Predicting outcomes from cognitive behavioural therapy for social anxiety disorder: A Bayesian network analysis

The identification of predictors of treatment response in cognitive behavioural therapy (CBT) for social anxiety disorder (SAD) could help to improve treatment outcomes. This study sought to identify which specific social anxiety symptoms measured pre-treatment were associated with post-treatment outcomes. A pre-registered retrospective cohort study was conducted including a sample of N = 1315 patients treated with CBT for SAD in routine clinical practice. The sample was partitioned into training and test subsets (random 50:50 split) to evaluate the model's out-of-sample performance. The outcome was a reliable and clinically significant improvement (RCSI) in anxiety after treatment. A Bayesian network model based on symptoms measured by the Social Phobia Inventory (SPIN) was developed in the training sample (n = 658) using 10-fold cross-validation for variable selection. Predictive accuracy was evaluated in an external test sample (n = 657) using an area under the receiver operating characteristic curve (AUC). Predictive accuracy reduced when comparing performance in the training (AUC = 0.67) to test samples (AUC = 0.58) with moderate out-of-sample prediction shrinkage (6.76 %). The variable selection process yielded a network of four key interrelated SAD symptoms: avoiding public speaking, fear of embarrassment, avoiding criticism and fear of being observed. Identifying important SAD symptoms at assessment enables these to be targeted during CBT to potentially help maximise treatment efficiency and effectiveness.