Clinical Trial of a Computer Diagnosis Program for Chest Pain

摘要

We evaluated the performance of a computer program designed to assist corpsmen inmanaging patients who present at sea with chest pain. The diagnostic accuracy of the program for three common and serious causes of chest pain--myocardial infarction (MI), angina, chest infection--as well as its accuracy for nonspecific chest pain, was compared with that of emergency room physicians at three Naval hospitals for 132 patients. The percentage of correct diagnoses by the computer program (72%) was not significantly different from that of the physicians (79%), although the accuracy of the physicians was better. The computer program was able to distinguish MI from the other diseases as well as the physicians (i.e. was as sensitive, 83% vs. 84%) and at the same time made fewer false diagnoses of MI (i.e. was more specific). The computer's ability to distinguish among cardiac illnesses was not as good as that of the physicians and it misdiagnosed more cases of angina. The computer program is suitable for use by medical personnel practicing in isolated locations and may help improve diagnostic accuracy in cases of MI. Its failures emphasize that it cannot substitute for medical personnel, but can be helpful to medical personnel when a patient's diagnosis is uncertain.