Phenotypic Profiles of Suicidal Ideation in Obsessive-Compulsive Disorder: An Interpretable Machine Learning Approach

Suicidal ideation in obsessive-compulsive disorder (OCD) is common and clinically significant, yet much of the existing literature conceptualizes suicide risk through the lens of comorbid depressive symptomatology. The present study examined whether other clinical features can identify clinically meaningful patterns associated with SI. Participants included 231 individuals with clinically significant OCD symptoms. SI was operationalized using Item 9 of the Beck Depression Inventory-II and binarized to reflect the presence or absence of suicidal thoughts. Depression severity scores were intentionally excluded from the predictive feature set, and three machine learning models (ElasticNet, Random Forest, and Explainable Boosting Machines) were evaluated using repeated nested cross-validation. All three algorithms showed comparable predictive performance. Given this overlap, the EBM was selected for interpretation due to its ability to model nonlinear relationships and interaction effects transparently. The model identified quality of life, obsessive-compulsive trait severity, somatic burden, and conscientiousness as prominent predictors of SI. Risk functions suggested nonlinear increases in estimated suicide risk at elevated levels of obsessive-compulsive traits and reduced quality of life. Additionally, interaction analyses indicated that severe obsessive-compulsive traits combined with elevated somatic burden were associated with higher estimated suicide risk than either factor alone. These findings suggest that interpretable machine learning can support clinically relevant phenotypic hypothesis generation. They also highlight somatic burden, functional impairment, obsessive-compulsive trait severity, and conscientiousness as potentially underappreciated targets for SI risk assessment in OCD, beyond the traditional focus on depressive comorbidity.