Case finding for population-based studies of rheumatoid arthritis: comparison of patient self-reported ACR criteria-based algorithms to physician-implicit review for diagnosis of rheumatoid arthritis.

To evaluate the interrater reliability of rheumatologist diagnosis of rheumatoid arthritis (RA) and the concordance between rheumatologist and computer algorithms for assessing the accuracy of a diagnosis of RA. Self-reported data regarding symptoms and signs for a diagnosis of RA were considered by a panel of rheumatologists and by computer algorithms to assess the probability of a diagnosis of RA for 90 patients. The rheumatologists' review was validated through medical record. The interrater reliability among rheumatologists regarding a diagnosis of RA was 84%; the chance-corrected agreement (kappa) was 0.66. Agreement between the rheumatologists' rating and the best-performing algorithm was 95%. Using rheumatologist's review as a standard, the sensitivity of the algorithm was 100%, specificity was 88%, and the positive predictive value was 91%. The validation of rheumatologist's review by medical record showed 81% sensitivity, 60% specificity, and 78% positive predictive value. Reliability of rheumatologists' assignment of a diagnosis of RA by using self-report data is good. Algorithms defining symptoms as either joint swelling or tenderness with symptom duration >or=4 weeks have a better agreement with rheumatologist's diagnosis than do ones relying on a longer symptom duration. These findings have important implications for health services research and quality improvement interventions pertinent to case finding for RA through self-report data.