Multiple comparisons for relative risk regression: extension of the kappa-ratio method

We propose a kappa-ratio method for multiple comparisons among disjoint groups in survival analysis, specifically among relative risk estimates from a proportional hazards regression analysis. The new method relates the magnitude of critical values to an overall measure of group heterogeneity. That is, large critical values result from small values of a chi-square statistic for testing between-groups homogeneity. Conversely, large values of the overall statistic lead to small critical values for individual comparisons, even if there are many groups. The adaptivity of the kappa-ratio method contrasts sharply with the conservatism of competing methods.