UC Davis

The analysis of usual, or long-term average, intake of nutrients, foods, and other dietary components is of interest both for population health monitoring as well as for studies of diet-health relationships. Usual diet is not directly observable, given multiple sources of measurement error in dietary recall methods and the existence of day-to-day variation in dietary intakes, necessitating the use of statistical models to estimate population distributions of usual intakes or estimate epidemiological associations. A higher amount of bias is associated with food frequency questionnaires (FFQ) than 24-hour dietary recalls (24HDR), while the main source of measurement error in the latter is within-individual variation (WIV) in intakes. Biomarkers of dietary intake may address some of these issues but are subject to their own sources of measurement error, including WIV in some cases, but further research is needed to validate these biomarkers, particularly of food intakes. In order to accurately assess dietary adequacy at the population level and characterize diet-health relationships, the degree and nature of within-individual variability in dietary intakes and biomarkers should be understood and accounted for during the design and analysis stages of nutrition studies. The present studies address several gaps in current research in this area.The first study addresses the estimation of population usual nutrient intake distributions when only a single 24HDR or food record was collected per study participant. Determining the proportion of a population at risk of inadequate or excessive nutrient intake is a crucial step in planning and managing nutrition intervention programs. Multiple days of 24-hour dietary intake data per subject allow for adjustment of modeled usual nutrient intake distributions for the proportion of total variance in intake attributable to within-individual variation (WIV:total). When only single-day dietary data are available, an external adjustment factor can be used; however, WIV:total may vary by population, and use of incorrect WIV:total ratios may influence the accuracy of prevalence estimates and subsequent program impacts. In this study, we compiled WIV:total estimates from publications and from reanalyses of existing datasets to describe variation in WIV:total across populations and settings. The potential impact of variation in external WIV:total on estimates of prevalence of inadequacy was assessed through simulation analyses using the National Cancer Institute 1-day method. WIV:total were extracted from 40 publications from 24 countries, and additional values were calculated from 15 datasets from 12 nations. Wide variation in WIV:total (from 0.02 to 1.00) was observed in publications and reanalyses. Few patterns by population characteristics were apparent, but WIV:total varied by age in children (under vs. over 1 year) and between rural vs. urban settings. Simulation analyses indicated that estimates of the prevalence of inadequate intake are sensitive to the selected ratio in some cases. Thus, selection of an external WIV:total estimate should consider comparability between the reference and primary studies with regard to population characteristics, study design and statistical methods. Given wide variation of observed ratios with few discernible patterns, the collection of 2 or more days of intake data in at least a representative subsample in population dietary studies is strongly encouraged. In the case of single-day dietary studies, sensitivity analyses are recommended to determine the robustness of prevalence estimates to changes in the variance ratio. In the second study, the issue of WIV in a candidate dietary biomarker was examined in the context of an epidemiological pregnancy cohort. Proline betaine (Pro-B) has consistently been identified in biomarker discovery studies as a marker of citrus intake, yet gaps remain in understanding the nature of WIV of this largely short-term biomarker and its quantitative relationship with usual citrus intake in free-living populations, including in pregnant woman. The objective of this study was therefore to identify and quantify sources of within- and between-individual components of variance in urinary Pro-B concentration during pregnancy in a Northern California population and assess the correlation between repeated biomarker measurements and reported usual intake of citrus fruit and juice in this population. Up to 5 repeated spot or 24-h urine specimens throughout pregnancy (n = 255) from a nested case-control sample of women (n = 107) aged 22-47 years participating in the Markers of Autism in Babies: Learning Early Signs (MARBLES) were analyzed for Pro-B using 1H-NMR spectroscopy. Linear or log mixed effects regression models were used to quantify within- and between-individual variance components and test potential temporal predictors of continuous or elevated (i.e. > 100 µmol/L or > 30 µmol/mmol creatinine) Pro-B concentration, adjusting for covariates. The correlation between averaged repeated biomarker measures and usual citrus intake during pregnancy reported by Block food frequency questionnaires was assessed using Pearson or Spearman correlations. This population of pregnant women reported a median citrus intake of 0.32 (IQR, 0.50) servings/day and a median daily probability of consumption of 0.29 (IQR, 0.35). The overall proportion of samples with elevated Pro-B was 0.42 or 0.35 using a non-normalized or creatinine-normalized data threshold, respectively. The proportion of variance in Pro-B concentration attributable to WIV ranged from 0.69 to 0.74 in unadjusted and adjusted models of non-normalized or creatinine-normalized Pro-B. Citrus season was a significant predictor of urinary Pro-B in most analyses (p<0.05), while gestational age was a significant predictor only of non-normalized Pro-B. Moderate correlations were found between reported usual citrus intake and averaged repeated urinary Pro-B measurements, which were stronger compared to correlations using only a single measurement per person. In conclusion, a high degree of WIV in urinary Pro-B during pregnancy, which was partially explained by citrus season, and moderate correlations with usual reported citrus intake was observed. Multiple samples per participant are likely needed to account for random WIV and/or seasonal variation in citrus intake in studies assessing usual citrus intake at the population level or associations between citrus intake during pregnancy and maternal or child health outcomes. In the third study, the issue of WIV in a biomarker of oxidative stress was investigated in the same MARBLES subsample described above. Oxidative stress is associated with a number of medical pathologies, and excessive rises in pregnancy may be involved in adverse birth outcomes and neurodevelopmental challenges in offspring. F2-isoprostanes are a commonly measured biomarker of global oxidative stress produced from the free radical-mediated peroxidation of arachidonic acid on lipid membranes, but WIV in their synthesis and urinary excretion may affect their utility as a marker of usual oxidative stress exposure in epidemiological studies. This study aimed to quantify within- and between-individual components of variance in the 8-isoprostane urinary metabolite, 2,3-dinor-8-isoprostane (2,3-dinor-8-isop), investigate the relative contributions of gestation progression and diet to this variability, and determine the effect of the number of samples per person on the estimated association between prenatal urinary isoprostanes and autism spectrum disorder (ASD). Linear mixed effects models were constructed to assess variance components and potential temporal predictors, including gestational age and the diet-related urinary biomarkers hippurate and proline betaine, of urinary 2,3-dinor-8-isop from repeated spot or 24-hour urine measurements throughout pregnancy (n=254). The estimated association between child ASD and prenatal urinary 2,3-dinor-8-isop excretion was compared between models using 1) all available samples throughout pregnancy, 2) all available samples by trimester, and 3) trimester-specific analyses using 1 sample per participant. WIV constituted between 44% and 48% of total variance across repeated measurements of 2,3-dinor-8-isop, after accounting for variation in urinary creatinine. Corresponding intraclass correlation coefficients (ICCs) reflected moderate reproducibility. Gestational age was a significant predictor of creatinine-normalized 2,3-dinor-8-isop concentration but accounted for a small portion (1%) of total variance. Neither dietary biomarker was associated with 2,3-dinor-8-isop level. Isoprostane levels did not differ between pregnancies resulting in children with autism versus no developmental concern, regardless of the number of specimens per person included. In conclusion, moderate within-person variability was observed in a marker of oxidative stress during pregnancy, which is likely to cause some attenuation of risk estimates in epidemiological analyses. WIV in oxidative stress status and biomarkers during pregnancy should be further studied to properly account for this variation in studies investigating prenatal oxidative stress in relation to child outcomes. Overall, from these studies it can be concluded that the issue of within-person variability in dietary intake measures remains a significant challenge to usual dietary intake assessment and is not necessarily solved by the use of biomarkers, which also may be subject to substantial variability that may attenuate observed epidemiological associations. This issue also extends beyond dietary measures to other biomarkers, as exemplified in a marker of oxidative stress, which may be responsive to fluctuating environmental/biological exposures. Careful consideration of the degree and sources of WIV in dietary recall and biomarker measures, which may differ in pregnancy, is warranted during the design and analysis of studies of diet for population health monitoring and investigations of the impact of diet on health and neurodevelopmental outcomes.