Analyses and measures of treatment effect
Mean changes and SD were computed for continuous variables for every study to calculate effect size. As for studies with no reported SD of the mean difference, following formula were used: SD change = square root [(SD baseline 2 + SD final 2) - (2 × 0.8 × SD baseline × SD final)]32. Standardized mean changes were used for variables pooled on the different scales. Between-study heterogeneity was assessed using the chi-squared (χ2) test and quantified using the I2 statistic, which represents the percentage of the total variation across studies that is attributable to heterogeneity rather than to chance. Significant heterogeneity was defined with a P-value of <0.05.
The random effects model was applied to compute the weighted mean differences (WMDs) with 95% confidence intervals (CIs) for estimating the overall effect. To evaluate whether the results could have been affected distinctly by a single study, an influence or sensitivity analysis was carried out33. Subgroup analysis was also performed, based on the gender of participant (male, female or both), baseline BMI (over 30 kg/m2 and 30 kg/m2 or lower), dose of soy protein (over 20 g/day and 20 g/day or lower), duration of studies (over 12 weeks and 12 weeks or lower), supplementation type (soy-based meal replacement, soy powder or capsule supplement) and study design (randomized cross over or parallel). Publication bias was assessed by Begg’s rank correlation test and Egger’s regression asymmetry test. Funnel plots also depicted the effect sizes (differences in means) against their corresponding standard errors. Statistical analyses were performed using STATA 11.2 software (StataCorp, College Station, Texas, USA).