Predicting depression treatment outcomes for cognitive behavioural therapy using machine learning: A systematic...

Area of Science:

• Psychiatry
• Computational Psychiatry
• Clinical Psychology

Background:

• Cognitive behavioural therapy (CBT) is an effective depression treatment, but patient responses vary.
• Machine learning (ML) offers potential for predicting individual treatment outcomes.
• Identifying predictors of CBT response is crucial for personalized medicine.

Purpose of the Study:

• To systematically review and synthesize findings from studies using ML to predict depression treatment outcomes.
• To assess the predictive performance and generalizability of ML models in CBT for depression.

Main Methods:

• Systematic literature searches across three databases.
• Narrative synthesis of ML methods and predictor variables.
• Meta-analysis of predictive performance using a common effect size (r) for internal and external cross-validation.

Main Results:

• Twenty-four studies (11,733 participants) were included; only 16.7% had low risk of bias.
• Common predictors included depression severity, functional impairment, sociodemographics, comorbidity, and adversity.
• ML models demonstrated moderate predictive accuracy (r=0.45) with adequate generalizability to external samples (5/7 studies).

Conclusions:

• ML methods show replicated evidence of predicting depression treatment outcomes.
• The generalizability of ML predictions to new samples is adequate.
• Further research is needed to improve the reporting of ML model-training details.