CHARACTERIZING THE JOHNS HOPKINS INFLAMMATORY BOWEL DISEASE COHORT USING OMOP CDM

Abstract

Abstract BACKGROUND Inflammatory Bowel Disease (IBD) is an increasingly prevalent condition that results in weight loss, bloody diarrhea, and poor quality of life. Traditional population-based research on patients with IBD has involved use of relatively shallow claims data that is deficient in precise demographics. IBD research using richly detailed electronic health record (EHR) data has been difficult due to the need for laborious manual chart review to extract information, but new methods and technology allow for rapid EHR data extraction. We present here the results of the extraction of IBD patient data from our EHR into the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM). METHODS This study is an observational cohort study using patient health care data from our EHR extracted into the OMOP CDM. Transforming data into the standard CDM format allows for information from different databases and coding schema to be compared and aggregated. We used a de-identified dataset of all patients seen at Johns Hopkins Medicine since 2016. Our cohort of IBD, Crohn’s disease (CD), and Ulcerative Colitis (UC) patients were stratified by age, sex, race, weight, and pregnancy status. Within the cohorts, we obtained demographics, comorbidities, symptoms, labs, medications, and other data. We validated the extraction by comparison with EHR queries using Slicer/Dicer (SD). RESULTS There were 11,357 IBD patients including 6,512 CD patients and 4,224 UC patients. There were more female patients for both CD and UC (56.4% v 57.5%) and more white patients for both CD and UC (55.4% v 56.7%) compared to black (23.5% v 22.6%) and Asian patients (5.6% v 4.9%). CD and UC patients were treated with systemic steroids (49.1% v 51.5%), 5-ASA (24.9% v 47.2%) anti-TNF inhibitors (24.0% v 12%), IL12/23 inhibitors (8.4% v 2.7%), and JAK inhibitors (0.3% v 1.8%). Using SD queries, the percent differences were determined between the OMOP and EHR data, revealing a 0.54% difference in the number of CD patients and 10.36% difference in the number of UC patients (Table 1). CONCLUSION Using OMOP CDM to analyze clinical characteristics and outcomes for patients with IBD is a valuable method for obtaining de-identified patient data, reducing patient risk, and facilitating rapid creation of on-demand cohorts for multi-institutional research studies. This advantage is balanced by the lack of access to row-level data to validate cohorts. Local validation of data extraction through SD queries can allow for rapid creation of cohorts for big data research. The most significant differences between SD queries and OMOP CDM was in the calculation of prevalence of medications and race, suggesting that differing data extraction coding methods could result in queries that underestimate rates of treatment due to proprietary local coding schema.