No abstract

No abstract

We show how to make an unsupervised discrimination of disease and nondisease states by measuring information and using newer notions of inductive reason. We also present a new theory of group-based reference values that is based on measuring information uncertainty. We use data on the isoenzymes creatine kinase-MB (CK-MB) and lactate dehydrogenase-1 (LD1) and on the percentage of LD1 from 101 patients with acute myocardial infarction (AMI) and from 41 patients with suspected, but unfounded, infarction (non-AMI). Calculating the Shannon entropy, a concept from information theory, of the data base allows determination of a difference in entropy values ("effective information"), which determines decision cutoff values that produce binary-base patterns yielding the fewest classification errors. Redundancy in testing is important because it provides the information to approach a goal of errorless discrimination by coding the test results and meeting the conditions of the "Noisy Channel Theorem" of information theory. This redundancy improves the predictive value of diagnosis by isolating the area of equivocation to evident patterns. Results for CK-MB and LD1 are 99% correct in assigning cases to AMI and non-AMI categories; adding %LD1 increases the proportion of errorless binary patterns from 25% to 90%.