How to Stratify Post-ED Suicide Risk in Mental Health Visits
How should clinicians distinguish short-term risk for nonfatal versus fatal suicidal events after an ED visit for a mental health disorder?
Patients discharged from the ED after a mental health visit remain at elevated risk for both nonfatal and fatal suicidal events over the next 180 days. This guide applies to ED encounters for individuals with documented mental health disorders and focuses on how to weigh the current visit and recent clinical history differently for nonfatal versus fatal risk.
-
Start with demographic risk stratification
Use age, sex, race, ethnicity, and insurance as the initial risk frame because demographics alone distinguished later nonfatal and fatal events with AUCs of 0.705 and 0.680, respectively. In this study, fatal events were more common among older adults and males, while nonfatal events were more common among adolescents and females.
-
Add mental health features from the ED visit
Incorporate mental health diagnoses and note-derived mental health features available during the first 2 days of the ED episode rather than relying on demographics alone. Adding the 54 ED mental health features improved model performance to AUC 0.838 for nonfatal events and 0.752 for fatal events, showing that current-visit clinical information materially improves discrimination.
-
Check for suicidal ideation and recent suicide-related care
Give particular attention to suicidal ideation during the ED visit and to any suicide-related event in the prior 180 days if the concern is repeat nonfatal suicidal behavior. In this study, suicidal ideation at the ED visit was the top predictor of nonfatal events, and a suicide-related event in the prior 180 days ranked among the top 3 predictors for nonfatal outcomes.
-
Reweight fatal-risk assessment toward lethality-associated features
Do not assume the strongest predictors of nonfatal events are the strongest predictors of suicide death. For fatal outcomes, male sex, depressive disorders, and white race were the top predictors, and sleep disorders and opioid-related symptoms were considerably more predictive of fatal events than of nonfatal events.
-
Incorporate 180-day clinical history across domains
Use a broad 180-day lookback that includes prior mental health diagnoses, physical health conditions, prior ED or inpatient suicidal presentations, treatment in any setting for suicidal symptoms or behavior, and prior prescriptions for medications with potential suicidal or depressive adverse effects. The full model combining ED features with 180-day historical mental health, physical health, and treatment features performed best, reaching AUC 0.874 for nonfatal events and 0.787 for fatal events.
Clinical Considerations
- The study population was limited to individuals with documented mental disorders, so this workflow may miss people who attempt suicide without a recorded mental health diagnosis.
- Model performance may not generalize to other health systems, patient populations, or countries without comparable EHR and note-derived data.
- The models were not compared against clinician suicide risk assessments.
Bottom Line
After an ED mental health visit, use more than demographics and distinguish fatal-risk signals such as male sex, depression, sleep disorders, and opioid-related symptoms from nonfatal-risk signals such as suicidal ideation and recent suicide-related care.
