Use of a Bayesian algorithm in the computer-assisted diagnosis of appendicitis.

One hundred consecutive patients with acute right lower quadrant abdominal pain were prospectively evaluated with a computerized Bayesian diagnostic algorithm. An accuracy rate of 92 per cent was obtained. Computer recommendations would have resulted in a negative exploration rate of 9 per cent, as compared with the rate of 19 per cent which was actually obtained. Even though our clinical management of these patients was in keeping with accepted standards, the Bayesian program would have avoided eight unnecessary operations. In all instances in which the patient presented with appendicitis, the computer correctly predicted that appendicitis was present. Computer-assisted diagnostic programs using a Bayesian approach may have some role in the evaluation of right lower quadrant abdominal pain. The technique presented herein describes a means of developing a database of conditional probabilities without reliance on large patient surveys. Even with this refinement, the Bayesian approach to diagnosis remains complex. The development of this type of program requires close interaction between computer scientists and surgeons. Nevertheless, the approach does appear promising and it may well be worth the considerable effort required to initiate such a system. The exact role for Bayesian diagnostic analysis cannot be predicted at this point. Certainly it should have no greater importance than a routine laboratory test. Perhaps the results of Bayesian analysis in this setting might assume a diagnostic significance similar to that of the white blood cell count. The work of DeDombal has done much to eliminate the physician reluctance seen with earlier programs. It has become increasingly apparent that computers may perform many clinically useful functions without infringing upon the art of medicine. The computer assisted diagnosis of acute abdominal pain may well constitute one such function.