Investigations on MCP-Mod Designs
Julia Krzykalla · Open access LMU (Ludwid Maxmilian's Universitat Munchen) · 2015
The present master's thesis investigates specic applications of the MCP-Mod approach which is a unication of the two approaches typically applied in the matter of dose-nding, the multiple comparison procedures and the modelling of a parametric dose-response function.By the combination of both, one benets from the advantages of the continuous modelling, but improves the validity of the results by basing the analyses not only on one pre-specied model but on a set of suitable models.The MCP-Mod approach by Bretz et al. (2005) has been designed for normally distributed outcomes collected in a basic study design.An enhancement by Pinheiro et al. (2014) makes the approach applicable to a broader range of outcome types, particularly for binary endpoints.As a binary data setting is the underlying scenario for the investigations in the practical part of the thesis, a description of this generalized version is as well included.Furthermore, a third approach is presented which is based on the same idea: the approach by Klingenberg (2009).The rst aim of this thesis is the comparison of the naive application of the original MCP-Mod approach with its generalized version and the Klingenberg approach for the case of a binary endpoint via simulations.Aspects for the comparison are the achieved power, the preservation of the type-I error and the precision of the target dose estimate.The simulations reveal that the rst mentioned approach leads to a loss in power and a potential ination of the type-I error whereas the other two methods show good performances in both, the testing and the estimation part.Secondly, the thesis investigates two dierent approaches for the combination of target dose results of separate trials with the aim of obtaining a common dosage proposal if adequate.The rst approach is to pool the data of the separate trials and perform the analysis based on the combined data set.For the second approach, the trials are analyzed separately and the results are combined only afterwards.The two approaches are judged by the same criteria as considered in the rst part.Simulations show that for an inconvenient combination of trial-specic design aspects, the pooled analysis approach without adjustments may lead to an ination of the type-I error while the second approach produces good results for all of the investigated aspects.Evidently, the type-I error ination of the pooled analysis approach can be avoided by adapting the determination of the p-value.