1355-P: Encounter Biases Using Electronic Health Records to Identify Type 2 Diabetes Incidence

by sugaradmin



Introduction and Objective: The use of electronic health records (EHR) to identify incident conditions in observational studies is growing. Health care encounters are related to health status and possibly other unobserved characteristics, which can lead to a phenomenon termed “encounter bias”. Several prior studies discussed potential encounter bias between encounter types and outcomes such as all-cause mortality and complications.Methods: We applied a well-established algorithm (SUPREME-DM) to the EHR data of Kaiser Permanente Southern California to identify type 2 diabetes (T2D) incidence in 2003-2019. We classified T2D incidence by encounter types: outpatient encounters only (OP), inpatient encounters only (IP), and hybrid with both IP and OP encounters (Hyd). We examined the potential encounter bias by encounter types in socio-demographics, biomarkers, and comorbidities at diabetes diagnosis as well as all-cause mortality following diagnosis.Results: We identified 323,920 incident cases of T2D (OP: 85.3%; IP: 4.5%; Hyd: 10.2%). Compared to individuals identified through IP encounters, individuals identified through OP encounters were younger (mean age: 56.9 vs. 63.9), had a higher mean BMI (32.9 vs. 29.6), and had a higher mean HbA1c (7.6 vs. 6.0). Differences between IP and Hyd groups were relatively small. However, those groups accounted for most of the mortality (30-day unadjusted rates: 9.2% and 9.9%; 2-yr unadjusted rates: 20.4% and 25.2%). Individuals identified though OP encounters had lower all-cause mortality rates (30-day unadjusted rates: 0.12%; 2-yr unadjusted rate: 2.13%), rates comparable to the general US adult population. Conclusion: We found important encounter biases in using a large EHR data warehouse to identify incidence of T2D: encounter types were associated with sizable differences in the follow-up outcome signals. Future studies using EHR as the central source to identify incidences of T2D need to take caution in encounter types of incident cases.

Disclosure

B. Han: None. X. Li: None. D.A. Cohen: None. E.L. Estrada: None. M. Habib: None. J.P. Martin: None. K. Reynolds: Research Support; Merck & Co., Inc. R. Sturm: None. H. Zhong: None. C. Nau: Research Support; National Institute of Diabetes and Digestive and Kidney Diseases.

Funding

National Institutes of Health (1R01DK132252-01)



Source link