Identifiability of cure models

Cure models can be used for censored survival data in which a fraction of the observations do not exhibit the event of interest despite long-term follow-up. In this paper we investigate the identifiability of two forms of the cure model, a standard cure model based on a mixture distribution and a non-mixture proportional hazards (PH) model with long-term survivors. In the standard cure model, except for the case where the conditional survival function is independent of covariates and the mixture probability is an arbitrary function of a covariate we show that the parameters of the standard cure model are identified. In the non-mixture PH model, we show the model is identifiable if the distribution function is specified.