Abstract:
|
Grouping observations is used to yield cost efficient prevalence estimates of diseases/biomarkers, and to protect confidentiality of participants' health status. For simple random samples, statistical methods have been developed for estimating prevalences of a single trait and of more than one trait from only group tested samples. However, we believe there has not been thorough research into accounting for complex sample designs with group testing. We employed sample weighted-generalized estimating equations for estimating the prevalences of categorical traits when the study samples are from complex sample designs. We also incorporate the possibility of misclassification in the testing of the traits and allow for simultaneous testing of multiple traits from the same grouped samples. We consider five different grouping methods to address sample weighting and cluster sampling aspects of complex designed samples, and we study analytically and empirically through simulations and a National Health and Nutrition Examination Survey (NHANES) data example the statistical properties of prevalence estimation and their variance estimation along with resulting confidence interval coverage.
|