(P38-10-26) What’s Left Behind: Examining Unaddressed Bias in Usual Intake Distributions Estimated With the National Cancer Institute Method

Objectives: Intake measurements derived from self-report (SR) dietary assessments are subject to systematic and random errors. Comprehensive adjustment for these errors is feasible only for the few nutrients (energy, protein, sodium, potassium) with corresponding recovery biomarkers (doubly labeled water, urinary nitrogen, sodium, and potassium), but adjustment solely for effects of random error can be done by statistical modeling of replicate measurements. We examined unaddressed bias in distributions of usual intake estimated via the National Cancer Institute (NCI) method.

Methods: The Interactive Diet and Activity Tracking in AARP study (IDATA, 1151 consented participants) administered up to 2 Food Frequency Questionnaires (FFQ), 2 4-day food records (FR), and 6 automated 24-h dietary recalls (24HR), plus up to 2 replicates of each biomarker mentioned above. We compared usual intake distributions derived from each SR instrument to those derived from the corresponding biomarkers, for the overall population and separately for the subpopulations with BMI < 25 and with BMI≥25. We computed the relative bias in percentiles, defined as (p^th percentile from SR – p<sup>th percentile from biomarker)/ p<sup>th percentile from biomarker, and averaged across percentiles of each SR-based distribution to obtain Average Relative Bias (ARB).

Results: ARBs (%, Men/Women) for energy were -14/-15, -25/-28, and -16/-17 for 24HR, FFQ, and FR. Average ARBs across the three non-energy nutrient amounts were -11/-3, -22/-15, and -15/-8. Average ARBs across the three nutrient densities were +4/+13, +7/+18, and +2/+10. When ARBs differed in the BMI-defined subpopulations, the BMI≥25 distribution showed a larger ARB. ARBs for selected low, middle, and high percentiles from SR-based distributions were often markedly different from each other. Even after accounting for random errors, biases in estimated prevalence of inadequacy/excess intake persist, differing by dietary component, personal characteristics, and relative positions of cutoff values in the true usual intake distribution.

Conclusions: These findings can inform sensitivity analyses for estimated usual intake distributions calculated via the NCI method for dietary components lacking recovery biomarkers, to help researchers gauge the robustness of future analyses.

Funding Sources: NCI (employer)