A framework based on combining multiple comparisons procedures (MCP) and modeling (Mod) techniques for dose finding studies, known as MCP-Mod, was proposed by Bretz et al. (2005) to handle normally distributed, homoscedastic dose-response data. An extension of MCP-Mod was later developed for applications with non-normal dose-response observations, being named gMCP-Mod. Both MCP-Mod and gMCP-Mod require prior specification of parameters for candidate models, the so called “guesstimates.” In situations where guesstimates are not readily available, uncertainties or inconvenience could be introduced into the dose response signal testing step of (g)MCP-Mod, potentially impacting on correctly identifying significant dose-response model(s).

We will describe a resampling based permutation test as an alternative to the MCP step in (g)MCP-Mod. The proposed approach avoids the requirement of guesstimates and therefore reduces the uncertainty/inconvenience in model specification in the MCP step, while still strongly controlling the family-wise error rate (FWER). The permutation test is based on the exchangeability of dose levels under the null hypothesis of a flat dose-response shape. This dose-exchangeability allows reference distributions for the permutation tests to be built based on resampling and used to control FWER. The proposed procedure was evaluated through simulation studies and the results indicate that it provides a robust alternative for the MCP step in (g)MCP-Mod, in terms of statistical power.