FRI0194 Identifying Patients with Rheumatoid Arthritis in Primary Care Electronic Medical Records

Background Rheumatology research in primary care populations has been hampered by an inability to efficiently identify rheumatology patients. Electronic medical records (EMRs) are a rich data source that can be used for both research and quality improvement. However methods to accurately identify patients with target diseases need to be developed. Objectives To determine whether rheumatoid arthritis (RA) patients can be accurately identified within primary care EMR data. Methods We performed a retrospective chart abstraction study using a random sample of 9500 adult patients from a total of 73,014 patients within the Electronic Medical Record Administrative data Linked Database (EMRALD) in Ontario, Canada (representing 83 family physicians). We first identified “confirmed” RA patients by manually reviewing each patient's EMR (our reference standard). We developed a free-text searching algorithm by applying a clinically derived list of terms to search the free text in the problem list and past medical history of the patient profile. The algorithm was refined iteratively by adding any new RA-related terms (idiosyncratic descriptors) discovered. We then applied computer searches of various combinations of structured and semi-structured EMR fields to also identify relevant laboratory tests, prescriptions, diagnosis codes (714) and the presence of rheumatology consult letters. Accuracy for differentiating between RA and non-RA patients was assessed. We computed and compared the sensitivity, specificity, and predictive values for different approaches using multiple combinations of computerized searches of EMR fields for RA case ascertainment. Results We identified 121 RA and 9379 non-RA patients in our random sample (prevalence: 1.27%) for this validation exercise. Using only diagnosis codes (714) alone had a 59.6% sensitivity and 46.1% positive predictive value (PPV). Identifying cases using a free text-searching algorithm for RA in the problem list and past medical history fields text resulted in a 74.4% sensitivity, 99.9% specificity, 90.0% PPV, and 99.7% negative predictive value. The addition of laboratory tests, prescriptions and diagnosis codes did not improve the accuracy over the free-text searching algorithm. Conclusions We established the feasibility and accuracy of an EMR-based algorithm for identifying RA patients within primary care EMR data. Despite the high PPV, primary care EMRs are likely to miss some RA patients due to incomplete population of