Ten RCTs (23,152 participants) were included in the review; trial sample size ranged from 207 to 9,306 participants. Intervention follow-up ranged from 2.8 to six years.
Risk of diabetes was statistically significantly lower with intervention groups versus control groups (RR 0.66 95% CI 0.55 to 0.80, ten trials), significant statistical heterogeneity was found (p<0.001). The effect was stronger with non-pharmacological interventions than with pharmacological ones (p<0.05).
A lower risk of fatal and non-fatal stroke was found for intervention groups versus control groups, but this was of borderline statistical significance (RR 0.76, 95% CI 0.58 to 0.99, three trials). No significant statistical heterogeneity was found (p<0.872). A similar result was obtained when rosiglitazone was removed; statistical significance was lost when only trials demonstrating diabetes prevention were examined (RR 0.76, 95% CI 0.57 to 1.01).
There were no statistically significant differences between interventions in all-cause mortality; sensitivity and subgroup analyses did not significantly alter the findings.
No evidence for publication bias was found.