In this paper we analyse the problem of the modelling of individual transitions in presence of an incomplete sampling scheme.

We analyse the model in which the latent durations Ti are i.i.d. generated by a distribution F. The statistician observes Yi= min (ti, Ci) and Ai=1I {Ti <= Ci} where ci is a censoring time. The prior probability on F is a Dirichlet process Di (n0F0).

This paper proposes to use a semi parametric regression method, named Sliced Inverse Regression (SIR hereafter), to analyse ambulatory blood pressure monitoring data.