On the use and computation of likelihood ratios in clinical chemistry

Abstract

The clinical relevance of likelihood ratios (L-values) for revising the physician's diagnostic probabilities has been recognized. However, the calculation of L-values, particularly in the case of quantitative or mixed quantitative-binary test results, raises problems that have not yet been addressed. Based on a very general assumption that yields a simple functional form for the likelihood ratio, a method is developed that allows such calculations regardless of the nature and the number of clinical laboratory tests to be interpreted simultaneously. Hence the notion of predictive value (posterior probability) is extended from binary or dichotomized tests to quantitative tests, and from univariate to multivariate clinical laboratory results. The simplicity and flexibility of this approach eliminates difficulties in computation arising from the addition of new data to an existing data base. It is hoped that this method will now allow L-values to be reported along with the original test results in daily laboratory practice.