Statistical analysis

Binary outcomes were summarized as risk ratios (RRs) with 95% confidence intervals (CIs), whereas continuous outcomes were summarized as mean differences (MDs) with 95% CIs. For continuous outcomes, the mean, standard deviation (SD), and number of participants analyzed per group were extracted. When medians with ranges or interquartile ranges were reported, these summary statistics were converted to means and SDs using validated methods to enable inclusion in the meta-analyses [17,18].

A frequentist network meta-analysis was conducted using a random-effects model as the primary analytical approach. Placebo and usual care were combined into a single control node. Global inconsistency was evaluated using the design-by-treatment interaction model, whereas local inconsistency was assessed using the node-splitting approach. The probability of each intervention ranking as the most effective for each outcome was estimated, and treatment rankings were summarized using the surface under the cumulative ranking curve (SUCRA). Conventional pairwise meta-analyses were also performed for all available direct comparisons to summarize direct evidence and support the assessment of consistency between direct and indirect estimates. For outcomes with a sufficient number of studies, comparison-adjusted funnel plots were constructed to explore potential small-study effects. Funnel plot asymmetry was additionally evaluated using Egger’s regression test applied to the comparison-adjusted effect sizes. All analyses were performed using Stata/SE (version 17.0; StataCorp, College Station, TX, USA) with the network meta-analysis command suite [19]. Statistical significance was set at p < 0.05.