Models Derived From Electronic Health Records Effective in Predicting Suicide Risk

According to a health systems analysis published in the American Journal of Psychiatry, prediction models using health record data and responses to self-report questionnaires significantly outperform existing suicide risk prediction methods.

Electronic health records available at 7 health systems serving a combined population of nearly 8 million members were accessed as part of the analysis between January 2009 and June 2015. Patients were included in the study following visits to mental health clinics or primary care clinics at which a mental health diagnosis was recorded. More than 300 demographic and clinical predictive factors were extracted from records for up to 5 years before each qualifying visit, including age, sex, race, ethnicity, current and past mental health and substance use diagnoses, past suicide attempts, past injury or poisoning diagnoses, and dispensed prescriptions for mental health medication. 

Logistic regression models predicting suicide attempt and death were developed using penalized least-absolute shrinking and selection operator (LASSO) variable selection in a random sample of 65% of eligible visits. The predictive models were subsequently validated using the remaining 35% of visits.  

During the study period, a total of 2,960,929 patients aged 13 years or older (mean age, 46 years; 62% female) made 10,275,853 specialty mental health visits and 9,685,206 primary care visits with mental health diagnoses. Health system records and state death certificate data identified 24,133 suicide attempts and 1240 suicide deaths, respectively, in the 90 days following an eligible visit. Mental health specialty visits with risk scores in the top 5% of the predictive model accounted for 43% of subsequent suicide attempts and 48% of suicide deaths. Of patients scoring in the top 5%, 5.4% attempted suicide and 0.26% died by suicide within 90 days. Primary care visits with scores in the top 5% accounted for 48% of subsequent suicide attempts and 43% of suicide deaths.

Prediction models incorporating both health record data and self-report questionnaire responses were thus highly effective in predicting suicide attempt and suicide death risk in patients following qualifying visits. These models substantially outperform existing tools for suicide prevention, indicating a potential need for modification of currently available interventions.

Reference

Simon GE, Johnson E, Lawrence JM, et al. Predicting suicide attempts and suicide deaths following outpatient visits using electronic health records [published online May 24, 2018]. Am J Psychiatry. doi:10.1176/appi.ajp.2018.17101167