Predicting Dropout in Psychotherapy for Major Depressive Disorder: A Machine Learning Approach to Identifying At-Risk...

Area of Science:

• Psychiatry and Mental Health
• Machine Learning in Healthcare
• Clinical Psychology

Background:

• Psychotherapy dropout for major depressive disorder (MDD) hinders treatment success.
• Early identification of at-risk patients is essential for effective intervention.
• Machine learning, especially with resampling, may improve prediction of patient dropout.

Purpose of the Study:

• To evaluate machine learning models for predicting psychotherapy dropout in MDD patients.
• To assess the impact of resampling techniques on predictive performance.
• To identify key predictors of dropout in a clinical population.

Main Methods:

• Utilized data from a randomized controlled trial of MDD outpatients (n=290).
• Applied multiple machine learning models and resampling strategies (e.g., SMOTE) to predict dropout.
• Externally validated the best models on an independent dataset (n=96).

Main Results:

• Models exhibited low overall predictive performance, with limited improvement during external validation.
• SMOTE resampling consistently enhanced F1-scores compared to no resampling.
• Random forest with SMOTE achieved the highest F1-score (0.303); key predictors included younger age, unemployment, comorbid disorders, lack of a partner, and moderate depression severity.

Conclusions:

• Current models show limited predictive utility, potentially due to restricted predictor sets.
• Findings offer a proof of concept for machine learning in dropout prediction, highlighting methodological challenges.
• Identified predictors are clinically accessible, suggesting future models with richer data could aid therapists in proactive patient engagement.