But for some soldiers, those three brief words signify something darker and far more personal. One of the most troubling statistics to emerge from the wars in Iraq and Afghanistan that followed the Sept. 11, 2001, terrorist attacks has been the suicide rate among military personnel. In 2012, for the first time, more active-duty service members killed themselves (349) than died in combat (295). While suicide affects all military branches, its consequences within the Army have been felt most keenly.
For decades, the Army’s suicide rate hovered far below that of civilians in the same demographic. But that changed in 2008, when data showed that 129 soldiers took their own lives the previous year, about 30 deaths per 100,000—the highest rate of suicide among soldiers since the service began tracking the statistic in 1980. Yet worse was to come. The Army suicide rate, which had been rising since 2004, continued to climb through 2012.
In response to the crisis, the Army in 2009, together with the National Institute of Mental Health, created the Study to Assess Risk and Resilience in Servicemembers, known as the STARRS program. It found that suicide rates for soldiers who deployed to Iraq and Afghanistan more than doubled from 2004 to 2009 to more than 30 deaths per 100,000 active-duty soldiers, during which period the suicide rate for the civilian population held steady at 19 per 100,000.
But the trauma of combat does not explain why the suicide rate for soldiers who did not deploy nearly tripled during that period, from about 10 per 100,000 to between 25 and 30 per 100,000.
We wanted to identify groups of soldiers with the highest risk of suicide, and we have found we can do that. That’s the really exciting part because if you can’t find them, you can’t help them.
dr. michael schoenbaum, nimh
“We probably went into this thinking in terms of ‘war is hell,’ and if you send people to war, then bad things happen to those people. But the suicide rate rose as much in soldiers who had not ever been deployed as it did in those who did,” says Dr. Michael Schoenbaum, senior advisor for mental health services, epidemiology and economics at NIMH’s Office of Science Policy, Planning and Communications.
Schoenbaum was also a scientific principal for the STARRS study, which began in 2009 with $50 million in funding from the Army under the leadership of then-secretary Pete Geren. Initially planned as a five-year effort to study risk and protective factors for suicidality within the Army, the program received a one-year, no-cost extension, which expired in June.
A BIG DATA CHALLENGE
The first phase of STARRS, conducted as a partnership between the Army, NIMH, and several academic institutions, involved analyzing Army and Defense Department personnel and administrative records for more than 1 million soldiers who served on active duty since 2004. That included some 1.1 billion data points from 39 Army and Defense Department databases. Sociodemographic variables such as gender, age, race, religion, education and family status were comingled with other Army data, including health records, Army entry characteristics and service records, which include rank, length of service, demotion history and any involuntary extensions of a soldier’s active-duty term, known as stop-loss orders, and other data points.
Altogether, the project included “hundreds of terabytes of data,” Dr. Kenneth Cox, Army STARRS scientific liaison and medical informatics consultant, told Government Executive. To put that in perspective, the printed collection of the U.S. Library of Congress would comprise only about 10 terabytes.
The teams packaged data in Hadoop clusters, computational groupings designed for storing and analyzing large data sets, and analyzed them with a mixture of open source and proprietary data management and statistical analysis software. Mixing those data sets together and analyzing them, the researchers hoped to use the Army’s existing information to identify which soldiers might be most at risk for attempting suicide.
“The metaphor I use sometimes is looking for needles in a haystack,” Schoenbaum says.
The Army’s active-duty population is currently around 500,000 people, although it has reached as high as 700,000 in recent years. In any given year there will be 150 to 160 suicides, “so in a haystack of 500,000 people, you’ll have 150 or so needles,” Schoenbaum explains.
“We went into this hoping that it would be feasible to use existing data—data from existing systems, like personnel and health systems and other kinds of computer systems the Army and DOD maintain routinely,” he says. “We hoped it would be possible to use data from those systems to build smaller haystacks. We wanted to find and identify groups of soldiers, based on common characteristics and experiences, with the highest risk of suicide, and we have found we can do that. That’s the really exciting part because if you can’t find them, you can’t help them.”
Using existing data, risk modeling and computer algorithms, STARRS can narrow the haystack to almost 5,000 soldiers at the highest risk for suicide. Data suggest those soldiers—1 percent of the Army’s active-duty personnel—could account for up to 25 percent of suicides, making some combination of targeted intervention and treatment both feasible and potentially very effective.
The Army, however, is still figuring out how to digest the data.
Speaking at a May analytics conference in Washington, Roy Wallace, the Army’s assistant deputy chief of staff, points out the moral dilemma service leaders face. Based on statistical trends, the Army could expect approximately 150 suicide deaths in the coming year, far fewer than the nearly 5,000 active-duty soldiers STARRS identifies as high risk.
“So there are thousands of false positives, which presents us with a moral dilemma in dealing with these individuals,” Wallace says. Because sensitivity and privacy are paramount, Cox says the Army will not “identify specific individuals” or “take actions or make decisions on a business level” yet.
In terms of identifying individuals at risk, “we are not going to do anything until we’re very sure that we won’t make things worse,” he says.
Making things worse is a very real possibility, according to Schoenbaum.
A suicide analysis of U.S. veterans published in The Journal of the American Medical Association Psychiatry in April suggested that vets forced out of the military for misconduct or other problems were twice as likely to take their own lives. They also are ineligible for the psychiatric care available to honorably discharged vets.
Hypothetically, in downsizing, the Army could use a risk algorithm to remove active-duty soldiers with the highest risk for suicide.
“Aside from being unethical, it would actually, quite probably, make things worse for everyone because people aren’t stupid,” Schoenbaum says. “If they learned that the Army is using risk algorithms to draw down, they’ll try to figure out what factors light up on the risk algorithm, and what leads to the red flag. What will end up happening if information about predicted risk is used in punitive ways instead of to help people, then what you’ll probably do is drive the problem underground.”
Over the next year or two, Col. Kevin Bigelman, chief of the operations and training division within the Army Resiliency Directorate, says the service will update its suicide prevention training with knowledge gleaned through the study.
Findings from the STARRS project has verified some previous research, displaced a few myths and offered new insights into the relationship between suicide and accidental deaths. For example, STARRS confirmed that women have a consistently lower risk of suicide than men, and that young soldiers have an elevated risk before, during and after deployment. Another expectation confirmed: During deployments, married soldiers and those with additional dependents are at much lower risk for suicide than unmarried soldiers with no dependents. The risk evens out, though, for married and unmarried soldiers who have either never deployed or previously deployed.
Outside of war, accidental deaths, like those resulting from vehicle accidents or training exercises, are traditionally the Army’s biggest killer. Between 2004 and 2009, 1,331 soldiers died in such accidents. In 2014, STARRS published research that said “predictors of accident deaths (e.g., being male, delayed rank progression) also identified risk for cases of completed suicide in Army soldiers from the same cohort.” Many of the same risk factors the STARRS team uses to algorithmically predict which active-duty personnel are at high risk for suicide—delayed promotions, low level of education, inadequate training and health concerns, for example—mirrored those that put Army personnel at increased risk of being seriously injured or killed in accidents.
Accidental deaths did not appear to be related to direct combat action. After 2004, accident deaths increased most among enlisted soldiers who never deployed. Yet accident deaths for previously deployed, enlisted soldiers held steady. Perhaps most surprisingly, “unadjusted odds of accident death declined” among currently deployed, enlisted soldiers, according to STARRS’ 2014 research.
“Along the way, we did some checking, and that same model that seems to identify soldiers at risk for suicide is also very good at predicting soldiers at risk for death by accident or serious injury,” Cox says. “Many of the risk factors turn out to apply to other things we don’t want to happen besides suicides.”
That data could prove vital in future Army policy. Accidental deaths in the Army typically occur twice as often as suicide deaths, Cox says.
“People think accidents are unavoidable, but many, if not all, accidents can be prevented, if people have taken the right steps,” Cox says. “The major thing is we might be able to impact both these negative adverse effects.”
STARRS’ 2014 research also indirectly contradicts the notion that exposure to combat-related trauma is the exclusive driver of increased Army suicides. Over the Army’s general soldier population, suicide risk is slightly higher among those currently and previously deployed than those who never deployed, but the significant rise in suicide rate among those never deployed during the Army’s War on Terror campaign suggest many more factors are in play.
Perhaps what most surprised the researchers was that the data showed “no consistent association of either stop-loss orders or accession waivers” to suicide deaths. Relaxed waiver qualifications during the buildup were thought to explain the rise in Army suicides, yet STARRS’ research attributes its higher association with suicide to missing waiver data for “almost all officers and for most soldiers with more than five years of Army service—both groups with low suicide risk.
“Within the subsample of enlisted soldiers with five or fewer years of service, which covers the difficult recruiting periods associated with the troop surge in Iraq and for which waiver data were essentially complete, no statistically significant association was found between suicide risk and receiving a medical, substance use, or conduct waiver,” the STARRS study says.
Suicide rates are highest among deployed junior enlisted E1 and E2 soldiers (83 deaths per 100,000 person-years of active-duty Army service) and those soldiers who were demoted within the previous two years across all deployment categories (50 deaths per 100,000 person-years of active-duty Army service).
Data suggests that a young soldier’s first deployment significantly increases his or her risk of suicide compared with never having deployed, Shoenbaum says. After soldiers deploy for the first time, the risk “never went back to predeployment rate after they return from theater.”
That itself isn’t particularly surprising, but Schoenbaum says STARRS data suggest soldiers on second and third deployments do not experience an increased suicide risk. That is not to say soldiers on their second, third or more tours of duty and their families are not facing added stressors— they are—but they’re also resilient, he adds.
“What we found is after a first deployment, if you stay in the Army and the Army chooses to deploy you again, the people who get a second or third deployment, on average, they are very resilient,” Schoenbaum says. “The suicide rate doesn’t go up further among those with one deployment on their second deployment, and the third deployment also does not increase your risk relative to the second deployment,” he says.
“Whatever process the soldiers use to stay in the Army and the Army [uses to] decide who to keep, those filters work,” Schoenbaum says. “At the [macro level], what we could say quite confidently is those multiple deployments are not what is driving up the Army’s suicide rate.”
MORE DATA TO COME
Most of STARRS’ research to date is based on the trove of information contained within administrative and personnel records. Thousands of soldiers whose records were contained within the first batch of STARRS data were already deceased. While extraordinarily useful, another more individualized aspect of the effort will continue to bear fruit for years to come. Called the STARRS longitudinal study, it aims to more directly study soldiers over time.
In 2011 and 2012, Army STARRS conducted surveys with 110,000 soldiers, 57,000 of whom were in their first weeks of service. These were not short surveys, but two-and-a-half-hour sessions consisting of interviews and computer-based cognitive performance tests.
Of those soldiers, 43,000 also provided blood samples. Blood samples could contain bio markers that indicate stress levels, for example, and future algorithms might be able to factor in such information.
The Army’s future risk algorithms may incorporate administrative and personnel records combined with these more nuanced data sets for more accurate insights and results, but for now it’s a waiting game.
We are not going to do anything until we’re very sure that we won’t make things worse.
dr. kenneth cox, army
“We got this incredibly rich information on 110,000 people, all alive,” Schoenbaum says. “But to study negative outcomes like suicide, we need time to pass. We have to wait and unfortunately, the reality of it is, some of those people we interviewed went on, or will go on, to have adverse outcomes. The research involves looking to see what we can learn from what they told us at the beginning that might be predictive about who goes on to have bad outcomes.”
In other words, some soldiers must die to advance the science.
Because most soldiers stay with the Army for several years, the only way to continue adding to the data will be following those soldiers throughout their careers. The Army is exemplary at tracking soldiers during their active-duty career, but that effectiveness doesn’t carry over once soldiers separate. STARRS collected next-of-kin and contact information voluntarily from most of the 110,000 soldiers it surveyed, making it easier to follow soldiers through their lives. But what might prove even more valuable to the longitudinal study is a budding partnership with the Veterans Affairs Department.
The initial STARRS funding expired June 30, but the STARRS longitudinal study has received funding for a five-year period, according to Scott Ludtke, acting executive director for Army STARRS. Funding will come from the Office of the Assistant Secretary for Health Affairs.
Ludtke says a memorandum of agreement between the Army, the Defense Department and the National Institutes of Health will ensure continued research under the STARRS longitudinal study. Importantly, VA will be a part of it.
“We are going to collaborate and continue and extend the data platform and research program established with Army STARRS,” Ludtke says. “VA will also participate and collaborate. The extent of involvement is not known at this time.”
If information about predicted risk is used in punitive ways . . . then what you’ll probably do is drive the problem underground.
dr. michael schoenbaum, nimh
STARRS’ continued research could play a major role in responding to President Obama’s executive order to improve access to mental health services for veterans, service members and military families. The National Research Action Plan, unveiled in 2013, is a formalized response to that order, creating a partnership between DOD, VA, and the departments of Health and Human Services and Education.
“Hopefully we can work with VA and add this kind of information to the platform,” Schoenbaum says. “The point of following (soldiers) forward is important in terms of maximizing value of what we can learn.”
More work—and insights—are yet to come, and results may carry over to the other military branches, which have their own suicide reduction efforts underway.
Cox calls STARRS a “major informatics achievement,” yet he says that description undervalues the importance of what has been learned so far.
Dr. Christopher Ivany, chief of the Behavioral Health Division at the Office of the Army Surgeon General, says STARRS has “confirmed previously known risk factors” and “in some cases raised new risk factors that were not well-known.” Those findings were all made using data that already existed on a DOD or Army server somewhere. The STARRS longitudinal study, containing a deeper pool of data per soldier, has exceptional potential to help the Army reduce its suicide and accident rates.
“Much of the impact is to be determined as we sift through finds and work with databases and risk modeling that is ongoing on the Army side,” Ivany says. “We still have potential and more benefits we’ll see in the future as we work in the risk models the Army STARRS team has provided.”
If war is hell, computer algorithms and all the data in the world may never change it. But the STARRS project suggests the Army is a step closer to predicting suicide, as opposed to reacting to it, by forecasting risk factors that drive soldiers to take their own lives. The expected net effect won’t only be a reduction in suicide rates, but an improvement in the overall mental health of the Army, from top brass down to the most junior enlistee.
Frank Konkel is the editorial events editor for Government Executive Media Group and a technology journalist for its publications. He writes about emerging technologies, privacy, cybersecurity, policy and other issues at the intersection of government and technology. Frank also runs Nextgov's Emerging Tech blog. He began writing about technology at Federal Computer Week and previously reported on local and state issues at daily newspapers in his home state of Michigan. Frank was born and raised on a dairy farm and graduated from Michigan State University.