Cure rate models for heterogeneous competing causes

• Published in: Statistical methods in medical research Vol. 32; no. 9; p. 1823
• Main Authors: Brandão, Márcia, Leão, Jeremias, Gallardo, Diego Ignacio, Bourguignon, Marcelo
• Format: Journal Article
• Language: English
• Published: England 01.09.2023
• Subjects: Algorithms; Brazil; Computer Simulation; Humans; Likelihood Functions; Melanoma - diagnosis; Models, Statistical; Skin Neoplasms; Survival Analysis; Brazil; melanoma data set; expectation–maximization algorithm; power series distribution; mixtures; Concurrent causes
• ISSN: 1477-0334
• DOI: 10.1177/09622802231188514

Cure rate models have been widely studied to analyze time-to-event data with a cured fraction of patients. In this type of model, the number of concurrent causes is assumed to be a random variable. However, in practice, it is natural to admit that the distribution of the number of competing causes is different from individual to individual. Our proposal is to assume that the number of competing causes belongs to a class of a finite mixture of competing causes distributions. We assume the number of malignant cells follow a mixture of two power series distributions and suppose that the time to the event of interest follows a Weibull distribution. We consider the proportion of the cured number of competing causes depending on covariates, allowing direct modeling of the cure rate. The proposed model includes several well-known models as special cases and defines many new special models. An expectation-maximization algorithm is proposed for parameter estimation, where the expectation step involves the computation of the expected number of concurrent causes for each individual. A Monte Carlo simulation is performed to assess the behavior of the estimation method. In order to show the potential for the practice of our model, we apply it to the real medical data set from a population-based study of incident cases of cutaneous melanoma diagnosed in the state of São Paulo, Brazil, illustrating that the model proposed can outperform traditional models in terms of model fitting.