Background Cognitive Bias Modification (CBM) is a potentially effective intervention for social anxiety disorder (SAD). However, the relative efficacy of CBM subtypes remains uncertain. Methods We co-produced a living systematic review of controlled studies evaluating stand alone CBM in participants with SAD and in non-clinical participants undergoing a social stressor, seeking to triangulate these. Databases were searched until October 2025 and two independent reviewers screened, extracted and critically appraised studies. The primary outcome was reduction in SAD symptoms, but we collected data on other mental health outcomes. Random-effect network meta-analysis was conducted. Credibility was assessed using the Confidence In Network Meta-Analysis framework. Results CBM for attention (CBM-A) demonstrated the largest reduction in SAD symptoms relative to control CBM-A although this estimate was imprecise (Standardised Mean Difference (SMD) = −0.18, 95% CI: −0.41 to 0.05, 95% PI: −1.02 to 1.04). Opposite CBM-A, training attention towards negative stimuli, performed similarly (SMD -0.17, 95% CI: −0.65 to 0.31, 95% PI: −0.97 to 1.21). CBM for interpretation (CBM-I) reduced SAD symptoms (SMD = −0.47, 95% CI: −0.84 to −0.09, 95% PI: −0.53 to 1.46) and cognitive bias versus its task-matched control condition without active bias training. The SAD network showed high inconsistency and 95% of studies were at moderate or high risk of bias. Three comparisons involving CBM-A were rated as moderate confidence, all others were low or very low. Sparse data precluded NMA of outcomes in non-SAD participants. Conclusion Inconsistency and imprecision in the network preclude firm conclusions regarding CBM efficacy for SAD. CBM-A showed a small effect but its similarity to opposite CBM-A suggests benefits may not depend on training direction. Future studies should explore the mechanism of CBM and reporting on potential side-effects. We will update this living systematic review as part of the GALENOS project. Registration: PROSPERO-ID:CRD42024601380.
Kennett et al. (Mon,) studied this question.