Kelly McSweeney

Jan 19th 2017

How Behavioral Science Experts Use Predictive Analytics for Suicide Prevention


In behavioral science, predictive analytics has great potential to help us anticipate the decisions people will make in the future. Data mining, statistical algorithms, machine learning and artificial intelligence can all be used to develop models that suggest future trends and behavior patterns among segments of the population — including those with increased tendencies to attempt suicide.

While data analytics are commonly used by companies to predict consumer behavior (e.g., how likely a person is to buy a product or repay a loan), they can also be used by mental health professionals. Ultimately, being able to predict human behavior could mean the difference between life and death. This is why behavioral scientists at agencies such as the U.S. Department of Veterans Affairs (VA) and the Department of Defense (DoD) are starting to use predictive analytics as an essential tool for suicide prevention efforts.

What Is Predictive Analytics?

Predictive analytics is a technique that uses big data to forecast the future. Today, there is so much data available, it can seem overwhelming at first. But with powerful computing and advanced algorithms, interesting patterns emerge. Thomas H. Davenport, an expert in predictive analytics, wrote in the Harvard Business Review, “All we have to do is gather the right data, do the right type of statistical model, and be careful of our assumptions.”

While an algorithm could never replace certain human traits, such as empathy, data models can help health care providers and agencies deploy their limited resources as effectively as possible. In particular, predictive analytics can identify the highest-risk cases for suicide so that every effort is made to save those lives before it is too late.

Suicide Prevention Efforts

In 2015, the Veterans Health Administration (VHA) published a study that evaluated whether predictive modeling could be used to identify patients at risk for suicide so that the highest-risk veterans could receive early intervention. The researchers used data mining techniques typically applied to consumer research for marketing and sales support. However, this time they analyzed VHA clinical records to identify groups of veterans who are at the highest risk to take their own lives. In the paper, the authors concluded, “Predictive modeling can identify high-risk patients who were not identified on clinical grounds.”

The researchers used data already available through the VA’s electronic health records collected as part of ongoing patient–provider encounters. The records include information, such as demographics, diagnoses and medications, and services received, such as inpatient mental health hospitalizations. All of this data can indicate a certain risk level for suicide.

The lead author of the paper, Dr. John F. McCarthy, Director for Serious Mental Illness Treatment Resource and Evaluation Center at the VA’s Office of Mental Health Operations, tells us, “This work didn’t involve anything extra for providers or patients. It took advantage of data that are collected on an ongoing basis, as part of delivering healthcare.”

Dr. Ira R. Katz, Senior Consultant for Mental Health Program Evaluation at the VA’s Office of Mental Health Operations, and one of the coauthors of the study, adds, “There’s nothing new about the risk factors that go into the model — mental health conditions, pain, chronic disease, hospitalizations, specific medications — they can each define risk.” Ultimately, he says, the model helps providers interpret the information in the health records.

The U.S. Army has a similar initiative, with its ongoing Study to Assess Risk and Resilience in Servicemembers (STARRS). In this study, researchers compiled existing data from 37 different army and DoD sources, which adds up to more than 1.1 billion records representing more than 1.6 million soldiers. They also included information about duration of service (for example, how many months a soldier was deployed in Afghanistan), and new data from questionnaires, neurocognitive tests and biomarkers that were identified from blood samples.

Outlook for the Future

By analyzing massive amounts of data, researchers are not only able to identify which patients were at the highest risk for suicide but they can even pinpoint when the risk is highest (within 26 weeks of being seen as an outpatient by mental health specialists, according to STARRS). According to the VA Suicide Prevention Program, “Veterans in the top 0.1 percent of risk (who have a 43-fold increased risk of death from suicide within a month) are identified before clinical signs of suicide are evident in order to save lives before a crisis occurs.” These highest-risk veterans then receive an especially high level of care, including additional follow-up visits and individualized care plans.

Predictive analytics could be used by any group looking to aid suicide prevention efforts, but military groups are especially well-suited for the task. Not only do they have access to sufficiently large amounts of data but taking care of soldiers and veterans is an especially urgent task. According to the latest report from the VA, the risk for suicide is 21 percent higher among veterans when compared with civilians.

Certainly, when it comes to such a deeply personal issue, the human factor will never be eliminated. However, just as automation has led to a revolution in manufacturing, it can also help mental health professionals accomplish more with their existing resources. Predictive analytics is an exciting new tool in behavioral science. This technique could save many lives by helping health care providers better understand trends in human behavior.

Predictive analytics can already help health care providers identify patients who are at high risk for suicide, and the more that the models are used, the better they will become. While a relative newcomer in the broad spectrum of health care, it is developing rapidly.

As Dr. Katz explains, “Moving forward, we intend to see how adding additional information from specific tests, from providers’ observations, etc. — can add to the predictive value of the model, so at the same time we’re using it to enhance care, we’re also using care to enhance the model.”