Long-term health histories embedded in electronic health records (EHRs) can be useful for predicting patients at high risk for suicidal behavior, and that data could be used as an early warning system, according to a study by Boston Children’s Hospital researchers.
Researchers with Boston Children's Hospital Informatics Program sought to determine whether longitudinal historical data, commonly available in EHR systems, can be used to predict patients’ future risk of suicidal behavior.
The results of the study were recently published in The American Journal of Psychiatry.
The research team, led by Yuval Barak-Corren of the Predictive Medicine Group at Boston Children’s Hospital Informatics Program, developed a model using EHR data from a large health care database for inpatient and outpatient visits spanning 15 years, from 1998 through 2012. Patients with three or more visits were included in the study. And, the researchers defined suicidal behavior using expert clinical consensus review, supplemented with state death certificates.
Among the study population, 1.2 percent of the cohort met the case definition for suicidal behavior, according to the study.
The researchers’ model achieved 33 percent to 45 percent sensitivity, or true positive rate, and 90 percent to 95 percent specificity, or true negative rate. In addition, the model predicted patients’ future suicidal behavior on average of three to four years in advance.
According to the study authors, the strongest predictors of future suicidal behavior risk identified by the model include both well-known risk factors, for example, substance abuse and psychiatric disorders, as well as less conventional risk factors, such as certain injuries and chronic conditions. This indicates that “a data-driven approach can yield more comprehensive risk profiles,” the study authors wrote.
“These findings suggest that the vast quantities of longitudinal data accumulating in electronic health information systems present a largely untapped opportunity for improving medical screening and diagnosis,” the study authors wrote. “Beyond the direct implications for prediction of suicide risk, this general approach has far-reaching implications for the automated screening of a wide range of clinical conditions for which longitudinal historical information may be beneficial for estimating clinical risk.”
