Résumé :
|
[BDSP. Notice produite par INIST-CNRS R0xn7Qso. Diffusion soumise à autorisation]. Background : Unbiased estimation of the prevalence of diseases and other conditions is important but can be expensive, especially for conditions which do not necessarily lead to contact with health services. A two-phase population survey may seem an attractive option when there is a relatively cheap, although fallible, test for disease status available : the test is used in the first phase of the survey but in the second, only a subsample are classified by the relatively expensive, gold standard. Previously the cost efficiency of such studies compared with simple, one-phase random sample designs was investigated empirically and some questions remain unclear. Methods : A simple formula for the maximum reduction in cost or standard error that can be achieved by two-phase sampling compared with simple random sampling is derived mathematically. A formula for the minimum reduction is also given and the influence of prevalence on efficiency explained. Results : The main result shows that the sensitivity and specificity of the first stage test set an absolute limit on the efficiency of two-phase designs ; in particular, two-phase sampling can never be justified on efficiency grounds alone if the test is not accurate enough.
|