An Influence Diagram for Assessing GVHD Prophylaxis after Bone Marrow Transplantation in Children

Graft-versus-host disease (GVHD) represents one of the major complications of allogeneic bone marrow transplantation (BMT). Nevertheless, the occurrence of mild GVHD could be desirable in high-risk leukemic patients, due to a relapse-preventing effect known as the graft-versus-leukemia (GVL) effect. Given that different prophylactic interventions are avail able to prevent GVHD development, the decision problem consists in assessing both type and dosage of the drugs in order to avoid or induce GVHD, according to the specific patient's condition. The decision problem can be represented and solved by using an influence dia gram. The choice of this formalism allows using new available methods for building and updating the model of the decision problem. The qualitative structure of the model and the conditional probabilities were first derived by combining literature results with a medical expert's judgement. More specifically, probabilities were initially assigned as ranges rather than as point values. Then, conditional probabilities were updated, by using a learning algorithm, as new cases became available. The authors analyzed 50 cases of pediatric patients affected by either malignant or nonmalignant diseases, undergoing BMT and re ceiving GVHD prophylaxis. They used the first 25 cases to adjust the initially assigned conditional probabilities, then checked the model obtained on the remaining patients. The overall performance for GVHD prediction was about 80%. Key words: Bayesian belief net works; influence diagrams; therapy planning; graft-versus-host-disease. (Med Decis Making 1994;14:223-235)