gms | German Medical Science

If the efficacy of a new investigational treatment is investigated in a placebo-controlled trial, it is not sufficient to demonstrate superiority by a statistically significant test result. The recent Points to Consider on the Choice of Non-Inferiority Margin, published in 2004 by the European Committee for Proprietary Medicinal Products [Ref. 1], points out that for demonstrating superiority of a test product over placebo not only statistical significance but also clinical relevance has to be shown. This key requirement was also stressed by Brown [Ref. 2] stating that observed effect sizes have to be both statistically significant and clinically relevant for the proof of efficacy. One approach that aims to combine statistical and clinical significance is the incorporation of a minimum clinically relevant effect in the test problem by considering shifted hypotheses [Ref. 3]. However, this concept might result in a considerable increase in sample size.

An alternative approach for assessing the clinical relevance of a statistically significant treatment effect is to perform a responder analysis [Ref. 1]. Here each patient is classified whether or not he has achieved a pre-defined meaningful treatment effect. Notwithstanding the fact that responder analyses often reflect medical judgement and, therefore, are recommended in various medical CPMP guidelines [Ref. 4], [Ref. 5], [Ref. 6], a critical scientific discussion concerning the appropriateness of dichotomization was recently started by Senn [Ref. 7]. In defence of the dichotomy, the reply by Lewis [Ref. 8] presented a number of supporting arguments. Planning considerations for responder analyses are addressed in a recent paper [Ref. 9].

A probably more intuitive approach to judge the clinical relevance is by considering the magnitude of the treatment effect on the original scale [Ref. 10]. In the presentation we focus on the assessment of clinically significance by point estimates and corresponding confidence intervals. Different criteria proposed in the literature for judging the clinical relevance of therapeutic response are presented and the relating power and sample size considerations are provided. We will show that requiring in addition to statistical significance an observed treatment effect greater than an a priori defined minimum clinically relevant effect is a feasible method for the assessment of clinical relevance. The consequences are illustrated for the example of a clinical trial in depression.