An expert system for assigning patients into clinical trials based on Bayesian networks

Assigning patients into clinical trials is a knowledge and data intensive task. Eligibility determination for entry into a clinical trial is based upon specific inclusion and exclusion criteria. This paper investigates the use of an excerpt system to assist the physician through this task. This expert system uses Bayesian networks, a probabilistic method that can take advantage of pre-existing statistical knowledge. The paper also describes the feasibility of such a system by presenting the implementation of three clinical protocols. The experimental results reveal that the approach is feasible. The system gives correct eligibility scores when all evidence is available but also predicts eligibility when there is missing evidence. The system directs the physician to the protocols the patient is most eligible for, according to the current evidence. The system has the ability of learning its prior and conditional probabilities (expert knowledge) from training examples.