Developing and Validating Models to Predict Suboptimal Early Glycemic Control Among Individuals With Younger Onset Type 2 Diabetes

Objective: Younger age at the time of type 2 diabetes onset increases individuals’ future complication risk. Proactively identifying younger-onset individuals at increased risk of not achieving early glycemic goals can support targeted initial care. Design and Methods: Individuals (ages 21–44) newly diagnosed with type 2 diabetes were identified and randomly assigned to training (70%) and validation (30%) datasets. Least absolute shrinkage and selection operator regression models were specified to identify key predictors (assessed at diagnosis) of suboptimal glycemic control (HbA1c≥8%) within 1 year after diagnosis using the training dataset. The full model included 48 candidate predictors. We also developed additional more streamlined models using more widely available predictors (transferable model), a smaller number of available predictors (simplified transferable model), and a bivariate model with HbA1c as the sole predictor (HbA1c-only model). Model-based predicted risk scores were used to stratify individuals in the validation dataset. Results: The cohort included 10,879 individuals. All of the models, including the HbA1c-only model, performed comparably. All had good discrimination (C-statistics ranging from 0.71 to 0.73) in the validation dataset. Conclusions: When predicting the risk of not achieving glycemic goals, the HbA1c-only model had comparable performance to the more complex prediction models. This simple risk stratification requires no computation and could be implemented simply by looking at the diagnosis HbA1c value. This practical approach can be used to identify newly diagnosed younger adults who may need extra attention during the critical early period after diagnosis.