In this thesis we introduce semi-parametric models for dimension reduction in a standard censoring scheme. The model considered here is a single-index model used to prevent us from the curse of dimensionality. An appealing feature of these models is that they include classical models used in that kind of context as particular cases. Therefore, in the first part, we present a single-index model for the conditional density problem and in the second part, we study a single-index model for the recurrent event process. Both of these model can be seen as a generalisation of the popular Cox model. Moreover, in both work, we introduce a new procedure designed to circumvent the weak performances of Kaplan-Meier estimator in the right-tail of the dis...