Partitioning the Effects of Concomitant Variables in Survival Analysis

Using the Kalbfleisch-Prentice marginal likelihood based on Cox's regression and life model, a computationally simple method is presented for testing general linear hypotheses of the form H o : ~ §= Q. The method, which does not require estimating §, is based on partitioning the asymptotic global chi-square statistic of Cox to produce the appropriate test statistic. Under H o ' this statistic also has an asymptotic chi-square distribution. An examination of its performance in small and moderate samples reveals the method to be useful, particularly for testing the significance of subsets of the observed concomitant variables.