Cohort effect in survival analysis: a change-point perspective

Olivier Bouaziz (1) , Grégory Nuel (1)

1 MAP5 - UMR 8145 - Mathématiques Appliquées Paris 5

In a survival analysis context we suggest a new method to treat the so-called cohort effet as a breakpoint problem where each stratum has its own baseline hazard function and proportional hazard parameter. Our approach combines various baseline hazard estimators (exponential, Weibull, piecewise constant, and non parametric) with the proportional hazard model into a EM-based segmentation approach using constrained HMM. The method provides an automatic procedure to estimate the hazard function in each stratum as well as the full posterior distribution of breakpoint locations. The new method is applied both to simulated data and to a diabetes cohort where it proves its consistence and usefulness.

Mots clés: Cohort effect; Constrained HMM; Cox model; EM; Survival analysis

Domaines: Applications [stat.AP]

Olivier Bouaziz, Grégory Nuel. Cohort effect in survival analysis: a change-point perspective. 2016. ⟨hal-01287075v1⟩