Assessing the relative sensitivity to change of rheumatoid arthritis activity measures: is the type of treatment an important third variable?

Abstract

Observational studies and meta-analyses of controlled clinical trials have been used to identify which measures of rheumatoid arthritis activity are most sensitive to change. These analyses often pool studies of different drugs, although it is not known if arthritis activity measures are differentially responsive to different drugs. In meta-analyses, estimates of the relative sensitivity to change of different measures may also be confounded by differences in drug efficacy, if studies of different drugs contribute different measures to the meta-analysis. To determine if the type of treatment acts as an important effect modifier or confounder in studies of the relative sensitivity to change of arthritis activity measures, we computed effect sizes for four measures (weighted tender joint count, grip strength, duration of morning stiffness, and erythrocyte sedimentation rate) used in each of 16 trials of five different disease-modifying antirheumatic drugs (methotrexate, sulfasalazine, cyclosporin A, intramuscular gold, and D-penicillamine) in rheumatoid arthritis. In a complete factorial analysis of variance, effect sizes differed significantly among drugs (p = 0.0006), but differed only marginally among measures (p = 0.08). No interaction was detectable between drugs and measures. These results suggested that effect modification by drugs was not present, but that pooled estimates of the sensitivity to change of different measures may be confounded in meta-analyses, if trials of more efficacious drugs contribute different measures than trials of less efficacious drugs. In a similar analysis of 26 trials of nine nonsteroidal anti-inflammatory drugs, we found significant differences in effect sizes among measures (p < 0.0001), but no differences among drugs (p = 0.96), and no interaction between drugs and measures. This study suggests that pooled analyses of the relative sensitivity to change of arthritis activity measures based on trials of different disease-modifying drugs may be confounded by drug effects, but confounding by drug effects is unlikely if these meta-analyses are based on trials of different nonsteroidal anti-inflammatory drugs. Although the power of these analyses to detect small interaction effects was limited, effect modification by drugs was not observed, indicating that the measures we examined were not strongly differentially responsive to different drugs.