Targeted Learning for Optimal Patient Assignment to Psychotherapy.

Objective: Previous studies often fell short in identifying differences in treatment effects between psychotherapeutic frameworks. Instead of focusing on the overall treatment effects, we aimed to identify the effects of individually optimal treatment choice [cf. treatment personalization].

Method: We used a causal-inference machine learning (i.e., targeted learning) framework to estimate effects from observational data obtained from the Finnish Psychotherapy Quality Registry, which includes adult patients diagnosed with various mental disorders (n = 2255). Our objective was to estimate the difference in average treatment outcomes between the optimal individualized treatment and a randomly allocated treatment (i.e., the average of all treatment options). Outcomes were changes in self-assessed symptom scores and clinician-assessed functioning. In addition, we estimated counterfactual total-population outcomes for psychodynamic, solution-focused, cognitive-behavioral, and integrative or cognitive-analytic therapies.

Results: Compared to the average treatment effects, the counterfactual optimal treatment produced 0.28-0.29 standard deviations larger benefits for all the outcomes (confidence intervals between 0.20-0.39). Assuming all patients underwent psychotherapy within a single framework, treatment effects on symptom scores were similar across frameworks, but some differences emerged for change in therapist-assessed functioning.

Conclusion: Identifying optimal treatment rules for psychotherapy frameworks is feasible and may significantly improve outcomes.