Comparing diagnostic tests with missing data

摘要: When missing data occur in studies designed to compare the accuracy of diagnostic tests, a common, though naive, practice is base comparison sensitivity, specificity, as well positive and negative predictive values on some subset that fits into methods implemented standard statistical packages. Such are usually valid only under strong completely at random (MCAR) assumption may generate biased less precise estimates. We review models use dependence structure observed cases incorporate information partially categorized observations analysis show how they be fitted via two-stage hybrid process involving maximum likelihood first stage weighted least squares second. indicate computational subroutines written R used fit proposed illustrate different strategies with observational collected thre...