MATCH: client-therapist matching with machine learning

Area of Science:

• Clinical Psychology
• Health Informatics
• Computer Science

Background:

• Therapist assignment significantly impacts counseling success.
• Current methods often rely on simple availability, potentially missing optimal matches.

Purpose of the Study:

• To develop and evaluate a framework for matching clients to therapists using archival data and algorithms.
• To improve client outcomes compared to assignment to the next available therapist.
• To incorporate therapist availability constraints into the matching process.

Main Methods:

• Utilized high-dimensional survey data (CCAPS-62, OQ-45) from 10,100 therapy courses (2014-2019).
• Estimated therapist-client outcome associations using predictive modeling.
• Implemented constrained matching algorithms respecting therapist availability.
• Evaluated performance via simulations and statistical tests.

Main Results:

• The model predicted therapist-client pairs associated with an average improvement of 0.145-3.65 OQ-45 points (d=0.11).
• This improvement was statistically significant (95% confidence interval).
• The method demonstrated potential for improving outcomes with minimal impact on wait times and administrative changes.

Conclusions:

• Data-driven client-therapist matching can significantly improve counseling outcomes.
• Accounting for therapist availability constraints is crucial for practical implementation.
• This approach offers a scalable solution for enhancing mental health service effectiveness.