Prescriptive Predictors of Mindfulness Ecological Momentary Intervention for Social Anxiety Disorder: Machine...

Area of Science:

• Psychiatry and Mental Health
• Digital Health
• Machine Learning in Healthcare

Background:

• Social anxiety disorder (SAD) treatment is hindered by stigma and cost, necessitating scalable, brief interventions.
• Predictive models are crucial for optimizing treatment allocation for SAD.
• Classical regression may not fully capture complex patient-treatment interactions.

Purpose of the Study:

• To apply precision medicine and machine learning (ML) to identify predictors of successful optimization to a self-guided mindfulness intervention (MEMI) versus self-monitoring (SM).

Main Methods:

• 191 participants with probable SAD were randomized to MEMI or SM.
• Machine learning models (random forest, support vector machines) were trained using 17 baseline predictors.
• Optimization was defined as higher probability of SAD remission at posttreatment and 1-month follow-up.

Main Results:

• ML models significantly outperformed logistic regression, achieving AU-ROC values of .71–.72.
• Key predictors of MEMI optimization included higher trait mindfulness, lower SAD severity, university education, no psychotropic medication, higher generalized anxiety, diagnosed depression/anxiety, and Chinese ethnicity.
• These predictors consistently identified individuals with higher remission probability.

Conclusions:

• ML models demonstrate moderate effectiveness in identifying prescriptive predictors for SAD treatment optimization.
• Identifying client strengths, weaknesses, and ethnicity can improve prediction of response to scalable treatments.
• A "prescriptive predictor calculator" could guide resource allocation and stratified care for SAD, potentially using MEMI as a pre-intensive therapy option.