Background: Recent advances in Cognitive Bias Modification (CBM) for problematic drinking, including Alcohol Use Disorder, alongside methodological refinements, warrant an update of the Individual Participant Data (IPD) Bayesian meta-analysis. This study integrates new datasets, focuses on alcohol CBM, and applies a two-stage IPD framework to examine CBM's effects on cognitive bias, alcohol consumption, and relapse. Methods: A two-stage IPD Bayesian meta-analysis was performed, supplemented by frequentist sensitivity analyses. The first stage estimated study-specific CBM effects, the second stage pooled these to examine within-study moderators (adherence, addiction severity) and between-study moderators (CBM type, control condition, additional therapy, training congruency, and context). Results: 23 studies with 8297 participants were included. CBM showed a small unadjusted effect on bias (d = −0.18, 95% CrI −0.32, 0.00, BF₁₀ = 10.88) and relapse (log OR = −0.26, 95% CrI −0.38, −0.14, BF₁₀ = 155.07; number needed to treat = 18.7), but not on alcohol consumption (d = 0.003, 95% CrI 0.00, 0.06, BF₁₀ = 0.08). Effects were attenuated after adjusting for moderators. Moderator analyses revealed that face-to-face CBM context (unadjusted and within-study adjusted) and additional psychological therapy (within-study adjusted) were associated with greater bias reduction, and higher training intensity (unadjusted) was related to better relapse prevention. Frequentist sensitivity analyses largely supported these findings. Conclusions: CBM reduced alcohol-related biases and relapse risk. Although overall evidence was no longer supported after adjusting for moderators, moderator analyses suggest CBM can be effective under specific conditions (e.g., face-to-face delivery, alongside therapy, higher training intensity). These findings underscore the need for refined, context-sensitive CBM protocols in alcohol interventions.
Pan et al. (2026) studied this question.