Online Program
Methods for Responder Analysis of Patient-Reported Outcomes
Alexandra I Barsdorf, Pfizer Inc Keywords: ROC Clinically important difference (CID) is a useful measure of change in a patient-reported outcome (PRO) that is considered meaningful to the patient. We demonstrate two estimation methods for CID and report how CID estimation is affected by correlation between the target PRO and anchor and by response categories. First, a distribution-based method is demonstrated that derives the CID as a score corresponding to a priori benchmark values for standardized effect sizes (0.2, 0.5, and 0.8). Second, an anchor-based method is presented that defines responders using an anchor correlated with the outcome. Alternatively, the anchor-based method is based on a receiver operating characteristic curve to optimally differentiate between responders and nonresponders. Both exact solutions to the CIDs and Monte Carlo simulations are reported using a latent bivariate normal distribution assumption between the outcome for CID and anchor. Finally, cumulative distribution function plots of change from baseline of a PRO are presented for a visual display of the entire distribution of responses, rather than only a point or interval estimate of the CID. These methods are illustrated on existing data.
|