Assessment of clinical enzyme methodology: A probabilistic approach

Abstract

A number of techniques are available to assess the clinical value of enzyme methodologies including regression and discriminant analyses, expert systems, neural networks, and probabilistic. Each has its adherents but the probabilistic approach appears to be the most commonly used technique. This approach uses the fourfold contingency table that categorises subjects by both the presence or absence of the target disease - as defined by a gold standard test - and by a test result being above or below a chosen decision threshold. From this classification can be defined the test's sensitivity and specificity. By altering the decision threshold across the entire range of test values a series of sensitivity:specificity pairs can be tabulated. These may be plotted as 1 - specificity (or false positive rate or fraction) versus sensitivity (or true positive rate or fraction) to create a receiver operator characteristic (ROC) curve. ROC curves can provide the accuracy of the test (the area under the curve with associated confidence intervals), the rule-in and rule-out decision thresholds, and the clinical power of the test (likelihood ratio). However, a review of three years' publications in a peer-reviewed journal indicated that much of this essential data is usually absent. It is argued that such publications should include the decision thresholds used, the area under the curve and its standard error, a statistical assessment of the difference between two or more ROC plots, the rule-in and rule-out decision thresholds (indicating if these change with time after the onset of disease), and the relevant likelihood ratios.