An Interpretable Model With Probabilistic Integrated Scoring for Mental Health Treatment Prediction: Design Study.

Area of Science:

• Psychiatry and Mental Health
• Machine Learning in Healthcare
• Clinical Decision Support Systems

Background:

• Machine learning (ML) in healthcare offers decision-making potential but struggles with explainability, confidence, and robustness.
• Existing ML models often lack context-based interpretability for clinical applications.

Purpose of the Study:

• To design and evaluate a novel, inherently interpretable ML model for clinical psychopathological treatment assessments.
• To enhance clinical explainability and trust through a transparent, hierarchical model structure.
• To address model confidence and robustness using probabilistic methods like Monte Carlo dropout.

Main Methods:

• Developed a novel ML model structure for psychopathological treatment assessments.
• Incorporated graphical interpretation of model outputs to improve explainability.
• Trained and validated the ML model using patient questionnaire data and demographics (N=1088) from a Danish web-based service.

Main Results:

• Achieved a balanced accuracy of 0.79 on the test set.
• Demonstrated precision ≥0.71 across all four prediction classes (depression, panic, social phobia, specific phobia).
• Obtained Area Under the Curve (AUC) scores of 0.93, 0.92, 0.91, and 0.98 for the respective classes.

Conclusions:

• Successfully demonstrated an ML model for mental health treatment with graphical interpretation of prediction probabilities.
• The model's output aids clinicians in understanding competing treatment options and prediction uncertainty.
• The ML model, with 79% balanced accuracy, is expected to be clinically valuable for patient screening and informing clinical interviews.