As the studies used a variety of continuous data scales to evaluate clinical outcomes, a unit-less measure of treatment effect size (ES) was applied to pool the results across themultiple controlled trials. As recommended by the Cochrane Collaboration,16 we used the standardised mean difference (SMD) as summary measure, which is applicable for inter- pretation as the ES originally proposed by Cohen.17 Clinically, an ES of 0.2 is considered small, 0.5 as moderate (and would be recognised clinically) and .0.8 as large.8 Accordingly, the ES (SMD) that we used was Hedges’ adjusted g value, which is very similar to Cohen’s d value, although with an adjustment for small sample bias.18 To pool the mean different weight reduction of individual study group, the weighted mean difference was applied.19 Random effects meta-analysis20 was used if the studies were heterogeneous, for which the Cochran Q test was used to assess the degree of heterogeneity21; the a risk for this analysis is set to 0.1 (p,10%).16 22 Quantification of the effect of heterogeneity was assessed by means of I2, which ranges from 0% to 100%; I2 shows the percentage of total variation across studies due to heterogeneity, and may be used to assess the consistency of evidence.23 These analyses were carried out using the software provided by the Cochrane Collaboration, Review Manager (RevMan V.4.2).18 For further illumination of the quality (ie, magnitude and intensity) of the intervention,12 we applied dose–response efficacy estimates following meta-regression analyses, with two subsequent a priori defined weight change differences (% point weight change as magnitude and % point weight change per week as intensity, respectively) as independent variables. Meta- regression analyses should be weighted to take account of both within-trial variances of treatment effects and the residual between-trial heterogeneity (ie, heterogeneity not explained by the covariates in the regression). We therefore applied the random effects meta-regression,24 using the individual study weight (based on the inverse variance) following the random effects model presented by DerSimonian and Laird.20 All meta- regression analyses were calculated using SAS statistical package V.8.
RESULTS