# Validating an Electronic Health Record Algorithm for Diabetes Screening Eligibility in the Emergency Department

## TL;DR

This study validated an electronic health record algorithm to identify emergency department patients eligible for diabetes screening based on ADA guidelines.

## Contribution

The study provides empirical validation of a diabetes screening algorithm for emergency department settings.

## Key findings

- The algorithm had acceptable sensitivity (0.69) and specificity (0.91) for identifying eligible patients.
- The positive predictive value was 0.75, and the negative predictive value was 0.88.
- The AUC of 0.74 suggests the algorithm has acceptable accuracy for diabetes screening in the ED.

## Abstract

While the American Diabetes Association (ADA) screening guidelines have been used widely, the way they are implemented and adapted to a particular setting can impact their practical application and usage. Our primary objective was to validate a best practice advisory (BPA) screening algorithm informed by the ADA guidelines to identify patients eligible for hemoglobin a1c (HbA1c) testing in the emergency department (ED).

This cross-sectional study included adults presenting to a large urban medical center’s ED in May 2021. We used sensitivity, specificity, likelihood ratios, and predictive values to estimate the algorithm’s ability to correctly identify patients eligible for diabetes screening, with manual chart review as the reference standard. Eligibility criteria targeted patients at risk for diabetes who were likely unaware of their elevated HbA1c. We also calculated the area under the receiver operating characteristic curve (AUC).

In May 2021, 2,963 (77%) of the 3,850 adults admitted to the ED had a routine lab ordered. Among those, 796 (27%) had a BPA triggered, and of those 631 (79%) had an HbA1c test completed. The algorithm had acceptable sensitivity (0.69, 95% confidence interval [CI] 0.66–0.72), specificity (0.91, CI 0.89–0.92), positive predictive value (0.75, CI 0.72–0.78) and negative predictive value (0.88, CI 0.86–0.89). The positive likelihood ratio (7.39, CI 6.35–8.42) was adequate, and the negative likelihood ratio (0.34, CI 0.30–0.37) was informative. The AUC of 0.74 (CI 0.72–0.77) suggests that the algorithm had acceptable accuracy.

Findings suggest that an electronic health record-based algorithm informed by the ADA guidelines is a valid tool for identifying patients presenting to the ED who are eligible for HbA1c testing and may be unaware of having prediabetes or diabetes. The ease of workflow integration and high yield of potentially undiagnosed diabetes and prediabetes makes the BPA algorithm an appealing method for diabetes screening within the ED.