Sensitivity of Single-domain Versus Multiple-domain Outcome Measures to Identify Responders in Chronic Low-back Pain

Abstract

A composite responder index for chronic low-back pain (CLBP) has recently been proposed to evaluate the efficacy of CLBP treatments in clinical trials. We compared the responsiveness of this composite measure with a number of single-item responder definitions. We pooled data from 2 placebo-controlled studies of etoricoxib in CLBP to evaluate 5 response criteria: 30% pain intensity (PI) reduction; 50% PI reduction; 20 mm absolute reduction (100 mm PI visual analog scale); patient global assessment of response to therapy (PGART); and the composite criteria of 30% reduction in PI+30% improvement in PGART of disease status+no worsening in function. We used bootstrap analysis and logistic regression to assess the ability to differentiate etoricoxib and placebo, and the κ coefficient to assess agreement among the responder criteria. The criterion of a 20 mm improvement in PI resulted in the greatest proportion (71.5%) of patients being classified as responders and all criteria separated etoricoxib from placebo (P≤0.0001). PGART had the highest discriminant ability (odds ratio 5.90), and was significantly (P<0.05) more discriminant than the 20 mm and ≥30% improvements and the composite criteria. After adjusting for all other measures, only PGART continued to show a significant treatment effect for etoricoxib versus placebo (P=0.0003). Kappa values contrasting the composite criteria and the single-item measures ranged from 0.59 to 0.85. These findings do not support the superiority of a composite index over single-item ratings of PI and PGART ratings, but do suggest that PGART ratings may be more responsive to treatment, perhaps because they measure something in addition to change in PI.