Methods of synthesis
The data were pooled in inverse-variance meta-regression analysis, using hierarchical linear mixed models. The percentage reduction in the outcome was used as the dependent variable. Regression models controlled for correlation between observations within the same study and for differences between studies in the duration of the intervention. The restricted maximum likelihood method was used to estimate model parameters, and denominator degrees of freedom were calculated using the method of Berkey et al. (see Other Publications of Related Interest). Some clinical differences between the studies were investigated in single-covariate regression analyses. Further analyses were conducted to investigate the effect of weight change on glycaemia and lipid outcomes.