Thirteen trials (n=17,627 patients) were included in the review. Sample sizes ranged from 57 to 5,238 patients. Eight trials reported adequate sequence generation, allocation concealment, double-blinding and losses to follow-up and withdrawals. One trial was open-label; the rest used double-blinding.
There was a statistically significant higher risk of pneumonia or lower respiratory tract infection observed with thiazolidinedione treatment compared with placebo or other active treatment (RR1.40, 95% CI 1.08 to 1.82; I2=0%; 10 RCTs, n=17,627). There was also a statistically significant higher risk of pneumonia or lower respiratory tract infection as serious adverse events (RR 1.39, 95% CI 1.05 to 1.83; I2 =0%; seven trials, n=16,083). Sensitivity analyses with the results combined using the Peto odds ratio and random-effects models yielded estimates similar to those found using the fixed-effect model.
Higher risks of pneumonia and lower respiratory tract infections were observed for pioglitazones compared with other active treatment or placebo (RR 1.63, 95% CI 1.09 to 2.46; five trials, n=6,129). There were no significant differences observed for rosiglitazone and active treatments or placebo (RR 1.26, 95% CI 0.90 to 1.76; eight trials, n= 11,498). No statistically significant heterogeneity was observed for these analyses.
No evidence of publication bias was observed with visual appraisal of the funnel plots. To reverse the higher risks of pneumonia or lower respiratory tract infection, 11 negative studies with an average sample size of 1,400 participants (fail-safe n) would have to be included.
Based on the average control rate in the trials, the number needed to harm for any pneumonia or lower respiratory rate infection associated with thiazolidinedione use was 239 (95% CI 117 to 1,191).