Title: Identification of Factors Associated with Agreement Between Rheumatoid Arthritis Electronic Health Record (EHR) Medication Lists and Prescribed Disease Modifying Treatment Plans
Authors: Ilinca D. Metes1, Heather Eng2, June Feng2, Balasubramani K. Goundappa2, Stephen R. Wisniewski2 and Marc C. Levesque1
University of Pittsburgh, Pittsburgh, PA, School of Medicine, Department of Medicine, Division of Rheumatology and Clinical Immunology1, School of Public Health, Department of Epidemiology2
Background/Purpose: Federal government programs incentivize the use of Electronic Health Records (EHR) including accurate medication lists. The accuracy of a patient’s EHR medication list is essential for understanding the efficacy of treatment plans and reducing adverse events. However, studies indicate that many EHR medication lists are inaccurate. Therefore, it was our aim to determine the level of agreement and to determine the factors associated with agreement between a RA patient’s EHR medication list and their disease modifying treatment plan.
Methods: Subjects (n = 1,029) were drawn from the University of Pittsburgh RA Comparative Effectiveness Research (RACER) registry (2010-13). For each outpatient visit (n = 5,236) a “Gold Standard” medication list (encompassing oral disease modifying anti-rheumatic disease modifying drugs (DMARDS), biologics, and corticosteroids) was created using retrospective (n = 4,251) and prospective (n = 985) data collected through chart review of physician notes and/or patient interviews. The “Gold Standard” was compared to the EHR medication list and the measure of agreement was determined using Cohen’s kappa coefficient. Patient, physician and clinic characteristics were compared between visits where medications matched vs visits where medications did not match using chi square for categorical variables and t-tests or Wilcoxon signed rank tests for continuous measures. To evaluate the characteristics associated with drug matches, a longitudinal regression using a generalized estimating equation (GEE) model was used to account for repeated observations within a patient, after controlling for baseline significant factors.
Results: Comparison of the “Gold Standard” and EHR medication lists resulted in 8,124 true positive, 95,984 true negative, 638 false positive, and 1,829 false negative medication matches. Retrospective (kappa = 0.86) and prospective (kappa = 0.85) data was combined to give an overall kappa = 0.86, with oral DMARDS having a kappa = 0.89, biologics a kappa = 0.85, and corticosteroids a kappa = 0.78. Based on the GEE model, the odds ratio (OR; 95% CI) of oral DMARD, biologic and corticosteroid medication matches was decreased by having a longer prescription list (0.94; 0.93-0.96), a high % decrease in medications compared to the last visit (0.97; 0.97-0.98), having high RA disease activity by the RAPID3 (0.38; 0.29-0.51), CDAI (0.32; 0.25-0.42), or DAS28 (0.76; 0.72-0.81), and having more comorbidities as measured by the Charlson (0.90; 0.83-0.97).
Conclusions: For RA patients in a usual care setting, there was a high level of agreement between their disease modifying drug treatment plan and their EHR medication list, with the level of agreement differing slightly between oral DMARDs, biologics and corticosteroids. A longer medication list, a high % decrease in overall medications since the last visit, being in high disease activity by the RAPID3, CDAI, or DAS28, and having more comorbidities as measured by the Charlson increased the likelihood of drug discrepancies. Using these results, our future research will focus on reducing discrepancies in the EHR medication list by implementing targeted interventions.