Diagnosis of acute myocardial infarction from two measurements of creatine kinase isoenzyme MB with use of nonparametric probability estimation.

Abstract

By using bivariate probability estimation for the diagnosis of acute myocardial infarction (AMI) we show how to overcome the difficulties encountered for patients whose clinical presentation is atypical and those encountered when multiple isoenzyme determinations are treated by univariate methods. We use the values for creatine kinase isoenzyme MB measured at the time of admission and 12 h later to estimate the Bayes factors in favor of AMI. The Bayes factors are compiled into a table that the clinician can use to estimate the posterior probability that a patient has AMI. The table of Bayes factors is based on data for a sample of 802 non-AMI patients and 180 AMI patients. Further to validate the method, we randomly chose 200 of the non-AMI and 50 of the AMI patients as an evaluation sample, then used the remaining 602 non-AMI and 130 AMI patients to recompute the Bayes factors. These Bayes factors were used to find the probability of AMI for each of the 250 patients in the evaluation sample. The method resulted in only one false positive and no false negatives. For the misclassified patient the measurements at admission and 12 h later were 1 and 11 U/L; the posterior odds were 15 to 1 in favor of AMI, but in fact the patient was non-AMI.