USE OF IRWIN RESTRICTED MEAN AS AN INDEX FOR COMPARING SURVIVAL IN DIFFERENT TREATMENT GROUPS - INTERPRETATION AND POWER CONSIDERATIONS

In the analysis of survival data from clinical trials and other studie s, the censoring generally precludes estimation of the mean survival t ime. To accommodate censoring, Irwin (1949) proposed, as an alternativ e, estimation of the mean lifetime restricted to a suitably chosen tim e T. In this article we consider the use of Irwin's restricted mean as an index for comparing survival in different groups, using as an exam ple published data from a randomized clinical trial in patients with p rimary biliary cirrhosis. Irwin's method, originally based on the actu arial survival estimator, is extended to incorporate covariates into t he analysis through the use of piecewise exponential models. For compa ring two survival curves, the logrank test is known to be optimal unde r proportional hazards alternatives. However, comparison of restricted means may outperform the logrank test in situations involving nonprop ortional hazard rates. We examine the size and power of these two proc edures under various proportional and nonproportional hazards alternat ives, with and without covariate adjustment. For survival curves that separate early in time the censored data generalization or: the Wilcox on test is known to exhibit high power, and we examine how the compari son of restricted means performs relative to this procedure also. (C) Elsevier Science Inc. 1997.