Early identification of psychotherapeutic change: AI-derived predictors from routine data

Measurement-based care has been shown to be an efficacious way to improve therapy outcomes. The use of advanced statistical approaches may assist in optimizing the feedback therapists receive. That is, algorithms developed within measurement-based care programs enhance their performance in predicting therapy outcomes. In this proof-of-concept article, we analyzed data from 9,591 clients who were treated at university counseling centers and completed the Behavioral Health Measure on a reoccurring basis. The sample included male and female clients of diverse ethnicities. We utilized two AI-derived approaches: ant colony optimization algorithm to derive a content-valid item subset that maximized the prediction of clinically significant change status, followed by logistic Lasso regression to identify the most influential item-level predictors. Early in treatment (i.e., the first three to five sessions), Behavioral Health Measure items were tested to determine which were the best predictors of therapy outcomes. The results demonstrated that five items, including self-esteem, general anxiety, substance use, cognitive attention, and overall work/school life purpose, were strong predictors of outcomes over the course of early treatment. These findings suggest that measurement-based care platforms or programs might benefit from more detailed attention to specific signals or patterns in the data that assist therapists and clients monitoring treatment progress. (PsycInfo Database Record (c) 2026 APA, all rights reserved).