DEVELOPMENT AND VALIDATION OF AN ALGORITHM TO IDENTIFY ASPIRIN USE FROM ELECTRONIC MEDICAL RECORDS AND CLAIMS DATA
Author(s)
Valerie Haley, PhD1, Michael Head, MS1, Sravanthy Myneni, B.Tech1, Shiva Vojjala, MS1, Emily Durden, PhD2, Steven Caproni, PharmD2, Lynda D. Lisabeth, PhD3, Seemant Chaturvedi, MD4, Kaitlyn Hopkins, BS1, Vincent J. Willey, PharmD1.
1Carelon Research, Wilmington, DE, USA, 2Bayer Healthcare Pharmaceuticals, Whippany, NJ, USA, 3University of Michigan, Ann Arbor, MI, USA, 4University of Maryland School of Medicine, Baltimore, MD, USA.
1Carelon Research, Wilmington, DE, USA, 2Bayer Healthcare Pharmaceuticals, Whippany, NJ, USA, 3University of Michigan, Ann Arbor, MI, USA, 4University of Maryland School of Medicine, Baltimore, MD, USA.
OBJECTIVES: Because millions of adults take low-dose aspirin to prevent cardiovascular disease (CVD), accurately measuring aspirin exposure is important in CVD research. We developed and validated an algorithm to identify aspirin use from electronic health records (EHR) and pharmacy claims data.
METHODS: We studied 9,480 U.S. adults with a hospitalization or emergency department visit for transient ischemic attack or non-cardioembolic ischemic stroke between 2016 and 2024 who had EHR data in the Healthcare Integrated Research Database (HIRD®). Within a window from stroke onset to 90 days post-discharge, we identified aspirin exposure by conducting keyword searches for aspirin use and allergy in unstructured EHR text and by querying encounter, start, and stop dates in structured EHR medication data. We validated the EHR algorithm against manual EHR review in a stratified random sample of 100 patients. Pharmacy claims for aspirin and other antiplatelets were integrated to estimate the prevalence of antiplatelet use.
RESULTS: The algorithm had a positive predictive value of 98.5% (91.7 to 100%) and negative predictive value of 94.3% (80.8 to 99.3%). The prevalence of aspirin use was 21.8% based on pharmacy claims, 76.9% based on EHR, and 78.7% using both EHR and claims. Specific start and stop dates were frequently unavailable. No antiplatelet therapy was found for 13.8% of patients, while single, dual, and 3+ antiplatelet therapies were used for 51.7%, 33.8%, and 0.7% of patients respectively.
CONCLUSIONS: Aspirin use was accurately classified using documentation found in the EHRs. However, underestimation of exposure is possible if clinicians associated with the available EHRs were unaware or didn’t document aspirin use. Integration of EHR structured medication lists and unstructured notes with administrative claims data is an acceptable method to ascertain evidence of exposure to aspirin and other over-the-counter medications commonly used for treatment and provides a more complete treatment history, facilitating clinically relevant analyses.
METHODS: We studied 9,480 U.S. adults with a hospitalization or emergency department visit for transient ischemic attack or non-cardioembolic ischemic stroke between 2016 and 2024 who had EHR data in the Healthcare Integrated Research Database (HIRD®). Within a window from stroke onset to 90 days post-discharge, we identified aspirin exposure by conducting keyword searches for aspirin use and allergy in unstructured EHR text and by querying encounter, start, and stop dates in structured EHR medication data. We validated the EHR algorithm against manual EHR review in a stratified random sample of 100 patients. Pharmacy claims for aspirin and other antiplatelets were integrated to estimate the prevalence of antiplatelet use.
RESULTS: The algorithm had a positive predictive value of 98.5% (91.7 to 100%) and negative predictive value of 94.3% (80.8 to 99.3%). The prevalence of aspirin use was 21.8% based on pharmacy claims, 76.9% based on EHR, and 78.7% using both EHR and claims. Specific start and stop dates were frequently unavailable. No antiplatelet therapy was found for 13.8% of patients, while single, dual, and 3+ antiplatelet therapies were used for 51.7%, 33.8%, and 0.7% of patients respectively.
CONCLUSIONS: Aspirin use was accurately classified using documentation found in the EHRs. However, underestimation of exposure is possible if clinicians associated with the available EHRs were unaware or didn’t document aspirin use. Integration of EHR structured medication lists and unstructured notes with administrative claims data is an acceptable method to ascertain evidence of exposure to aspirin and other over-the-counter medications commonly used for treatment and provides a more complete treatment history, facilitating clinically relevant analyses.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
RWD76
Topic
Real World Data & Information Systems
Topic Subcategory
Health & Insurance Records Systems
Disease
SDC: Cardiovascular Disorders (including MI, Stroke, Circulatory)