Stefan Wellek1, Maria Blettner. 1. Institute of Medical Biostatistics, Epidemiology and Informatics, University Medical Center of the Johannes Gutenberg University Mainz.
Abstract
BACKGROUND: An increasing number of clinical trials are being performed to show the absence of relevant differences between the effects of two treatments. The primary care physician makes use of the results of so-called equivalence studies, at least indirectly, practically every day. Equally important are active control clinical trials in which the efficacy of a new treatment has to be proven through demonstrating non-inferiority as compared to a standard treatment. METHODS: Explanation of basic principles and statistical techniques with reference to the original literature; selective searches in the medical literature. RESULTS: First of all, a suitable distributional parameter must be chosen that can be considered a reasonable measure of dissimilarity of the population effects of the treatments under comparison. The simplest approach to the statistical demonstration of equivalence or non-inferiority is to calculate confidence intervals for that parameter. To keep the required number of subjects for equivalence and non-inferiority studies as low as possible, statistical tests should be used which are optimized with respect to power. CONCLUSION: Data from equivalence and non-inferiority studies need to be assessed for statistical significance no less than data that are generated to show that two treatments have different effects. A negative result in a traditional two-sided test does not suffice for statistically proving equivalence.
BACKGROUND: An increasing number of clinical trials are being performed to show the absence of relevant differences between the effects of two treatments. The primary care physician makes use of the results of so-called equivalence studies, at least indirectly, practically every day. Equally important are active control clinical trials in which the efficacy of a new treatment has to be proven through demonstrating non-inferiority as compared to a standard treatment. METHODS: Explanation of basic principles and statistical techniques with reference to the original literature; selective searches in the medical literature. RESULTS: First of all, a suitable distributional parameter must be chosen that can be considered a reasonable measure of dissimilarity of the population effects of the treatments under comparison. The simplest approach to the statistical demonstration of equivalence or non-inferiority is to calculate confidence intervals for that parameter. To keep the required number of subjects for equivalence and non-inferiority studies as low as possible, statistical tests should be used which are optimized with respect to power. CONCLUSION: Data from equivalence and non-inferiority studies need to be assessed for statistical significance no less than data that are generated to show that two treatments have different effects. A negative result in a traditional two-sided test does not suffice for statistically proving equivalence.
Authors: Joseph J Eron; Benjamin Young; David A Cooper; Michael Youle; Edwin Dejesus; Jaime Andrade-Villanueva; Cassy Workman; Roberto Zajdenverg; Gerd Fätkenheuer; Daniel S Berger; Princy N Kumar; Anthony J Rodgers; Melissa A Shaughnessy; Monica L Walker; Richard J O Barnard; Michael D Miller; Mark J Dinubile; Bach-Yen Nguyen; Randi Leavitt; Xia Xu; Peter Sklar Journal: Lancet Date: 2010-01-12 Impact factor: 79.321
Authors: Bernd Röhrig; Jean-Baptist du Prel; Daniel Wachtlin; Robert Kwiecien; Maria Blettner Journal: Dtsch Arztebl Int Date: 2010-08-09 Impact factor: 5.594
Authors: Daniel F Martin; Maureen G Maguire; Gui-shuang Ying; Juan E Grunwald; Stuart L Fine; Glenn J Jaffe Journal: N Engl J Med Date: 2011-04-28 Impact factor: 91.245
Authors: Peter Markus Spieth; Anne Sophie Kubasch; Ana Isabel Penzlin; Ben Min-Woo Illigens; Kristian Barlinn; Timo Siepmann Journal: Neuropsychiatr Dis Treat Date: 2016-06-10 Impact factor: 2.570