LAUSR

Well-defined diagnostic gold standards are rarely available for the diagnosis of rare diseases such as some tropical infections. Algorithms for reliable test evaluation without gold standard are therefore desirable. On the basis of previous work of Gart & Buck (1966), Rogan & Gladen (1978), and Hui & Walter (1980), a modeling approach for the comparison of diagnostic tests without a gold standard - and thus without reliably characterized reference sample material - is established especially for situations not fulfilling the requirement of conditional independence of the diagnostic tests evaluated. In the approach introduced, each test is conducted within two populations with different values of prevalence and the test results are used to define two virtual reference tests. The model is useful for the comparison of tests for rare tropical diseases where no gold standard can be provided.