Validity of area-under-the-curve analysis to summarize effect in rheumatoid arthritis clinical trials

There is a continuing interest in increasing the statistical efficiency of the analysis of clinically meaningful endpoints in rheumatology. One issue that is attracting increasing attention is whether the conventional practice of only reporting the outcome at the end of the study (EOS) might be replaced or complemented by a longitudinal summary that better reflects the clinical course of the disease. The area under the curve (AUC) is a summary measure that integrates serial assessments of a patient's endpoint over the duration of the study. We evaluated the utility of AUC as a summary measure for the analysis and reporting of two RA trials: (i) methotrexate combined with cyclosporine versus methotrexate and placebo in partial methotrexate responders in relatively late disease, and (ii) prednisone plus methotrexate plus sulfasalazine versus sulfasalazine alone in relatively early disease. We replicated the published results of each trial first using the conventional EOS and then AUC summaries. For each patient, the changes from baseline over time were transformed into a summary measure by calculating AUC using the trapezium rule and then standardizing it by the study duration. Using an approach similar to the index of responsiveness to change, we scaled treatment differences derived from EOS and AUC summary measures by their standard deviation of the control group. This signal-versus-noise ratio captures the treatment discrimination ability of each summary measure. Compared to EOS and within each treatment group, the AUC summary reported smaller effects (i.e., change from baseline) with reduced errors in the estimates. AUC measures preserved discriminant validity in treatment comparisons and reported smaller but more precise treatment effect estimates. In the COBRA trial with rapidly-acting medications, AUC seemed to be more sensitive than EOS to detect treatment difference. With slow acting medications and in relatively late disease patients as in the cyclosporine trial, EOS was more sensitive to detect treatment difference than was AUC. In this setting, AUC, however, still seemed to be more sensitive than EOS for the two responsive-to-change endpoints: tender joint counts and pain by visual analog scale. AUC integrates repeated assessments during the trial duration into summary measures. Compared to EOS, the report of RA trial results using AUC summary provides smaller estimates of treatment effects but with better precision. AUC summary is likely to preserve treatment group discrimination taking into account the appropriate onset and offset of the drug action. Trial reports using AUC summary have smaller effect sizes. For trials with long acting medications and short duration similar to the cyclosporine trial, AUC still preserves treatment discrimination but may not be as sensitive as EOS. The calculations of AUC require some additional work in the analysis of each endpoint.