Classifying radiographic progression status in early rheumatoid arthritis patients using propensity scores to adjust for baseline differences

Various methods are used to measure radiographic joint damage in patients with rheumatoid arthritis (RA), but determining proportions of responsive patients is difficult. A key problem in observational studies when assessing damage outcomes is incorporating time to treatment initialization and adjusting for observed baseline differences. We examined five different definitions to select an appropriate index to classify radiographic damage in RA patients as progressive or nonprogressive. In addition, we compared different times from symptom onset to treatment and their effects on patient radiographic categorization. Propensity scores to adjust for baseline differences, including time since symptom onset, were used to match those treated early with those treated later using the stratification, radius, nearest neighbor and kernel methods. The mean effect of treatment on the treated was computed for each matching method. Observational data were analyzed for 185 early RA patients from the Western Consortium study followed six to sixty months (mean thirty-one months). For the selected index, 75 patients were categorized as nonprogressors; they had significantly lower disease activity, more clinical improvement and were treated earlier than the progressors. Of those treated within three months of symptom onset, 57% were classified as radiographically progressive versus 35% of those treated later (P = 0.0058). However, after propensity score adjustment for baseline differences, we noticed nonsignificant (P > 0.05) nonprogression in patients given earlier treatment. We conclude that propensity score analysis reduced but did not remove all bias.