Abstract
The demonstration of varying treatment efficacy among different subsets of patients is an important part of the analysis of clinical trials. The paper commences by clarifying the meaning of interaction and by reviewing valid procedures of analysis in the presence of interaction. Suitable descriptive measures of treatment—covariate interactions are (ratios of) hazard ratios to which generalized Patel—Hoel tests for qualitative or ordinal categorical covariates are related. Both the hazard ratios and the tests are given in a formulation for use with different scoring systems such as u- or e. Furthermore, the procedures are related to common k-sample tests and thus are suitable for proportional and for converging hazards. A worked example is included.

This publication has 33 references indexed in Scilit: