Background The ability to assess rheumatoid arthritis (RA) disease activity using administrative data could provide a potential mechanism to identify patients with high disease activity who may be candidates for enhanced RA treatment. The Veterans Affairs rheumatoid arthritis (VARA) registry is a prospective observational cohort that collects RA disease activity measurements including the 28 joint count disease activity score (DAS28) at twelve US Veterans Affairs (VA) Medical Centers.
Objectives This study used VA administrative data to develop a model to predict DAS28 in US Veterans.
Results We identified 1,582 US Veterans with RA who fulfilled enrollment criteria. The average age was 62.7±11.0 years, 91% male, with a disease duration 13.7±12.1 years with 78% and 74% seropositive for rheumatoid factor and anti-CCP antibodies respectively. Of the 1,275 variables evaluated, 175 had prevalence of 1% or more for consideration in the model. Of these 175 candidate variables, 32 remained in the final model. The key variables associated with disease activity were related to number of clinic visits (rheumatology, primary care, and ancillary services), numbers of laboratory tests, medication changes, imaging episodes, and surgery. The association of the predicted and observed DAS28 was relatively limited with unadjusted and adjusted R-square as 0.24 and 0.22 respectively. The model was very sensitive for detection of high/moderate disease (sensitivity =91.3%) but the specificity was only 25.3%.US Veterans with RA enrolled in the VARA registry with administrative data available for the year prior to DAS28 measurement and were evaluated. Patients were excluded if they had cancer, organ transplantation, or other autoimmune diseases. We identified 1,275 administrative data elements with suspected association with disease activity to investigate for association with DAS28 score. Least absolute shrinkage and selection operator (LASSO) was used to develop a model for the prediction of DAS28 as a continuous variable. The sensitivity, specificity, positive and negative predictive values were calculated. For these calculations, we defined high/moderate disease as DAS28>3.2 and low disease as DAS28≤3.2 in a categorical manner.
Conclusions While there were multiple administrative elements that correlated with disease activity measured by DAS28, it was not possible to develop a strong predictive model using available administrative data. These observations suggest that the continued direct observation of disease activity remains the most effective measure to identify RA patients with high disease activity. However the transient nature of RA disease activity may predicate the need to refine our model to account for disease duration. Additionally a single point-in-time measurement may not be indicative of patient disease activity over a long period of time. The development of other methods that can extract information from provider assessment of disease from electronic medical records has the potential to meet this critical need.
Disclosure of Interest G. Cannon Grant/research support from: Amgen, C.-C. Teng Grant/research support from: Amgen, N. Accortt Shareholder of: Amgen, Employee of: Amgen, D. Collier Shareholder of: Amgen, Employee of: Amgen, M. Trivedi Shareholder of: Amgen, Employee of: Amgen, B. Sauer Grant/research support from: Amgen