Artikel
Evaluating non-inferiority in a gold standard design using nonparametric methods
Suche in Medline nach
Autoren
-
Kerstin Rubarth
- Charité - Universitätsmedizin Berlin, Institut für Biometrie und klinische Epidemiologie, Berlin, Germany
-
Paavo Sattler
- Technische Universität Dortmund, Fakultät für Statistik, Dortmund, Germany
-
Frank Konietschke
- Charité - Universitätsmedizin Berlin, Institut für Biometrie und klinische Epidemiologie, Berlin, Germany
| Veröffentlicht: | 15. September 2023 |
|---|
Gliederung
Text
Introduction: Non-inferiority studies are widely used in medical research, for example, when the efficacy of a newly developed drug is compared to the efficacy of a standard treatment, while providing a better safety profile. However, when investigating efficacy in a two-arm design, only an indirect proof of efficacy can be obtained since no direct comparison with the new treatment to a placebo treatment is conducted. Therefore, the so-called gold-standard-design [1], involving an additional group treated with placebo, offers the possibility to simultaneously test assay sensitivity (the ability of a trial to distinguish an effective treatment from a placebo treatment) as well as non-inferiority of the new treatment compared to the standard treatment. Many authors proposed parametric procedures for analyses in this setting [2], [3]. However, in many practical applications, sample sizes are quite small; outcomes are highly skewed or are measured on an ordinal scale. For these situations, non-parametric rank-based methods based on relative effects should be preferred over traditional parametric methods. Munzel [4] developed theory for non-parametric analyses in gold standard designs, defining the effect measures, the so-called relative effects, by using a sum of weighted distribution functions. These relative effects are no model constants but depend on the sample size of the trial. Furthermore, Munzel suggests inferring hypothesis in this framework via confidence intervals obtained by using the Fieller theorem, which has some constraints for practical applications.
Methods: Therefore, we propose to define relative effects by using unweighted sums of distribution functions in order to obtain model constants, which do not depend on the sample size as proposed by Brunner, Konietscke and Bathke [5]. Furthermore, we present alternatives to the confidence interval procedure by Munzel and conduct an extensive simulation studies in terms of type-I error rates as well as power under alternatives.
Results and conclusion: Besides the advantage that by using our definition of relative effects, we obtain model constants, our approach exhibits a better performance and should therefore be preferred over the procedure by Munzel.
The authors declare that they have no competing interests.
The authors declare that an ethics committee vote is not required.
References
- 1.
- Hauschke D, Pigeot I. Establishing efficacy of a new experimental treatment in the gold standard design. Biom J. 2005;47(6):782-6; discussion 787-98. DOI: 10.1002/bimj.200510169
- 2.
- Pigeot I, Schäfer J, Röhmel J, Hauschke D. Assessing non-inferiority of a new treatment in a three-arm clinical trial including a placebo. Stat Med. 2003;22(6):883-99. DOI: 10.1002/sim.1450
- 3.
- Kieser M, Friede T. Planning and analysis of three-arm non-inferiority trials with binary endpoints. Stat Med. 2007;26(2):253-73. DOI: 10.1002/sim.2543
- 4.
- Munzel U. Nonparametric non-inferiority analyses in the three-arm design with active control and placebo. Stat Med. 2009;28(29):3643-56. DOI: 10.1002/sim.3727
- 5.
- Brunner E, Bathke A, Konietschke F. Rank and Pseudo-Rank Procedures for Independent Observations in Factorial Designs. Heidelberg, Germany: Springer; 2019. (Springer Series in Statistics). DOI: 10.1007/978-3-030-02914-2
