Ten RCTs (n=13,715) and two observational studies (n=31,679) were included in the review. All RCTs were double-blinded and six had adequate allocation concealment, the remaining four were unclear.
The incidence of fractures was significantly higher in the thiazolidinedione groups compared to control groups (OR 1.45, 95% CI: 1.18, 1.79, p<0.001, 10 RCTs). Statistical heterogeneity was moderate (I2 = 27%). The analysis was dominated by two large reasonable quality RCTs. Subgroup analysis of the RCTs which reported data separately for men and women, showed that women in thiazolidinedione groups had a significantly higher risk of fracture compared to controls (OR 2.23, 95% CI: 1.65, 3.01, p<0.001, n=4,400) but that there was no significant difference between the groups for men (OR 1.00, 95% CI: 0.73, 1.39, p=0.98, n=7,001). In neither case was there any statistically significant heterogeneity (I2 = 0%). NNH varied from 55 (95% CI: 34, 103) in young newly diagnosed women to 21 (95% CI: 14, 39) in postmenopausal women. Results from the observational studies supported those of the RCTs.