Improving the responsiveness of rating scales

Rating scales that measure health status are used frequently by clinicians to evaluate endpoints in clinical trials and in outcomes research. To be useful as an indicator of intervention success, an outcome measure must be reliable, valid, and responsive. Responsiveness is the ability of a tool to accurately detect important changes over time and is less frequently studied than the measurement characteristics of reliability and validity. There are several ways of quantifying responsiveness, including the effect size statistic and the standardized response mean.1,2⇓ The general concept in these statistics is that a mean change in score from time 1 to time 2 is divided by some measure of the score dispersion (e.g., the standard deviation at time 1 or for the change score itself). The interpretation of responsiveness of an outcome measure is critical for understanding the potential impact of an intervention on a population. A thorough discussion of the different methods of calculating responsive statistics can be found in articles by Kazis et al.2 and Cohen.3 In this issue of Neurology , O’Connor et al.4 consider the problem of evaluating the effectiveness of neurorehabilitation in a cohort of over 1,400 inpatients using the Barthel Index (BI) to measure the impact of rehabilitation. They conclude that the BI total score has good responsiveness to change in this population, but that floor and ceiling effects observed in some of the individual items imply that the BI may actually underestimate the effect of rehabilitation on patient outcomes. This finding is an important illustration of …