Leveraging Machine Learning to Predict Mental Health Referral Follow-Up Among US Military Personnel

Area of Science:

• Military Health
• Machine Learning in Healthcare
• Mental Health Services Research

Background:

• Noncompliance with mental health referrals is a significant barrier to care for U.S. military personnel.
• Operational deployments and military stressors exacerbate mental health challenges, impacting treatment access and increasing costs.
• Identifying service members at risk for noncompliance can enable targeted interventions to improve care.

Purpose of the Study:

• To develop machine learning (ML) models to predict noncompliance with mental health referrals.
• To identify key predictors of referral noncompliance among active-duty military personnel.

Main Methods:

• Retrospective data analysis of 14,289 active-duty personnel referred for mental health care via Periodic Health Assessment (PHA) from 2016-2020.
• Predictive models were developed using demographics, health screenings, medical history, and prior healthcare utilization.
• Noncompliance was defined as failure to follow through with referrals within 90 days.

Main Results:

• 34.0% of the study sample were noncompliant with mental health referrals.
• Extreme gradient boosting (XGBoost) models achieved high performance (AUC ≈ 0.80).
• Prior healthcare utilization (previous visits, mental health diagnoses), alcohol screening, and age were the strongest predictors of noncompliance.

Conclusions:

• Machine learning models show significant potential for identifying military personnel at risk of noncompliance with mental health referrals.
• These models can support targeted interventions to improve mental health care follow-up rates.
• Future research should focus on validating these models and exploring the underlying mechanisms of noncompliance.