Topic: Title: Diagnostic checks for mixture cure models with interval censored data Authors:  Sylvie Scolas - Universite catholique de Louvain (Belgium) [presenting]
Anouar El Ghouch - The University catholique de Louvain (Belgium)
Catherine Legrand - Universite Catholique de Louvain (Belgium)
Abderrahim Oulhaj - United Arab Emirates University (United Arab Emirates)
Abstract: In elderly population, studying the time until Mild Cognitive Impairment (MCI) can be crucial since it is considered as a precursor of Alzheimers disease. However, not everyone will develop signs of MCI. We refer to this as the presence of a cure fraction. Another characteristic of these data, also found in other medical studies, is that patients are followed up at scheduled interviews, leading to interval-censored event-times, in addition to the possibility of being right-censored. When considering diagnostic tools in survival models, the idea of residual checking, widely used in linear regression, must be adapted to take account of right-censoring. Proposals have been made, mainly in the context of right-censored Cox and AFT regression models. For example, martingale residuals are commonly used to detect non-linearity of a covariate. While cure models are gaining into popularity, literature on residual-based model checking in this context is very sparse, and to the best of our knowledge nonexistent in the case of interval-censored data and cure. Clearly, an adaptation to mixture cure models is not straightforward and worth further study. We discuss the extension of existing methods to the presence of interval-censoring, and the difficulties encountered when applying these ideas to the cure mixture model. We illustrate our results based on simulations and on real data from an Alzheimer study.