Using Bayesian networks to predict survival of liver transplant patients.

The relative scarcity of grafts available for liver transplantation highlights the need to identify patients likely to have good outcomes after treatment. We used transplant information from the United Network for Organ Sharing database to construct a Bayesian network model to predict 90-day graft survival. The final model incorporated a set of 29 pre-transplant variables, and it achieved performance, as measured by area under the receiver operating characteristic curve, of 0.674 by cross-validation and 0.681 on an independent validation set. The results showed a positive predictive value of 91%, while the negative predictive value was lower at 30%. With additional refinement and validation, our model may be useful as an adjunct to clinical experience in identifying patients most likely to have good outcomes following liver transplantation.