Predicting depression outcome in mental health treatment: a recursive partitioning analysis

Recursive partitioning was applied to a longitudinal dataset of outpatient mental health clinic patients to identify empirically factors and interactions among factors that best predicted clinical improvement and deterioration in symptoms of depression across treatment.

Sixty-two variables drawn from an initial patient survey and from chart review were included as covariates in the analysis, representing nearly all of the demographic, treatment, symptom, diagnostic, and social history information obtained from patients at their initial evaluations. Trees estimated the probability of participants' having depression at their last assessment, improving to a clinically significant degree during treatment, or developing a new onset of significant depressive symptoms during treatment.

Initial pain, the presence of anxiety, and a history of multiple types of abuse were risk factors for poorer outcome, even among patients who did not initially have significant depressive symptoms.

By examining multiple-related outcomes, we were able to create a series of overlapping models that revealed important predictors across trees. Limitations of the study included the lack of cross-validation of the trees and the exploratory nature of the analysis.