Diagnostic accuracy of pancreatic enzymes evaluated by use of multivariate data analysis

Abstract

We analyzed pancreatic enzyme data from 508 patients with suspected pancreatitis by neural network analysis, by an Expert multirule generation protocol, and by receiveroperator characteristic (ROC) curve analysis of a single test result. Neural network analysis showed that use of lipase provided the best means for diagnosing pancreatitis. Diagnostic accuracies achieved by using amylase only, lipase only, and amylase and lipase in combination were 76%, 82%, and 84%, respectively. Use of the Expert rule generation protocol provided a diagnostic accuracy of 92% when rules for single and multiple samplings were combined. ROC curve analysis for initial enzyme activities showed the maximal diagnostic accuracy to be 82% and 85% for amylase and lipase, respectively; use of peak enzyme activities yielded accuracies of 81% and 88%, respectively. The evaluation of laboratory test data should include analysis of the diagnostic accuracy of laboratory tests by multivariate techniques such as neural network analysis or an Expert systems approach. Multivariate analysis should allow for a more realistic assessment of the diagnosis accuracy of laboratory tests because all the available data are included in the evaluation.