Abstract
The emphasis currently being placed by the FDA on global clinical judgments recorded on ordered-category rating scales, such as the Clinical Global Impression (CGI) or Clinician Interview-based Impression of Change (CIBIC), has raised serious concerns among industry statisticians about the general acceptability of parametric statistical methods in controlled clinical trials that rely on such measurements for the evaluation of treatment effects. To address these concerns here, simulation methods were used to generate repeated measurements having characteristics of CGI or CIBIC ratings in a randomized parallel-groups design, and repeated measurements analysis of variance (ANOVA), analysis of covariance (ANCOVA), trend tests, and endpoint analyses were applied to the highly discontinuous ordered-category data. The parametric tests of significance were documented to provide appropriate protection against Type I errors and to have power substantially superior to that achieved by dichotomizing outcomes into “improved” versus “not improved” for simple chi-square tests. These results are presented in support of statisticians who are not likely to be surprised at the robustness of the parametric methods, but who have feared that previously existing documentation has addressed the issue with insufficient specificity for the types of rating-scale data and types of repeated measurement designs that are commonly employed in controlled clinical trials.