Predicting Suicides Among US Army Soldiers After Leaving Active Service

Area of Science:

• Military health
• Public health
• Data science

Background:

• Military suicide rates significantly increase post-service.
• Early identification of high-risk service members is crucial for intervention.

Purpose of the Study:

• Develop a predictive model for suicide risk in US Army soldiers transitioning to civilian life.
• Utilize administrative data for predicting suicides 1-120 months post-service.

Main Methods:

• Prognostic study using a consolidated administrative database of US Army soldiers (2010-2019).
• Machine learning models (Lasso, Super Learner) trained to predict suicide.
• Validation performed on a 30% hold-out sample.

Main Results:

• A Lasso model accurately predicted suicides up to 10 years post-service.
• Key predictors included sociodemographics, career characteristics, and psychopathology.
• The top 10% highest-risk soldiers accounted for 30-46% of all suicides.

Conclusions:

• Administrative data can effectively predict long-term suicide risk in transitioning soldiers.
• Model-informed prevention strategies show positive net benefit.
• Further research needed on intervention cost-effectiveness.