Mean changes versus dichotomous definitions of improvement

Abstract

In recent years, when reporting the results of clinical trials for chronic disease, including rheumatologic conditions, use has been made of dichotomous definitions of improvement, but it is to be expected that continuous definitions would offer improved discrimination between treatment groups. Nevertheless, a well-constructed dichotomous outcome (usually a composite) has advantages of clinical sense and specificity and may, under a variety of realistic conditions, have power that closely approximates that of standard continuous outcome measures. This has been seen for established dichotomous outcome definitions for two rheumatologic conditions, rheumatoid arthritis (RA) and ankylosing spondylitis (AS). Simulation studies performed using multivariate normal generated data that approximates actual trial data for each of RA and AS patients demonstrate the relative power of several dichotomous and continuous outcomes in realistic situations for each of RA and AS. Although the continuous outcomes are typically more powerful than the dichotomous ones, there are some situations in which the power of a well-defined dichotomous outcome approaches or even exceeds that of a continuous outcome based on mean change.