| Abstract: |
Few health policy issues in the U.S. are scrutinized as intensely as the distribution of organs for transplantation, with much effort currently underway to examine the efficacy of existing organ allocation algorithms. The design of efficient organ allocation policies depends strongly upon accurate estimates of the survival benefit of transplantation. Many organ allocation policies order patients on the wait-list based on time-dependent health status measures (internal time-dependent covariates), such that priority is given to candidates at the greatest risk for wait-list mortality. It is of great interest to measure the transplant benefit by levels of such health status measures to identify the status levels that represent either futile or unnecessary transplants. In the presence of observational data, the survival benefit of transplantation has, to date, been quantified through the parameter corresponding to a binary time-dependent transplant indicator variable. Parameters from a standard time-dependent analysis using existing methods (i.e., separate transplant indicator for each status level) are difficult to interpret because they apply while the patient is at a particular level; i.e., they account for the patient's current condition, but do not account for the possibility that the patient's condition may worsen. We propose a novel method for estimating the effect of a time-dependent treatment by levels of an internal time-dependent covariate. The method yields parameter estimates which, rather than applying to the patient's current health status, average over future potential changes in health status. The proposed method is applied to end-stage liver disease data obtained from a national organ failure registry. © 2009 American Statistical Association. |
