Identifying past-year self-reported suicidality in outpatients with somatic symptom disorder using an interpretable machine-learning model: a multicenter study with an online calculator

Comprehensive Summary

Wang et al. studied whether routinely collected outpatient data could be used to identify past-year self-reported suicidality in patients with somatic symptom disorder, a group in which suicide risk can be easily missed because care often stays focused on physical complaints. The researchers analyzed a multicenter cross-sectional registry from three hospitals in Ganzhou that included 899 adults with DSM-5 somatic symptom disorder. The data were split into training and test sets, 11 final predictors were selected from psychometric, laboratory, and vital sign measures. Eight machine-learning models were compared before the top model was further explained with SHAP and turned into a web calculator. About 19.9% of participants reported past-year suicidality, and all models performed well in the test set. The RANGER random forest model performed best, with an AUC of 0.978, accuracy of 0.967, sensitivity of 0.927, and specificity of 0.977. Insomnia severity was the strongest contributor to predicted risk, followed by mindfulness questionnaire scores and neurocognitive status, while questionnaire-only models still performed strongly and lab-only models were preforming notably weaker. The authors argue that this model works best as a practical triage tool for identifying SSD outpatients who may need more structured suicide assessment rather than as a model that predicts future suicidal behavior.

Outcomes and Implications

This study is impactful because somatic symptom disorder patients often present in outpatient settings where suicidality may be overlooked, so a tool built from information already collected in routine care could make risk screening more realistic. The findings suggest that sleep disturbance, anxiety, and lower cognitive performance may be especially useful signals when deciding who needs closer suicide risk evaluation. Clinically, the model and its web calculator could help support outpatient triage using accessible inputs like qGooduestionnaires, labs, and vital signs. The authors are clear that it still needs external validation and prospective testing before it should be adopted in routine practice. For now, it looks more like a promising support tool than something ready for immediate broad implementation.