点击显示 收起
1 From the Division of Nutrition and Physical Activity, National Center for Chronic Disease Prevention and Health Promotion, and the Division of Laboratory Sciences, National Center for Environmental Health, Centers for Disease Control and Prevention, Atlanta.
2 Address reprint requests to C Ballew, CDC Division of Nutrition and Physical Activity, MS K-26, 4770 Buford Highway NE, Atlanta, GA 30341. E-mail: ckb2{at}cdc.gov.
ABSTRACT
Background: Inadequate vitamin A status has been a potential nutritional problem for some segments of the US population, particularly children and the poor.
Objective: We evaluated serum retinol concentration by using population-representative data from 16058 participants aged 4 to 90 y in the third National Health and Nutrition Examination Survey, 1988–1994.
Design: We used multivariate regression to examine the simultaneous associations of sociodemographic, biologic, and behavioral factors with serum retinol concentration.
Results: In children, serum retinol concentrations were greater with greater age, body mass index, serum lipids, and the use of supplements containing vitamin A. In adults, male sex, serum lipids, alcohol consumption, and age were positively associated with serum retinol concentration in most racial/ethnic strata. Household income was not associated with serum retinol concentration in children; associations were inconsistent in adults. The prevalence of serum retinol <0.70 µmol/L was very low in all strata; the prevalence of serum retinol <1.05 µmol/L was 16.7–33.9% in children aged 4–8 y and 3.6–14.2% in children aged 9–13 y, depending on sex and racial/ethnic group. The prevalence of serum retinol<1.05 µmol/L was higher in non-Hispanic black and Mexican American children than in non-Hispanic white children; these differences remained significant (P < 0.0001) after covariates were controlled for. Among adults, nonwhite women were significantly (P < 0.0001) more likely than white women to have serum retinol <1.05 µmol/L after covariates were controlled for.
Conclusions: Clinically low serum retinol concentration is uncommon in US residents aged 4 y, although racial/ethnic and socioeconomic differences in serum retinol concentration still exist.
Key Words: Adults • children • females • males • blacks • whites • Mexican Americans • vitamin A • serum retinol • sociodemographic factors • multivariate analysis • NHANES III • third National Health and Nutrition Examination Survey • United States • prevalence
INTRODUCTION
Vitamin A deficiency has been a nutritional problem for some segments of the US population in the past, particularly children and the poor (1–5). Low vitamin A intakes continue to be a potential public health issue (6). The third National Health and Nutrition Examination Survey 1988–1994 (NHANES III) provides data that are critical for examining the distribution of serum retinol concentrations in the US population, for estimating the prevalence of low serum retinol concentration, and for exploring some of the sociodemographic and lifestyle factors associated with variation in these concentrations.
SUBJECTS AND METHODS
Sample and measurements
NHANES III was a complex, stratified probability sample intended to be representative of the noninstitutionalized population of the United States. Detailed descriptions of the planning, ethical approval, and conduct of the survey have been published (7). Our analysis is based on public-release data tapes (8–10). Blood was obtained by venipuncture from all participants (11). Serum retinol was measured with use of an isocratic, reversed-phase HPLC (Waters Chromatography Division, Milford, MA). Analytic protocols for other serum analytes and laboratory quality-assurance procedures were described elsewhere (9, 12).
The NHANES III household interview included probes for the multiple use of dietary supplements. Participants were asked to list all supplements that they took in the 30 d before the interview. Brand names were recorded and, if possible, bottles were examined to determine the nutrient content of each supplement reported. Supplement products were linked to a nutrient content database (10). We calculated the total supplemental vitamin A intake (preformed retinol + ß-carotene/6) from supplements in µg retinol equivalents (RE)/d for each participant. Participants who did not take supplements containing vitamin A, 72% of the sample, were assigned a value of 0 for this variable. The range of total vitamin A obtained from supplements was 10–148624 µg RE/d. We categorized supplemental vitamin A intake as 0, 1–999 µg RE/d (5% of the sample), or 1000 µg RE/d (23%).
We did not include dietary intake of vitamin A in our analysis because participants provided only a single 24-h dietary recall. Preformed retinol and carotenoids with provitamin A activity are distributed unevenly in the diet. Although a single day of recall from a large sample may be used to characterize the population distribution of intakes, many days of recall are required to estimate an individual's usual dietary vitamin A intake accurately (13, 14). Software is available to compute the estimated usual intake of nutrients based on the ratio of inter- to intraindividual variation in intake derived from participants who provide more than one day of recall (15, 16). However, after evaluating the results obtained by applying the software to the 2 d of data provided by 5% of the participants and after consultating with one of the developers of the software (personal communication, A Carriquiry, 1999), we concluded that it was not possible to compute an accurate estimate of usual dietary vitamin A intake for individual participants in NHANES III.
The NHANES III laboratory sample for serum retinol analysis included 22940 individuals aged 4 to 90 y. We excluded participants who had conditions at the time of the examination that were temporary, reflected abnormal physiologic status, and would alter serum retinol concentration. We also excluded participants with missing data for covariates known to affect serum retinol concentration [alcohol use, smoking, estrogen use, serum lipids, body mass index (BMI)], and household income. Of participants with serum retinol measurements, 1279 had serum biochemistry values indicative of possible acute or chronic disease that could affect serum retinol concentration (elevated liver enzyme activities or C-reactive protein or the presence of antigens indicating active hepatitis), 4635 were missing data for serum biochemistry variables, 323 women were pregnant, 183 women were missing data for pregnancy status or estrogen use, 3094 participants were missing data on alcohol use or smoking status, 2042 were missing household income, and 169 were missing data on serum lipids or BMI. Some participants met more than one exclusion criterion. Our final sample for this analysis was 16058 subjects. Participants were classified in the data set as non-Hispanic white, non-Hispanic black, Mexican American, or other (including respondents of non-Mexican Hispanic, American Indian–Alaska Native, Pacific Island, and Asian origin). We included participants of other racial/ethnic origins in the tabulations of total population values but there were too few participants of other origins (n = 736 after all other exclusions) to allow subgroup analysis (7).
Alcohol use was self-reported in participants aged 12 y. We calculated the average number of drinks per week from the reported number of days each participant consumed alcohol and the usual number of drinks per occasion. One drink was defined in the questionnaire as 12 oz (408 mL) beer, 4 oz (204 mL) wine, or 1 oz (51 mL) of liquor. Serum cotinine was measured in all participants aged 4 y. We used serum cotinine as an index of smoking behavior (17). Cotinine quantifies tobacco exposure and captures cigar and pipe smoke as well as cigarette smoke exposure (18, 19). A serum cotinine concentration >85 nmol/L (>15 µg/dL) is likely to indicate active smoking rather than passive exposure (17). Women were classified as taking exogenous estrogens if they reported using birth control pills or hormone replacement therapy in the 30 d before the examination.
The poverty-income ratio (PIR) is based on total household income adjusted for household size and the cost of living at the time of the survey (7). It is used as an eligibility criterion for participation in federal and state economic assistance programs and as an index of relative socioeconomic status in the NHANES III survey (7). We used 3 PIR strata: <1.3, 1.3–3.5, and >3.5. In this analysis, education of adult participants and education of the head of the household for children proved to be highly correlated with PIR so we used only PIR as a marker of socioeconomic status.
We set critical cutoff points for serum retinol concentration at <70 µmol/L (<20 mg/dL), generally accepted as indicating likely inadequate vitamin A status, and <1.05 µmol/L (<30 µg/dL), interpreted as possibly responsive to greater intake of vitamin A (3, 20). The age strata used in this analysis are those currently used by the Institute of Medicine for the new dietary reference intakes (21): 4–8, 9–13, 14–18, 19–30, 31–50, 51–70, and >70 y. Serum retinol was not measured for participants aged <4 y in NHANES III.
Statistical analysis
We performed all analyses with SUDAAN software (release 7.5, 1997) to take into account the complex weighted and stratified nature of the sample (22). Values could not be computed for some cells and some computed values were statistically unreliable because of small cell size (<30) or large CVs (>30% of the point estimate) (7, 22). Statistically unreliable values are reported but flagged, consistent with the reporting conventions of the National Center for Health Statistics (7).
We did not perform univariate tests of significance of differences in serum retinol concentration by sex, racial/ethnic group, or age stratum for 2 reasons. First, because of the large sample in NHANES III, differences of small magnitude achieve substantial statistical significance. Second, potentially confounding variables were not distributed equally among the strata. Because supplement use, PIR, exogenous estrogen use, smoking status, serum lipids, and BMI have been reported to affect serum retinol concentration and because these factors varied by sex, age, and racial/ ethnic stratum in the NHANES III sample, we examined the association of these variables with serum retinol concentration by using multivariate regression analysis. PIR and supplement use were included in all models. Of the other independent variables, only those that achieved the P < 0.01 level of statistical significance were retained in the models. We used multiple linear regression to describe the simultaneous relations among sociodemographic and lifestyle factors and serum retinol concentration and to control for known or suspected confounders (23). We used multiple logistic regression to identify characteristics predicting risk of serum retinol <1.05 µmol/L (<30 µg/dL) (23).
We tested the continuous independent variables for linear relation with serum retinol by examination of residuals (23). The association of serum retinol concentration with age was linear but associations with serum lipids, BMI, and serum cotinine were not. Various transformations did not render the associations linear but rank ordering of serum lipid values, BMI, and serum cotinine values produced linear associations with serum retinol concentration.
We tested the independent variables for colinearity (23) and found it to be negligible. We tested for interactions among the major independent categorical variables (sex and racial/ethnic group) and the other independent variables in the multivariate regression models (23). We found a significant sex x age interaction (P < 0.001) in adults but not in children aged <14 y. This suggested that we should perform separate analyses for children and adults. In addition, some girls aged 14–18 y reported taking birth control pills; in children aged 14–18 y, the 95th percentile of cotinine was >85 nmol/L, indicating likely active smoking; and some children aged 14–18 y reported drinking alcoholic beverages regularly. Therefore, we analyzed participants aged <14 y, sexes combined, excluding the very few users of tobacco, alcohol, or birth control pills in this age group (n = 29). We analyzed males and females aged 14 y separately; these models included self-reported alcohol use and serum cotinine concentration for both sexes and estrogen use for females. In both males and females, we found significant interactions among racial/ethnic group and several of the other independent variables so we performed separate regression analyses for each of the 3 racial/ethnic groups by sex.
RESULTS
For all sex, age, and racial/ethnic strata, the 5th percentile value of serum retinol was >0.70 µmol/L (20 µg/dL) (Table 1). Eight strata had 1st percentile values <0.70 µmol/L but only 5 of these were statistically reliable estimates. For children aged <14 y, BMI, serum total cholesterol, serum triacylglycerols, and age were positively associated with serum retinol concentration (Table 2). The use of supplements containing 1000 µg RE/d was only marginally significantly associated with serum retinol concentration in children (P < 0.05). PIR did not achieve significance in the multivariate model. When other significant sources of variation were controlled for, non-Hispanic black and Mexican American children had significantly lower serum retinol concentrations than did non-Hispanic white children. There was no significant sex difference in serum retinol concentration in children aged <14 y.
View this table:
TABLE 1. Serum retinol distribution by sex, age, and race/ethnicity in participants in the third National Health and Nutrition Examination Survey, 1988–19941
View this table:
TABLE 2. Multiple linear regression analysis of factors associated with serum retinol concentration in children aged <14 y in the third National Health and Nutrition Examination Survey, 1988–19941
In participants aged 14 y, the significant predictors of serum retinol concentration varied by sex and racial/ethnic group (Table 3). In females of all racial/ethnic groups, taking birth control pills or hormone replacements was positively associated with serum retinol concentration.
View this table:
TABLE 3. Multiple linear regression analysis of factors associated with serum retinol concentration in participants aged 14 y in the third National Health and Nutrition Examination Survey, 1998–19941
In participants of all ages except for Mexican American men, PIR had no significant effect on the adjusted mean serum retinol concentration after other significant variables were controlled for with multiple linear regression techniques (Table 4). The use of supplements containing vitamin A had a slightly greater effect. Significant racial/ethnic differences persisted after control for all the variables included in the regression models; non-Hispanic black and Mexican American participants had lower serum retinol concentrations than did non-Hispanic white participants in all comparisons. The racial/ethnic differences in participants aged 14 y were greater than those in children.
View this table:
TABLE 4. Adjusted mean serum retinol concentrations in participants in the third National Health and Nutrition Examination Survey, 1988–19941
The prevalence of serum retinol <0.70 µmol/L was <2% in all strata and for all but one stratum the prevalence estimate was statistically unreliable; in effect, the prevalence estimates were statistically indistinguishable from zero (data not shown). The prevalence of serum retinol <1.05 µmol/L was highest in children aged 4–8 y, between 16.7% and 33.9%, depending on sex and racial/ethnic group (Table 5). The prevalence of serum retinol <1.05 µmol/L was between 3.6% and 14.2% in children aged 9–13 y. In participants aged 14 y, estimates of the prevalence of serum retinol <1.05 µmol/L were between 0% and 6.4%, but all were statistically unreliable and indistinguishable from zero.
View this table:
TABLE 5. Prevalence of serum retinol <1.05 µmol/L in participants in the third National Health and Nutrition Examination Survey, 1988–19941
In children aged <14 y, the risk of having serum retinol <1.05 µmol/L was significantly greater with lower BMI (P 0.001) and with lower serum total cholesterol, serum triacylglycerols, and age (all P 0.0001, data not shown). When these variables were controlled for, non-Hispanic black and Mexican American children were significantly more likely than were non-Hispanic white children to have serum retinol <1.05 µmol/L [odds ratio (OR) = 1.94, 99% CI: 1.37, 2.74; 1.89, 99% CI: 1.31, 2.71, respectively; P 0.0001]. In children, PIR and the use of supplements containing vitamin A were not significant predictors of the risk of having serum retinol <1.05 µmol/L.
In males aged 14 y, the risk of having serum retinol <1.05 µmol/L was significantly greater with lower serum triacylglycerol concentrations (P 0.0001). Mexican American males in the lowest PIR group were substantially more likely than were other males to have serum retinol <1.05 µmol/L (OR = 15.20, 99% CI: 5.67, 40.89; P 0.0001). Neither racial/ ethnic group nor PIR was a significant independent predictor of serum retinol <1.05 µmol/L and use of supplements containing vitamin A was not a significant predictor.
Among females aged 14 y, the risk of having serum retinol <1.05 µmol/L was greater with lower serum total cholesterol (P 0.01) and serum triacylglycerol (P 0.0001) concentrations. Non-Hispanic black and Mexican American females were more likely than were non-Hispanic white females to have serum retinol <1.05 µmol/L (OR = 3.36, 99% CI: 1.41, 8.03; 7.11, 99% CI: 3.35, 15.08, respectively; P 0.0001). PIR and supplement use were not significant predictors of serum retinol <1.05 µmol/L.
DISCUSSION
Nearly all children with serum retinol concentrations <0.70 µmol/L (<20 µg/dL) and many children with concentrations of 70–1.04 µmol/L (20–29 µg/dL) respond to vitamin A supplementation with increases in serum retinol; thus, serum retinol <1.05 µmol/L is considered indicative of potentially suboptimal vitamin A status (3, 20).
Nearly 15% of children aged 3–11 y in the first National Health and Nutrition Examination Survey, 1971–1974; 20–30% of children aged 3–11 y in the second National Health and Nutrition Examination Survey, 1976–1980 (NHANES II); and 25–40% of Mexican American children aged 3–11 y in the Hispanic Health and Nutrition Examination Survey, 1982–1984 (HHANES) had serum vitamin A concentrations <1.05 µmol/L (3). We found that 16.7–33.9% of children aged 4–8 y in the NHANES III survey had serum vitamin A concentrations <1.05 µmol/L and that the prevalence was higher in non-Hispanic black and Mexican American children than in non-Hispanic white children. We also found that the prevalence of serum retinol <1.05 µmol/L in nonwhite children aged 9–13 y was between 7.3% and 14.2%.
It is difficult to make direct comparisons among the surveys because different age strata and different vitamin A assays were used (3). Nevertheless, it appears that the prevalence of potentially suboptimal concentrations of serum retinol (<1.05 µmol/L) is still high in children in the United States, especially minority children. However, in no sex, age, or racial/ethnic stratum of NHANES III did the prevalence of clinically low serum retinol (<0.70 µmol/L) reach 2% and in many strata the estimates of the prevalence of serum retinol <0.70 µmol/L were not significantly different from zero. Spannaus-Martin et al (24) reported that 4% of a sample of 77 preschool children of low socioeconomic status in Iowa had serum retinol <0.70 µmol/L. Because such individuals were rare in the NHANES III sample, we cannot arrive at a statistically reliable population estimate.
In children aged <14 y in NHANES III, small but significant racial/ethnic differences in serum retinol concentration persisted after age, BMI, serum lipids, the use of supplements containing vitamin A, and PIR were controlled for. Non-Hispanic black and Mexican American children were more likely than were non-Hispanic white children to have serum retinol <1.05 µmol/L. The risk of potentially suboptimal serum retinol was also associated with low BMI and low serum lipid concentrations. Income expressed as the PIR was not a significant predictor of serum retinol concentration in children, nor did it explain the racial/ethnic differences.
Our results differ from those of Looker et al (4), who examined racial/ethnic differences in mean serum retinol in children aged 3–11 y in NHANES II and HHANES. They found that racial/ ethnic differences in mean serum retinol became nonsignificant after PIR and supplement use were controlled for. Looker et al (4) included only children in the highest and lowest PIR categories and the earlier data sets included only yes or no responses about the use of dietary supplements. We included children of all PIR strata and we were able to quantify the amount of vitamin A obtained from supplements. In addition, we were able to perform multivariate analysis because oversampling of young and minority participants provided larger samples in NHANES III than in earlier surveys. It is possible that differences in analytic approaches between our study and theirs may account for some of the discrepancy. Although means in NHANES III are higher than those reported in the earlier surveys, it is difficult to rigorously evaluate a potential time trend in serum retinol concentration because of changes in the serum retinol measurement method (3). We also found very few children with serum retinol <0.70 µmol/L, whereas Looker et al (4) found appreciable numbers of children with such low concentrations in the earlier surveys. One possible interpretation is that vitamin A status has improved for many children in the United States since the 1970s, although some children are still at risk of potentially suboptimal vitamin A status, and those are disproportionately minority children.
There have been several previous investigations of the correlates of serum retinol concentration in adults, many reporting conflicting results (25–29). The NHANES III sample was large, nationally representative of the general population and, after certain exclusions, representative of the apparently healthy population. The size of the sample and the oversampling of some strata (children, minorities, and the elderly) provides statistical power to explore the effects of and potential interactions among several independent variables simultaneously. In multivariate analysis, we found that in most sex and racial/ethnic strata, serum retinol concentration increased with age, was positively correlated with serum lipids, and was positively correlated with self-reported alcohol consumption patterns but was not significantly associated with a biomarker of smoking intensity. PIR was generally not a significant correlate of serum retinol. Taking the covariates into account, we found that racial/ethnic group differences in adjusted mean serum retinol concentration persisted in adults as they did in children. We also found that low-income Mexican American men and nonwhite women were at significantly greater risk of having potentially suboptimal serum retinol concentrations than were other groups, even after significant covariates were controlled for.
Serum retinol was positively correlated with taking >1000 µg RE supplemental vitamin A/d in non-Hispanic white and Mexican American adults but not in non-Hispanic black adults. Some investigators reported a direct effect of vitamin A supplementation on serum retinol concentration in clinical trials (30, 31), although in some cases, the effect may be confined to participants who had relatively low serum retinol concentrations before supplementation (32, 33). The NHANES III data are from a cross-sectional survey of free-living individuals rather than from a clinical trial. We found that supplement use was positively associated with serum retinol concentration after income, smoking, and alcohol consumption were controlled for but we were not able to control satisfactorily for other factors such as dietary intake of preformed retinol and ß-carotene or for physical activity, which have been shown to covary with supplement use (34–37) and might affect serum retinol concentration.
Because vitamin A is stored in the liver and released as needed to maintain serum concentrations within a relatively wide range, serum retinol concentration is not a sensitive measure of vitamin A status until stores are depleted (38). We conclude, on the basis of these survey data from 1988–1994, that relatively few adult Americans were at risk of vitamin A deficiency at that time. However, we cannot determine how many may have had low reserves, and some individuals may have been at risk of inadequate vitamin A status. Some children aged 4–8 y and some minority children aged 9–13 y had potentially suboptimal vitamin A status. NHANES III did not provide serum retinol data for children aged 4 y and the proportion of these children who may be at risk is unknown. Monitoring of vitamin A status in children, and especially in those aged 4 y, who were previously identified as a high-risk group, is still an important public health priority in the United States.
ACKNOWLEDGMENTS
Clifford L Johnson and Cynthia Ogden, National Center for Health Statistics, Centers for Disease Control and Prevention, offered advice and support and provided data sets for this analysis. Jacqueline D Wright, National Center for Health Statistics, Centers for Disease Control and Prevention, provided programming guidance in the analysis of nutrient intakes from dietary supplements. Alicia Carriquiry from the Statistical Laboratory of Iowa State University consulted on the use of SIDE/IML software. Laboratory analyses of serum retinol were performed by Dan Huff, Carolyn Hodge, and Patricia Yeager, Division of Laboratory Sciences, National Center for Environmental Health, Centers for Disease Control and Prevention.
REFERENCES