Validation of an Automated Data Collection Method of Inflammatory Bowel Disease Medications in the National Veterans Affairs Healthcare System

Background: The national Veterans Health Administrative (VHA) database contains clinical data of about 21 million United States (US) veterans and approximately 60,000 living veterans with inflammatory bowel disease (IBD). Validating the accuracy of pharmacy records to detect the use of specific IBD medications will help conduct epidemiological studies using this large dataset of patients. We aimed to determine accuracy of detecting IBD medications through pharmacy records and compare it with manual chart review (MCR) in the VHA database, which has not been performed before. Methods: We identified IBD patients between October 1, 2015, and November 30, 2021, from the VHA National Corporate Data Warehouse (CDW) by ICD-10-CM diagnosis codes [(K51 ulcerative colitis (UC) or K50 Crohn’s disease (CD)] based on a previously validated methodology for IBD cases. Cases were identified as UC or CD if there were two encounters including one outpatient encounter with the same diagnosis. Cases that did not meet the criteria were categorized as IBD unclassified (IBDU). Medication use was defined as one or more prescription fills for IBD medication based on VA pharmacy records by automated data collection (ADC). Medication data for specific IBD medications (mesalamine, sulfasalazine, balsalazide, azathioprine, mercaptopurine, methotrexate, cyclosporine, adalimumab, infliximab, golimumab, certolizumab, vedolizumab, tofacitinib, upadacitinib, ozanimod, ustekinumab, risankizumab) was extracted by prescription fill class and CPT codes. MCR was conducted for a random subset of 250 patients by 3 independent reviewers. To increase the strength of our validation methodology each medication was identified by at least 2 reviewers making it a set of 6 concordance pairs. Concordance proportions are reported for medications. Results: We identified 57,823 IBD patients with one or more prescriptions for IBD medication and randomly selected 250 patients, 134 UC (53.6%), 85 CD (34%) and 31 IBDU (12.4%) from VA facilities throughout the country. The average concordance proportion between prescription records by ADC and MCR was 82.6% (SD 29.8). The average concordance among all 6 reviewer pairs was 77.2% (SD 35.5). Concordance based on individual drug and IBD diagnosis will be conducted. Conclusions: In this first study to systematically assess the accuracy of pharmacy records to detect the use of IBD medications in the national VHA dataset, we were able to detect IBD medications with 83% accuracy. Having a validated and accurate method to detect the use of IBD medications will be essential to study IBD in one of the largest national cohorts of IBD patients.