Relations between effectiveness of a diagnostic test, prevalence of the disease, and percentages of uninterpretable results. An example in the diagnosis of jaundice.

The relations between effectiveness, the percentages of uninterpretable results of a test, and the prevalence of the disease are studied in the example of the diagnosis of jaundice. Ten hepatologists and ten hepatobiliary surgeons were interviewed, and nineteen articles were reviewed. Accuracies of ultrasonography, endoscopic retrograde cholangiography, and transhepatic cholangiography, as well as of three strategies combining these tests, were ranked by hepatologists in an order that differed from chance, and by surgeons in an order that did not differ from chance. Analyses of published data, taking into account the percentages of uninterpretable results, showed that for a high prevalence of extrahepatic cholestasis, as in jaundiced patients seen by surgeons, there is no significant difference between the respective effectiveness of each test or strategy. We concluded that effectiveness must take into account the percentages of uninterpretable results and must be expressed as a function of prevalence.