Early detection of necrotizing pancreatitis by computer.

In 59 patients admitted with a clinical picture compatible with acute pancreatitis, 88 features, recorded within the first day, were retrospectively collected and stored in a database programmed on a NCR DM V personal computer. On the basis of the final diagnosis, the patients were divided into two groups, according to Becker's grading of acute pancreatitis. A third group comprised the false-positives. Using 14 significant differences (p less than 0.025) between the three groups, a Bayesian analysis was programmed for early prediction, in terms of percentage probability, of each of the three final diagnoses. A retrospective assessment of computer performance was carried out in the patients entered the initial database, and further 38 underwent a prospective test. In the detection of necrotizing pancreatitis the procedure gave 95.8% accuracy, 91.9% sensitivity and 98.3% specificity. Previous diagnosis of pancreatitis, estimate of the degree of necrosis, data to be used for the processing, are discussed. Performances, similar in retrospective and prospective series, proved higher as compared to diagnostic peritoneal lavage. Comparison with Ranson's multiple criteria or computed tomography is inappropriate. The system seems to be reliable feasible and adaptable. The choice of treatment is difficult in acute pancreatitis. A computer-aided analysis may provide a better approach, than an unaided clinical estimate, to several decisional problems in acute pancreatitis.