Abstract:
|
The Division of Health and Nutrition Examination Surveys at the National Center for Health Statistics conducts internal method comparison studies whenever laboratories undergo instrumental or methodological measurement changes. Difference plots and Deming regressions guide decision making and if substantial differences are observed, then adjustment equations are released with the data for analysts planning to combine data from multiple survey cycles. This project aims to assess the robustness of the methodology by analyzing blinded pseudo-crossover-studies created from publicly available National Health and Nutrition Examination Survey data. Specifically, we address three questions: are adjustment equations overproduced, is it sufficient to rely on single measurements by each method versus duplicate measurements, and how are overall trend estimates influenced by the inclusion of adjustment equations? Preliminary results suggest that adjustment equations are not overproduced; that in most cases single measurements suffice, but duplicate measurements are less biased; and that the effect of adjustment equations on trends is minimal with some overly precise standard error estimates.
|