|
Abstract:
|
The National Health and Nutrition Examination Survey (NHANES) is often used to monitor trends in various nutritional biomarkers. While NHANES tries to maintain the same biomarker assay over time, there are times when an assay may change. To minimize the impact these assay changes on fluctuations of a biomarker’s concentration over time, “bridging studies” have been designed to develop statistical models to facilitate trend assessment. The statistical models range from simple regression "adjustment equations" to multiple imputation models. Using a combination of simulations and real data, this paper explores the conditions under which the various statistical models may lead to reasonable statistical inferences concerning biomarker concentration trends and whether aspects of the bridging study design can mitigate the problems of the simpler statistical adjustment models.
|
