In recent years, Galen and Gambino have popularized the concept of predictive value, formulas based on Bayes’ theorem that help demonstrate the impact of disease prevalence on interpretation of laboratory test results (Table 1-1). Prevalence is the incidence of the disease (or the number of persons with the disease)in the population being tested. Briefly, predictive value helps dramatizethe fact that the smaller the number of persons with a certain disease in the population being tested, the lower will be the proportion of persons with an abnormal test result who will be abnormal because they have the disease in question (i.e., the higher will be the proportion of false positive results). For example, if test Y has a sensitivity of 95% and a specificity of 95% fordisease Z (both of which would usually be considered quite good), and if theprevalence of disease Z in the general population is 0.1% (1 in 1,000 persons), the predictive value of a positive (abnormal) result will be 1.9%. This meansthat of 100 persons with abnormal test results, only 2 will have disease Z, and 49 of 50 abnormal test results will be false positive. On the other hand, if the prevalence of disease Z were 10% (as might happen in a group of persons referred to a physician’s office with symptoms suggesting disease Z), the predictive value would rise to 68%, meaning that 2 out of 3 persons with abnormal test results would have disease Z.

Influence of disease

Table 1-1 Influence of disease prevalence on predictive value of a positive test result

Predictive value may be applied to any laboratory test to evaluate the reliability either of a positive (abnormal) or a negative (normal) result. Predictive value is most often employed to evaluate a positive result; in that case the major determinants are the incidence of the disease in question for the population being tested and the specificity of the test. However, predictive value is not the only criterion of laboratory test usefulness and may at times be misleading if used too rigidly. For example, a test may have excellent characteristics as a screening procedure in terms of sensitivity, low cost, and ease of technical performance and may also have a low positive predictive value. Whether or not the test is useful would depend on other factors, such as the type and cost of follow-up tests necessary in case of an abnormal result and the implications of missing a certain number of persons with the disease if some less sensitive test were employed.
There may be circumstances in which predictive value is misleading or difficult to establish. If one is calculating the predictive value of a test, one must first know the sensitivity and specificity of that test. This information requires that some accurate reference method for diagnosis must be available other than the test being evaluated; that is, a standard against which the test in question can be compared (a “gold standard”). This may not be possible. There may not be a more sensitive or specific test or test combination available; or the test being evaluated may itself be the major criterion by which the diagnosis is made. In other words, if it is not possible to detect all or nearly all patients with a certain disease, it will not be possible to provide a truly accurate calculation of sensitivity, specificity, or predictive value for tests used in the diagnosis of that disease. The best one could obtain are estimates, which vary in their reliability.