The SF-36 Arthritis-Specific Health Index (ASHI) II. Tests of validity in four clinical trials

OBJECTIVE. The SF-36 Arthritis-Specific Health Index (ASHI) was constructed to improve the responsiveness of the SF-36 Health Survey to changes in the severity of arthritis through the use of arthritis-specific scoring algori thms. This study compared the responsiveness of the ASHI and other generic scales and summary measures scored from the SF-36 in clinical trials of hea lth outcomes for patients with arthritis. METHODS. Longitudinal data for patients (n = 835) participating in four pla cebo-controlled trials were analyzed. Study participants had at least a 6-m onth history of moderate to severe osteoarthritis or rheumatoid arthritis o f the knee or hip. All had undergone a washout period of 3 to 14 days befor e baseline assessment to bring about a flare state in osteoarthritis or rhe umatoid arthritis symptoms. Their average age was 60 years, and 72% were fe male. Responders and nonresponders were classified on the basis of physicia n assessments of changes in arthritis severity, with blinding as to treatme nt group; treated and untreated (placebo) groups were also compared. For th e SF-36 ASHI, generic physical (PCS) and mental (MCS) component summary mea sures and each of eight subscales scored from the SF-36 (acute version) cha nge scores were computed by subtracting scores before treatment from scores at 2-week follow-up. To evaluate empirical validity, analyses of variance were performed. For each measure, an F-ratio was computed for the compariso n between clinically defined groups of responders and nonresponders and bet ween groups of patients assigned to placebo versus drug therapy. Relative v alidity (RV) coefficients were computed for the ASHI in comparison with PCS , MCS, and the best SF-36 scale to determine which was more responsive. RESULTS. In analyses of each of the four trials and all trials combined, RV coefficients for the ASHI were higher than those for both of the generic S F-36 summary measures and for the most valid SF-36 scale (Bodily Pain), wit h only one exception. Across 40 tests of validity in distinguishing treated from untreated patients, the ASHI was 5% to 19% more valid than the best S F-36 scale (RV = 1.05-1.19; RV = 1.10 in all trials combined). The generic summary measures (PCS and MCS) were much less valid in these tests (RV = 0. 67 and 0.27, respectively). In analyses of responders and nonresponders, RV coefficients for the ASHI ranged from 0.70 to 1.22 (RV = 1.04 in all trial s combined), in comparison with the best SF-36 subscale, which was always B odily Pain. RV coefficients were lower for PCS (RV = 0.75) and much lower t han the MCS (RV = 0.18) in comparisons of treatment outcomes based on all t rials combined.