Using Electronic Health Records to Assess Suicide Risk

A research study found that electronic health record (EHR) information can serve as an “early warning system” to alert clinicians about patients who should be assessed for suicide risk. Researchers conducted an analysis of historical EHR data to develop a model that identified patients who were at risk for attempting or dying by suicide, and suggested that this model could be used to predict future suicide risk. They cautioned, however, that clinical decisions should not be made solely on the basis of a patient’s medical history.

The study analyzed EHR data on patients who had made three or more visits to a health care system. Based on diagnoses and demographic characteristics recorded in the EHR, the researchers developed a model that could predict a patient’s risk for attempting or dying by suicide. Patients who did attempt or die by suicide were identified by case reviews and death certificates.

The model identified 45 percent of patients who attempted or died by suicide. These patients were flagged as being at risk an average of four years before the suicidal behavior occurred. Only 10 percent of patients identified as being at risk did not attempt or die by suicide. While the model achieved similar results for men and women, it was most effective when it was used for gender-specific age groups. For example, it identified 54 percent of women ages 45 to 65 who would attempt or die by suicide within four years of the assessment.

The study identified conditions that are not usually included in suicide assessments, such as hepatitis C, which helped predict suicide risk. According to the authors, the success of the model was based on combining these risk factors with more commonly used indicators of risk, such as depression. Assessments that relied only on commonly used risk factors were less effective. For example, histories of depression and substance abuse only predicted 34 percent of the patients who attempted or died by suicide. Models that used a single risk factor were even less powerful, such as a history of depression alone, which only identified 29 percent of the patients who would attempt or die by suicide.

Barak-Corren, Y., Castro, V. M., Javitt, S., Hoffnagle, A. G., Dai, Y., Perlis, R. H., . . . Reis, B. Y. (2017). Predicting suicidal behavior from longitudinal electronic health records. American Journal of Psychiatry, 74(2), 154–162.