Differential diagnosis of jaundice: a pocket diagnostic chart

Based on extensive clinical and clinical chemical information (107 different items) from 1002 jaundiced patients, we developed a diagnostic algorithm which was evaluated on a test sample of another 110 jaundiced patients. A primary classification into categories of obstructive jaundice (probability of obstruction greater than or equal to 0.80), non-obstructive jaundice (probability of obstruction less than or equal to 0.20), and of doubtful causes of jaundice (probability of obstruction: 0.20-0.80) was attempted. Among 234 patients in the data base who were classified as obstructive, 220 (94%) proved to be so, as did 36 (97%) of 37 in the test sample. The corresponding figures for non-obstructive jaundice were 463 (96%) of 483 patients correctly classified in the data base and 47 (92%) of 51 patients in the test sample. Altogether 69% of the patients in the data base and 75% of those in the test sample were correctly classified, in 27% and 20% the cause of jaundice was doubtful, and only 4% and 5%, respectively, were misclassified. A slight majority of the patients in whom the algorithmic diagnoses were doubtful proved obstructive. A close correlation was found between the preliminary diagnoses made by the algorithm and by the clinicians. A secondary classification of the patients by the algorithm into benign versus malignant causes of obstructive jaundice performed equally well in the data base and the test sample.