Early experiences with a program designed for the diagnosis of thoracic disorders

Abstract Studies of the reasoning processes of clinicians have led to proposals to use computers to and clinical decision making. The aims of such aids being to improve accuracy, consistency and cost-effectiveness of clinical decisions. The program described in this paper deals with the diagnosis of thoracic disorders in patients whose chest X-ray is substantially normal and who are complaining of one or more of the symptoms cough, shortness of breath, coughing up blood or chest pain. Standard observations such as the patient's age, height and weight and the responses to a few more detailed questions about the patient's presenting symptoms are recorded. The program provides a probability-based system dealing with 21 different diagnoses or a combination of two diagnoses. The probabilities of the occurrence of the different symptoms for the various diseases were initially allocated subjectively by an expert and have subsequently been modified by more objective experience. The prior probabilities of the different diseases were obtained from previous clinic records. The program uses a method of logical inference based on Bayes' theorem to calculate the possible diagnoses. The program serves two purposes. Firstly as an educational system for students, guiding them in taking a history and showing them a method of thinking more logically about diagnosis. Secondly the system is used as a diagnostic tool for new patients in the chest clinic, recommending simple tests to help confirm or deny the suggested diagnoses.