Expert system support using Bayesian belief networks in the prognosis of head-injured patients of the ICU.

Abstract

The present study concerns the construction and operation of a Bayesian analytical system, namely a Bayesian belief network (BBN) for the prognosis at 24 h of head-injured patients of the intensive care unit. The construction of a BBN incorporates the maintenance of a large database including all the critical variables corresponding to the specific clinical domain. This database is processed to provide the necessary libraries of conditional probability values. BBNs permit the combination of prognostic evidence in a cumulative manner and provide a quantitative measure of certainty in the final decision. The user views the changes at each step, thus being capable of deciding upon the necessary pieces of information in order to reach a certain belief threshold. The system produces results that are compatible with the opinions of medical experts regarding the prognosis of patients exhibiting certain patterns of clinical or laboratory data.