Data synthesis and statistical analysis:
A meta-analysis was performed using mean differences (MDs) and their standard deviations (SDs) for CRP (mg/l), or these values were computed using the data were extracted from included articles. Based on Cochrane Handbook (40) the mean change from baseline in the CRP concentrations and its SD for both intervention and comparison groups were used to compute the effect size. Hozo method (41) was used to estimate mean of CRP concentrations if median or range was reported. Moreover, SDs were computed by multiplying standard errors (SEs) by square root of the sample size in studies which reported SEs. The summary of the overall effects and heterogeneity was estimated by using the DerSimonian and Laird random effects model (42), if a heterogeneity test was statistically significant. Statistical heterogeneity of intervention effects was measured using Cochran’s Q test and I-squared statistic. Heterogeneity was considered substantial if p-value for the Cochran’s Q test was ≤0.10 or value of the I-squared statistic was ≥50% (43).
Subgroup analyses was performed to explore sources of heterogeneity which included soy isoflavones dosage or soy isoflavones plus soy protein dosage, study design, intervention duration, mean of baseline CRP concentration, health status of participants, sample size, geographical region, age of participants, body mass index (BMI), quality assessment of articles and year of study publication. Meta-regression was used to examine the impact of moderator variables on MDs and compute adjusted effect of soy isoflavones dosage and soy isoflavones plus soy protein dosage on MDs after controlling for other variables. Visual inspection of Begg’s funnel plot, Begg’s rank correlation, and Eggar’s weighted regression tests were employed to evaluate the presence of publication bias in the meta-analysis (44, 45). Sensitivity analysis was conducted using the leave-one out method to specify the effect of each study on the overall effect size. Ninety-five percent confidence intervals were provided for all calculated effect sizes. All analyses were conducted using STATA 15 software (Stata Corp, Collage Station, TX).