Tools for Suicide Risk Identification at Discharge May Be on Horizon

Characteristics of Profitable Hospitals Identified
Characteristics of Profitable Hospitals Identified
Those admitted to a hospital with a psychiatric diagnosis were 63% likely to die by suicide.

It may be possible to develop a tool to identify patients at the highest risk of death by suicide at the time of hospital discharge by looking for specific clinical features and analyzing the language in their narrative discharge notes, suggests a new study.

An estimated 12.6 per 100,000 individuals die by suicide annually in the U.S., and suicide represents the second leading cause of death among adolescents and young adults ages 15-24.

Thomas H. McCoy Jr, MD, and his colleagues at Massachusetts General Hospital in Boston retrospectively analyzed clinical data from 458,053 individuals discharged from Massachusetts General Hospital and Brigham and Women’s Hospital from January 2005 through December 2013. The patients represented 845,417 hospital discharges and were followed a median 5 years.

The overall rate of death from any cause during those 9 years was 18%, and the 235 individuals who died by suicide represented 0.1% of the full cohort. Expanding this group to the 2,026 who died by suicide or accidental death comprised 0.4% of the total.

Those admitted with a psychiatric diagnosis were 63% likely to die by suicide and 84% likely to die an accidental death before accounting for other factors. Those with any lifetime psychiatric visits had a 14-15% greater risk of suicide, and those who were male, white, or had an emergency department visits within the year before admission were more than twice as likely to die by suicide. Any lifetime psychiatric visits also increased the risk of accidental death by 16%, and patients with ED visits in the past year were nearly three times more likely to die an accidental death.

After adjusting for these and other sociodemographic characteristics and clinical features, patients were 30% less likely to die by suicide if the narrative notes in their chart had positive valence as identified by a previously validated tool. For example, words such as “glad, pleasant and lovely” reflected positive valence whereas words such as “gloomy, unfortunate and sad” reflected negative valence. Negative and positive valence were measured separately since both may be reflected in notes simultaneously.

Combining analysis of valence in the notes with demographic and clinical features allowed the researchers to develop a prediction model for those at higher risk of suicide.

“We find that incorporating a simple natural language processing strategy improved the ability to estimate risk for suicide and accidental death,” the authors wrote. “There is likely substantial opportunity to better capture words reflecting emotion using more curated data sets and thereby improve discrimination further.”

Still, the limitations and challenges of such a tool mean a need for further refinement and replication as well as risk-benefit analyses, they added.

“Developing a risk stratification model represents only part of a continuum from defining a clinical problem to validation and model presentation,” wrote the authors. “In the absence of prior risk prediction models among medically hospitalized patients, our model using coded data only may be considered a starting point with which efforts at improvement may be compared.”

Related Articles


McCoy TH Jr, Castro VM, Roberson AM, et al. Improving prediction of suicide and accidental death after discharge from general hospitals with natural language processing. JAMA Psychiatry. 2016;73:1064-1071.