Abstract
OBJECTIVE: To evaluate whether the Area Deprivation Index (ADI) contributes to predicting type 2 diabetes development in youth with prediabetes compared with a machine learning (ML) model built with other data elements. RESEARCH DESIGN AND METHODS: Patient encounters (n = 665) from an electronic medical record were used to build supervised ML models to predict type 2 diabetes development within 1 year of prediabetes diagnosis. The ADI was constructed using patients' census block data. Two models, trained on data with and without ADI, were built. The model selection resulted in logistic regressions with 1) HbA1c only and 2) HbA1c + ADI as the best models from each data set. RESULTS: A total of 181 patient encounters (27.2%) developed type 2 diabetes. The area under the receiver operating characteristic curve of the HbA1c-only model was 0.68 and of the HbA1c + ADI model, 0.73. CONCLUSIONS: The addition of ADI to the model resulted in the best performance in predicting youth-onset type 2 diabetes development.