Cytotoxic Chemotherapy Monitoring Using Stochastic Simulation on Graphical Models

Abstract

This paper describes a Bayes Network approach to the modelling of growth or response curves, with application to the monitoring of cytotoxicity in breast-cancer patients under chemotherapy cycles. The approach uses experience of past cycles of therapy of the patient at hand to perform an adaptive adjustment of the parameters of a patient’s specific model of the toxicity evolution. This adaptive process allows more accurate patient-specific predictions, and hence a more rational dosage planning. We use a stochastic simulation algorithm, called Gibbs sampling, to perform the necessary inference calculations on the graphical model. We also describe result obtained on real data, using our newly developed GAMEES program for Gibbs sampling on Bayesian graphical models.