The complexity of disease processes is due in large measure to (a) their temporal evolution and (b) the multiplicity of manifestations. This is clearly evident in organ transplantation. Following listing as a candidate for receipt of an organ, the patient (1) may receive the transplant, (2) may die while waiting, (3) may be removed from the waiting list for other reasons, or (4) may still be waiting at the end of the period of observation. The events may all have different time courses. In addition, the fate of the patient is best summarized by survival from the time of listing, whether or not a transplant was received.

Our initial approach to the assessment of competing risks was to use multinomial logistic regression for a fixed time frame, an approach that required complete follow up in that interval. The analysis of survival following listing employed a fully parametric time to event model that allowed variation over time of the contributions of individual risk factors, the Bailey's modification of the Makeham model (Elandt-Johnson & Johnson, Survival Models and Data Analysis, pp. 350, 371 [Wiley, 1980]). To better deal with the time course of the competing risks, this model was generalized to accommodate competing risks.

We will describe the theoretical bases of the models employed, illustrate their application to liver and heart transplantation, and evaluate their performance.