A two-tier proportional hazards regression model with change-points in the baseline function

Abdullah Oueslati (1) , Olivier Lopez (1)

1 LSTA - Laboratoire de Statistique Théorique et Appliquée

In this article, we consider a new regression model for counting processes under a proportional hazards assumption. This model is motivated by the need of understanding the evolution of the booking process of a railway company. The main novelty of the approach consists in assuming that the baseline hazard function is piecewise constant. Hence, the parameters of the model can be separated into two different types: parameters that measure the influence of the covariates, and parameters from a (multiple) change-point model for the baseline. Cox's semiparametric regression can be seen as a limit case of our model. We develop an iterative procedure to estimate the different parameters, and a test procedure that allows to perform change-point detection in the baseline. Our technique is supported by simulation studies and a real data analysis, which show that our model can be a reasonable alternative to Cox's regression model, particularly in the presence of tied event times.

Mots clés: proportional hazards regression; change-point detection; iterative procedures; dynamic programming; revenue-management; survival analysis; recurrent event data analysis

Domaines: Statistiques [math.ST]; Théorie [stat.TH]

Abdullah Oueslati, Olivier Lopez. A two-tier proportional hazards regression model with change-points in the baseline function. 2012. ⟨hal-00655891⟩