Natural classes of treatment response

Abstract

This study examines an approach to identifying patterns of treatment response over time. Treatment response profiles are identified by cluster analyzing a repeated measure of patient performance collected at intervals during treatment. The procedure is demonstrated in Study 1 using monthly urinalysis results of 103 patients entering methadone maintenance treatment. The internal, external, and face validities of derived treatment response profiles are evaluated. A logistic regression model predicting treatment response is then constructed from intake variables found to correspond with the treatment response profiles. Study 2 replicates the procedures on an independent sample. Treatment response profiles facilitate the analysis of treatment response offering advantages over common measures of treatment outcome, such as performance at follow-up, change in performance from treatment entry to follow-up, or performance summed across treatment.