Entropy and information content of laboratory test results.

Abstract

This article introduces the use of information theoretic concepts such as entropy, S, for the evaluation of laboratory tests results, and it offers a new measure of information, 1 - S, which tells us just how far toward certainty a laboratory test result can predict a binary outcome. The derived method is applied to the serum markers troponin I and prostate-specific antigen and to histologic grading of HER-2/neu staining, to cytologic diagnosis of cervical specimens, and to the measurement of tumor thickness in malignant melanoma. Not only do the graphic results provide insight for these tests, they also validate prior conclusions. Thus, this information theoretic approach shows promise for evaluating and understanding laboratory test results.