Twenty-nine RCTs (20,254 patients) were included in the review: 18 trials of rosiglitazone (11,756 patients) and 11 trials of pioglitazone (8,498 patients). The quality of the included trials was generally good; only one trial scored less than 3 on the Jadad scale. Trial sample size ranged from 105 to 5,238 patients. Length of follow-up ranged from three to 39 months.
Compared with control, with TZDs there was a statistically significantly greater risk of heart failure (OR 1.59, 95% CI 1.34 to 1.89, I2=0%; 10 trials), severe heart failure (OR 1.47, 95% CI 1.16 to 1.87, I2=0%; 10 RCTs) and peripheral oedema (OR 2.04, 95% CI 1.85 to 2.26, I2=48%; 27 RCTs). There was a higher risk of heart failure with rosiglitazone (OR 2.73, 95% CI 1.46 to 5.10, I2=0%; eight trials) than with pioglitazone (OR 1.51, 95% CI 1.26 to 1.81, I2=0%; two trials). Results using risk ratios and for NNH were presented in the review.
Subgroup analysis indicated no significant differences on the basis of patient type or length of follow-up. Sensitivity analysis with different meta-analysis models showed similar results to the main model. Multivariable meta-regression indicated that baseline risk of heart failure was significantly associated with risk of heart failure (p=0.01). There was evidence of publication bias with the heart failure outcome (Egger’s test p<0.0001) and oedema outcome (Egger’s test p<0.0001).