Nonparametric Mixture Cure Models with Cure Partially Known

• Activity Number: 270 - Nonparametric and Semiparametric Statistical Inference for Cure Models
• Type: Invited
• Date/Time: Tuesday, July 30, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Journal of Nonparametric Statistics
• Author(s): M. Amalia Jácome* and Wende Safari and Ignacio López-de-Ullibarri
• Companies: Universidade da Coruña and Universidade da Coruña and Universidade da Coruña
• Keywords: Bandwidth; Bootstrap; Censored data; Kernel estimation; Local maximum likelihood; Survival analysis

Abstract

Mixture cure models assume that the population is made up of two distinct groups: those who will experience the event of interest and those who will not (the uncured and cured patients, respectively). The main goal of these models is to estimate the cure probability, possibly depending on a set of covariates. A common assumption of traditional cure models is that cured and uncured subjects cannot be distinguished within the censored observations. Hence, the cure indicator is usually modeled as a latent variable. However, sometimes this assumption is not entirely valid, when some censored individuals can be considered to be cured or long-term survivors. One typical example is the case when individuals are assumed to be cured if their survival time is larger than a given cutoff or cure threshold (e.g., 5 years for recurrence in some types of cancer). Here we propose a completely nonparametric method for the estimation of the cure probability in mixture cure models when cure is partially known. Some properties are given, and a bandwidth selector is proposed. The practical performance of the procedure is shown with a dataset of patients with sarcoma.