[Estimation by the Hui and Walter method of the sensitivity and specificity of a diagnostic test in the absence of a reference test: results of a simulation study].

Hui and Walter proposed an approach to directly estimate the parameters of diagnostic tests (sensitivity and specificity for dichotomous tests) and the prevalence of the disease in the absence of a reference error-free test, and without imposing any constraint. We performed a simulation study to investigate the finite-sample properties of the proposed estimators for the situation of three tests. We obtained the following results: i) in screening situations (prevalence < or = 0.10), severe bias and poor precision in the parameter estimates occur and this approach should not be used except for studies with a very large sample size (n > or = 2500); ii) in diagnostic situations (prevalence > or = 0.30), the approach is much more reliable and very little or no bias occurs even for small sample sizes (n = 50, 100), except when both the prevalence is very high (p > or = 0.90) and the performances of the tests mediocre (sensitivity and specificity lower than 0.80).