Background/Purpose: Research using electronic health records (EHR) may offer advantages over traditional observational studies, including lower costs and greater generalizability to a broader patient population; however, EHR data may be more biased as a result of the high prevalence of missing data. Little research is available that directly evaluates the potential benefits and weaknesses of each study design within a single population. We examined differences in baseline demographics and disease outcomes between RA patients enrolled in a cohort study with more standardized data collection and patients whose data comes purely from the EHR within the same health system. We also compared the availability of measures and prevalence of missing data between the two groups and explored differences in longitudinal predictors of RA disease activity.

Methods: We included individuals with an RA diagnosis (ICD-9 code 714.0) and at least 2 rheumatology clinic visits within 12 months between 2013-2017 from the EHR of a public hospital (n=377 patients, n=2,269 visits). Approximately half were also enrolled in an RA cohort study. In order to examine if longitudinal differences in disease activity were present between cohort and non-cohort groups, mixed effects models were used to evaluate the association between sex, race/ethnicity, age, body mass index (BMI), smoking status and medication on Clinical Disease Activity Index (CDAI) score. Interaction between covariates and participant status (cohort vs. non-cohort) was also assessed.

Results: No significant baseline differences between cohort (n=187) and non-cohort (n=190) participants were found with respect to sex, age, race/ethnicity, smoking status, or disease activity measures (Table). Variables with a higher prevalence of missing data in non-cohort individuals compared to cohort individuals included language (14% vs. 0%), BMI (14% vs. 4%), smoking status (18% vs. 6%), and certain disease activity measures (21-22% vs. 3-6%). Black, non-Hispanic race/ethnicity was associated with a higher CDAI score over the study period compared to white, non-Hispanic individuals in non-cohort participants, while no association was found in cohort participants (p-interaction = 0.07).

Conclusion: Non-cohort participants from the EHR were comparable to a research cohort drawn from the same health system across some variables, but demonstrated more severe disease trajectories in racial/ethnic minorities. While challenges remain given the prevalence of missing data for specific variables in the EHR, utilizing EHR data repositories may inform our understanding of disease trajectories for RA patients who are not adequately captured in research cohorts.

Table 1. Demographic and disease characteristics of RA cohort and non-cohort participants from the EHR of a public hospital in California, 2013-2017.

Table values represent: N (%) or Mean (SD)