Cure Models for Estimating Hospital-Based Breast CancerSurvival

Abstract


Objective: Research on cancer survival is enriched by development and application of innovative analyticalapproaches in relation to standard methods. The aim of the present paper is to document the utility of a mixturemodel to estimate the cure fraction and compare it with other approaches.
Methods: The data were for 1,107patients with locally advanced breast cancer, who completed the neo-adjuvant treatment protocol during 1990-99at the Cancer Institute (WIA), Chennai, India. Tumour stage, post-operative pathological node (PN) and tumourresidue (TR) status were studied. Event free survival probability was estimated using the Kaplan-Meier method.Cure models under proportional and non-proportional hazard assumptions following log normal distribution forsurvival time were used to estimate both the cure fraction and the survival function for the uncured.
Results:Event free survival at 5 and 10 years were 64.2% and 52.6% respectively and cure fraction was 47.5% forall cases together. Follow up ranged between 0-15 years and survival probabilities showed minimal changesafter 7 years of follow up. TR and PN emerged as independent prognostic factors using Cox and proportionalhazard (PH) cure models. Proportionality condition was violated when tumour stage was considered and it wasstatistically significant only under PH and not under non PH cure models. However, TR and PN continued tobe independent prognostic factors after adjusting for tumour stage using the non PH cure model. A consistentordering of cure fractions with respect to factors of PN and TR was forthcoming across tumour stages using PHand non PH cure models, but perceptible differences in survival were observed between the two.
Conclusion: IfPH conditions are violated, analysis using a non PH model is advocated and mixture cure models are useful inestimating the cure fraction and constructing survival curves for non-cures.

Keywords