Military service members experience a sharp increase in suicide risk after returning to civilian life. Being able to Identify service members at high risk for suicide — before they leave service — could potentially help target preventive interventions.
In 2024, the Veterans Administration and Department of Defense issued the Clinical Practice Guideline for Assessment and Management of Patients at Risk for Suicide calls for the development and validation of “AI algorithms for predicting actionable suicide risk (person-level)…[and] corresponding clinical tools to ensure that algorithms can be used to inform decision making at the point of care.” A recent study led by Chris J. Kennedy, PhD and colleagues from the Center for Precision Psychiatry at Mass General Hospital, have demonstrated that artificial intelligence (AI) can be used to predict suicide risk among U.S. Army veterans with moderate to good accuracy using data available before they leave active service.
Study Overview
For this study, a consolidated administrative database was created including all regular US Army soldiers who left the service between 2010 and 2019. Suicides were identified in this cohort using the National Death Index.
Machine learning models were trained to predict suicides over the next 1 to 120 months using a random subset (70%) of the cohort. Validation of the model was implemented using the remaining 30% of the sample. Predictors used in this study were derived from administrative records available before leaving military service and included sociodemographic variables, Army career characteristics, mental health risk factors, indicators of physical health, information on social networks and supports, and relevant psychosocial stressors.
In this group of 800,579 soldiers (84.9% male; median age at discharge, 26 years), 2084 suicides were identified (51.6 per 100 000 person-years).
Key Findings
The AI model’s predictive accuracy ranged from excellent (area under the curve of 0.87) for suicides occurring in the first month after leaving service to good (0.72) for suicides occurring in the 10 years after leaving the service. Individuals in the top 10% of risk scores accounted for 30.7% to 46.6% of all suicides across different time horizons. Individuals in the top 25% of risk scores accounted for 55.5% of all suicides occurring across the 10-year horizon.
By far, the most robust predictor of risk was being male. However, none of these factors on their own can meaningfully predict suicide. More sophisticated modeling using AI allows researchers to evaluate more complex combinations of risk factors. The study identified several key factors associated with increased suicide risk:
- Sociodemographic variables: Male gender, non-Hispanic White ethnicity, younger age
- Army career characteristics: Combat-related duties, less than 20 years of service, non-honorable discharge
- Mental health factors: Alcohol-related outpatient visits, inpatient hospitalization for psychiatric illness, disorder inpatient admissions, suicidal ideation while in service
Implications for Suicide Prevention
The current study from Kennedy and colleagues demonstrates that a model based on administrative data available in the medical record at the time of leaving active Army service has good accuracy in predicting suicides over the subsequent ten years. Ultimately this research could significantly impact how the military approaches mental health support for transitioning service members. Using machine learning to estimate suicide risk could potentially facilitate quick decision-making regarding suicide risk, inform disposition from high-acuity settings (e.g., inpatient psychiatry and emergency departments), and recognize which service members may need enhanced care during their transition out of the service.
Furthermore, the model’s ability to predict suicide risk before soldiers leave active service, using information that is available in the medical record, opens up possibilities for targeted suicide prevention interventions for high-risk individuals. However, the researchers emphasize that further studies are needed to develop interventions based on the model’s predictions and to determine the cost-effectiveness of implementing these interventions.
Co-author Santiago Papini, PhD, states, “By identifying at-risk individuals early, we may be able to provide more targeted and timely interventions, potentially saving lives”.
In May 2018, the Department of Veterans Affairs (VA) began a multi-year effort to implement the new Federal electronic health record (EHR). The Federal EHR will ultimately simplify the experience for Veterans and their health care teams, and the VA is actively working on integrating this type of predictive modeling into their healthcare system. This innovative approach could potentially revolutionize how the military addresses mental health support for service members transitioning to civilian life.
As the Honorable Lloyd J. Austin III, U.S. Secretary of Defense, stated, “Suicide prevention is a long-term effort. Change will not happen overnight, but we have no time to spare”. This AI model represents a significant step forward in ongoing efforts to protect the mental health of our service members and veterans.
Read More
Kennedy CJ, Kearns JC, Geraci JC, Gildea SM, Hwang IH, King AJ, Liu H, Luedtke A, Marx BP, Papini S, Petukhova MV, Sampson NA, Smoller JW, Wolock CJ, Zainal NH, Stein MB, Ursano RJ, Wagner JR, Kessler RC. Predicting Suicides Among US Army Soldiers After Leaving Active Service. JAMA Psychiatry. 2024 Dec 1; 81(12):1215-1224.
In the News
New AI model helps predict veteran suicide risk (Medical Xpress)
