Estimates of the discriminant ability of definitions of improvement for juvenile rheumatoid arthritis.

Abstract

To investigate the ability of various definitions of improvement to distinguish between patients with juvenile rheumatoid arthritis (JRA) treated with active drug from those given placebo in randomized trials. A core set of 6 response (outcome) variables for use in JRA has been reported. These core variables were combined into a number of "definitions of improvement" for the purpose of classifying individual patients as either "clinically significantly improved" or "not improved." We used a large dataset from randomized controlled trials to test the discriminant ability (sensitivity to change) of the definitions. We calculated the proportion of patients classified as "improved" by each definition in each of the treatment and control groups. Effect sizes were weak in 4 of the treatment regimens used (D-penicillamine, hydroxychloroquine, auranofin, and very low dose methotrexate) and no definition discriminated well between drug and placebo treated groups. Definitions that required 20 to 30% improvement in 3 to 4 of the 6 core set variables showed statistically significant differences in the proportions of patients who were classified as improved in the group treated with low dose methotrexate (10 mg/m2 body surface area/wk) compared to placebo. A definition resembling the Paulus criteria used in adult RA trials (4 of 6 core set variables improved by > or = 20%) performed well, as did the definition selected previously as the best for JRA (3 of 6 core set variables improved by > or = 30%, not more than one worsening by > 30%). Definitions that require 20 to 30% improvement in 3 to 4 core outcome variables are sensitive to change and are able to clearly distinguish between treated and control groups when an effective drug is being tested. Further testing of their validity is under way.