1413-P: Youth-Onset Diabetes—Who Is Missed by EHR-Based Surveillance

Background: Efficient population-level surveillance of youth-onset diabetes is essential for public health planning. Early surveillance approaches predated the use of electronic health records (EHR) and required manual chart reviews of all potential cases. Approaches utilizing EHR data present the potential for significant gains in cost efficiency, but the degree of precision remains unclear.

Objective: Identify characteristics of diabetes patients <20 years in Colorado who were missed or misclassified by an EHR algorithm compared with patients identified by a traditional approach including manual chart review.

Methods: We compared incident diabetes patients diagnosed in 2014-2018 in Colorado by SEARCH for Diabetes in Youth and ascertained using manual chart review to determine diabetes status, type, and date of diagnosis to those identified by date of first diabetes diagnosis per EHR data extracts. Both approaches used similar and multiple case sources (networks of providers and hospitals) designed to cover all of Colorado.

Results: SEARCH ascertained 2206 patients with incident diabetes in 2014-2018. The EHR-based algorithm correctly identified incident diabetes status for 81% with 8.7% not identified as a diabetes case, and 10.3% assigned the wrong incident year. Patients missed or misclassified by the algorithm were more likely to have type 2 diabetes (36.1% incorrect or missing) or other diabetes types (36.4% incorrect or missing), to identify as Hispanic of any race (24.0% incorrect or missing) or other race (45.5% incorrect or missing) or were first identified by SEARCH at sources that did not use EHRs (75.8% incorrect or missing).

Conclusion: Algorithmic misclassification from EHR system data using date of first diagnosis may be more profound for minority race/ethnic groups and those without type 1 diabetes. Improved understanding of who is missed or misclassified by such approaches may assist with bias correction efforts to provide valid, generalizable estimates of the diagnosed diabetes population.