Abstract

The performance of diagnostic tests can be assessed by a number of methods. These include sensitivity, specificity,positive and negative predictive values, likelihood ratios and receiver operating characteristic (ROC) curves. This paper describes the methods and explains which information they provide. Sensitivity and specificity provides measures of the diagnostic accuracy of a test in diagnosing the condition. The positive and negative predictive values estimate the probability of the condition from the test-outcome and the condition's prevalence. The likelihood ratios bring together sensitivity and specificity and can be combined with the condition's pre-test prevalence to estimate the posttest probability of the condition. The ROC curve is obtained by calculating the sensitivity and specificity of a quantitative test at every possible cut-off point between 'normal' and 'abnormal' and plotting sensitivity as a function of 1-specificity. The ROC-curve can be used to define optimal cut-off values for a test, to assess the diagnostic accuracy of the test, and to compare the usefulness of different tests in the same patients. Under certain conditions it may be possible to utilize a test's quantitative information as such (without dichotomization) to yield diagnostic evidence in proportion to the actual test value. By combining more diagnostic tests in multivariate models the diagnostic accuracy may be markedly improved.