Does test quality influence the outcome of algorithmic classification of jaundiced patients?

A simple diagnostic algorithm for preliminary classification of jaundiced patients on the basis of 12 clinical and clinical chemical variables included s-alkaline phosphatases, s-aspartate aminotransferase, and s-bilirubin. When reclassifying 985 finally diagnosed jaundiced patients from our data base nearly 70% of the patients were correctly classified and only 3.5% were misclassified as obstructive or nonobstructive cases, whereas 27% were classified as having a doubtful cause of jaundice. In order to test the possible dependence of the algorithm on the analytical quality of the clinical chemical data we simulated inaccuracies of ± 40% for s-alkaline phosphatases, ± 20% for s-aspartate aminotransferase, and ± 25% for s-bilirubin and increased imprecisions (CV%) from 4 to 14, 4 to 12, and 3 to 8, respectively. These major pertubations, however, had only very little effect on the performance of the algorithm, which lead us to the conclusion, that preliminary classification of jaundiced patients is quite insensitive to analytical error.