Novel, Powerful Tool Pinpoints Veterans at Suicide Risk

Liam Davenport

June 22, 2015

A novel algorithm that takes into account numerous factors associated with suicide is able to identify groups of at-risk veterans who may benefit from targeted suicide prevention strategies, say US researchers.

Working from data on more than a million patients in the Veterans Health Administration (VHA), they were able to identify individuals at up to 80 times greater risk for suicide than the overall population.

Crucially, the resulting algorithm, which could be applicable to other healthcare systems, was able to pinpoint more than three times more at-risk individuals than current efforts to flag patients via the Veterans Affairs suicide prevention program.

"The most direct clinical application of predictive modeling would be to allow targeting of selective clinical and preventive services," the authors write.

"One strategy would enhance clinical care, possibly through a program where each patient's provider(s), probably with support from a care manager, would re-evaluate care plans, identify and address barriers to delivering evidence-based care, implement and monitor the outcomes of any needed changes in treatment, and repeat this process as necessary."

They note that another approach would be to implement interventions such as patient education, expressions of caring, regularly scheduled contact, outreach in response to missed appointments, facilitated access to services, and training in coping skills.

The study was published online June 11 in the American Journal of Public Health.

Risk Factors

Coauthor Michael Schoenbaum, PhD, senior advisor for mental health service, epidemiology, and economics at the National Institute of Mental Health, Rockville, Maryland, explained that currently, there are several ways in which an individual at risk for suicide is identified.

"One of them is...people attempt suicide and survive, and are then treated by the medical system for the injuries they've inflicted on themselves; so you identify somebody at suicide risk because they've survived a suicide attempt," Dr Schoenbaum told Medscape Medical News.

"Another way is because people may be brought to or bring themselves to a medical setting without having harmed themselves yet, but they tell the medical care system that they are thinking about harming themselves, that they are at risk for suicide."

"The third way that we identify people is via some interaction between an individual and a clinician in which the clinician does his or her best to try to identify, based on what they see in the interaction with the patient, whether somebody is at risk of suicide."

Dr Shoenbaum continued: "Our concern has been that those signals aren't enough, that we're missing a lot of the people we really need to be helping."

"So, at a high level, the question we were taking on in this study was, are there other ways to do this, maybe better ways to do this? in any case, additional ways beyond the ways I've described?"

The researchers used data from the VHA National Patient Care Database to construct three samples: a model development, a model validation, and a prediction sample model.

The development and validation samples contained data on all patients who died from suicide between 2008 and 2011 and had received VHA services in the previous 2 years, and a random 1% of patients who survived and had received VHA services in the previous 2 years.

Randomly assigning half of the patients and control individuals to the development and validation samples, the team created samples containing 3180 patients and 1,056,004 control persons.

They discovered that individuals who committed suicide were more likely than control persons to have the following characteristics:

  • To be young, male, and unmarried; to live in a rural area; to have a history of or be at risk for homelessness

  • To have no service-connected disabilities

  • To have been diagnosed with mental health conditions, pain, sleep disorders, and traumatic brain injury; to have used VHA mental health services

  • To have had psychiatric hospitalizations; to have received mental health residential care and emergency department or urgent care

  • To have used psychotropic medication; and to have previously attempted suicide

Developing a predictive algorithm on the basis these differences, the investigators found that in the highest 0.01% stratum for calculated risk for suicide in the development and validation samples, the risk for suicide in the next 12 months was 82 and 60 times greater, respectively, than the rate in the overall sample.

In the highest 0.10% stratum for calculated suicide risk, the rates for suicide were 39 and 30 times greater, respectively, than in the overall sample. In the highest 1.00%, the risks were 14 and 12 times greater, respectively; those in the higher 5.00% were 6.3 and 5.7 times greater, respectively.

Powerful Tool

Crucially, the algorithm was able to identify more at-risk individuals than the current VA flagging system. For the top 0.01%, the flagging system identified 31% of at-risk individuals, accounting for 3 of 7 suicides.

For the top 0.10%, 21% of at-risk individuals were flagged, which translated into 16 of 35 suicides. For the top 1.00%, 8.8% of individuals were flagged, accounting for 38 of 176 suicides.

Discussing why the algorithm was able to hone in so many more individuals at risk for suicide, Dr Shoenbaum said that there are a number of difficulties with relying on associations between suicide and single characteristics or experiences, such as depression.

"One of them is that the vast majority of people with any given risk factor, even powerful risk factors like depression or posttraumatic stress disorder, don't have the outcome," he noted.

"The second is that when one looks backwards, when one starts with people who have the outcome, it turns out that only a minority and never a majority have any given risk factor in common."

"Together, these things mean that if you want to take a public health perspective and you're responsible for a population, and you want to find the people in that population who, at a point in time, you should be focusing on to prevent or treat suicidality, you don't get a lot of leverage by looking at one risk factor at a time, or even by using a checklist of risk factors that a clinician might implement in an individual patient encounter."

Dr Schoenbaum said the algorithm was so powerful in identifying patients at risk for suicide because they were able to examine data on hundreds of characteristics that are empirically associated with suicide.

"We were able to identify a relatively small fraction of the larger population that accounted for a disproportionately high fraction of all the suicides."

However, Dr Shoenbaum stressed that the clinical flagging system currently used by the Veterans Administration (VA) remains a valuable tool, despite having identified a relatively small fraction of individuals at higher risk for suicide.

"This not the fault of the VA's clinical method," he said, adding: "The clinical flagging is important, and of course the VA should continue to use it, and they should continue to act on it."

"What I would say is that the [algorithm] lets the VA identify many, many additional people that weren't previously on their radar screen."

Emphasizing the importance of the study, Dr Shoenbaum said: "We undertook this research to answer the question of whether it was possible to use VA patient care records to do this kind of predictive analysis."

"My view of our study is that we tested that question, and we answered that, yes, it is possible to do this analysis."

He continued: "Now the next steps are, Are we going to implement these methods? and then what are we going to do for the people that we identify? I guess I would say that's going to be an evolving discussion."

"The details are going to vary a lot, depending on who the groups are you are identifying. They're going to depend on the particular clinical setting where you can bring them in, and they are going to depend on a patient-by-patient basis exactly what the clinical care system finds out might be going on with these people, for the individual patient that needs to be addressed," Dr Shoenbaum added.

"The common denominator is going to be proactive engagement, proactive contact. Don't let people fall through the cracks, and don't wait until they happen to show up on their own and ask for help," he concluded.

"Groundbreaking" Research

In an interview with Medscape Medical News, coauthor Caitlin Thompson, PhD, deputy director for suicide prevention for the US Department of Veterans Affairs, Washington, DC, described the study as "groundbreaking."

She explained that although the VA's current suicide prevention program is an "excellent clinical way" to identify veterans who might be at high suicide risk and then provide an enhanced care delivery system, the novel algorithm will enhance those efforts.

Dr Thompson said: "This study is able to further identify those patients that may not be as readily evident to be at high risk for suicide."

"So by being able to find veterans who maybe aren't talking about their suicidal thoughts and feelings as openly as others, and those who may have never even sought mental treatment but are shown to have these factors that put them at high risk, it really is an additional gift in terms of them allowing our suicide prevention coordinators and VA clinicians to really do some interesting interventions with these folks even before they may have even talked about suicide."

Both Dr Shoenbaum and Dr Thompson agree that the current study is an important first step in that process, with Dr Thompson emphasizing that it "is going to be very iterative."

"This is our first look at this," she said, "and we are currently designing specific interventions for those individuals at varying points of being at high risk on the algorithm."

"Then we’ll be able to continue to refine and refine and hone both the algorithm and the specific interventions as we study the effects."

Dr Thompson pointed out that suicide rates in the VA system have been leveling off in recent years while those in the general population have been increasing.

"That speaks to us that we are doing something at VA that's working, but I think that what this is really going to help us do is to get these rates lower and lower, because we'll be able to do more targeted interventions with more people who might not have been clinically immediately apparent," she said.

Crucially, Dr Thompson also believes that the algorithm is applicable to medical systems other than the VA. She said: "I absolutely think it is fundamental work that I think other, larger medical facilities will be able to capitalize on."

The authors report no relevant financial relationships.

Am J Public Health. Published online June 11, 2015. Abstract


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.