HOW ACCURATE IS COMPUTER-AIDED DIAGNOSIS?

A computer program for diagnosis of acute abdominal pain (CAD-A) was compared with an amended program (DIAG) and with clinical performance. The programs were tested at three Scottish hospitals, representing a range of practice, in 6962 cases. Accuracy of CAD-A lay in the range 48-59%, and of DIAG 56-62%. Clinical accuracy was constant at 65%. These figures suggest that computer systems based on Bayes' formula have no useful role in the diagnosis of acute abdominal pain.