Intended for healthcare professionals

CCBYNC Open access

Comparative cardiovascular effects of thiazolidinediones: systematic review and meta-analysis of observational studies

BMJ 2011; 342 doi: (Published 17 March 2011) Cite this as: BMJ 2011;342:d1309
  1. Yoon Kong Loke, senior lecturer in clinical pharmacology1,
  2. Chun Shing Kwok, medical student1,
  3. Sonal Singh, assistant professor of medicine2
  1. 1School of Medicine, Health Policy and Practice, University of East Anglia, Norwich NR4 7TJ, UK
  2. 2Johns Hopkins University School of Medicine, Baltimore, MD 21283, USA
  1. Correspondence to: Y K Loke y.loke{at}
  • Accepted 15 February 2011


Objective To determine the comparative effects of the thiazolidinediones (rosiglitazone and pioglitazone) on myocardial infarction, congestive heart failure, and mortality in patients with type 2 diabetes.

Design Systematic review and meta-analysis of observational studies.

Data sources Searches of Medline and Embase in September 2010.

Study selection Observational studies that directly compared the risk of cardiovascular outcomes for rosiglitazone and pioglitazone among patients with type 2 diabetes mellitus were included.

Data extraction Random effects meta-analysis (inverse variance method) was used to calculate the odds ratios for cardiovascular outcomes with thiazolidinedione use. The I2 statistic was used to assess statistical heterogeneity.

Results Cardiovascular outcomes from 16 observational studies (4 case-control studies and 12 retrospective cohort studies), including 810 000 thiazolidinedione users, were evaluated after a detailed review of 189 citations. Compared with pioglitazone, use of rosiglitazone was associated with a statistically significant increase in the odds of myocardial infarction (n=15 studies; odds ratio 1.16, 95% confidence interval 1.07 to 1.24; P<0.001; I2=46%), congestive heart failure (n=8; 1.22, 1.14 to 1.31; P<0.001; I2=37%), and death (n=8; 1.14, 1.09 to 1.20; P<0.001; I2=0%). Numbers needed to treat to harm (NNH), depending on the population at risk, suggest 170 excess myocardial infarctions, 649 excess cases of heart failure, and 431 excess deaths for every 100 000 patients who receive rosiglitazone rather than pioglitazone.

Conclusion Among patients with type 2 diabetes, use of rosiglitazone is associated with significantly higher odds of congestive heart failure, myocardial infarction, and death relative to pioglitazone in real world settings.


Troglitazone, the first thiazolidinedione, was withdrawn from the market because of liver toxicity.1 Muraglitazar, a dual peroxisome proliferator activated receptor (PPAR) agonist, failed to achieve regulatory approval because of concerns about adverse cardiovascular events.2 Rosiglitazone and pioglitazone are the available thiazolidinediones in North America, but meta-analyses of randomised controlled trials have suggested an increased risk of ischaemic cardiovascular events with rosiglitazone.3 4 In contrast, meta-analysis of trials of pioglitazone indicates the possibility of an ischaemic cardiovascular benefit.5 Robust evidence also shows that both drugs increase the risk of congestive heart failure and fractures, but whether any meaningful difference exists in the magnitude of risk between the two thiazolidinediones is not known.6 7 The European Medicines Agency has recommended the suspension of marketing authorisation for rosiglitazone, whereas the US Food and Drug Administration has allowed the continued marketing of rosiglitazone with additional restrictions.8

No long term trials with cardiovascular outcomes have directly compared these two drugs. Clinical trials have strict selection criteria that may exclude participants at high risk of adverse events, and adverse cardiovascular outcomes can be rare in such trials.9 On the other hand, population based observational studies resemble clinical practice, where patients may have risk factors for cardiovascular disease or comorbidities. Therefore, consideration of the evidence from carefully conducted observational studies is essential to determine if any difference in cardiovascular events or mortality exists between the two drugs.

Our objective was to systematically determine the comparative effects of rosiglitazone and pioglitazone on cardiovascular outcomes (myocardial infarction and congestive heart failure) and mortality from observational studies in patients with type 2 diabetes. We aimed specifically to calculate the pooled odds ratios for adverse cardiovascular events with rosiglitazone compared with pioglitazone—that is, the relative likelihood of cardiovascular harm if rosiglitazone was used rather than pioglitazone.


Eligibility criteria

We selected controlled observational (non-randomised) studies that reported on cardiac outcomes in patients receiving rosiglitazone compared with pioglitazone. We included studies of a cohort or case-control design that enrolled participants with type 2 diabetes mellitus. The primary outcome of interest was myocardial infarction. Secondary outcome measures were congestive heart failure and overall mortality. Eligible studies had to present one of the following: odds ratio, relative risk, hazard ratio, or sufficient raw data to enable calculation of the odds ratio where not otherwise reported.

Search strategy

We searched Medline and Embase by using Ovid SP (from inception to the end of September 2010), with the search terms (pioglitazone or rosiglitazone or thiazolidinedione$).mp and (myocardial-infarction or cardiovascular or cardiac or heart).mp and (cohort or case-control or observational or retrospective).mp. We did not use any language restrictions, but we limited the search to human studies. Additionally, we signed up with PubMed to receive automated electronic notification of any new articles containing the above search terms. To identify unpublished studies, we reviewed the regulatory authorities’ websites (US Food and Drug Administration and European Medicines Agency), as well as the study registers of the drug manufacturers GlaxoSmithKline and Takeda. We checked the bibliographies of included studies and recent review articles for additional relevant articles.

Study selection and data extraction

Two reviewers (CSK and YKL or SS) checked all titles and abstracts for studies that could potentially meet the inclusion criteria. We retrieved full reports of these potentially eligible studies for detailed assessment by two reviewers (CSK and YKL), who then independently extracted information on study design, drug use, study location, characteristics of participants, and relevant outcomes on to a preformatted spreadsheet. Any uncertainties or discrepancies between the two reviewers were resolved through consensus after re-checking of the source data and consultation with the third reviewer. We also contacted authors if any areas of uncertainty needed clarification.

Where different timings and durations of thiazolidinedione use were reported in the study participants, we pre-specified that data would be preferentially extracted from the participants with current or most recent use, until cessation of treatment. We also aimed to extract risk estimates pertaining to overall use in the entire cohort rather than in any specific subgroups.

Risk of bias

In accordance with the recommendations of the Cochrane Adverse Effects Methods Group, we checked the methods of selection of participants (including baseline characteristics and adjustment for confounders), nature of follow-up, ascertainment of drug use, and definition and monitoring of adverse outcomes.10 To counter selective reporting bias, we contacted authors when relevant cardiovascular outcomes were potentially measured but were not reported or were stated to be non-significant. We used a funnel plot to assess publication bias.

Data analysis

We used RevMan 5.0.25 (Nordic Cochrane Centre) to do random effects meta-analysis using the inverse variance method for pooled odds ratios. We used the fixed effects model for sensitivity analysis. We assumed similarity between the odds ratio and other relative measures such as relative risk, rate ratios, or hazard ratios because cardiovascular events and deaths were rare events.11

Where possible, we aimed to pool adjusted odds ratios from the primary studies; otherwise, we used raw outcome data to yield unadjusted odds ratios. In view of the potential diversity of study designs, we grouped the studies for the analysis according to studies for which only the unadjusted odds ratios were available, with no correction for baseline differences or confounding, and those for which we were able to extract odds ratios adjusted for potential confounders.

For consistency in direction of risk comparisons, we used the odds ratio to assess the magnitude of risk for rosiglitazone use compared with that for pioglitazone use. For studies that reported the odds ratio for pioglitazone compared with rosiglitazone, we used the reciprocal of the point estimate and the bounds of the confidence intervals.

We estimated the number needed to treat to harm per year (NNH) (and 95% confidence interval) by applying the pooled odds ratio from the meta-analysis to the annual rate of the event in different populations.12 The NNH is the number of patients with type 2 diabetes who need to be treated with rosiglitazone rather than pioglitazone for one additional patient to have an adverse outcome.

Statistical heterogeneity

We used the I2 statistic to assess statistical heterogeneity. I2 values of 30-60% represented a moderate level of heterogeneity.13


Figure 1 shows the process of selection of studies. We retrieved 16 observational studies involving 810 000 thiazolidinedione users (429 000 patients taking rosiglitazone and 381 000 taking pioglitazone), after a detailed review of 189 citations.14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 Fifteen studies reported on the outcome of myocardial infarction, eight studies reported on the outcome of congestive heart failure, and eight studies reported on mortality.


Fig 1 Flow diagram of process of selection of articles for meta-analysis. GSK=GlaxoSmithKline

We found 12 retrospective cohort studies and four case-control studies. The mean or median follow-up time ranged from 105 days to 7.1 years. Four studies reported duration of thiazolidinedione use, with a range of 215 to 450 days. The mean age of participants ranged from 54 to 76 years across studies, but most study participants were generally aged above 60 years; only two studies reported the average age of their participants as under 60 years. An average of 55% of patients were male across 15 studies. Table 1 shows the main characteristics of the studies and participants; table 2 shows the outcomes, interventions, and quality assessments.

Table 1

 Design and characteristics of included studies

View this table:
Table 2

 Drug use, study outcomes, and potential sources of bias

View this table:

Risk of bias

The included studies were broadly similar in terms of ascertainment of drug use and cardiovascular outcomes (table 2); they relied mainly on computerised diagnostic codes, pharmacy claims databases, and retrospective chart reviews. Few researchers made attempts to verify drug history directly with the patients or to check the validity of the prescriptions data source; only one study was deemed to have specific validation of drug use.14 Most studies reported the accuracy of outcome ascertainment on the basis of historical validation studies; only two studies specifically cross checked or validated outcomes for this analysis.15 23 Both of these studies showed a limited degree of misclassification.15 23

None of the studies provided details about the severity and consequences of the cardiac adverse events. Almost all the studies used a wide variety of variables to adjust for potential confounders. Two cohort studies checked specifically for similarities between the rosiglitazone and pioglitazone populations and did not find any major differences in the characteristics evaluated.17 19 We were able to use adjusted risk estimates for most studies, except for two studies for which we calculated odds ratios from the raw data.18 21

Myocardial infarction

Compared with pioglitazone, use of rosiglitazone was associated with a significantly increased odds of myocardial infarction from 15 studies (pooled odds ratio 1.16, 95% confidence interval 1.07 to 1.24; P<0.001) (fig 2). We found a moderate level of heterogeneity (I2=46%) for the pooled results for myocardial infarction, which stemmed from combining the unadjusted and adjusted studies together for the overall estimate.


Fig 2 Meta-analysis of odds ratio for myocardial infarction with rosiglitazone versus pioglitazone

Heart failure

Based on the pooled results of eight studies, the odds of congestive heart failure were statistically significantly higher for rosiglitazone than for pioglitazone (odds ratio 1.22, 1.14 to 1.31; P<0.001), with moderate statistical heterogeneity (I2=37%) (fig 3).


Fig 3 Meta-analysis of odds ratio for heart failure with rosiglitazone versus pioglitazone

Overall mortality

The odds of death were statistically significantly higher for rosiglitazone than for pioglitazone when we pooled eight studies, with an odds ratio of 1.14 (1.09 to 1.20; P<0.001) (fig 4). We found no evidence of statistical heterogeneity for this outcome (I2=0%).


Fig 4 Meta-analysis of odds ratio for overall mortality with rosiglitazone versus pioglitazone

Number needed to treat for harm

In a low risk population (age 45-64 years) with type 2 diabetes but no previous history of myocardial infarction, the underlying incidence of myocardial infarction was 1.08% per year.30 Use of rosiglitazone here would result in an annual NNH of 587 (95% confidence interval 392 to 1339). This can be equated to 170 excess myocardial infarctions for every 100 000 patients who received rosiglitazone rather than pioglitazone.

In a US cohort of patients with type 2 diabetes, the baseline incidence of heart failure was found to be 3.08% per year.31 Use of rosiglitazone here would result in an annual NNH of 154 (110 to 241). This can be equated to 649 excess cases of heart failure for every 100 000 patients who received rosiglitazone rather than pioglitazone.

In a large French registry study of patients with type 2 diabetes and atherosclerosis, the underlying mortality rate was found to be 3.15% per year.32 Use of rosiglitazone here would result in an annual NNH of 232 (163 to 360). This can be equated to 431 excess deaths for every 100 000 patients who received rosiglitazone rather than pioglitazone.

Sensitivity analysis

Meta-analysis using the fixed effects model yielded estimates that were similar in direction and magnitude to those from the random effects model for myocardial infarction (odds ratio 1.15, 1.10 to 1.21), heart failure (1.23, 1.17 to 1.29), and overall mortality (1.14, 1.09 to 1.20).

In view of potential patient selection bias arising after publication of a meta-analysis in May 2007 that showed increased myocardial risk with rosiglitazone,33 we did a post hoc analysis by excluding the single study that had a substantial proportion of patients recruited after May 2007.17 This did not appreciably change the direction and magnitude of the estimates for myocardial infarction (odds ratio 1.17, 1.08 to 1.27), heart failure (1.21, 1.10 to 1.33), and overall mortality (1.13, 1.04 to 1.24). Further exclusion of another study (recruitment dates 2002 to 2008) did not appreciably change the odds ratios for myocardial infarction (1.19, 1.09 to 1.29), heart failure (1.18, 1.06 to 1.33), and overall mortality (1.14, 1.06 to 1.22).19

Assessment of publication bias

The funnel plot showed that risk estimates stemmed mostly from large, precise studies that seemed to be fairly well distributed, with no definite evidence of asymmetry (fig 5).


Fig 5 Funnel plot based on odds ratio for myocardial infarction and study precision


Our results suggest a modest but statistically significant increase in the odds of myocardial infarction (approximately 16%), congestive heart failure (approximately 23%), and mortality (approximately 14%) with use of rosiglitazone compared with those for pioglitazone use in real world studies among patients with type 2 diabetes. The consistency in the magnitude of increased risk for the different cardiac outcomes, as well as mortality, indicates that this is unlikely to be a chance finding. Other strengths of our analysis include the large number of thiazolidinedione users (around 810 000) and the absence of substantial statistical heterogeneity, which suggests that the risk is maintained across most populations and is unaffected by geographical variations.

Comparison with other studies

Our synthesis of evidence from observational studies extends the findings of a cardiovascular hazard with rosiglitazone from meta-analysis of clinical trials to real world settings and suggests the possibility of a cardiovascular difference between the two drugs.3 4 5 Adjusted indirect comparisons of the risk estimates from meta-analysis of myocardial infarction and heart failure in randomised controlled trials shows that rosiglitazone is associated with an increased relative risk of 1.58 (95% confidence interval 1.14 to 2.20) for myocardial infarction and 1.48 (1.01 to 2.18) for heart failure, compared with pioglitazone.3 4 5 34 The direction of effect for both outcomes is consistent with our analysis, whereas the relatively lower point estimates seen in our analysis may reflect the generally more conservative nature of estimates of harm that has been noted with observational studies.35 Participants in trials may differ from those in observational studies, because most of the observational studies recruited a wider, more generalisable range of patients by not enforcing rigid inclusion and exclusion criteria relating to comorbid conditions. Of the 16 included studies, two excluded patients with existing cardiac conditions (so that they could study incident disease)20 23 and only one excluded patients who had comorbidities such as renal or liver disease.

As both drugs are known to cause heart failure, the increased risk of congestive heart failure associated with rosiglitazone compared with pioglitazone represents its differential cardiovascular toxicity. Two interpretations for their differential effects on myocardial infarction are possible. One possibility is that these findings represent an ischaemic cardiovascular benefit with pioglitazone. However, conclusive evidence on ischaemic cardiovascular benefit with pioglitazone is lacking; a meta-analysis of trials yielded a relative risk of 0.81 (0.64 to 1.02) for myocardial infarction. The other possibility is a greater ischaemic cardiovascular hazard with rosiglitazone, consistent with evidence from clinical trials.3 4

Possible biological mechanism

The precise biological mechanisms responsible for these differences in cardiovascular risk and mortality are uncertain. Significant differences have been found between the thiazolidinediones in lipid metabolism; rosiglitazone causes greater elevations of triglycerides and low density lipoprotein cholesterol than does pioglitazone.36 Pioglitazone had a significantly more favourable effect on triglycerides, high density lipoprotein cholesterol, low density lipoprotein particle concentration, and low density lipoprotein particle size than did rosiglitazone. Whereas pioglitazone has shown some potential benefit in preventing progression of atherosclerosis,37 rosiglitazone failed to show any significant effect in preventing atherosclerosis in a recent study.38 The more powerful renal PPARγ agonistic effect of rosiglitazone, leading to more fluid retention, may explain its greater risk of congestive heart failure.39 40

Clinical and policy implications

Our findings have important implications. Rosiglitazone is still available on a restricted basis in the United States and Canada.8 41 However, for patients who need thiazolidinedione treatment, continued use of rosiglitazone may lead to excess heart attacks, heart failure, and mortality, compared with pioglitazone. The size of the effect on public health may be considerable, given the data from June 2009 showing that about 3.8 million prescriptions for rosiglitazone were dispensed annually in the United States.42 However, other adverse effects are associated with both the thiazolidinediones, such as the doubling of risk of fracture in women.7 Concerns also exist about a modest increase in the risk of bladder cancer with pioglitazone after long term use in an observational study and a higher percentage of bladder cancers with pioglitazone relative to comparator arms in long term randomised controlled trials.43 Further studies are needed to investigate these other adverse events, as clinicians need to balance these risks and benefits against those of emerging alternative agents such as incretin mimetics that may or may not be safer than thiazolidinediones.

Limitations of study

Our analysis has some limitations, relating mainly to the quality of the primary studies. Misclassification of outcomes and drug use may occur in observational studies that rely on healthcare databases and discharge codes. However, any potential misclassification of drug use and outcomes would affect both thiazolidinediones equally. Non-randomised data are susceptible to selection bias and residual confounding. However, investigators of the two largest cohort studies found little difference in the baseline demographics and cardiovascular risk of patients who used pioglitazone and rosiglitazone.17 19 Both drugs are from the same class and were licensed for similar indications. Until May 2007, no reasons existed why any specific group of patients would have been systematically channelled towards one thiazolidinedione or the other. Our risk estimates did not change despite exclusion of the two studies that recruited after May 2007.17 19 Potential exists for bias in selection of outcomes for analysis. However, although selective reporting favouring significant beneficial outcomes may occur, one cannot assume that reporting bias is similarly focused on significant findings of harm. The converse may occur with competing interests that emphasise interpretation and reporting of safety in a manner that is favourable to rosiglitazone.44 Finally, we had insufficient data to assess effects on stroke or death from cardiac causes.


Our results show that among patients with type 2 diabetes, use of rosiglitazone is associated with a modest but statistically significant increase in the odds of myocardial infarction, congestive heart failure, and death compared with patients receiving pioglitazone in real world settings. Clinicians, patients, and regulatory authorities should carefully consider these results in the context of the available information on the thiazolidinediones’ benefits on glycaemic control and harm relating to different outcomes.

What is already known on this topic

  • Both rosiglitazone and pioglitazone are known to increase the risk of congestive heart failure

  • However, the risk of ischaemic cardiovascular events seems to have been mainly associated with rosiglitazone rather than pioglitazone

  • Any distinct differences between the cardiovascular effects of the thiazolidinediones have yet to be fully clarified

What this study adds

  • In observational studies, rosiglitazone is associated with increased odds of congestive heart failure, myocardial infarction, and death compared with pioglitazone


Cite this as: BMJ 2011;342:d1309


  • Contributors: YKL, CSK, and SS developed the concept and protocol for the review. CSK and YKL abstracted and analysed data. YKL, CSK, and SS wrote the manuscript. YKL had full access to all of the data in the study, takes responsibility for the integrity of the data and the accuracy of the data analysis, and is the guarantor.

  • Funding: SS is supported by grant number 1KL2RR025006-03 from the NCRR, a component of the National Institutes of Health (NIH), and NIH Roadmap for Medical Research. The design and conduct of the study; collection, management, analysis, and interpretation of the data; and preparation, review, and approval of the manuscript was independent of any sources of funding. Its contents are solely the responsibility of the authors and do not necessarily represent the official view of NCRR or NIH.

  • Competing interests: All authors have completed the Unified Competing Interest form at (available on request from the corresponding author) and declare that (1) they have no relationships with any company that might have an interest in the submitted work in the previous three years; (2) their spouses, partners, or children have no financial relationships that may be relevant to the submitted work; and (4) they have no non-financial interests that may be relevant to the submitted work.

  • Data sharing: No additional data available.

This is an open-access article distributed under the terms of the Creative Commons Attribution Non-commercial License, which permits use, distribution, and reproduction in any medium, provided the original work is properly cited, the use is non commercial and is otherwise in compliance with the license. See: and


View Abstract