Diagnostic testing: What do those results really mean? (Proceedings)
Diagnostic testing is an integral part of the practice of veterinary medicine but are all test results equal?
Diagnostic testing is an integral part of the practice of veterinary medicine – but are all test results equal? Let's consider what constitutes a clinical test. A clinical test can be anything that predicts health status better than chance. When applied to apparently healthy individuals for the purpose of early disease detection it is considered a screening test. When applied to individuals for the purpose of confirming or ruling-out a specific diagnosis it is considered a diagnostic test.
No test is perfect – so what we can do is characterize the “correctness” of a test through reference based evaluation (i.e., compare to a gold standard) or through non-reference based evaluation (i.e., latent class Bayesian analysis). Traditional reference based evaluations typically compare test results from two tests on two groups, one that is clearly diseased and one that is clearly non-diseased. While this does provide some insight into test performance, as from this test sensitivity and specificity can be estimated, these are not typically the patients to which tests are applied. Rather we use tests to help us determine the most likely differential for patients that fall somewhere in the middle – neither clearly diseased nor clearly non-diseased. Alternatively, tests can be characterized using non-reference based latent class analysis (i.e., no gold standard) to determine a more accurate (i.e., unbiased) estimate of test sensitivity and specificity.
Test sensitivity and specificity are generally considered to be constant and are evaluated before the test is applied. However, in practice what we really want to know is how to interpret test results after it is applied. This is the predictive value of a test – given a test result the probability of true disease status. The key here is that predictive values change – they are closely related to prevalence and cannot be applied beyond the population in which they were established. Thus it is important for practitioners to be aware of common diseases in their given practice region.
What if there is no single “best” diagnostic test? In this instance we can employ a multiple tests (i.e., 2 or more tests) and interpret them in series or in parallel. When testing in series, a patient must prove that it is truly diseased as any negative test is interpreted as the patient being negative. This will improve the overall specificity of the testing strategy and the positive predictive value. When testing in parallel the patient must prove that it is truly healthy as any positive test is interpreted as the patient being positive. This, then, will improve the overall sensitivity of the testing strategy and the negative predictive value. Choosing a test to apply depends upon the reason(s) for testing the patient and its perceived value. In general, using a highly sensitive test will identify all infected or diseased individuals, will help rule-out a particular condition, and will increase confidence in a negative result (e.g., early in the diagnostic process). Alternatively, using a highly specific test will help rule-in a particular condition and will increase confidence in a positive result (e.g., when you want to confirm a disease).
While diagnostic testing is a key component in patient management – not all tests are equal. Practitioners must be aware of test sensitivity and specificity, the expected prevalence of a condition in a given practice region, understand how prevalence can impact predictive values, and recognize that the value of diagnostic tests is highly variable among different diseases.