Adjusting survival curves for imbalances in prognostic factors

Abstract

A new method for comparing the survival of two or more groups of patients adjusting for factors distributed unevenly between the groups is presented. This is a development of previous methods, and provides a graphical counterpart to Mantel's adjusted chi-square statistic. The method can be used to retrospectively stratify for prognostic factors, and to provide additional validation and interpretation of multivariate results, including those based on Cox's proportional hazards model. Like Mantel's adjusted chi-square statistic, the method adjusts at every event, based on the numbers of patients still at risk in each of the groups, and is thus able to show up time-dependent effects: factors can be seen to be relevant during certain periods of the study only. The method presented thus allows curves to be drawn as they would have been expected to look, had the prognostic factors been evenly distributed between the groups.