Predicting Inadequate Early Glycemic Control among Adults with Younger-Onset Type 2 Diabetes

The prevalence of type 2 diabetes (T2D) is increasing among individuals younger than 45 years, who are at an increased risk of T2D-related complications, making the lasting benefits conferred by early glycemic control particularly critical. The ability to proactively identify these younger-onset individuals at increased risk of inadequate early glycemic control could support more targeted and tailored initial care. The study sample was randomly split into training (70%) and validation (30%) sets to develop and test the model. We used logistic LASSO (Least Absolute Shrinkage and Selection Operator) regression to identify predictors of inadequate glycemic control (HbA1c ≥8%) at 1-year following diagnosis among adults ages 21-44. Potential predictors were assessed at diagnosis. We used terciles of predicted risk to stratify our cohort into low (LR), intermediate (IR) and high-risk (HR) groups and compared characteristics between them using chi-square and ANOVA tests. Of the 10,897 individuals with new T2D, the mean age was 37.9 ± 5.3 years and 33% were Latinx. The LASSO model selected 46 out of 97 candidate variables. The resulting model yielded a c-statistic of 0.73 and had good calibration. We observed differences in socioeconomic, clinical, and care utilization characteristics between the risk groups. Those in high-risk group were more likely to be Latinx (45% HR vs. 30% IR vs. 27% LR, p<0.01), residents of deprived neighborhoods (33% HR vs. 23% IR vs. 22% LR, p<0.01), and have an HbA1c ≥ 10% ( 74% HR vs. 9% IR vs. 0% LR, p<0.01) but less likely to have received a flu shot ( 22% HR vs. 37% IR vs. 42% LR, p<0.01) or be active users of the online patient portal (44% HR vs. 65% IR vs. 75% LR, p<0.01). Using a machine-learning developed predictive model, we characterized adults <45 years at low, intermediate or high risk for inadequate early glycemic control. Such models can support healthcare systems by leveraging EHR data to risk stratify younger-onset individuals and inform the intensity of initial T2D care. Disclosure P. Mishra: None. S. Alexeeff: None. R. W. Grant: None. A. Gopalan: None.