Pioglitazone and Risk of Cardiovascular Events in Patients With Type 2 Diabetes Mellitus:  A Meta-analysis of Randomized Trials FREE

Thiazolidinediones are agonists of the peroxisome proliferation–activated receptor γ (PPAR-γ), which regulate transcription of a variety of genes encoding proteins involved in glucose homeostasis and lipid metabolism.^1- 2 By virtue of their efficacy in achieving glycemic control, the thiazolidinediones pioglitazone and rosiglitazone are both widely used to treat patients with type 2 diabetes mellitus. Although these agents can cause peripheral edema and congestive heart failure,^3- 4 their beneficial effects on glucose metabolism and insulin sensitivity have stimulated interest that thiazolidinediones might reduce ischemic cardiovascular complications of diabetes mellitus.

However, a recent meta-analysis of 42 trials comparing rosiglitazone with placebo or active comparators in more than 27 000 patients with diabetes suggested that treatment with rosiglitazone was associated with an increased risk of myocardial infarction and cardiovascular death.⁵ Furthermore, a previous analysis had demonstrated that muraglitazar, an investigational dual agonist of both α- and γ-isoforms of PPAR, was also associated with an excess incidence of death and major cardiovascular events in patients with diabetes.⁶

Whether pioglitazone shares the same risks for ischemic cardiovascular complications has not been adequately evaluated. The Prospective Pioglitazone Clinical Trial in Macrovascular Events (PROactive) studied more than 5000 patients with diabetes at high risk for macrovascular complications, and reported that treatment with pioglitazone produced a nonsignificant reduced risk for coronary and peripheral vascular events.⁷ For a secondary end point, a composite of death, myocardial infarction, or stroke, a statistically significant benefit was observed. However, pioglitazone increased the incidence of congestive heart failure (although not of mortality associated with heart failure). Moreover, it was unclear whether findings among the patients in PROactive could be extrapolated to lower-risk populations of diabetic patients without established vascular disease.

The objective of the present meta-analysis was to systematically evaluate the effect of pioglitazone on the incidence of ischemic cardiovascular complications, using the complete set of patient-level, time-to-event data from the randomized controlled trials of pioglitazone in diabetes mellitus.

A database containing individual patient data collected during eligible clinical trials of pioglitazone was transferred by its manufacturer (Takeda, Lincolnshire, Illinois) to the Cleveland Clinic Cardiovascular Coordinating Center, an academic research organization in Cleveland, Ohio, for independent analysis (Figure 1). All studies for which the data were finalized within the manufacturer's clinical development database for pioglitazone were reviewed. The criteria for eligibility for this meta-analysis were that the trial be randomized, double-blinded, and controlled with placebo or active comparator.

A total of 19 trials enrolling 16 390 patients met these criteria and formed the basis of this analysis (Table 1). Two additional trials within the manufacturer's database were excluded and were not transmitted to the Cleveland Clinic either because the study was not finalized (phase 4 trial of 315 patients) or because the study database had only been recently received by Takeda and had not yet been translated from French (phase 3 trial of 299 patients). Preliminary review by Takeda of the events in those 2 trials detected only 2 potential end points—serious heart failure in the placebo group of the first study and 1 death due to hemorrhagic stroke in the pioglitazone group of the second study.

An additional 20 completed trials enrolling a total of 3014 patients were identified that had not been conducted by the manufacturer and for which databases were not housed at Takeda. These trials had been conducted in Japan (7 studies, 1026 patients), Italy (4 trials, 414 patients), Germany (2 trials, 308 patients), other countries in Europe or South Africa (4 trials, 758 patients), China (1 trial, 236 patients), Mexico (1 trial, 244 patients), or the United States (1 trial, 28 patients). Because study databases and time-to-event data were unavailable, we excluded these trials from our meta-analysis. In lieu of availability of the databases, Takeda identified from the study reports 10 primary end points and 1 heart failure event among 1586 pioglitazone-treated patients compared with 9 primary end points and 2 heart failure events among 1428 control patients.

Protocols for each of the individual trials used for this meta-analysis had been approved by local institutional review boards, and all patients gave written informed consent. Approval of our institutional review board was not required for this meta-analysis because only deidentified data were used.

Patient inclusion and exclusion criteria for individual trials varied. In general, studies included adult patients with type 2 diabetes mellitus and inadequate glycemic control and excluded patients with undue safety risks. The primary objective of most of the trials was to determine the efficacy of pioglitazone, in combination or comparison with insulin, metformin, sulfonylureas, or rosiglitazone, in improving glycemic control. Six trials had as their primary end point other assessments of safety or efficacy: hepatic toxicity (OPI-506), triglyceride levels (GLAI), changes in carotid intima-medial thickness (Carotid Intima-Media Thickness in Atherosclerosis Using Pioglitazone [CHICAGO]),⁸ cardiovascular outcomes among patients with established vascular disease (PROactive),⁷ walking distance among patients with mild cardiac disease (OPI-520), or heart failure progression among patients with advanced congestive heart failure (OPI-504).

We specified the composite end point of death from any cause, nonfatal myocardial infarction, or nonfatal stroke as the primary outcome of this meta-analysis. This composite end point represents irreversible ischemic events and is widely used for cardiovascular outcome trials of chronic therapies.⁹ A prespecified secondary end point was serious heart failure. The composite outcome of death and serious heart failure was also determined to avoid potential bias introduced by censoring of patients due to death.

Because most of the trials included in this meta-analysis were designed primarily to assess glycemic control with pioglitazone, vascular ischemic end points were not prospectively defined in a uniform fashion across the clinical trial database. For PROactive and Carotid Intima-Media Thickness in AthCHICAGO, the end point events of death, myocardial infarction, and stroke were adjudicated by blinded clinical events committees according to definitions specified in the trial protocols. For CHICAGO and OP-504, serious heart failure was adjudicated. For those studies with central clinical events committees, adjudicated end points were used for the current meta-analysis. For all other trials, end points were identified from safety data collected as part of routine adverse event reporting for each study.

Specific preferred terms from the Medical Dictionary for Regulatory Activities version 9.0 (Northrop Grumman Corp, Reston, Virginia) were mapped to the end points of myocardial infarction, stroke, and heart failure. Study reports or databases were searched manually or electronically for the preferred terms associated with myocardial infarction and stroke among adverse event (AE) listings. Heart failure end points were defined as serious by identification of mapped preferred terms that had been reported as serious adverse events (SAE).

All end points were analyzed according to time-to-event (defined as the period [in days] from the first dose of study medication to first occurrence of the event). Patients who did not experience an event were censored at the date of last known follow-up. Hazard ratios (HRs) and 95% confidence intervals (CIs) were calculated using a Cox proportional hazards model to compare the effect of pioglitazone vs the control therapy for each trial. Studies of similar duration were collapsed together (3-6 months, > 6 months-12 months, > 12 months-24 months, and > 24 months) and used as a stratification variable in the primary analysis of the combined studies. A fixed-effects approach was used to combine the estimates across the duration strata and statistical heterogeneity across all the trials was tested with the I² statistic.

Hazard ratios were also calculated for demographic subgroups, including sex, age (< 65 years vs ≥ 65 years) and body mass index (< 30 vs ≥ 30 [calculated as weight in kilograms divided by height in meters squared]) and the interaction between subgroup and treatment group was tested. Kaplan-Meier curves were generated to show the survival curves by treatment group for the composite of all-cause mortality, nonfatal myocardial infarction, or nonfatal stroke, and separately for serious heart failure. To assess the consistency of the effect, HRs were computed for subgroups defined by duration of study drug treatment or type of control therapy. SAS statistical software was used for all analyses (SAS version 8.2, Cary, North Carolina).

The 19 trials included in this meta-analysis are summarized in Table 1 and classified according to the planned duration of study drug treatment. Actual time taking prescribed study medication was less than 6 months, greater than 6 months to 12 months, greater than 12 months to 24 months, and greater than 24 months, in 26%, 18%, 16%, and 40% of the total 16 390 patients, respectively. Pioglitazone monotherapy was compared with placebo in 3 small trials (865 patients), sulfonylureas in 6 trials (5125 patients), metformin in 1 trial (1164 patients), and rosiglitazone in 1 trial (735 patients). Pioglitazone was tested in combination with a sulfonylurea, insulin, or metformin in 8 trials (8501 patients). The largest of the trials was PROactive, which enrolled 5238 patients (32% of the entire population in this meta-analysis and 55% of the patient-years) observed for a mean of 34.5 months.

Only limited information regarding demographic and baseline characteristics was collected in a uniform fashion across the trial databases (Table 1). Patients were older in the PROactive trial (mean age, 62 years) than in the other pioglitazone studies (mean age, 57 years), with less racial diversity (99% vs 78% white race, respectively).

Table 2 lists the clinical outcome measures in the pioglitazone and control groups of the 19 individual trials. Summary event rates and HRs are reported in Table 3. The primary composite end point of death, nonfatal myocardial infarction, or nonfatal stroke occurred in 375 of 8554 patients (4.4%) randomized to receive pioglitazone and 450 of 7836 patients (5.7%) treated with control therapy (HR, 0.82; 95% CI, 0.72-0.94; P = .005). Individual components of the primary end point were all reduced by a similar magnitude with pioglitazone treatment, with HRs ranging from 0.80 to 0.92 (Table 3). Progressive separation of time-to-event curves became apparent after approximately 1 year of therapy (Figure 2A).

Serious heart failure was reported in 200 (2.3%) of pioglitazone-treated patients and 139 (1.8%) of control patients (HR, 1.41; 95% CI, 1.14-1.76; P = .002; Table 3), with no further separation of time-to-event curves after approximately 1.5 years (Figure 2B). There was no evidence of heterogeneity across the trials for either end point (I ² = 0%, P = .87 for the composite end point and I ² = 0%, P = .97 for heart failure). Combining the HRs using a random-effects approach resulted in the identical result. Similarly, there was no heterogeneity in either end point for strata defined by sex, age, or body mass index (Table 4).

The composite of serious heart failure or death was not significantly increased among patients receiving pioglitazone (HR, 1.11; 95% CI, 0.96-1.29; P = .17). Multivariate adjustment for the limited patient baseline and demographic data available (age, sex, race, body mass index, and systolic blood pressure) produced only minor changes in estimates of the HRs for any of the end points (data not shown).

Outcomes for the primary ischemic composite and the secondary severe heart failure end points classified according to types of trials are illustrated in Figure 3A and Figure 3B, respectively. Patients enrolled in the PROactive trial had higher absolute event rates for both the primary and secondary end points compared with patients in all of the other 18 trials, reflecting the high-risk criteria for inclusion in PROactive (established coronary artery, cerebrovascular, or peripheral vascular disease). Nevertheless, HRs for the treatment effect of pioglitazone were similar in magnitude for PROactive as compared with the other trials. Concordance of the HRs was also observed for the trials classified according to the duration of study drug treatment or the type of control therapy.

Cardiovascular disease remains the most frequent cause of death and morbidity among patients with diabetes mellitus.^10- 11 Although agonists of nuclear PPAR-γ receptors are effective in achieving glycemic control in patients with diabetes mellitus, at least 2 agents with this mechanism of action have been associated with adverse cardiovascular outcomes.^5- 6

The current analysis therefore assessed the influence of pioglitazone, a thiazolidinedione PPAR-γ agonist widely used among patients with type 2 diabetes mellitus, on cardiovascular end points. Using patient-level time-to-event data derived from the complete database of randomized controlled double-blinded trials of pioglitazone, this meta-analysis demonstrated that therapy with pioglitazone is associated with a significantly lower risk of death, myocardial infarction, or stroke among a broad population of patients with diabetes (HR, 0.82). The magnitude and direction of this protective effect of pioglitazone was homogeneous across trials of different durations ranging from 4 months to 3.5 years, across studies using a variety of control or concomitant diabetic therapies, and among trials of patients with or without established vascular disease. Consistent with previously observed effects of thiazolidinediones on edema,^3- 4 the incidence of serious heart failure was increased by pioglitazone (HR, 1.41), although without an associated increase in mortality. These findings suggest that the net clinical cardiovascular benefit with pioglitazone therapy is favorable, with an important reduction in irreversible ischemic events that is not attenuated by the risk of more frequent heart failure complications.

Although strict control of hyperglycemia appears to diminish microvascular complications, little progress has been made in identifying therapies for type 2 diabetes that convincingly reduce the macrovascular complications leading to cardiac disease.¹² PPARs are ligand-inducible nuclear transcription factors that increase expression of genes involved in the regulation of lipoprotein metabolism, glucose homeostasis, and inflammation.^2,13 Thiazolidinedione agonists of PPAR-γ increase insulin sensitivity and glucose uptake in adipose and muscle tissue, suppress hepatic gluconeogenesis, and diminish fasting glucose, glycosylated hemoglobin, and plasma insulin levels. Moreover, some of these drugs may have salutary effects on lipid profiles, blood pressure, and inflammatory markers.^14- 18

Despite diverse sites of action that would be expected to favorably impact the pathogenesis of atherosclerosis among diabetic patients, development of this class of agents, as well as of the dual PPAR-α and PPAR-γ agonists, has been marked by a disappointing string of failures in preclinical and clinical testing. Several PPAR agonists produced multispecies, multiorgan malignancies or direct hepatic, renal, skeletal muscle, or cardiac toxicity in animal models. The first thiazolidinedione in clinical use, troglitazone, was removed from the market 3 years after its approval because of cases of drug-induced liver failure.¹⁹ A meta-analysis of phase 2 and 3 trials of muraglitazar, an investigational dual PPAR-α and PPAR-γ agonist, demonstrated an increased risk of cardiovascular complications (relative risk, 2.2-2.7) among diabetic patients receiving that agent.⁶ The development program for another dual agonist, tesaglitazar, was terminated due to renal toxicity.²⁰

Pioglitazone and rosiglitazone were approved by the US Food and Drug Administration on the basis of those agents' ability to reduce levels of blood glucose and glycosylated hemoglobin. Neither of the drugs was investigated in single trials of sufficient size to definitively evaluate their effects on cardiovascular mortality and morbidity. In the largest randomized study of rosiglitazone, the Diabetes Reduction Assessment with ramipril and rosiglitazone Medication (DREAM) trial, a nonsignificant trend toward a greater incidence of cardiovascular death, myocardial infarction, or stroke (1.2% vs 0.9%, P = .2) was observed among over 5000 patients treated with rosiglitazone.²¹ When the data from DREAM were combined in a meta-analysis of all available randomized trials of 24 weeks or more of rosiglitazone therapy, a significant excess risk of myocardial infarction (odds ratio 1.43, P = .03) and a strong trend toward increased cardiovascular death (odds ratio 1.64, P = .06) were confirmed.⁵ More recently, an unplanned interim analysis of 4458 patients enrolled in an ongoing trial of rosiglitazone was insufficient to determine if this agent was associated with an adverse effect on coronary ischemic events.²²

In light of the findings of the rosiglitazone meta-analysis, it was important to establish whether pioglitazone carries a liability with regard to cardiovascular complications. Results of the PROactive trial suggested that treatment with pioglitazone was beneficial from the cardiovascular standpoint, although significant differences were not observed in the prespecified primary end point (death, myocardial infarction, stroke, acute coronary syndrome, leg amputation, or coronary or leg revascularization).⁷ The current meta-analysis of data from the pioglitazone database presented here constitutes reasonably strong evidence that this agent does, in fact, reduce the risk of cardiovascular ischemic end points among patients with type 2 diabetes mellitus. These findings extend the observations of PROactive in a larger population and to lower-risk patients without established vascular disease. This analysis also provides reassuring information that although fluid retention and heart failure are more frequent with pioglitazone treatment, the offsetting risks do not appear to negate the beneficial effects of the drug on irreversible ischemic and fatal end points.

It is not clear why these 2 thiazolidinediones should have disparate effects on cardiovascular outcomes. Various PPAR agonists can yield markedly different patterns of gene modulation,²³ resulting in complex and largely unknown differences in effects on metabolic pathways. Although pioglitazone and rosiglitazone have similar effects on glycemic control, for example, pioglitazone produces greater reductions in serum triglycerides and increases in high-density lipoprotein cholesterol levels.^18,24 The 15% relative increase in high-density lipoprotein observed with pioglitazone is similar in magnitude to that which has been associated with coronary atheroma regression²⁵ or reduction in the incidence of coronary heart disease^26- 27 with other lipid-modifying agents. The findings of this study illustrate that drugs of the same “class” may in fact have quite different therapeutic profiles and highlight the potential hazards involved in using surrogate end points such as glycosylated hemoglobin rather than assessing safety and efficacy in relation to unequivocal clinical end points.

This study has important limitations. As the majority of trials pooled for this analysis were not originally intended to assess cardiovascular outcomes, end points were not uniformly adjudicated nor assessed by standard definitions. Furthermore, we excluded a group of trials for which the study databases and time-to-event data were unavailable. However, only a small number of events occurred in these unavailable trials precluding the possibility of a meaningful effect on the outcome of the meta-analysis. Important baseline or demographic variables predictive of outcome were not collected in a consistent manner across the trials, limiting the ability to adjust apparent treatment effects with multivariate models. Nevertheless, because all of the trials used for this analysis were double-blinded and randomized, potential biases introduced by these limitations should be minimized.

In conclusion, the findings of this meta-analysis provide evidence of a favorable effect of pioglitazone on ischemic vascular complications, which is distinct from the efficacy of thiazolidinediones in reducing blood glucose levels.