Physicians' diagnoses compared with algorithmic differentiation of causes of jaundice.

Clinical data were collected in 194 cases of jaundiced patients treated at the "Adolfo Lopez Mateos" ISSSTE Hospital in Mexico City from July 1985 to July 1986. A copy of the clinical history of each patient was given to each of four physicians--one recently graduated from medical school, another in his first year of gastroenterology, and two others who were experienced gastroenterologists. The same clinical data were processed by a computer set up to use a modified Danish COMIC algorithm. All physicians and the computer technician were blinded to the "gold standard" pathologic diagnoses, with which their diagnoses were compared. Accuracy rates of the physicians in distinguishing intrahepatic (medical) from extrahepatic (surgical) jaundice were 78%, 86%, 86%, and 91%, and the accuracy of computer-assisted diagnoses was 96%. Chi-squared analysis of the diagnoses of three of the physicians and those of the computer showed significant differences (p between 0.1 and 0.01). For the diagnoses of the remaining physician, however, no significant difference was found after chi-squared continuity correction.