Introduction: Medical readiness continues to be a significant concern for the military. DoD policy directs medical authorities to refer service members to the Disability Evaluation System (DES) when the course of further recovery is relatively predictable or within 1 year of diagnosis, whichever is sooner. The Medical Evaluation Readiness Information Toolset (MERIT) is an application that leverages artificial intelligence within a clinical decision support tool to provide clinicians with predictions of a service member's likelihood of referral to the DES for the top 24 medical conditions that result in separation from the service, which represent more than 90% of all referral cases to the DES since 2000.
Materials and methods: Data spanned 19 years and contained records for over 3 million army service members. The MERIT team incorporated a novel approach using a Gamma window function to weight recent medical data more than older medical data in the creation of a "Disease Severity Index" (DSI) that summarized the progression of a health deterioration process per medical condition code. Time-dependent medical encounter data were aggregated into an individual-level DSI. The identified features including the DSI were used in logistic regression and random forest models to predict whether a service member is likely to be referred to the DES. Models were constructed for each of the top 24 unfitting medical conditions.
Results: MERIT produced a set of high-performing classification models with area under the receiver operating characteristics curves across all conditions exceeding 0.919 using logistic regression for all conditions.
Conclusions: This project demonstrated with a high degree of accuracy that MERIT, using a combination of ICD codes and personnel records, can be used to develop an individual risk profile for each service member.
© The Association of Military Surgeons of the United States 2023. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com.