|
Abstract:
|
EPA is interested in understanding the relationship between usual (long-term average) fish consumption and blood organic mercury concentration. The publicly available NHANES survey files have dietary intake data from two 24-hour periods and measurements of blood total mercury and inorganic mercury. Recent NHANES releases also have organic mercury measurements. For earlier releases, calculation of organic mercury concentrations from total and inorganic concentrations using subtraction gives some negative values and is complicated by non-detects. As a result, organic concentrations were imputed using a Bayesian model. The National Cancer Institute (NCI) provides SAS macros for calculating regression calibration estimates of usual fish consumption; however, those macros can be very slow when using many predictors and may not converge. To complete the calculations within a reasonable time, we have used various approximations when imputing the organic mercury concentrations and fitting the NCI-model. This presentation presents the modeling results and compares them to a Bayesian model that combines the estimation of usual fish consumption and imputation of organic mercury.
|
