Examining the minimum important difference

The increasing use of constructed scales and indices in clinical science has preceded in many cases a clear understanding of how to appraise the importance of the differences or changes that are thereby observed. For example, in the design of clinical trials which employ such scales as outcome measures it may be difficult to specify what constitutes a clinically significant shift in means, a key factor in sample size calculations. Determination of the minimum important difference relative to specific outcome measures has historically been based on informal and/or intuitive arguments. In this paper we propose a formal statistical framework for these considerations, based on a previously published validation study design which captures patients' perceptions in comparative self-reported assessments. We begin by adopting a mixed-effect model to represent the comparative assessments as composites of individual self-ratings on an underlying continuous scale. We then present two basic approaches for assessing the relation between the hypothesized latent scale and the outcome scale(s) under consideration, taking the latent scale as a plausible benchmark against which observable changes on the outcome scale can be judged.