Identifying prognostic markers of mortality in transplantation is essential for determining patients at high-risk of death and optimizing medical management. Nevertheless, an important part of the mortality may not be due directly to end-stage renal disease and/or kidney transplantation. Moreover, it is often impossible to individually determine the death causality, for example a cancer can be related to the chronic disease and immunosuppressive drug exposure or this cancer would have an independent link.

In survival analysis, one solution is to distinguish between the expected mortality of one general population (estimated on the basis of mortality tables) and the excess mortality attributable to the pathology, by using an additive relative survival model. The main objective of such models is to estimate the “net survival”, survival which would be observed if the only possible death is related to the transplantation.

We have adapted this concept in order to evaluate the capacity of a marker to predict the mortality attributable specifically to transplantation.

We illustrate this method of net time-dependent ROC curves with the analysis of kidney transplant recipients of the DIVAT (Nantes). From 1230 patients and by using the mortality tables of the general population, we have validated the score of Hernandez et al. (Transplantation 2009) to predict the all-cause mortality for a prognostic time at 10 years post transplantation (AUC=0.68, CI95%= [0.62, 0.74]). However, this score seems to not predict the mortality specifically related to end stage renal disease in the ten years post transplantation (AUC=0.65, IC95%= [0.56, 0.72]). We have modified this score by recalculating each variable weight. This modified score allows to do a better prognostic of the excess mortality (AUC=0.73, IC95%= [0.64, 0.80]).

« Back to 2013 American Transplant Congress