Predictive models for diabetes patients in medicaid

Predictive modeling can be used to identify persons who are at increased risk for adverse health outcomes. We used demographic, medical, and pharmacy claims data to create a gender-specific model for fee-for-service Medicaid based on 2 states' data that can assist with the identification of persons with an elevated future risk of hospitalization, elevated claims expense, or death. Depending on age and the outcome of interest, the area under the receiver operating characteristic curve for this predictive modeling tool across 2 states' diabetes populations ranged from 0.608 to 0.834. We conclude that this analysis yielded a level of accuracy comparable to other predictive models that can be used to target patient enrollment in population-based care management.