Abstract:
|
In the United States, approximately 17,000 people are waiting for a liver transplant. These patients are prioritized for transplant according to medical need, as characterized by a composite score of liver functionality called MELD. The MELD score is an integer from 6 to 40, with higher scores indicating poorer liver function and hence higher short-term mortality in the absence of transplant. Estimating the precise relationship between MELD and mortality is challenging for a number of reasons: (1) survival times are censored, (2) censoring by transplant is informative of survival time in violation of a traditional assumption, (3) data are collected as repeated measurements, and (4) very high MELD scores are rare. I introduce a fully nonparametric approach for repeated-measurements survival data that combines bootstrapping and inverse probability censoring weighted Kaplan-Meier modeling. I use it to estimate the bias-corrected 90-day without-transplant survival probability (with confidence bands) as a function of MELD score. This function has an important role to play in determining exception scores, defining the pediatric MELD score, and optimizing geographic allocation districts.
|