Abstract:
|
Cure models are useful tool to analyze and describe a population contains both unobservable cured and uncured patients. The current existing R package uses expectation-maximization (EM) algorithm to estimate cure models and bootstrap method to estimate variance. One problem is the convergence of EM algorithm is highly dependent on the observed sample, and it may give unreliable estimator and unstable bootstrap standard error. In this study, we propose a robust procedure for cure models in survival analysis by incorporate a weakly informative prior in the logistic part in cure model. Our simulation results indicated the feasibility of our method and the advantage of this robust procedure in terms of reducing bias and mean squared error and stabilizing bootstrap variances estimates. We also provided real data illustrations of the proposed method. In addition, we developed an R package rcure to implement our proposed method. Other features of this package include estimate prognostic accuracy, i.e. AUC, k-index and c-index, with bootstrap confidence intervals.
|