Toward precision psychological rehabilitation: predicting CBT efficacy in post-stroke depression using machine...

Area of Science:

• Neuroscience
• Psychiatry
• Medical Informatics

Background:

• Post-stroke depression (PSD) is a common complication impacting recovery.
• Cognitive Behavioral Therapy (CBT) is a potential treatment, but predicting response is challenging.

Purpose of the Study:

• To evaluate CBT's effectiveness for PSD.
• To develop and validate an interpretable machine learning model for predicting individual CBT response in PSD patients.

Main Methods:

• Retrospective analysis of clinical and psychological data from 120 PSD patients receiving CBT and 123 controls.
• Comparison of symptom changes (PHQ-9, GAD-7, GSE) between groups.
• Development of a random forest classifier to predict CBT response, validated using SHAP values and ablation analyses.

Main Results:

• The CBT group showed significantly greater improvement in depressive symptoms compared to the control group.
• The random forest model achieved high predictive performance (AUC = 0.897, accuracy = 0.861).
• Key predictors for CBT response included baseline depressive severity (PHQ-9), anxiety levels (GAD-7), self-efficacy (GSE), and social support (SSRS).

Conclusions:

• CBT is associated with improved depressive symptoms in PSD patients.
• The developed machine learning model shows promise for predicting treatment response in PSD.
• Further prospective, multi-center validation is required before clinical application of the predictive model.