(PO18-08-22) A Comparison of Methods to Adjust Risk Models for Measurement Error in Dietary Exposure

The ePoster content is available to NUTRITION 2022 LIVE ONLINE attendees who are logged in.<br><br>If you are registered, log in and then come back here to use this button.

Presenting Author(s)

Joy Hutchinson, MSc, RD
– PhD Student, University of Waterloo, Kitchener, Ontario, Canada

Co-Author(s)

Sharon I. Kirkpatrick, PhD, RD
– Associate professor, University of Waterloo, Waterloo, Alberta, Canada
MW

Michael Wallace, PhD
– University of Waterloo, Waterloo, Ontario, Canada
KD

Kevin W. Dodd, PhD
– Mathematical Statistician, National Cancer Institute, Rockville, Maryland, United States

Disclosure(s):

Joy Hutchinson, MSc, RD: No relevant financial relationship(s) with ineligible companies to disclose.

Objectives: To compare approaches to adjust for measurement error when exploring the relationship between the sodium-to-potassium ratio and hypertension and blood pressure.

Methods: Using National Health and Nutrition Examination Survey (NHANES) data from 2005-2010, Zhang et al. (2013) examined relationships between the sodium-to-potassium ratio (Na:K) and hypertension and blood pressure. They used the Iowa State University (ISU) Method to account for measurement error assuming the dietary intakes assessed with up to two 24-hour recalls (24HR) per person were unbiased for long-term (usual) intake. We replicated their analytic dataset (n=10467 adults ³20y). Systolic and diastolic blood pressure values were calculated as the mean of three readings; a binary variable for hypertension status was derived from blood pressure values and self-reported health information. Logistic (hypertension) and linear (blood pressure) models were adjusted for age, sex, and race/ethnicity. The primary exposure variable, Na:K, was obtained from calibration models that adjusted for random variation in the 24HR. The National Cancer Institute (NCI) Method used regression calibration from a joint model - fit with two methods, maximum likelihood and Markov Chain Monte Carlo (MCMC) - for sodium and potassium. A third model used the ISU Method to univariately model each component. All calibration models adjusted for the same covariates as the regression models plus nuisance effects. Standard errors for estimates and differences (e.g., MCMC - ISU) were computed using Balanced Repeated Replication to account for NHANES’ survey design.

Results: Both NCI implementations gave nearly equivalent results that sometimes differed from the ISU results. A one-unit increase in Na:K was associated with a mean systolic blood pressure increase of 2.82 mmHg, se .57 (ISU) vs. 3.87 mmHg, se .87 (MCMC) (t-test pdiff=0.005), and a mean diastolic blood pressure increase of 0.80 mmHg, se .43 (ISU) vs. 0.61, se .68 (MCMC) (pdiff=0.495). A one-unit increase in Na:K was associated with an increase of 0.41, se .11 (ISU) vs. 0.63, se .18 (MCMC) (pdiff=0.008) in the log odds of hypertension.

Conclusions: The choice of modeling approaches for 24HRs may affect estimated error-corrected relationships between health outcomes and dietary intake.

Funding Sources: Vanier Canada Graduate Scholarship