Service Retention Among Coast Guard Members Seeking Behavioral Healthcare

Introduction Behavioral health conditions (BHC) can reduce service member retention. This analysis sought to identify demographic and diagnostic factors among BHC care-seeking Active-Duty United States Coast Guard (ADCG) that were predictive of discharge before completion of obligated service. Methods A four-year retrospective cohort study of ADCG personnel was conducted. Five machine-learning (ML) algorithms and logistic regression were applied to data on ADCG who sought outpatient care for BHC in 2016. Covariates examined as possible mediators of early service termination included diagnosis group, gender, rank grouping, and race. Results Only 26.4 of every 1,000 members who sought BHC care did not complete their service obligation. Diagnosis group did not predict early service termination, whereas senior enlisted rank was associated with early termination. The ML algorithms best predictive of early discharge from service were bagging classifier and decision tree classifier. Logistic regression performed as well as the two leading algorithms. Conclusions Specific ML models can be used to identify personnel groups at risk for early separation, such as senior enlisted personnel. Traditional epidemiologic methods demonstrate value in predicting service member separation.