Clinical Correlates of Suicidal Ideation in Adults With Obsessive-Compulsive Disorder: An Interpretable Machine Learning Study
- 1Yale University
Abstract
IntroductionDepression accounts for much of the association between obsessive-compulsive disorder and suicidal ideation in clinical samples, but whether other features carry independent information is unclear. We examined what non-depression features reveal about suicidal ideation.
MethodsAdults with obsessive-compulsive disorder (N = 231) completed measures of quality of life, anxiety, physical symptoms, obsessive-compulsive symptoms and traits, and personality. Elastic net, random forest, and explainable boosting machines identified suicidal ideation (Beck Depression Inventory-II item 9) without depression measures in nested cross-validation. Secondary analyses compared these features with depression severity.
ResultsSixty-six participants (28.6%) endorsed suicidal ideation; 61 of these (92.4%) reported thoughts without intent to act. All models showed similar discrimination (area under the curve = .77-.78), as did depression severity alone (.77). Adding the features to depression severity yielded a small estimated gain (difference = .02, 95% confidence interval: -.03 to .06), and flexible models offered no reliable advantage. Most features, including trait anxiety and obsessive-compulsive symptom severity, were associated with suicidal ideation on their own, but these associations attenuated under mutual adjustment. Lower quality of life remained the clearest correlate.
ConclusionNon-depression features identified suicidal ideation about as well as depression severity, consistent with substantial shared, distress-related information. Clinicians should ask about suicidal ideation directly, whatever a patients depressive symptoms, and consider overall distress in case formulation.
Related
- Reduced cortical thickness in obsessive-compulsive disorder and candidate molecular correlates
- Concentrated Exposure Therapy vs Internet-Based Cognitive Behavioral Therapy for Obsessive-Compulsive Disorder: A Randomized Clinical Trial.
- Intensive cognitive behavioral therapy for obsessive-compulsive disorder: a meta-analysis of randomized controlled trials
- Remission Trajectories During Low-Intensity Cognitive Behavioural Therapy for Obsessive–Compulsive Disorder: A 2-Year Naturalistic Follow-Up Study
- An Open-Label Multidose Psilocybin Intervention for Obsessive-Compulsive Disorder.
BibTeX
@misc{zaboski2026clinical,
title = {{Clinical Correlates of Suicidal Ideation in Adults With Obsessive-Compulsive Disorder: An Interpretable Machine Learning Study}},
author = {B. A. Zaboski and E. F. Mattera and C. A. Pittenger},
howpublished = {medRxiv},
year = {2026},
month = may,
doi = {10.64898/2026.05.31.26354549},
url = {https://www.medrxiv.org/content/10.64898/2026.05.31.26354549v2},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Clinical Correlates of Suicidal Ideation in Adults With Obsessive-Compulsive Disorder: An Interpretable Machine Learning Study
AU - B. A. Zaboski
AU - E. F. Mattera
AU - C. A. Pittenger
PY - 2026
DA - 2026/05/31
PB - medRxiv
DO - 10.64898/2026.05.31.26354549
UR - https://www.medrxiv.org/content/10.64898/2026.05.31.26354549v2
AB - IntroductionDepression accounts for much of the association between obsessive-compulsive disorder and suicidal ideation in clinical samples, but whether other features carry independent information is unclear. We examined what non-depression features reveal about suicidal ideation. MethodsAdults with obsessive-compulsive disorder (N = 231) completed measures of quality of life, anxiety, physical symptoms, obsessive-compulsive symptoms and traits, and personality. Elastic net, random forest, and explainable boosting machines identified suicidal ideation (Beck Depression Inventory-II item 9) without depression measures in nested cross-validation. Secondary analyses compared these features with depression severity. ResultsSixty-six participants (28.6%) endorsed suicidal ideation; 61 of these (92.4%) reported thoughts without intent to act. All models showed similar discrimination (area under the curve = .77-.78), as did depression severity alone (.77). Adding the features to depression severity yielded a small estimated gain (difference = .02, 95% confidence interval: -.03 to .06), and flexible models offered no reliable advantage. Most features, including trait anxiety and obsessive-compulsive symptom severity, were associated with suicidal ideation on their own, but these associations attenuated under mutual adjustment. Lower quality of life remained the clearest correlate. ConclusionNon-depression features identified suicidal ideation about as well as depression severity, consistent with substantial shared, distress-related information. Clinicians should ask about suicidal ideation directly, whatever a patients depressive symptoms, and consider overall distress in case formulation.
ER -