Survivorship analysis when cure is a possibility: A monte carlo study

Abstract

Parametric survivorship analyses of clinical trial commonly involves the assumption of a hazard function constant with time. When the empirical curve obviously levels off, one can modify the hazard function model by use of a Gompertz or Weibull distribution with hazard decreasing over time. Some cancer treatments are thought to cure some patients within a short time of initiation. Then, instead of all patients having the same hazard, decreasing over time, a biologically more appropriate model assumes that an unknown proportion (1 - π) have constant high risk whereas the remaining propotion (π) have essentially no risk. This paper discusses the maximum likelihood estimation of π and the power curves of the likelihood ratio test. Monte Carlo studies provide results for a variety of simulated trials; empirical data illustrate the methods.