Using and Interpreting Diagnostic Tests with Dichotomous or Polychotomous Results

Abstract

In order to use and interpret diagnostic tests, we need to know the operating characteristics of the test (sensitivity, specificity, predictive values and likelihood ratios), and the prevalence of the disease. Sensitivity is the probability that a sick person tests positive, and a very sensitive test is useful when negative, to rule out the disease. Specificity is the probability of a healthy person to test negative, and a very specific test is useful when positive, to rule in the diagnosis. The positive predictive value is the probability that a person whose test is positive has the disease; the negative predictive value is the probability that a person whose test is negative is healthy. The predictive values depend on the prevalence of disease (pretest probability) so that, for the same predictive values, the probability of disease is higher in settings with a higher prevalence of disease than in those with a lower prevalence (Bayes’ rule). The likelihood ratios of a test, which may be calculated from its sensitivity and specificity, are stable for different prevalences, can deal with both dichotomous and polychotomous (multilevel) test results, and can also be used to calculate the post-test probability. For a valid diagnostic study, consider the standards for reporting of diagnostic accuracy (QUADAS) criteria.