1277-P: Prediction of HbA1c Test Eligibility in Emergency Department

Abstract

Introduction: Diabetes screening using best practice advisory (BPA) alerts integrated into the emergency department (ED) workflow may efficiently identify at-risk patients disproportionally affected by diabetes. The objectives of this study were to 1) validate a BPA algorithm informed by the American Diabetes Association (ADA) screening criteria using elements from the electronic medical record (EMR) to identify eligible patients for hemoglobin A1c (HbA1c) testing, and 2) compare the characteristics of the ADA-based algorithm with one informed by the United States Preventative Services Task Force (USPSTF) recommendations. Methods: This cross-sectional study involved adults admitted to the University of Illinois Health ED in May 2021 with labs ordered (n=2963). Sensitivity, specificity, positive predictive value (PPV), and negative predictive values (NPV) estimated the BPA algorithm’s ability to correctly identify patients eligible for diabetes screening using manual EMR chart review as the reference standard. Receiver operator characteristic (ROC) curves and area under the curve (AUC) were calculated to compare ADA and USPSTF-informed algorithms. Results: The sensitivity, specificity, PPV, and NPV of the ADA model were 0.70 (95% CI: 0.66-0.75), 0.91 (0.90-0.92), 0.76 (0.73-0.79), and 0.88 (0.86-0.89), respectively. Compared to the ADA model, the USPSTF model’s sensitivity was worse (0.25 (0.22-0.27)) and specificity was slightly higher (0.94 (0.93-0.95)), though PPV remained similar (0.75 (0.71-0.79)) and NPV was lower (0.64 (0.62-0.66)). The ROC AUC for the ADA and USPSTF-based models were 0.91 (0.89-0.93) and 0.86 (0.84-0.89), respectively. Conclusion: Despite moderate to low sensitivity, the PPVs and large AUCs are acceptable due to the high prevalence of at-risk populations presenting to the ED. Findings suggest that a BPA algorithm informed by the ADA guideline is an adequate tool to identify eligible patients for diabetes screening in the ED, especially given its ease of workflow integration. Disclosure M.Smart: None. J.Lin: Research Support; Novo Nordisk, Xeris Pharmaceuticals, Inc. T.Lee: None. B.T.Layden: Consultant; Bayer Inc. Y.Eisenberg: None. K.K.Danielson: None. A.Kong: None. Funding Novo Nordisk (G2111)