Comparison of Bayesian Network and
Decision Tree Methods for Predicting
Access to the Renal Transplant Waiting List
Sahar BAYAT
a, 1
, Marc CUGGIA
a
, Delphine ROSSILLE
a
, Michèle KESSLER
b
,
Luc FRIMAT
c
a
INSERM U936, Université Rennes 1, IFR 140, Rennes, France
b
Service de Néphrologie, Hôpitaux du Brabois, Nancy, France
c
EA 4003, Nancy Université, France
Abstract. The study compares the effectiveness of Bayesian networks versus
Decision Trees for predicting access to renal transplant waiting list in a French
healthcare network. The data set consisted in 809 patients starting renal
replacement therapy. The data were randomly divided into a training set (90%) and
a validation set (10%). Bayesian network and CART decision tree were built on
the training set. Their predictive performances were compared on the validation set.
The age variable was found to be the most important factor in both models. Both
models were highly sensitive and specific: sensitivity 90.0% (95%CI: 76.8–100),
specificity 96.7% (95%CI: 92.2–100). Moreover, the models were complementary
since the Bayesian network provided a global view of the variables’ associations
while the decision tree was more easily interpretable by physicians. These
approaches provide insights on the current care process. This knowledge could be
used for optimizing the healthcare process.
Keywords. decision support, decision tree, Bayesian network, healthcare network,
renal transplant waiting list
1. Introduction
Incidence and prevalence of End-Stage Renal Disease (ESRD) requiring Renal
Replacement Therapy (RRT) (hemodialysis, peritoneal dialysis or kidney
transplantation) continue to increase. Kidney transplantation is associated with longer
survival and lower long-term cost. But, given the graft shortage, transplantation is not
commonly achievable. On the other hand, the selection criteria of the potential
transplant recipient diverge from one centre to another. Ideally, placement on the
waiting list should be solely based on medical factors. Previous studies showed that
female gender, old age, distance from transplantation centre and private ownership of
dialysis facilities have been associated with poor access to kidney transplant waiting
lists [1, 2]. Previously [3, 4] we showed that access to the renal transplant waiting list
in NEPHROLOR, a French healthcare network, is primarily associated with age and
medical factors. The factors were identified by conventional statistical methods and
1
Corresponding Author: Sahar Bayat, Département d’Information Médicale (DIM), Hôpital Pontchaillou,
2 rue Henri Le Guilloux, 35033 Rennes, France; E-mail: sahar.bayat@chu-rennes.fr.
Medical Informatics in a United and Healthy Europe
K.-P. Adlassnig et al. (Eds.)
IOS Press, 2009
© 2009 European Federation for Medical Informatics. All rights reserved.
doi:10.3233/978-1-60750-044-5-600
600