Figure 3.

Pooled effect estimates are shown as diamonds, one each for trials conducted in otherwise healthy, dyslipidaemia, metabolic syndrome criteria, type 2 diabetes mellitus and their combination (total). Paired analyses were applied to all cross-over trials (10) and one substudy. Data are expressed as mean differences with 95% CI, using generic inverse-variance random effects models. Interstudy heterogeneity was tested by using the Cochran Q statistic (Chi²) at a significance level of p<0.10 and quantified by the I² statistic. FBG, fasting blood glucose; RCT, randomised controlled trial.