Predicting treatment response in body dysmorphic disorder with interpretable machine learning

A 2025 preprint applies interpretable machine learning to predict treatment response and remission in body dysmorphic disorder. It examines which baseline factors drive those predictions.

Posted to arXiv in March 2025 by Omar Costilla-Reyes and Morgan Talbot, the preprint extends the analysis from the 2024 Journal of Affective Disorders study of smartphone-delivered cognitive behavioral therapy for body dysmorphic disorder. Body dysmorphic disorder involves preoccupation with perceived flaws in appearance, and predicting who will respond to treatment is a long-standing clinical question. The authors evaluate several predictive models for treatment response and remission, with an emphasis on methods whose reasoning can be inspected. Interpretable models matter here because a clinician can examine and question a prediction that rests on a few readable rules or weights, which is harder with a black-box model.

Treatment credibility, the degree to which a participant finds the treatment believable at the outset, emerged as the strongest predictor in the evaluated models, ahead of baseline symptom severity and comorbid conditions. Decision trees exposed credibility thresholds that could be inspected, while logistic regression and support-vector machines remained competitive with more complex approaches. The authors argue that transparent models can help target modifiable factors such as credibility. Because credibility can be addressed at the start of care, for example through how a treatment is introduced and explained, the finding gives clinicians and researchers a concrete, readable lead for improving engagement, and it shows how interpretable methods can turn predictive modeling into practical insight for mental-health research.

The preprint belongs to Dr. Costilla Reyes’s digital mental health outcome modeling work, which began during his 2020–2021 postdoctoral fellowship at MIT’s Institute for Medical Engineering and Science. It follows the 2024 smartphone CBT study and relates to the 2026 study of comorbid anxiety and depression improvement in smartphone-delivered interventions, also with Talbot.