Algorithmic Diagnosis of Jaundice

Abstract

Extensive clinical and clinical chemical information was collected from 1002 jaundiced patients. By applying Bayes' theorem and logistic discriminant analysis, a diagnostic algorithm was developed based upon 21 of the 107 variables collected. This algorithm permitted a probabilistic classification of jaundiced patients into four diagnostic categories: acute non-obstructive, chronic non-obstructive, benign obstructive and malignant obstructive jaundice. Of the 985 patients with a final diagnosis a correct probabilistic diagnosis (obstruction vs. non-obstruction) was suggested by the algorithm in 867 patients (88%). Adopting a probability limit of 0.80, 683 patients (69%) were correctly classified, 34 patients (3.5%) were wrongly so, and 268 patients (27%) could not be classified with a probability above 0.80 (doubtful cases). The algorithm was also tested in a further series of 110 jaundiced patients and found to perform equally well: 88 patients classified, 22 patients remaining doubtful. Patients with doubtful diagnoses should be referred to a non-invasive test such as ultrasound examination, whereas patients with definite diagnoses can be referred to invasive tests (liverbiopsy, direct cholangiography) as appropriate. The diagnostic algorithm seems to be a valuable aid for the preliminary differential diagnosis of the jaundiced patient and can be used in the planning of a diagnostic strategy for the individual patient.