Comparing patterns of recent mental health service use for predicting suicidal events following emergency department mental health visits in the United States: A national cohort study

Objective: To examine patterns of recent service use to predict non-fatal suicidal events shortly following emergency department (ED) visits for mental health. Methods: For this retrospective cohort study, we used Optum electronic health record data from 2,445,597 ED mental health episodes (2015–2022) for persons aged ≥ 10 years. We then constructed a series of logistic regression models to evaluate how six permutations of characterizing prior 180-day mental health service use predicted acute non-fatal suicidal events within 180 days of ED discharge beyond demographic characteristics and ED mental health diagnoses. Model performance was assessed by area under the receiver operating curve (AUC). Results: Overall, 7.2% (n = 176,000) of episodes resulted in an acute suicidal event within 180 days. Model performance improved from demographic characteristics and ED mental health diagnoses (AUC = 0.76) when past 180-day service use variables were added, but with minimal differences between a binary variable of any mental health service use compared to monthly counts, weighted slopes, or interactions (AUC’s 0.78–0.82). The final model containing demographics, ED mental diagnosis, and past 180-day service use yielded an AUC of 0.83. The most predictive past service use variables were any inpatient or ED event for self-harm or suicide ideation (OR = 5.45, 95% CI = 5.37, 5.54) and any ED mental health visit (OR = 1.87, 95% CI = 1.84, 1.90). Conclusions: As part of evaluating suicide risk in ED settings, information about recent acute care for mental health, suicidal ideation, or self-harm use significantly contributes to the short-term prediction of non-fatal suicidal events.