The water-soluble B vitamins folate, B-6 and B-12 are important cofactors in one-carbon metabolism, participating in methylation reactions and DNA synthesis (1). Vitamin C functions as a water-soluble antioxidant due to its high reducing power. Its best characterized function is in the synthesis of collagen connective tissue protein through the hydroxylation of proline and lysine residues of procollagen (2). While clinical deficiencies are rare for these vitamins, interest in these nutrients persists; “suboptimal” folate status is known to increase the risk of neural tube defects (3), vitamin C in combination with other supplements (vitamin E, zinc, and beta-carotene) has been shown to slow the progression of age-related macular degeneration (4), and “suboptimal” vitamin B or C status may modulate chronic diseases such as cardiovascular disease, cancer, and/or cognitive function (5,6). Furthermore, in the era of post-folic acid fortification, public health concerns are no longer limited to low folic acid intakes, but extend to the safety of high intakes (7), which are largely driven by supplement use (8).

Different biochemical indicators have been measured in the US population as part of the NHANES 2003–2006 to assess the status of water-soluble vitamins. The CDC’s Second National Report on Biochemical Indicators of Diet and Nutrition in the US Population (Second Nutrition Report) used these data to provide a descriptive analysis of the nutritional status of Americans by age, sex, and race-ethnicity. These analyses however, provide only limited interpretation of relative differences in nutritional status by demographic subgroup (9).

The relationship between diet and health is of great public health interest. Nutritional biomarkers are considered mediators of this relationship, avoiding reliance on biased self-reports of diet (10,11). Aside from diet, various genetic, biological, and lifestyle variables influence biomarkers. However, the associations between these variables and biomarkers are understudied. Data from national nutrition surveys in the United States, United Kingdom, Canada, and Mexico have shown that socioeconomic status and/or lifestyle variables are related to nutritional biomarkers (Table 1). Most of these studies were limited in scope, investigating the influence of 1 or few variables on 1 or 2 biomarkers (12–30). Few studies have assessed the relationship of multiple biomarkers with socioeconomic variables (31,32), smoking (33), alcohol consumption (34), or BMI (35,36). To our knowledge, no studies have examined the combined association of sociodemographic and lifestyle variables with biomarkers of water-soluble vitamin status.

To fill this knowledge gap and in order to examine whether demographic differentials in nutritional status found in the Second Nutrition Report were confounded by certain variables, we applied a systematic modeling approach to questionnaire and laboratory data from the adult US population participating in NHANES 2003–2006 to assess the association of 10 preselected sociodemographic (age, gender, race-ethnicity, education, and income) and lifestyle (supplement use, smoking, alcohol consumption, BMI, and physical activity) variables with 8 biomarkers of water-soluble vitamin status. These results will provide a foundation to researchers who develop predictive regression models addressing specific hypotheses. Companion publications in this journal supplement address the same questions for other biomarker classes featured in the Second Nutrition Report.

The NHANES, designed and carried out by the National Center for Health Statistics (NCHS) ⁵ at CDC, collects cross-sectional data on the health and nutritional status of the civilian non-institutionalized US population (37). Since 1999, it has been conducted as a continuous survey with data released in 2-y cycles. The 2003–2006 survey cycles obtained a stratified, multistage, probability sample designed to represent the American population on the basis of age, sex, and race/ethnicity. Data collection consisted of a screening visit, during which sample persons were identified; an interview during which a wide battery of health related questions were asked; and an examination consisting of direct standardized physical examinations, including body measurements and blood and urine collection, carried out in a mobile examination center. All respondents gave their informed consent, and the NHANES protocol was reviewed and approved by the NCHS Research Ethics Review Board. Interview and examination response rates for each survey period are publically available (38).

A description of the civilian non-institutionalized US population by the variables studied using NHANES 2003–2006 can be found in Supplemental Table 3. Most of the continuous variables (age, PIR, smoking, alcohol consumption, BMI, and exercise duration) were at best moderately significantly correlated (|r| ≤0.43) with the biomarkers of water-soluble vitamin status; MMA showed a moderate significant correlation with age only (r = 0.33) (Table 2). Based on the magnitude of the statistically significant Spearman correlations, age and smoking were generally the strongest correlates of biomarker concentrations, with levels increasing with increasing age and decreasing with increasing exposure to cigarette smoking (except for tHcy which was positively correlated to smoking).

Bivariate methods (model 1) were used to test for significant differences among variable categories. Of the demographic variables, age (except for PLP), sex (except for 4PA and B-12) and race-ethnicity (except for VIC) were significantly associated with most biomarkers, with age and race-ethnicity, separately, accounting for the largest variability in most biomarkers (Table 3). The socioeconomic variables education (except for B-12 and MMA) and PIR (except for B-12, tHcy and MMA) were also significantly associated with most biomarkers, but other than for PLP, they did not account for much of the variability in biomarker concentration. All 5 lifestyle variables were significantly associated with all biomarker concentrations, except for MMA, which was only significantly associated with alcohol consumption and physical activity (Table 4). Supplement use and smoking, separately, accounted for the largest variability in biomarker concentrations, while the other 3 variables explained only little of the biomarker variability.

In multiple regression models, the chunk of sociodemographic variables (model 2) explained up to 27% of the variability in biomarker concentrations: 2% (B-12), 6% (PLP), 8% (VIC), 13% (4PA and MMA), 14% (S-FOL), 15% (RBC-FOL), and 27% (tHcy) (Supplemental Table 4). Together, the chunks of sociodemographic and lifestyle variables (model 3) explained up to 29% of the variability: 7% (B-12), 15% (MMA), 22% (VIC), 23% (PLP), 25% (4PA), 26% (S-FOL and RBC-FOL), and 29% (tHcy). Adjusting for sociodemographic variables generally led to a mild attenuation of beta coefficients, while additionally adjusting for lifestyle variables more acutely diminished the association with sociodemographic variables, suggesting that sociodemographic variables may capture some unmeasured association that was shared with lifestyle variables.

Because the log transformations may obscure the interpretation of the beta coefficients, we estimated the percent change in biomarker concentrations (change in μmol/L for VIC which was not log transformed) associated with each covariable (Table 5). As noted with the beta coefficients, the estimated effect of most of these variables changed between models 1 and 3, suggesting that at least some of the association measured in the unadjusted model may be a result of confounding with variables not included in the model. For example, the estimated percent change for S-FOL concentrations for persons who were older by 10 y fell from 9.9% in model 1 to 6.8% in model 3, for women vs. men from 13.8% to 5.8%, and for NHB vs. NHW from −23.6% to −13.0%. Based on the full regression model 3, we observed significant associations for most biomarkers with age (8/8), sex (8/8), race-ethnicity (6/8 for NHB vs. NHW and 7/8 for MA vs. NHW), supplement use (8/8), smoking (7/8), and BMI (6/8); and for some biomarkers with PIR (5/8), education (1/8), alcohol consumption (4/8), and physical activity (5/8) (for a graphic representation, see also Supplemental Fig. 1–4). Age (being 10 y older) showed the strongest association with 4PA (15%), tHcy (10%), and MMA (9%); sex (being a women) with PLP (−21%), tHcy (−15%), and VIC (5.7 μmol/L); and race-ethnicity (being MA or NHB vs. NHW) with 4PA (−24% and −13%, respectively), MMA (−22% and −22%, respectively), and B-12 (20% and 15%, respectively). Supplement use and smoking both showed the strongest association with biomarkers of vitamin B-6 status (4PA 104% and −18%, PLP 79% and −28%, respectively), biomarkers of folate status (S-FOL 38% and −15%, RBC-FOL 24% and −12%, respectively), and VIC (16 and −11 μmol/L, respectively). Estimated vitamin concentrations in supplement users were up to twice as high compared to nonusers. As expected, the inversely correlated metabolites Hcy and MMA showed lower estimated concentrations in supplement users. Estimated vitamin concentrations in smokers were up to ~30% lower and tHcy concentrations were nearly 10% higher.

Using cross-sectional data for biomarkers of water-soluble vitamin status from a nationally representative sample of American adults participating in NHANES 2003–2006, we found that 1) age, sex, and race-ethnic differentials in biomarker concentrations remained significant, though the magnitude of the differentials was generally diminished after adjusting for key sociodemographic and lifestyle variables; and 2) of the variables studied, supplement use, smoking, and race-ethnicity were important correlates of biomarkers of water-soluble vitamin status, independent of the other sociodemographic and lifestyle variables in the model.

We used 3 different approaches to study the association between biomarkers and variables––correlations, bivariate regression, and multiple regression models and found good consistency across these approaches. Age and smoking emerged as the strongest individual correlates of the biomarkers. Using bivariate methods, age, race-ethnicity, supplement use, and smoking accounted for the largest portions of the variability in biomarker concentrations. Finally, using multiple regression models, age, sex, race-ethnicity, supplement use, and smoking continued to be significantly associated with nearly all biomarkers.

Our modeling estimated 79% higher PLP concentrations in supplement users compared to nonusers; this large difference can also be observed when comparing the prevalence of low PLP (<20 nmol/L) in supplement users (7.8%) compared to nonusers (19%) (data not shown). Morris et al. found similar prevalence estimates (11% in supplement users and 24% in nonusers) in NHANES 2003–2004 after adjusting for a similar list of variables plus self-reported diabetes status and intakes of protein and energy (19). Our analysis estimated S-FOL and RBC-FOL concentrations to be 39% and 24% higher, respectively in supplement users, similar to a recent NHANES 1999–2010 report (41% and 33% higher, respectively) (13). The larger proportional increase in S-FOL compared to RBC-FOL as a result of a folate dose––also noted after the introduction of folic acid fortification (13) and in response to long-term folic acid supplementation (47)—is likely a result of the much higher RBC folate concentration compared to serum. We estimated 21% higher B-12 concentrations in supplement users compared to nonusers, which is consistent with a prevalence of low B-12 (<200 ng/L) of 2.0% in supplement users compared to 3.1% in non-users (data not shown). Evatt et al. found a slightly bigger difference in low B-12 prevalence between users (1.7%) and nonusers (3.9%) for persons 18 y and older in NHANES III, however they assessed specifically B-12-containing supplement consumption (20). Our analysis estimated VIC concentrations to be 16.4 μmol/L higher in supplement users. Schleicher et al. found age-adjusted VIC concentrations to be 25 μmol/L higher in adults who consumed vitamin C-containing supplements as part of NHANES 2003–2004 (23).

Smoking was also significantly related in our analysis with most biomarkers, with smokers having lower vitamin concentrations compared to nonsmokers: VIC (~30%), PLP (29%), 4PA (18%), S-FOL (16%), and RBC-FOL (13%). Schleicher et al. reported 30% (men) and 33% (women) lower age-adjusted VIC concentrations for smokers vs. nonsmokers (23). Morris et al. showed that adjusted PLP concentrations of current smokers compared to those who never smoked were 25% and 22% lower in supplement users and nonusers, respectively (19). Using data from NHANES III, Mannino et al. found adjusted (sociodemographic variables and folate intake) RBC-FOL concentrations to be 16% lower in smokers compared to nonsmokers with low exposure to passive smoking (16). As expected due to the inverse relationship of tHcy with folate, vitamin B-6 and B-12, we found higher estimated tHcy concentrations (8%) in smokers. Similar observations were made in previous analyses of the US population (25,26,28,29). The 1994/1995 British National Diet and Nutrition Survey (NDNS) of people aged 65 y and over found an inverse relationship between smoking status and nutrient intake (VIC, B-6, and folate) and between smoking status and micronutrient indices (VIC, PLP, S-FOL, and RBC-FOL) after adjusting for intake and other covariates (33). Smoking itself may predispose to lower water-soluble vitamin status. In a recent analysis from two large Norwegian B vitamin intervention trials (NORVIT and WENBIT), Ulvik et al. showed that smoking status was directly associated with tHcy and inversely with S-FOL and PLP in a dose-response relationship (48). More interestingly, smokers with low serum cotinine (abstained from smoking for ≥3 d) had higher S-FOL and PLP concentrations compared to smokers with high serum cotinine. The authors suggested that the short-term effects may be related to acute smoking-induced oxidative stress; long-term effects among ex-smokers may reflect changes in diet and/or restoration of vitamin concentrations in tissue after smoking cessation.

Our analysis also showed a significant relation between race-ethnicity and most biomarkers. Compared to NHW, MA and NHB had lower S-FOL, RBC-FOL, 4PA, and MMA, but higher B-12 and VIC. Similar race-ethnic differentials were found in previous descriptive analyses of the US population (13,19,49), but also in an analysis by Kant et al. after they adjusted for socioeconomic status (and additionally for nutrient-specific intake) (32). The authors found lower folate intake and status in NHB, both pre- and post-fortification and suggested that ethnic-specific nutrition interventions would be needed to target at-risk ethnic groups and promote dietary changes. However, given the different frequencies of the MTHFR C677T genotype among the three major race-ethnic groups (50) and the fact that the BioRad radioassay—used in NHANES 1988–1994 and 1999–2006—responded differently to blood samples with the T/T genotype compared with either C/C or C/T genotypes (13), the association between RBC-FOL and race-ethnicity needs to be interpreted with caution. While we generally observed attenuation of the effect of race-ethnicity on biomarker concentrations with increasing adjustment, this was not the case for VIC. After adjusting for lifestyle variables, NHB had significantly higher VIC concentrations compared to NHW, whereas prior to adjustment the opposite was true, which may suggest confounding with at least 1 of the lifestyle variables, assuming the model is not misspecified.

Other sociodemographic or lifestyle variables assessed in this analysis had generally weaker and in some cases nonsignificant associations with biomarkers. After adjusting for sociodemographic and lifestyle variables, age was positively correlated with most biomarkers (negatively correlated with PLP) and women had better folate and vitamin C status, but lower vitamin B-6 and B-12 status, compared to men. Similar age and sex differences were reported in previous NHANES analyses (9,13,19,23,29,49,51). We confirmed previous findings of lower S-FOL (17,18,35,36) and higher RBC-FOL (18) concentrations with higher BMI. However, Tinker et al. only found BMI inversely associated with S-FOL among women who did not use folic acid-containing supplements, hypothesizing that cellular uptake and tissue distribution of folate may be altered by BMI which may be compensated by folic acid supplement use (18). Similar to findings by Walmsley et al. from the British NDNS (34), we also noted higher estimated PLP concentrations with higher alcohol consumption and no (RBC-FOL, 4PA, MMA, and VIC) or minimal (slightly lower S-FOL and B-12 and slightly higher tHcy) changes in all other biomarkers. This is expected based on results from a randomized intervention study of moderate alcohol consumption in postmenopausal women, which showed no (S-FOL, MMA) or small (B-12, tHcy) effects of 1 or 2 drinks/d over an 8-wk period (52).

To our knowledge, this is the first study that examined the association of demographic, socioeconomic, and lifestyle variables with all biomarkers available in the more recent continuous NHANES to interpret the status of 4 water-soluble vitamins: folate, vitamins B-6, B-12, and C. By applying a systematic modeling approach and limiting data driven decisions in the model building process we preserved the statistical properties of P-values and coefficients (46). Additionally, the hierarchical chunk regression modeling provided a natural way to systematically assess the magnitude of an estimated change in biomarker concentration with a change in a single covariate, holding all other variables constant, across biomarkers. Moreover, we applied the same approach to other classes of nutritional and dietary biomarkers allowing comparisons over a wide range of indicators (see other papers in this journal supplement; a summary table is presented in [46]). The large sample size in NHANES in combination with the use of 2 survey cycles that maximized the number of available biomarkers (i.e., tHcy, MMA, and VIC data are not available after 2006) allowed us to assess associations with a fair number of covariates in the same model.

Our analysis has limitations. The cross-sectional nature of NHANES prevented us from drawing any causal relationships between the biomarkers and variables in our study. Our results could be confounded by unmeasured biological and genetic factors. We did not test for interactions between variables due to limitations in degrees of freedom, nor did we maximize the predictive power of our descriptive model. Investigating nutrient-nutrient interactions (e.g., vitamin B-6 and protein intake, vitamin C and iron), studying how various health conditions or health risk factors are associated with nutritional biomarkers, or how dietary intake or intake of specific dietary supplements are associated with nutritional biomarkers or interact with variables included in our analysis was outside the scope of this study. Nutrient intake is known to be a major determinant of biomarker concentrations and both intake and biomarkers are indicators of nutritional status. We chose to describe how biomarkers were associated with certain variables after adjusting for sociodemographic and lifestyle variables and within that scheme dietary intake was more naturally an outcome variable than a covariate. A few studies have shown that associations of biomarkers with different variables remained unchanged after addition of the relevant nutrient intake to regression models (16,17,32,33). Regardless, our descriptive analysis cannot answer the question whether the associations we found are explained by intake or not. In summary, we conclude that supplement use, smoking, and race-ethnicity were associated with notable changes in concentrations of most biomarkers of water-soluble vitamin status, after adjusting for preselected sociodemographic and lifestyle variables. This analysis provides a foundation for future data analyses that set out to build predictive models to address specific hypotheses between nutritional status and health.