Abstract
BACKGROUND: A human diagnostician may harbour a special bias towards favourable positive or negative test results. The aim of the present analysis is to describe in quantitative terms how bias can affect the test characteristics of a human tester. METHODS: Whereas an unbiased tester would give absolute (100%) preference to true positive or true negative test results, and no (0%) preference to any false positive or false negative test results, a biased tester may harbour some preferences towards false positive or false negative tests. Such bias can be phrased in terms of a separate sensitivity-specificity matrix. The bias matrix multiplied with the original test matrix yields the biased test matrix. Similarly, the extent of ignorance by a human tester about the diagnosis is modelled as a separate sensitivity-specificity matrix, which captures the concordance between positive and negative diagnoses made by an ignorant and expert diagnostician. RESULTS: Increasing bias or ignorance result in decreasing test performance with decreasing positive predictive values until the test completely loses its discriminatory power. With more pronounced bias towards false test results, any positive test outcome may even become misinterpreted as predicting the non-existence of a given diagnosis. CONCLUSIONS: The proposed model helps to understand in quantitative terms, how bias and ignorance can alter a diagnostician's interpretation of test outcomes and result in diagnostic errors.