Comorbid anxiety predicts lower odds of MDD improvement in a trial of smartphone-delivered interventions

Interpretable machine learning shows that higher baseline anxiety predicts lower odds of improvement in major depressive disorder. The trial compared three smartphone-delivered interventions.

Published in the Journal of Affective Disorders with a February 2026 issue date, the study by Morgan B. Talbot, Jessica M. Lipschitz, and Omar Costilla-Reyes is a secondary analysis of a randomized trial that included several hundred participants with major depressive disorder (MDD). The trial compared three smartphone-delivered interventions: Project EVO, a cognitive-training app; iPST, a problem-solving therapy app; and Health Tips, an active control. The arXiv version lists Lipschitz and Costilla-Reyes as co-senior authors.

Comorbid anxiety, anxiety disorders occurring alongside depression, is common in MDD, which makes its effect on the outcomes of digital interventions a practical question. The authors asked whether baseline anxiety, measured with the GAD-7 (Generalized Anxiety Disorder 7-item) questionnaire, predicts improvement in depression. They applied classical machine-learning models, including decision trees, to identify baseline predictors of improvement in the weeks after enrollment. Baseline GAD-7 emerged as the strongest predictor of treatment outcomes: a decision tree indicated that a score of 11 or higher, consistent with at least moderate anxiety, was associated with lower odds of MDD improvement in all three intervention groups.

Secondary analysis of trial data lets researchers ask new questions of carefully collected evidence, and interpretable models keep the answers readable: a decision tree states its rule in plain terms, here a baseline anxiety score that separates participants with higher and lower odds of improvement. Identifying who is more likely to improve with a given smartphone intervention is a step toward stratified care, in which treatment is matched to the person from the start.

Dr. Costilla Reyes’s affiliation for this paper is Equ Healthcare; the work was completed while he was at MIT and continues the digital mental health outcome modeling he began there in 2020. An earlier version was posted to arXiv in September 2024 and is linked alongside the journal record. The study follows the 2024 smartphone CBT study and the 2025 preprint on body dysmorphic disorder.