Clinical Note–Extracted Psychosocial Factors for Predicting Suicide Attempt Among ED Patients With Suicidal Ideation

Importance The Joint Commission recommends universal suicide screening in emergency departments (EDs), which emphasizes the need to identify at-risk individuals. Existing suicide risk prediction models rely primarily on clinical data and demonstrate limited performance. The potential of incorporating psychosocial information to enhance predictive performance remains understudied. Objective To evaluate whether augmenting clinical data–based risk scores with psychosocial factors improves the prediction of suicide attempt (SA). Design, Setting, and Participants This retrospective prognostic study based on electronic health record data included 4661 ED patients discharged after presentation for suicidal ideation (SI) from middle Tennessee hospitals between June 1, 2018, and February 27, 2024. Main Outcomes and Measures The primary outcome was SA within 90 days of ED admission and time-to-event in days. Clinical data–based Vanderbilt Suicide Attempt and Ideation Likelihood (VSAIL) score and 6 psychosocial factors (homelessness, financial insecurity, chronic stress, social isolation, loneliness, and adverse childhood experiences) derived from clinical notes were integrated using a Cox proportional hazards regression model. Performance metrics included area under the receiver operating curve (AUROC), area under the precision-recall curve (AUPRC), positive predictive value (PPV), negative predictive value, sensitivity, and specificity. Performance was evaluated for models trained on (1) VSAIL, (2) psychosocial factors, and (3) VSAIL plus psychosocial factors. Results This study included 3382 Vanderbilt University Hospital (VUH) (mean SD age, 26.1 15.6 years; 1751 males 51.8%) and 1279 Regional Health Systems (RHS) (mean SD age, 34.5 18.0 years; 715 males 55.9%) ED visits for SI. Within 90 days, SAs were reported in 160 (4.7%) VUH and 34 (2.7%) RHS ED visits for SI. Compared with VSAIL alone, VSAIL plus psychosocial factors was associated with significantly increased median AUROC (VUH: 0.645 IQR, 0.645-0.645 vs 0.734 IQR, 0.719-0.747; P lt; .001; RHS: 0.547 IQR, 0.547-0.547 vs 0.680 IQR, 0.672-0.687; P lt; .001), AUPRC (VUH: 0.083 IQR, 0.083-0.083 vs 0.122 IQR, 0.111-0.137; P lt; .001; RHS: 0.029 IQR, 0.029-0.029 vs 0.054 IQR, 0.052-0.058; P lt; .001), and PPV (VUH: 0.093 IQR, 0.082-0.094 vs 0.143 IQR, 0.123-0.161; P lt; .001; RHS: 0.042 IQR, 0.040-0.043 vs 0.112 IQR, 0.096-0.129; P lt; .001) while maintaining specificities above 0.90. Chronic stress emerged as the strongest predictor of SA (β = 0.643 95% CI, 0.427-0.859; P lt; .001). Conclusions and Relevance In this prognostic study of patients discharged from the ED after presentation for SI, augmenting a clinical data–based suicide risk prediction model with clinical note–extracted psychosocial factors was associated with significantly higher predictive performance. These findings suggest that psychosocial factors can enhance risk stratification and support targeted interventions, such as therapies addressing chronic stress.