Statistical Analyses of Mutually Exclusive Competing Risks in Neonatal Studies

Abstract

Following the well-established approach on how to deal with competing risks in the situation of time-to-event endpoints, cumulative incidences have to be used to analyse each single category of an outcome in the situation of competing risks without a time-to-event structure as well. This can be easily done by applying a simple chi-square test. Nevertheless, these categorial outcomes are usually combined to get a composed dichotomous outcome to face the problem on how to deal with a significant chi-square omnibus test in the situation of more than 1 df, i.e. > 2x2 tables. The aim of this report is to question the practice of combined, i.e. composed dichotomized, endpoints because important information is lost and the real effect of interest in confirmatory phase III studies may only become apparent in explorative secondary analyses. It is shown – by using hypothetical data and by recalculation of published phase III studies’ results – how the use of a chi-square omnibus test and the scarcely known post-hoc testing answers the real question of interest within one primary confirmatory analysis. This method reveals insight into the actual effect of a new treatment or therapy on the event of interest in the presence of a mutually exclusive competing risk