Data Analysis

Data were pooled using the STATA version16.1 I/C software package (StataCorp LLC, College Station, TX, USA), using the restricted maximum likelihood model (REML) to estimate 95% confidence intervals (95% CI). Differences between groups for a particular PROM were pooled after computing the standardized mean difference (SMD) using Hedges’ g. Effect sizes were rated as small, medium or large based on SMD cut-offs of 0.2, 0.5 and 0.8, respectively [32]. The random effects model was used a priori due to the expected heterogeneity owing to the small sample sizes [33] of the TAK studies and diverse PROMs assessed. Statistical heterogeneity was assessed using I² statistic; values > 50% denoted significant heterogeneity [14]. In such cases, individual studies were excluded to assess whether this exclusion ameliorated the observed heterogeneity. Results were summarized descriptively wherever appropriate. Subgroup analyses were pre-planned for adult and childhood-onset TAK [34, 35]. Studies using version 2 of SF-36 were excluded to evaluate the impact of the version of SF-36 in a secondary analysis. Sensitivity analyses were based on study design (cross-sectional or cohort).