Predicting treatment response to cognitive behavior therapy in social anxiety disorder on the basis of demographics, psychiatric history, and scales: A machine learning approach

Abstract

Only about half of patients with social anxiety disorder (SAD) respond substantially to cognitive behavioral therapy (CBT). However, there has been little evidence available to clinicians or patients about whether any individual patient is more or less likely to have a positive response to CBT. Here, we used machine learning on data from 157 patients to examine whether individual patient responses to CBT can be predicted based on demographic information, psychiatric history, and self-reported or clinician-reported scales, subscales and questionnaires acquired prior to treatment. Machine learning models were able to explain about 26% of the variance in final treatment improvements. To assess generalizability, we evaluated multiple machine learning models using cross-validation and determined which input features were essential for prediction. While prediction accuracy was similar across models, the importance of specific features varied across models. In general, the combination of total scale score, subscale scores and responses to individual questions on a severity measure, the Liebowitz Social Anxiety Scale (LSAS), was the most informative in achieving the highest predictions that alone accounted for about 26% of the variance in treatment outcome. Demographic information, psychiatric history, personality measures, other self-reported or clinician-reported questionnaires, and clinical scales related to anxiety, depression, and quality of life provided no additional predictive power. These findings indicate that combining scaled and individual responses to LSAS questions are informative for predicting individual response to CBT in patients with SAD.