Following the approach proposed by Thierry Moreau in the framework of the Cox model ,we have developed a piecewise PH relative survival model, in which the disease-related hazard is expressed as a step function of the follow-up time.

The effect of the j-th covariate is represented by b _j + g _jk. As a consequence for a given covariate b _j the PH hypothesis can be tested by LRT with r-1 degrees of freedom. As the parameters g _jk are independent, using more than 2 time-segments allows estimating non-monotonic variations of the hazard ratio.