All Times EDT
Keywords: Patient-reported outcome; Meaningful within-patient change; Regression model; Repeated measure; Empirical cumulative distribution function
Objectives: The FDA recommends the use of anchor-based methods supplemented by empirical cumulative distribution function(eCDF) curves to establish a meaningful within-patient change(MWPC) for patient-reported outcome (PRO) measures. In practice, the estimates obtained from model-based and eCDF-based approaches may not closely align. To help interpret differing results, we investigate and compare these approaches. Methods: Both repeated measure linear regression model (RMM, the anchor-based method) and eCDF approaches were used to estimate a MWPC on a target PRO measure. We used an empirical data set (adapted from a real data set) and a simulated data set included 500 patients with up to 6 visits per patient, target PRO (range 0 to 10), and an anchor measure on patient global impression of change(PGIC) from 1 (much worse) to 5 (much better). Ninety-five percent confidence intervals for the MWPC were calculated by the bootstrap method (1,000 iterations). Results: The distribution of the PRO score changes affected the degree of concordance between RMM and eCDF-based estimates. PRO score changes from simulated normally distributed data led to greater concordance between the two approaches than PRO score changes from empirical data. Confidence intervals of eCDF-based estimates for the MWPC were usually much wider than that of RMM estimates, and the eCDF-based estimate noticeably varied across visits. Conclusions: A difference between RMM and eCDF-based estimates for MWPC could be explained by the fact that, unlike eCDF derived median, the RMM derived mean considers all available measurements across time. We recommend that the RMM approach be given preference over the eCDF approach because RMM integrates more information across a diverse range of PRO and PGIC scores and provide more precise estimates on MWPC. As such, the RMM approach is expected to more accurately assess the true relationship between target PRO measure and anchor measure in arriving at a MWPC estimate.