Psychiatrists have long known that a patient's risk for suicide spikes after hospitalization for a mental health disorder.
But being able to predict who may be at the most risk has proved elusive: Few patterns have emerged from the thousands of post-hospital suicides in the U.S. to determine who might need extra support once they've been discharged.
New evidence from a massive Army study, however, could lead to improvements in predicting who is at highest risk for committing suicide.
As part of the Army STARRS project (Study To Assess Risk and Resilience in Servicemembers), professor of health care policy at Harvard Medical School Ronald Kessler and others found that by analyzing all sorts of data on soldiers hospitalized for mental health conditions, they could develop an algorithm to determine who may be at highest risk.
According to the study published Wednesday in JAMA Psychiatry, more than half the suicides among soldiers that occurred after hospitalization from 2004 to 2009 happened among 5 percent of troops determined by the algorithm to be at highest risk.
In the five years covered by the researchers, 68 soldiers died by suicide within 12 months of being discharged from a hospital. The figures were relatively low compared with the number of hospitalizations for mental health issues during that time frame — 53,769 — but the rate of suicide among the population was high, nearly 264 per 100,000 person-years compared with 18.5 suicides per 100,000 Army-wide.
And among the 5 percent determined by the algorithm to be at highest risk, the suicide rate was 3,824 per 100,000 person-years.
That figure wasn't the only disturbing trend: Soldiers in the predicted high-risk group suffered seven unintentional injury deaths, 830 suicide attempts and 3,765 additional hospitalizations within 12 months of discharge.
"In the high-risk group, even though suicide in itself was three or four soldiers out of a hundred, at least 40 had something happen to them. These individuals are flirting with death. We saw that they do things that are dangerous ... it's an awful high-risk group," Kessler said.
By pulling together massive amounts of personnel data and combining it with health records, criminal records, legal issues, bank account information and more, researchers were able to determine characteristics and life factors that can predict suicide risk.
They found the strongest predictors included sociodemographic factors such as being male, enlisting at a later age, criminal offenses, weapons possession, prior suicide attempts or thoughts, and prior psychiatric treatment.
But while many of these factors previously have been known to those studying the issues of military suicides, they and other determinants have never been manipulated in such a way that the data could be used to generate a score, or predictor, that could be used by medical providers to understand their patient's propensity for attempting suicide, Kessler said.
The researchers believe that by gathering all available data on individual soldiers, they can create a database where each service member's risk can be predicted. This up-to-date statistic could be included in health records so providers are aware of potential problems if a service member is hospitalized.
"We believe we can take all the information we have for them, current and across their entire careers, and generate a score for them which says if they happen to show up in a mental hospital, here's the score we have for them and here's what their risk is," Kessler said.
With that knowledge, he said, physicians can construct treatment plans and monitor patients more closely than they otherwise would have.
The researchers also hope they can design a data and scoring system that would be tailored to individuals' circumstances and help guide treatment.
For example, Kessler said, psychiatrists and psychologists would treat a young soldier who has combat-related post-traumatic stress disorder and suffered a breakup with his girlfriend differently from an older service member who may be facing involuntary discharge and financial trouble.
The group is currently working with the Center for Army Analysis to develop a framework for the database. Kessler could not say when it would be available for use.
He added that the findings could have broader implications for civilian behavioral health treatment.
"The algorithm could presumably help target preventive interventions as well as treatment," he said.