This work may not be copied, distributed, displayed, published, reproduced, transmitted, modified, posted, sold, licensed, or used for commercial purposes. By downloading this file, you are agreeing to the publisher’s Terms & Conditions.

Original Research

Machine Learning Prediction of Quality of Life Improvement During Antidepressant Treatment of Patients With Major Depressive Disorder: A STAR*D and CAN-BIND-1 Report

Tejas Phaterpekar, MDS; John-Jose Nunez, MD, MSc; Emma Morton, PhD; Yang S. Liu, PhD; Bo Cao, PhD; Benicio N. Frey, MD, PhD; Roumen V. Milev, MD, PhD; Daniel J. Müller, MD, PhD; Susan Rotzinger, PhD; Claudio N. Soares, MD, PhD; Valerie H. Taylor, MD, PhD; Rudolf Uher, MD, PhD; Sidney H. Kennedy, MD; and Raymond W. Lam, MD

Published: November 15, 2023


Background: Quality of life (QoL) is an important patient-centric outcome to evaluate in treatment of major depressive disorder (MDD). This work sought to investigate the performance of several machine learning methods to predict a return to normative QoL in patients with MDD after antidepressant treatment.

Methods: Several binary classification algorithms were trained on data from the first 2 weeks of the Sequenced Treatment Alternatives to Relieve Depression (STAR*D) study (n = 651, conducted from 2001 to 2006) to predict week 9 normative QoL (score ≥ 67, based on a community normative sample, on the Quality of Life Enjoyment and Satisfaction Questionnaire–Short Form [Q-LES-Q-SF]) after treatment with citalopram. Internal validation was performed using a STAR*D holdout dataset, and external validation was performed using the Canadian Biomarker Integration Network in Depression-1 (CAN-BIND-1) dataset (n = 175, study conducted from 2012 to 2017) after treatment with escitalopram. Feature importance was calculated using SHapley Additive exPlanations (SHAP).

Results: Random Forest performed most consistently on internal and external validation, with balanced accuracy (area under the receiver operator curve) of 71% (0.81) on the STAR*D dataset and 69% (0.75) on the CAN-BIND-1 dataset. Random Forest Classifiers trained on Q-LES-Q-SF and Quick Inventory of Depressive Symptomatology–Self-Rated variables had similar performance on both internal and external validation. Important predictive variables came from psychological, physical, and socioeconomic domains.

Conclusions: Machine learning can predict normative QoL after antidepressant treatment with similar performance to that of prior work predicting depressive symptom response and remission. These results suggest that QoL outcomes in MDD patients can be predicted with simple patient-rated measures and provide a foundation to further improve performance and demonstrate clinical utility.

Trial Registration: identifiers NCT00021528 and NCT01655706

J Clin Psychiatry 2024;85(1):23m14864

Author affiliations are listed at the end of this article.

Volume: 85

Quick Links: Uncategorized

Continue Reading…

Subscribe to read the entire article


Buy this Article as a PDF