiglitazone and glyburide on cardiovascular function and glycemic control

in patients with type 2 diabetes. Diabetes Care 2002; 25: 2058-64

10. Kannel WB, Hjortland M, Castelli WP. Role of diabetes in congestive heart

failure: the Framingham Study. Am J Cardiol 1974; 34: 29-34

11. Tang WH. Do thiazolidinediones cause heart failure? A critical review. Cleve

Clinic J Med 2006; 73: 390-6

12. Mannucci E, Monami M, Di Bari M, et al. Cardiac safety profile of rosigli-

tazone: a comprehensive meta-analysis of randomized clinical trials. Int J

Cardiol 2009. doi: 10.1016/j.ijcard.2009.01.064

13. Johnson JA,Majumdar SR, Simpson SH, et al. Decreased mortality associated

with the use of metformin compared with sulfonylurea monotherapy in type 2

diabetes. Diabetes Care 2002; 25: 2244-8

14. Koro CE, Bowlin SJ, Weiss SR. Antidiabetic therapy and the risk of heart

failure in type 2 diabetic patients: an independent effect or confounding by

indication. Pharmacoepidemiol Drug Saf 2005; 14: 697-703

15. Hernandez AV, Walker E, Ioannidis JP, et al. Challenges in meta-analysis of

randomized clinical trials for rare harmful cardiovascular events: the case of

rosiglitazone. Am Heart J 2008; 156: 23-30

16. Sweeting MJ, Sutton AJ, Lambert PC. What to add to nothing? Use and

avoidance of continuity corrections in meta-analysis of sparse data. Stat Med

2004; 23: 1351-75

17. Bradburn MJ, Deeks JJ, Berlin JA, et al. Much ado about nothing: a com-

parison of the performance of meta-analytical methods with rare events.

Stat Med 2007; 26: 53-77

18. Hernandez AV, Westerhout CM, Steyerberg EW, et al. Effects of platelet

glycoprotein IIb/IIIa receptor blockers in non-ST segment elevation acute

coronary syndromes: benefit and harm in different age subgroups. Heart

2007; 93: 450-5

19. Jadad AR, Moore RA, Carroll D, et al. Assessing the quality of reports of

randomized clinical trials: is blinding necessary?Control Clin Trials 1996; 17: 1-12

20. Moher D, Liberati A, Tetzlaff J, et al. Preferred reporting items for systematic

reviews and meta-analyses: the PRISMA statement. Ann Intern Med 2009;

151: 264-9

21. Higgins JP, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat

Med 2002; 21: 1539-58

22. Egger M, Altman DG, Smith GD. Systematic reviews in health-care: meta-

analysis in context. 2nd ed. London: BMJ Books, 2001

23. Yusuf S, Peto R, Lewis J, et al. Beta blockade during and after myocardial infarc-

tion: an overview of the randomized trials. Prog Cardiovasc Dis 1985; 27: 335-71

24. DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials

1986; 7: 177-88

25. van Houwelingen HC, Arends LR, Stijnen T. Advanced methods in meta-

analysis: multivariate approach and meta-regression. Stat Med 2002; 21:

589-624

26. AgrawalA, SautterMC, JonesNP.Effects of rosiglitazonemaleatewhenadded to

a sulfonylurea regimen in patients with type 2 diabetes mellitus and mild to

moderate renal impairment: a post hoc analysis. Clin Ther 2003; 25: 2754-64

27. Albertini J-P, McMorn SO, Chen H, et al. Effect of rosiglitazone on factors

related to endothelial dysfunction in patients with type 2 diabetes mellitus.

Atherosclerosis 2007; 195: e159-66

28. Barnett AH, Grant PJ, Hitman GA, et al. Rosiglitazone in type 2 diabetes:

an evaluation in British Indo-Asian patients. Diabet Med 2003; 20: 387-93

29. Bhatt DL, Chew DP, Grines C, et al. Peroxisome proliferator-activated

receptor g agonists for the prevention of adverse events following percutane-

ous coronary revascularization: results of the PPAR Study. AmHeart J 2007;

154: 137-43

30. Cao Z, Zhou YJ, Zhao YX, et al. Rosiglitazone could improve clinical out-

comes after coronary stent implantation in non-diabetic patients with meta-

bolic syndrome. Chin Med J 2006; 119: 1171-5

31. Dailey GE, Noor MA, Park J-S, et al. Glycemic control with glyburide/metformin tablets in combination with rosiglitazone in patients with type 2

diabetes: a randomized double-blind trial. Am J Med 2004; 116: 223-9

32. DargieHJ,Hildebrandt PR,RieggerGA, et al. A randomized, placebo-controlled

trial assessing the effects of rosiglitazone on echocardiographic function and

cardiac status in type 2 diabetic patients with New York Heart Association

functional class I or II heart failure. J Am Coll Cardiol 2007; 49: 1696-704

33. Davidson JA, McMorn SO, Waterhouse BR, et al. A 24 week, multicenter,

randomized, double-blind, placebo-controlled, parallel-group study of the

efficacy and tolerability of combination therapy with rosiglitazone and sul-

fonylurea in African American and Hispanic American patients with type

2 diabetes inadequately controlled with sulfonylurea monotherapy. Clin Ther

2007; 29: 1900-14

34. Fonseca V, Rosenstock J, Patwardhan R, et al. Effect of metformin and rosi-

glitazone combination therapy in patients with type 2 diabetes mellitus:

a randomized controlled trial. JAMA 2000; 283: 1695-702

35. Gerstein HC, Yusuf S, Bosch J, et al. Effect of rosiglitazone on the frequency of

diabetes in patients with impaired glucose tolerance or impaired fasting glu-

cose: a randomised controlled trial. Lancet 2006; 368: 1096-105

36. Gomez-Perez FJ, Fanghanel-Salmon G, Barbosa JA, et al. Efficacy and safety

of rosiglitazone plus metformin in Mexicans with type 2 diabetes. Diabetes

Metab Res Rev 2002; 18: 127-34

37. Hedblad B, Zambanini A, Nilsson P, et al. Rosiglitazone and carotid IMT

progression rate in a mixed cohort of patients with type 2 diabetes and the

insulin resistance syndrome: main results from the Rosiglitazone Athero-

sclerosis Study. J Intern Med 2007; 261: 293-305

38. Hollander P, Yu D, Chou HS. Low-dose rosiglitazone in patients with insulin-

requiring type 2 diabetes. Arch Intern Med 2007; 167: 1284-90

39. Lebovitz HE, Dole JF, Patwhardhan R, et al., for the Rosiglitazone Clinical

Trials Study Group. Rosiglitazone monotherapy is effective in patients with

type 2 diabetes. J Clin Endocrinol Metab 2001; 86: 280-8

40. Phillips LS, Grunberger G, Miller E, et al. Once- and twice-daily dosing with

rosiglitazone improves glycemia control in patients with type 2 diabetes.

Diabetes Care 2001; 24: 308-15

41. Raskin P, Rendell M, Riddle MC, et al. A randomized trial of rosiglitazone

therapy in patients with inadequately controlled insulin-treated type 2 dia-

betes. Diabetes Care 2001; 24: 1226-32

42. Rosenstock J, Goldstein BJ, Vinik AI, et al. Effect of early addition of rosiglita-

zone to sulphonylurea therapy in older type 2 diabetes patients (>60 years):

the Rosglitazone Early vs SULphonylurea Titration (RESULT) study.Diabetes

Obes Metab 2006; 8: 49-57



43. Zhu XX, Pan CY, Li GW, et al. Addition of rosiglitazone to existing sulfo-

nylurea treatment in Chinese patients with type 2 diabetes and exposure to

hepatitis B or C. Diabetes Technol Ther 2003; 5: 33-42

44. Aronoff S, Rosenblatt S, Braithwaite S, et al. Pioglitazone hydrochloride

monotherapy improves glycemic control in the treatment of patients with type

2 diabetes: a 6-month randomized placebo-controlled dose-response study.

Diabetes Care 2000; 23: 1605-11

45. Berhanu P, Perez A, Yu S. Effect of pioglitazone in combination with insulin

therapy on glycaemic control, insulin dose requirement and lipid profile in

patients with type 2 diabetes previously poorly controlled with combination

therapy. Diabetes Obes Metab 2007; 9: 512-20

46. Einhorn D, Rendell M, Rosenzweig J, et al. Pioglitazone hydrochloride in

combination with metformin in the treatment of type 2 diabetes mellitus: a

randomized, placebo-controlled study. The Pioglitazone 027 Study Group.

Clin Ther 2000; 22: 1395-409

47. Erdmann E, Charbonnel B, Wilcox RG, et al. Pioglitazone use and heart fail-

ure in patients with type 2 diabetes and pre-existing cardiovascular disease:

data from the PRO active study (PROactive 08). Diabetes Care 2007; 30:

2773-8

48. Goldstein BJ, Rosenstock J, Anzalone D, et al. Effect of tesaglitazar, a dual

PPARa/g agonist, on glucose and lipid abnormalities in patients with type

2 diabetes: a 12-week dose-ranging trial. Curr Med Res Opin 2006; 22:

2575-90

49. Herz M, Johns D, Reviriego J, et al. A randomized, double-blind, placebo-

controlled, clinical trial of the effects of pioglitazone on glycemic control and

dyslipidemia in oral antihyperglycemic medication-naıve patients with type 2

diabetes mellitus. Clin Ther 2003; 25: 1074-95

50. Kipnes MS, Krosnick A, Rendell MS, et al. Pioglitazone hydrochloride in

combination with sulfonylurea therapy improves glycemic control in patients

with type 2 diabetes mellitus: a randomized, placebo-controlled study. Am J

Med 2001; 111: 10-7

51. Mattoo V, Eckland D, Widel M, et al. Metabolic effect of pioglitazone in

combination with insulin in patients with type 2 diabetes mellitus whose dis-

ease is not adequately controlled with insulin therapy: results of a six-month,

randomized, double-blind, prospective, multicenter, parallel-group study.

Clin Ther 2005; 27: 554-67

52. Rosenblatt S, Miskin B, Glazer NB, et al., and the Pioglitazone 026 Study

Group. The impact of pioglitazone on glycemic control and atherogenic

dyslipidemia in patients with type 2 diabetes mellitus. Coron Artery Dis 2001;

12: 413-23

53. Rosenstock J, Einhorn D, Hershon K, et al., and the Pioglitazone 014 Study

Group. Efficacy and safety of pioglitazone in type 2 diabetes: a randomised,

placebo-controlled study in patients receiving stable insulin therapy. Int J Clin

Pract 2002; 56: 251-7

54. Scherbaum WA, Goke B, and the German Pioglitazone Study Group. Meta-

bolic efficacy and safety of once-daily pioglitazone monotherapy in patients

with type 2 diabetes: a double-blind, placebo-controlled study. Horm Metab

Res 2002; 34: 589-95

55. Temple R, Ellenberg SS. Placebo-controlled trials and active-control trials in

the evaluation of new treatments. Ann Intern Med 2000; 133: 455-63

56. Singh S, Loke YK, Furberg CD. Thiazolidinediones and heart failure: a teleo-

analysis. Diabetes Care 2007; 30: 2148-53

57. Mannucci E, Monami M, Lamanna C, et al. Pioglitazone and cardiovascular

risk: a comprehensive meta-analysis of randomized clinical trials. Diabetes

Obes Metab 2008; 10: 1221-38

58. Richter B, Bandeira-Echtler E, Bergerhoff K, et al. Rosiglitazone for type 2

diabetes mellitus. Cochrane Database of Systematic Reviews 2007; Issue 3:

CD006063. doi: 10.1002/14651858.CD006063.pub2

59. Richter B, Bandeira-Echtler E, Bergerhoff K, et al. Pioglitazone for type 2

diabetes mellitus. Cochrane Database of Systematic Reviews 2007; Issue 4:

CD006060. doi: 10.1002/14651858.CD006060.pub2

60. Winkelmayer WC, Setoguchi S, Levin R, et al. Comparison of cardiovascular

outcomes in elderly patients with diabetes who initiated rosiglitazone vs

pioglitazone therapy. Arch Intern Med 2008; 168: 2368-75

61. Juurlink DN, Gomes T, Lipscombe LL, et al. Adverse cardiovascular events

during treatment with pioglitazone and rosiglitazone: population based co-

hort study. BMJ 2009; 339: b2942

62. Graham DJ, Ouellet-Hellstrom R, MaCurdy TE, et al. Risk of acute myo-

cardial infarction, stroke, heart failure, and death in elderlyMedicare patients

treated with rosiglitazone or pioglitazone. JAMA 2010; 304: 411-8

63. Guan Y, Hao C, Cha DR, et al. Thiazolidinediones expand body fluid volume

through PPARgamma stimulation of ENaC-mediated renal salt absorption.

Nat Med 2005; 11: 861-6

64. Wertz DA, Chang CL, Sarawate CA, et al. Risk of cardiovascular events and

all-cause mortality in patients treated with thiazolidinediones in a managed-

care population. Circ Cardiovasc Qual Outcomes 2010; 3: 538-45

65. Rutter MK, Parise H, Benjamin EJ, et al. Impact of glucose intolerance and

insulin resistance on cardiac structure and function: sex-related differences in

the Framingham Heart Study. Circulation 2003; 107: 448-54

66. Hutton JL. Number needed to treat and number needed to harm are not the

best way to report and assess the results of randomised clinical trials. Br J

Haematol 2009; 146: 27-30

67. McAlister FA. The ‘‘number needed to treat’’ turns 20 – and continues to be

used and misused. CMAJ 2008; 179: 549-53

68. Massel D. The number need to harm: is it too optimistic? Can J Cardiol 2003;

19: 1490-2

69. Loke YK, Singh S, Furberg CD. Long-term use of thiazolidinediones and

fractures in type 2 diabetes: a meta-analysis. CMAJ 2009; 180: 32-9

70. Nathan DM, Buse JB, Davidson MB, et al. Medical management of hyper-

glycemia in type 2 diabetes: a consensus algorithm for the initiation and

adjustment of therapy. A consensus statement of the American Diabetes Asso-

ciation and the European Association for the Study of Diabetes. Diabetes

Care 2009; 32: 193-203

71. Home PD, Pocock SJ, Beck-Nielsen H, et al. Rosiglitazone evaluated for cardio-

vascular outcomes in oral agent combination therapy for type 2 diabetes

(RECORD): amulticentre, randomised,open-label trial.Lancet 2009; 373: 2125-35

72. USFDADrug Safety andAvailability. Postmarket drug safety information for

patients and providers [online]. Available from URL: http://www.fda.gov/

Drugs/DrugSafety/PostmarketDrugSafetyInformationforPatientsandProviders/

ucm226976.htm [Accessed 2010 Dec 8]

73. European Medicines Agency press release. European Medicines Agency rec-

ommends suspension of Avandia, Avandamet and Avaglim [online]. Avail-

able from URL: http://www.ema.europa.eu/docs/en_GB/document_library/

Press_release/2010/09/WC500096996.pdf [Accessed 2010 Dec 8]

Correspondence: Dr Adrian V. Hernandez, MD, PhD, Assistant Professor of

Medicine, Department of Quantitative Health Sciences/JJN3-01, Cleveland

Clinic, 9500 Euclid Avenue, Cleveland, OH 44195, USA.

E-mail: [email protected]



thiazolidinediones and risk of heart failure in patients with or at high risk of type 2 diabetes...

Documents