点击显示 收起
1 From the Chronic Disease Nutrition Branch, Division of Nutrition and Physical Activity (CG and MKS), the Division of Diabetes Translation (BAB), the Division of Adult and Community Health (RD), the National Center for Chronic Disease Prevention and Health Promotion, Centers for Disease Control and Prevention, Atlanta, GA, and the Alaska Native Epidemiology Center, Anchorage (CB).
2 Address reprint requests to C Gillespie, Chronic Disease Prevention Branch, Division of Nutrition and Physical Activity, 4770 Buford Highway NE, Mailstop K26, Atlanta, GA 30341-3724. E-mail: cgillespie{at}cdc.gov.
ABSTRACT
Background: The biological variability in serum retinol concentrations has never been examined in a large sample, and its effect on population distribution estimates and the clinical assessment of vitamin A status is unknown.
Objective: We evaluated the biological CV of serum retinol and examined the effect of the CV on both population distribution estimates and clinical assessments of vitamin A status by using data from the third National Health and Nutrition Examination Survey, 1988–1994.
Design: We described the biological CV [(SD/
Conclusions: The actual population prevalence of inadequate vitamin A status may be 75% lower than the estimates previously reported. Confirmation of vitamin A status may be needed for persons in the United States with observed serum retinol concentrations near the recognized cutoff.
Key Words: Biological variation • CV • regression • regression to the mean • serum retinol • vitamin A • third National Health and Nutrition Examination Survey • NHANES III
INTRODUCTION
Population distributions of serum retinol and the prevalence of inadequate serum retinol concentrations have been estimated and described (1–5). The population distribution of serum retinol is affected by 2 distinct components of the total variance: interindividual (between-person) and intraindividual (within-person) variance. Intraindividual variation can distort quantile estimates above and below the median by creating a wider spread in the distribution, with more observations in the tails (6–8). The intraindividual variance itself consists of 2 components: analytic variance, or that which is inherent in the laboratory assay procedure, and biological variance, or physiologic changes in serum concentrations observed in an individual person from day to day. Another factor that influences both population prevalence estimates and individual serum concentrations is the regression to the mean effect. Individual persons are presumed to have an inherent physiologic range that follows a normal distribution and cycles around a theoretical set point, which represents the true mean. If a person has a particularly high or low measurement on one occasion, a second measurement will be expected to yield a more moderate result, or one closer to the population mean (9–11). Standard laboratory methods for measuring serum retinol and the variation inherent in the laboratory assay have been documented, but the effect of intraindividual biological variance on the distribution of serum retinol and the prevalence estimates of inadequate serum retinol has not been examined in a large, nationally representative sample (12). Without knowing the magnitude of biological variance, public health practitioners cannot be certain how accurately an estimated prevalence of vitamin A deficiency based on a sample with a single concentration reflects the true population status.
Individual clinical assessments of vitamin A status rely on an established reference range for serum retinol, with 1st and 99th percentiles of 0.87 and 4.01 µmol/L [(µmol/L)/0.03491 = µg/dL], respectively, based on a preliminary analysis of the 1988–1994 third National Health and Nutrition and Examination Survey (NHANES III) data (13). Concentrations < 1.05 µmol/L indicate potentially inadequate vitamin A status, and concentrations < 0.70 µmol/L indicate a frank deficiency (2, 13, 14). Without knowing the magnitude of individual biological variation, clinicians cannot accurately diagnose vitamin A status on the basis of a single test, and it is not known whether test results near the cutoff should be repeated.
We used data from NHANES III to describe the biological variation of serum retinol and the associations between the biological variation and other factors known or suspected to be associated with serum retinol concentrations. We evaluated the effect of the biological and analytic variation on individual clinical assessment of vitamin A status and the interpretation of population distributions of serum retinol concentrations.
SUBJECTS AND METHODS
Sample and measurement
NHANES III was a complex, stratified probability sample intended to be representative of the noninstitutionalized civilian population of the United States. The sample design and survey methods were described elsewhere (13). Each participant in the survey received a comprehensive physical examination at a mobile examination center (MEC), and blood specimens were collected from all examinees aged =" BORDER="0"> 1 y (n = 29 314) (15, 16). Serum retinol was measured via isocratic reversed-phase HPLC (HPLC Waters Chromatography Division, Milford, MA) on 22 940 examinees aged =" BORDER="0"> 4 y (16). Detailed descriptions of the blood handling procedure for retinol analysis were published previously (13).
NHANES III collected extensive data during the interview and examination that were pertinent to the analysis of serum retinol. Covariates known or suspected to be associated with variation in serum retinol concentrations include the use of retinol-containing supplements (17), the heavy use of alcohol (defined in this analysis as self-reported consumption of =" BORDER="0"> 5 alcoholic drinks per occasion more than once a month), smoking (assessed by measuring serum cotinine concentrations in this analysis), the use of exogenous estrogens, body mass index (BMI), and serum lipids (18–21). Participants were classified as non-Hispanic white, non-Hispanic black, Mexican American or "other" race-ethnicity. We used the poverty-income ratio (PIR) as a marker of socioeconomic status (13 9 y were classified in 3 PIR strata: 1.3, > 1.3–3.5, and > 3.5. Because of sample size considerations, participants aged 6–18 y were categorized into 2 PIR strata: 1.3 and > 1.3. We excluded from this analysis participants with conditions at the time of the examination that are known to affect serum retinol concentrations: elevated concentrations of C-reactive protein (> 2.0 mg/dL), the presence of antigens indicating active hepatitis, a history of liver disorder or liver disease [assessed on the basis of =" BORDER="0"> 2 liver enzymes (alanine aminotransferase, alkaline phosphatase, aspartate aminotransferase, and lactose dehydrogenase) being elevated beyond the upper limit of the reference range], and pregnancy (13, 18, 22–24).
We used 3 analytic population samples in this analysis: 1) the laboratory quality control (QC) data to estimate the analytic CV of serum retinol, 2) a subsample of NHANES III participants who received a second physical examination to estimate the biological CV (CVb), and 3) the primary NHANES III sample to estimate the effect of CVb and the analytic CV (CVa) on the population distribution of serum retinol and to assess its effect on the individual clinical assessment of vitamin A status.
The laboratory QC data used in this analysis consisted of blood samples taken from a group of volunteers who were not part of the NHANES III sample. Approximately 400 total blood samples, 5 from each MEC location, were split into 2 aliquots and sent to the laboratory as a blind QC measure; 159 of these duplicate samples were tested for serum retinol and used in this analysis to estimate the analytic CV.
During the NHANES III survey, a second physical examination was conducted on a nonrandom sample of 30 participants at each MEC location (5% of the primary NHANES III sample). Of those 2586 participants aged =" BORDER="0"> 6 y who had a second examination, 2259 (96%) subjects had complete data for serum retinol. Of those subjects, 93 participants were excluded because of missing data for the covariates of interest, and 161 subjects were excluded because of conditions known to affect serum retinol concentrations or for missing data for these conditions. One outlier was removed, and the final sample used to estimate the CVb of serum retinol was 2004 participants. We included participants of other racial-ethnic groups in the calculation of total population values, but there were too few participants of "other" race-ethnicities to allow for a subgroup analysis (n = 61).
Of those participants in the primary NHANES III examination, 21 398 examinees aged =" BORDER="0"> 6 y had complete data for serum retinol. Of those participants, 1817 were excluded because of missing data on the covariates of interest, 1062 were excluded because of conditions known to affect serum retinol concentrations, and 312 were excluded because of data for these conditions. Some participants were excluded on the basis of more than one criterion. Because a linear model was used to predict the regression to the mean effect, participants with serum retinol concentrations outside the range analyzed in the repeat subsample (range: 0.38–5.27 µmol/L) were excluded to avoid extrapolation beyond the range of the model (n = 21). The final sample used to estimate the population distribution of the CVb consisted of 18 425 participants. Participants of "other" race-ethnicities (n = 770) were included in the calculation of total population estimates.
Statistical methods
Although the CV is the most widely accepted method of reporting analytic variance in serum constituents, numerous methods have been incorporated in the assessment of biological variance (9, 25–42). In this analysis we extended the use of the CV to biological variance to account for the dependent relation between the variance and the concentration of serum retinol and incorporated a linear model to predict and account for the regression to the mean effect. In addition to enabling a more accurate prevalence estimate of inadequate serum retinol by adjusting for factors influencing the distribution, this method also allows for clinical assessment on the basis of a single serum sample.
The CV [(SD/
RESULTS
Analytic CV
The plot of the CVa of each split sample ordered by the absolute serum retinol concentration (range: 0.87–4.33 µmol/L, n = 159) is shown in Figure 1. There was no significant linear association between the CVa and the serum concentration observed in the QC sample. We used the mean CVa (
FIGURE 1.. Observed analytic variation (CVa) in serum retinol concentrations. The horizontal dashed line represents the mean CVa (1.58%).
Biological CV
The mean and quartile distribution of the CVb of serum retinol among the NHANES III participants with a repeat examination is shown in Table 1. Participants who had identical repeated retinol concentrations with an observed CVt of zero (n = 127; 6.3%) and those participants for whom the estimated CVa exceeded the observed CVt (n = 155; 7.7%) were assigned a value of zero for their CVb. The range of the CVb across all strata was 0–56.5% (data not shown). Because there were no significant differences in the CVb between the sexes in any of the strata (data not shown), only the totals are presented. Across all age and race-ethnicity strata, the median serum retinol CVb was < 7.5%, and the 75th percentile was < 15%. Overall and among adults, the mean CVb among non-Hispanic blacks and Mexican Americans was significantly higher than that in non-Hispanic whites (P < 0.05).
View this table:
TABLE 1. Mean and quartile distribution of the biological CV (CVb) of serum retinol among participants in the third National Health and Nutrition Examination Survey repeat examination
The final multiple linear regression models are shown in Table 2. Among adults, race-ethnicity was significantly associated with the rank order of the CVb for serum retinol. Non-Hispanic black and Mexican American adults showed a higher rank order of the CVb than did non-Hispanic whites (P < 0.01). The rank order of BMI was also significantly associated with the rank order of the CVb, indicating a 0.1-unit decrease in CVb rank for every unit increase in BMI rank (P = 0.0002). Of the children and adolescents aged 6–18 y, no factors were significantly associated with the CVb.
View this table:
TABLE 2. Multiple linear regression analysis of factors associated with the biological CV (CVb) of serum retinol among adult participants in the third National Health and Nutrition Examination Survey repeat examination
Effect of the CV on clinical evaluation and population prevalence estimates
The overall mean CVb for serum retinol among the NHANES III subsample with a repeat examination was 6.45%. The 95% CIs resulting from the linear regression model used to estimate and adjust for the regression to the mean effect are shown in Figure 2. At the clinical level, the linear regression model and the CV can be applied to a single serum retinol concentration (in µmol/L) to determine an individual’s predicted mean and 95% CI. The CVa of the specific laboratory should be added to the CVb to obtain a more accurate 95% CI (seeAppendix A for a detailed solution):
DISCUSSION
The CVa for serum retinol did not vary significantly across the concentration range with a mean of 1.58% and a maximum of 6.86%. Significant differences in the mean CVb between non-Hispanic whites and the other ethnic groups were observed. This was consistent with the results of the regression analysis, wherein the CVb was higher in the adult non-Hispanic black and Mexican American ethnic groups. These differences were most likely attributable to the increased prevalence of a serum retinol concentration < 1.05 µmol/L observed in these groups in the current analysis and previously reported estimates, which indicated lower mean serum retinol concentrations (5). The results of the regression analysis indicated that an increasing rank order of BMI was associated with a lower CVb in adults; however, the small magnitude of the parameter estimate indicated that this association was physiologically negligible. In children and adolescents aged 6–18 y, no factors were significantly associated with the CVb.
The total variance of the distribution of serum retinol and the prevalence estimates of extreme values are affected by intraindividual biological variance and the regression to the mean effect. In the current analysis, prevalence estimates of inadequate serum retinol were reduced after adjustment for the intraindividual portion of the variance and the regression to the mean effect in the distribution, and the actual population prevalence of inadequate serum retinol concentrations may be 75% lower than the estimates previously reported (5). Although the adjustment is large as a percentage of the prevalence estimate, the magnitude of the actual difference is small (2%), which indicated that, overall, an estimate based on a sample with a single serum measurement is accurate in estimating the population prevalence. In the non-Hispanic black and Mexican American racial-ethnic groups, the biological variability produced a larger prevalence adjustment (4.28% and 4.34%, respectively). In children aged 6–11 y, the prevalence adjustment was of a larger magnitude (12.21%), which indicated a higher level of biological variability and a less accurate prevalence estimate on the basis of a single serum concentration measurement in this age group.
Given the efficacy, availability, and relative safety of appropriate retinol supplementation in reversing subclinical vitamin A deficiency, confirmation of vitamin A status for persons with an observed concentration < 1.05 µmol/L does not appear to be as critical as that for those with an observed concentration above the cutoff, but whose 95% CI overlaps the cutoff. In general, the vitamin A status of a person with a single retinol concentration between 1.05 and 1.16 µmol/L should be confirmed, although this range will be affected by specific laboratory variance.
The CVb for serum retinol was estimated in this analysis by using the NHANES III subsample with a repeat examination. The small size of this sample and the nonrandom selection of the participants were considered when attempting to generalize the results of this study to the entire NHANES III sample, which was designed to represent the noninstitutionalized civilian population of the United States. We compared the demographic distribution of the NHANES III participants with a repeat examination with that of those participants without a repeat examination (data not shown). We found few statistically significant (P < 0.05) differences in the demographic distribution between the samples. The average age of the participants in the subsample was older than that of those in the primary NHANES III sample (mean age: 43 and 37 y, respectively; P < 0.0001). The proportion of non-Hispanic whites was significantly larger in the subsample than in the primary NHANES III sample (41% and 36%, respectively; P < 0.0001). The mean serum retinol concentration was higher in the participants in the research subsample than in those in the primary NHANES III sample (difference: 0.1 µmol/L; P < 0.0001). Although the differences were statistically significant, we do not believe that these differences between the subsample and the primary NHANES III sample indicated clinically significant physiologic differences nor were the differences large enough to invalidate the application of these results to the primary NHANES III sample or the generalizability and clinical applicability of these results to the US population.
Because the regression to the mean effect is driven by the sample mean, the results of the current study cannot be extrapolated beyond the population from which the sample was drawn. The same analysis performed within other populations, such as those in developing nations where the population mean serum retinol concentration is likely to be lower and the prevalence of inadequate vitamin A higher than in the United States, will not yield similar results.
APPENDIX A
To account for the regression to the mean effect, we used a linear regression model to predict the mean serum retinol concentration from the first serum concentration in the NHANES III subsample with a repeat examination:
ACKNOWLEDGMENTS
We thank Brenda G Lewis (National Center for Health Statistics, Centers for Disease Control and Prevention, Atlanta) for offering advice and support and for providing data sets for this analysis and Dan Huff, Carolyn Hodge, and Patricia Yeager (Division of Laboratory Sciences, National Center for Environmental Health, Centers for Disease Control and Prevention) for conducting the serum retinol measurements.
CG designed the study, performed and interpreted the statistical analysis, and composed the manuscript. CB and BAB assisted and advised in the study design and statistical analysis and interpretation. MKS assisted and advised the interpretation of results and the public health application and assisted with the manuscript composition. RD assisted and advised in the statistical design and analysis. None of the authors had any real or potential conflicts of financial or personal interest with the financial sponsor of this scientific project.
REFERENCES