Targeted Learning for Optimal Patient Assignment to Psychotherapy.

Area of Science:

• Psychiatry and Mental Health
• Machine Learning in Healthcare
• Causal Inference

Background:

• Previous research often failed to differentiate treatment effects across psychotherapy frameworks.
• Focusing on average treatment effects overlooks the potential for personalized treatment selection.

Purpose of the Study:

• To identify the effects of individually optimal treatment choice in psychotherapy.
• To estimate the difference in average treatment outcomes between optimal individualized treatment and randomly allocated treatment.

Main Methods:

• Utilized a causal-inference machine learning framework (targeted learning).
• Analyzed observational data from the Finnish Psychotherapy Quality Registry (n=2255 adult patients).
• Estimated counterfactual outcomes for psychodynamic, solution-focused, cognitive-behavioral, and integrative/cognitive-analytic therapies.

Main Results:

• The counterfactual optimal treatment yielded 0.28–0.29 standard deviations larger benefits across all outcomes compared to average treatment effects.
• Treatment effects on symptom scores were similar across frameworks when assuming a single framework for all patients.
• Some differences in treatment effects emerged for therapist-assessed functioning across frameworks.

Conclusions:

• Identifying optimal treatment rules for psychotherapy frameworks is feasible.
• Personalized treatment selection holds significant potential for improving patient outcomes in mental health.