Citation

Jain FA, Hunter AM, Brooks JO, Leuchter AF. Depress. Anxiety 2013; 30(7): 624-630.

Affiliation

Department of Psychiatry and Biobehavioral Sciences, University of California, Los Angeles, California.

Copyright

(Copyright © 2013, John Wiley and Sons)

DOI

10.1002/da.22045

PMID

23288666

Abstract

BACKGROUND: There are many prognostic factors for treatment outcome in major depressive disorder (MDD). The predictive power of any single factor, however, is limited. We aimed to develop profiles of antidepressant response and remission based upon hierarchical combinations of baseline clinical and demographic factors. METHODS: Using data from Level 1 of the Sequenced Treatment Alternatives to Relieve Depression trial (STAR*D), in which 2,876 participants with MDD were treated with citalopram, a signal-detection analysis was performed to identify hierarchical predictive profiles for patients with different treatment outcome. An automated algorithm was used to determine the optimal predictive variables by evaluating sensitivity, specificity, positive and negative predictive value, and test efficiency. RESULTS: Hierarchical combinations of baseline clinical and demographic factors yielded profiles that significantly predicted treatment outcome. In contrast to an overall 47% response rate in STAR*D Level 1, response rates of profiled patient subgroups ranged from 31 to 63%. In contrast to an overall remission rate of 28%, identified subsets of patients had a 12 to 55% probability of remission. The predictors of antidepressant treatment outcome most commonly incorporated into profiles were related to socioeconomic status (e.g., income, education), whereas indicators of depressive symptom type and severity, as well as comorbid clinical conditions, were useful but less powerful predictors. CONCLUSIONS: Hierarchical profiles of demographic and clinical baseline variables categorized patients according to the likelihood they would benefit from a single antidepressant trial. Socioeconomic factors had greater predictive power than symptoms or other clinical factors, and profiles combining multiple factors were stronger predictors than individual factors alone.

Language: en