Nearly every day, an active-duty service member takes his or her own life, causing waves of grief among families and peers.
And each day, on average, 22 military veterans in the United States commit suicide.
Roughly one in five Americans who commit suicide is or was in the service.
The military has been working aggressively since 2007 to try to stem suicides, but everyone agrees there will be no easy fix.
Now, however, big-data specialists, including the Newton software firm Attivio Inc., are collaborating with military suicide experts to try to address the problem by using social media to monitor veterans for signs of despondency.
The specialists said they are identifying key words and phrases that suggest someone is spiraling downward, while developing an analytics system that could examine thousands of online posts and alert medical specialists and family members when a veteran’s comments indicate he or she is at risk of committing suicide.
Relatively few people come out and say they are suicidal. But by tracking the postings of veterans who agree to participate in the system, organizers hope eventually to identify those at high risk, and to intervene early enough to make a difference.
The program, called the Durkheim Project after the man known as the father of sociology, Emile Durkheim, is developing algorithms to determine which phrases or combination of phrases are most predictive of suicide attempts.
“It’s the words they are using that’s the reliable signal,” said Chris Poulin, the director and principal investigator of the Durkheim Project and a predictive-analytics expert.
Although Poulin would not reveal the precise phrases the Durkheim Project has identified, suicide researcher Craig Bryan, a University of Utah psychologist who is advising the project, said that the coded language of the suicidal often includes phrases such as “You’d be better off without me,” “I messed everything up,” and “I can never be forgiven for my mistakes.”
Moreover, behaviors such as buying a gun or giving away belongings can help to identify at-risk veterans and are often reported on social media, Bryan said.
In the first phase of the project, which was just completed, Poulin and colleagues developed a language-driven predictive model to estimate suicide risk.
They based their research on doctors’ notes from the records of three groups of veterans: those who had committed suicide, those who had psychiatric challenges but were not suicidal, and those with no apparent psychiatric issues.
From the doctors’ notes, Poulin said, it was possible to distinguish between the three groups.
Those whose psychiatric problems were not life-threatening tended to talk about their eyesight, for example, or their preoccupations and personal hygiene. Those who were healthy talked to their doctors about muscle or joint pain or eating disorders. And the suicidal discussed their agitation and fears, along with their need for painkillers.
In the next phase of the Durkheim Project, Poulin and others will test the predictive quality of those insights among as many as 100,000 service members and veterans who agree to have their social media and mobile posts shared.
This kind of varied language, as well as the shorthand used on social media, can be extremely challenging to analyze, said Sid Probstein, the chief technology officer for Attivio, which is responsible for that analysis.
How those phrases change over time can also be a warning sign, Probstein said, so a huge amount of data has to be gathered from text messages, Twitter, Facebook, and other social media outlets and analyzed.
This kind of analysis of so-called “natural language” has been possible only recently, Probstein said, “so it’s really only in the last decade that you could imagine doing something like the Durkheim Project.”
Facebook, which helped debug the program, will assist in recruiting volunteers, Poulin said. The military provided a two-year start-up grant for the research.
Eventually, those whom the project identifies as at-risk will be automatically linked to resources in their area to get help, and their support network will be notified.
Users will be able to opt in and out of the system, so their privacy will not be compromised by the data analysis without their knowledge, said Poulin, who also worked with former colleagues at the Geisel School of Medicine at Dartmouth College.
“You don’t just get the person to opt in, but you get their friends and family to opt in, and let family know this person is suffering,” Poulin said.
If they do not have such support in their lives, there are scripts that can be read to at-risk people that have been shown to help, he said.
Even people who feel alone will leave a footprint on Facebook or via a text, Poulin said. “Their social network may be very small, but not so small that they don’t use a phone,” said Poulin.
Bryan, associate director of the National Center for Veterans’ Studies at the University of Utah, said the pressures on the military have increased exponentially over the past decade, as wars in Iraq and Afghanistan have dragged on and downsizing has meant service members doing more with less, increasing their stress and suicide risk, both during their service and after retirement.
Multiple deployments also take their toll on service members, who may be away from home for most of six or seven straight years, said Kelly Posner, director of the Center for Suicide Risk Assessments at Columbia University.
In combat, service members are distracted by the urgency of their tasks. Now that the wars are winding down and they are back home for good, “they’re having to face what they saw, what they did, what they didn’t do, their unmet mental health risks, their families that have not been together,” she said.
Suicide, she said, is the fourth-leading cause of death for those ages 18 to 64, and probably the most preventable, “but we need to keep working hard.”
By using observable behavior, the Durkheim Project may help the invisible become visible, Bryan said.
“That is really one of the missing links in suicide prevention, both inside and outside military.”Karen Weintraub
can be reached at firstname.lastname@example.org.