Twenty-two RCTs (n=1,746 participants, range 23 to 297) were included. Thirteen studies reported adequate method of sequence generation or allocation concealment. All studies used intention-to-treat analysis. Assessment of blinding was considered unnecessary as all studies assessed self-report measures of the main outcome.
The overall mean effect size (Hedge's g) indicated superiority of computerised CBT over control across all four disorders (g=0.88, 95% CI 0.76 to 0.99, NNT 2.15; 22 RCTs). Similar results were found for major depression (g=0.78, 95% CI 0.59 to 0.96; six RCTs), social phobia (g=0.92, 95% CI 0.74 to 1.09; eight RCTs), panic disorder (g=0.83, 95% CI 0.45 to 1.21; six RCTs) and generalised anxiety disorder (g=1.11, 95% CI 0.76 to 1.47; two RCTs). With the exception of panic disorder (I2=49.77%), no evidence of heterogeneity was found. A small non-significant indication of publication bias was detected (g=0.80).
No significant differences were found in effect sizes between studies that used wait list control groups and treatment as usual and other control groups.
Overall adherence was reported as good: a median of 80% (range 48% to 100%, 22 RCTs) of participants who began the programs completed all lessons. A median of 86% (range 70% to 100%, 10 RCTs) of patients reported that they were satisfied or very satisfied.
Computerised CBT and traditional face-to-face CBT were found to be equally beneficial (five RCTs).