Precision of composite measures of osteoarthritis efficacy in comparison to that of individual endpoints

Objective. Osteoarthritis (OA) clinical studies generally employ various endpoints to evaluate a spectrum of disease manifestations. Compared with individual endpoints, a composite measure might (1) provide a uniform outcome measure of OA efficacy with greater face validity, (2) address multiplicity, and (3) enhance precision. Combinations of endpoints were analyzed to investigate precision of composite measures of OA efficacy. Methods. We reanalyzed three 6 week, placebo controlled, double blind, parallel group studies (2 by the same protocol) of the cyclooxygenase-2 (COX-2) specific inhibitor rofecoxib. The average change from baseline at study weeks 2, 4, and 6 was assessed for 10 individual response variables, including patient and investigator global assessments, WOMAC 3.0V OA Index pain, stiffness and physical function subscales, graded study joint tenderness, and rescue analgesic use. Relationships among variables were evaluated using pairwise correlations and principal components analysis. The precision of variables to differentiate rofecoxib from placebo was evaluated using effect size (i.e., mean difference between rofecoxib versus placebo divided by pooled SD). Results. Correlations among all pairs of response variables ranged from 0.5 to 0.9, except those with tenderness (0.4 to 0.6) and those with analgesic use (0.2 to 0.4). The first principal component explained about 70% of the total variability, with weights generally similar (0.17 for rescue analgesic use, 0.25 for tenderness, and 0.30 to 0.37 for the others). These results indicate that nearly all measures are closely related. Based on these results, various linear combinations of the 9 endpoints were formed and their precision to discriminate active treatment from placebo was compared to that of the individual endpoints. Effect sizes of the individual endpoints ranged from 0.6 to 1.1; those of the composites from 0.7 to 0.9. The results were very consistent between study protocols. Conclusion. In comparison to individual endpoints, composite analyses of OA clinical endpoints do not increase precision to discriminate active treatment from placebo.