Fifteen RCTs (1,091 participants, range 23 to 130) were included in the meta-analysis. Study quality scores ranged from 18 to 62 (mean 38.4). No studies were judged to be of high quality (>65 points). No studies reported using independent blind assessors to evaluate outcomes. Drop-outs ranged from zero to 49%.
CBT compared with passive control groups: There was a statistically significant moderate to large improvement for CBT compared with a waiting-list control for primary tinnitus distress measures (Hedges’ g 0.70, 95% CI 0.56 to 0.84, fixed-effect model) and for mood measures (Hedges’ g 0.35, 95% CI 0.21 to 0.50, fixed-effect model). Results using a random-effects model were similar. There was evidence of statistical heterogeneity for both analyses (Ι²=26% and Ι²=40%).
CBT compared with active control groups: There was a statistically significant small to moderate improvement for CBT for tinnitus distress measures (Hedges’ g 0.44, 95% CI 0.16 to 0.72, fixed-effect model) and for mood measures (Hedges’ g 0.42, 95% CI 0.09 to 0.74, fixed-effect model). Results were similar using a random-effects model for tinnitus measures but became non-significant for mood measures. There was no evidence of statistical heterogeneity for tinnitus distress (Ι²=0), but heterogeneity was moderate for mood (Ι²=39%).
Overall there was a significant long-term effect of CBT compared to passive and active controls (Hedges' g 0.60, 95% CI 0.39 to 0.80, random-effects model) but heterogeneity was high (Ι²=91%). Meta-regression analysis found a statistically significant negative association between effect sizes and time, which indicated that effect sizes decreased slightly over time. Subgroup analyses of the effects of methodological quality and publication bias did not significantly alter the main findings.