(208) Misclassification of Metformin and Insulin Exposure Among Individuals with Type 2 Diabetes Using Integrated Electronic Health Record and Pharmacy Claims Data

Publication Number: 1200

Background: Studies utilizing electronic health record (EHR) databases often use order date as a proxy for when a patient was exposed to the effects of a medication under investigation. As such, medication exposure in EHR studies may be misclassified due to lags from order to fill date. The extent of this misclassification is poorly characterized.

Objectives: We aimed to quantify potential misclassification of medication exposure in EHR studies using linked EHR and pharmacy claims data. To do so, we used new prescriptions for metformin and insulin among an incident cohort of individuals with type 2 diabetes as an illustrative example.

Methods: Using Optum Integrated Claims EHR from 2010-2021, we identified individuals with integrated EHR and claims data using the “integrated” variable. Among these individuals, we identified those with incident type 2 diabetes through first-time prescription orders or fills for metformin or insulin after age 40. Individuals with prescription orders or fills within 365 days of enrollment were excluded to avoid prevalent users. To quantify temporal lags in medication exposure, we calculated the number of days elapsed from first order to first fill among individuals who filled prescriptions. To quantify the extent that non-adherence to prescription orders contributes to lags in medication exposure, we also calculated the number of orders placed prior to first fill.

Results: We identified 1,280,045 individuals with incident type 2 diabetes. 1,068,824 (83.5%) of these were incident users of metformin. 75,339 (7.0%) were excluded for a recorded fill before their first metformin order. Among the 235,196 individuals who had a metformin order with subsequent fill, the median number of days to fill was 80 (25th percentile: 0 days, 46th percentile: 30 days, 51st percentile: 90 days). The median number of metformin orders before first fill was 2 (IQR: 1 – 6). In the insulin analysis, 468,405 (36.6%) individuals were incident users. 37,614 (8.0%) were excluded for a recorded fill before their first insulin order. Among the 98,846 individuals who had an insulin order with subsequent fill, the median number of days to fill was 159 (25th percentile: 0 days, 40th percentile: 30 days, 46th percentile: 90 days). The median number of insulin orders before first fill was 4 (IQR: 1 – 11).

Conclusions: This study demonstrates that studies utilizing only EHR medication orders to ascertain outpatient medication use are subject to substantial misclassification by delays from prescription order to fill. Additionally, for many individuals, orders are placed several times before medications are finally filled. We recommend that studies using only EHR medications orders implement sensitivity analyses to assess whether delays in medication exposure alter conclusions.