EHR Data, Model Better Predict Patient Suicide Risk

Register now

Using data from electronic health records, Department of Veterans Affairs and National Institute of Mental Health researchers were able to identify VA patients with very high, predicted suicide risk that were not previously flagged by clinicians.

According to researchers who developed a suicide-risk algorithm by studying the VA patient population from fiscal years 2009-2011, these methods could help prevent civilian as well as veteran suicides. The study, published in the online issue of American Journal of Public Health, randomly divided the patient population in half — using data from one half to develop the predictive model and then testing the model using data from the other half. Both study samples included 3,180 suicide cases and 1,056,004 control patients.

Also See: Analysis Shows High Rate of Suicide for At-Risk Doctors

Researchers compared predicted suicide risk to actual mortality — suicide data came from the National Death Index — with the goal of assessing the performance of the predictive model. What they found was that their predictive model was more sensitive than a VA care system used to identify patients as being at high-risk for suicide based on information obtained during clinical encounters. In fact, in groups with the highest predicted suicide risk based on the model, less than one-third of patients had been identified clinically.

“This is valuable, because it gives the VA more extensive information about suicide risk,” said Michael Schoenbaum, senior advisor for mental health service, epidemiology and economics at NIMH and one of the co-authors of the study. “If the VA can identify small groups of people with a particularly high-risk of suicide, then they can target enhanced prevention and treatment services to these highest-risk individuals.”

“It’s particularly encouraging that these analyses use the types of data available to any large healthcare system,” added NIMH Director Thomas Insel, M.D.

For reprint and licensing requests for this article, click here.