Logo Logo
Switch Language to German
Schneider, Micha (6. August 2019): Dealing with Heterogeneity in Discrete Survival Analysis using the Cure Model. Department of Statistics: Technical Reports, No.224


Cure models are able to model heterogeneity which arises from two subgroups with different hazards. One subgroup is characterized as long-term survivors with a hazard equal to zero, while the other subgroup is at-risk of the event. While cure models for continuous time are well established, cure models for discrete time points are rarely prevalent. In this article I describe discrete cure models, how they are defined, estimated and can be applied to real data. I propose to use penalization techniques to stabilize the model estimation, to smooth the baseline and to perform variable selection. The methods are illustrated on data about criminal recidivism and applied to data about breast cancer. As one result patients with no positive lymph nodes, a very small tumor, which can be well differentiated from healthy cells and with ethnicity which is neither black or white have the best estimated chances to belong to the long-term survivors of breast cancer.