A partial review of cure models with an application to French cancer registries data to improve patients' access to insurance and credit

Abstract : Background Survival cure models are widely used in public health researches to analyze time-to-event data in which some subjects would never experience the event of interest; these subjects are said to be statistically cured. There are two types of cure models, the mixture cure modeland the non-mixture cure model which were first formulated respectively by Boag(1949)[1] and Yakovlev et al. (1993) [2]. These models have been intensively developed [3,4 among others] and have also been extended to the net survival framework [5-7 for instance].In cancersurvival analysis,net survival is a measure of survival in the hypothetical world wherecancer would be the only possible cause of death [8,9]. proportion of the subjects who are no longer at risk to die from their cancer i.e. the subjects without additional risk of death due to cancer (curedsubjects) ii) the time from which subjects can be supposed to be cured(thus the timeto surtax-free insurance). Methods : The principles Underlying the formulation of both the mixture and the non-mixture cure models were recalled and abrief review of the two types of models was provided. The extension of cure models to the net survival framework was exposed and the flexible non -mixture cure model based on net survival and developed by Andersson et al. (2011) was described. The later model was fitted to melanoma, colorectal and liver cancers data from the French cancer registries network. The data included all patients diagnosed between 1989 and 2010, aged between 15 and 74 at diagnosis and followed -up on June 31, 2013 for vital status. Cure time T was defined as the time when 90% of deaths due to cancer had occurred. T corresponded to the time at which the net survival reached a plateau at a non-zero value defined as the cure proportion P.T was referred to as the time from diagnosis to surtax-free insurance. Results : For melanoma, netsurvival reached a plateau at a cure proportion P of 88% for women and 82% for men. Cure times T were respectively 11.5 and 8.0 years after diagnosis. For colorectal cancer P was 57% for women and 51% for men,corresponding T were 7.5 and 8.4 years. T varied according to age, ranging from 7.3 years to 7.8 years forwomen and 8.2 to 8.6 years for men. For liver cancer, P varied according to age from 6 to 21% for women and 6 to 11% for men. T ranged from 3.4 to 5.1 years for women and 4.2 to 5.0 years for men. Conclusions : Cure models are useful tools to improve access to insurance and credit by allowing time to surtax free to rest on statistical evidence, and to be adjusted according to cure time.Cure time varied with cancer site, age and sex. It was lower than 10 years in various cases. Time to surtax free insurance should be reassessed for each site according to newly estimated time to cure. However the cure time as defined and estimated when using cure models is not entirely satisfactory and is subject to criticism. Further works on cure models are then needed to improve the estimation of the cure time.
Liste complète des métadonnées

Contributeur : Lnc - Université de Bourgogne <>
Soumis le : mardi 12 décembre 2017 - 11:20:48
Dernière modification le : dimanche 11 février 2018 - 01:14:29


  • HAL Id : hal-01661773, version 1



Olayidé Boussari, Gaëlle Romain, Morgane Mounier, Nadine Bossard, Laurent Remontet, et al.. A partial review of cure models with an application to French cancer registries data to improve patients' access to insurance and credit. SADA’ 2016, Nov 2016, Cotonou, Benin. 〈https://www.researchgate.net/profile/Castro_Hounmenou/publication/320486559_Parameter_estimation_in_nonparametric_nonlinear_mixed_effect_model_application_to_sparse_data_from_population_pharmacokinetic/links/59e808cca6fdccfe7f8b1b84/Parameter-estimation-in-nonparametric-nonlinear-mixed-effect-model-application-to-sparse-data-from-population-pharmacokinetic.pdf#page=20〉. 〈hal-01661773〉



Consultations de la notice