Main

In the United States, iron overload primarily results from hemochromatosis,1 an autosomal recessive condition, characterized by lifelong excessive absorption of iron. Iron accumulates in body organs, eventually causing inflammation and damage. Cirrhosis of the liver, liver cancer, heart failure, diabetes, impotence, arthritis, and other disorders can result from hemochromatosis,15 but early detection and treatment of this genetic condition can decrease morbidity and sometimes prevent the onset of disease.2,3,5 On the basis of screening studies using iron indices, hemochromatosis is estimated to occur in 2 to 5 per 1000 persons in white populations in the USA.6,7

Although linkage of hemochromatosis to the major histocompatability complex (HLA) region on chromosome 6 has been recognized for more than 20 years,8,9 the hemochromatosis (HFE) gene was first identified in 1996.10 Two missense mutations in HFE, C282Y and H63D, account for most of the cases among whites. Homozygosity for C282Y accounts for 52 to 100% of these cases,11 and a much smaller percentage of cases are attributed to C282Y/ H63D compound heterozygosity and H63D homozygosity.11 In the United States 0.26% (95% confidence intervals [CI], 0.12, 0.49) of persons are homozygous for the C282Y mutations, 2% are compound heterozygous (C282Y/ H63D), and 2% are homozygous for the H63D mutation.12 The prevalence of C282Y and H63D genotypes were similar for non-Hispanic whites.12 Thus, a higher proportion of people at risk for hemochromatosis is revealed by HFE genotypes than by biochemical tests.

Limited clinical observations in screening studies suggest that an estimated 40% to70% of persons with the C282Y homozygous genotype develop clinical evidence of iron overload.11 Simple heterozygosity for the C282Y or H63D mutation can be associated with increased levels of transferrin saturation (TS),1324 which is an initial indicator of hemochromatosis.1 In case-control studies of whites, simple heterozygosity for the C282Y mutations was associated with a 4-fold increase in the odds of iron overload, and simple heterozygosity for the H63D mutation was associated with a 1.6-fold increase.25 However, potential biases in selection and ascertainment of cases and controls, combined with a failure to consider factors such as age, race and ethnic group, sex, dietary factors, and alcohol consumption, may have decreased the accuracy of these estimates.25 In addition, although iron overload increases with age and is more prevalent in males with hemochromatosis,1 little is known about how age or sex affects the natural history of iron overload among persons with H63D or C282Y mutations. Finally, no study thus far has included racial and ethnic minorities in large enough groups to examine the association of HFE genotypes and transferrin saturation separately. The objective of the current study was to examine the association in the U.S. population by race and ethnic group, age, and sex.

MATERIALS AND METHODS

Survey design and participants

The National Health and Nutrition Examination Surveys (NHANES) are a series of national surveys that the National Center for Health Statistics, Centers for Disease Control and Prevention (CDC), began conducting in 1966 to estimate the prevalence and risk factors for common diseases in the U.S. population. NHANES III was conducted in two phases in 1988 to 1991 and 1992 to 1994.26,27 Each phase was nationally representative, and it over-sampled certain populations including non-Hispanic blacks and Mexican-Americans. Population weights were calculated to account for oversampling, multistage sampling design, and nonresponse to the household interview and the examination.

Cell lines from 8,205 participants in NHANES III were immortalized with Epstein-Barr Virus (EBV) to establish a DNA bank. Although CDC planned to collect DNA for storage, the decision to establish cell lines occurred after phase 1 had already begun. Overall, 15,427 individuals were interviewed as part of phase 2. Of these, 10,280 were 12 years and older. A total of 10,052 individuals 12 years and older were examined, and cell lines were available for 7,195 (71.6%) of those examined. For the final analytic sample, participants were excluded if they were pregnant, were missing transferrin saturation, and were not Mexican-American, non-Hispanic white or non-Hispanic black, leaving 6,702 participants. To assure that previously anonymized specimens remained anonymous, special procedures were required by the Institutional Review Board (IRB) of the National Center for Health Statistics. Because of these procedures, 1,529 participants were eliminated randomly to decrease the ability to link specimens. Two additional specimens failed to amplify for genotyping, which left a final sample of 5,171.

Informed consent

The IRB approved the NHANES survey as well as this specific analysis. Although participants in the survey signed an informed consent form, specific mention of genetic research was not included. Because of the scientific importance of this resource, the NHANES IRB approved a process to make DNA available to the research community for anonymized testing. For this purpose, anonymized testing was defined as the inability for anyone, including CDC staff, to link the results of the genetic tests back to the survey participant.28 Although participants were therefore not informed of their HFE genotype, they had been provided with iron metabolism laboratory test results during the course of the survey.

Transferrin saturation

Transferrin saturation (TS) was determined by dividing the concentration of serum iron (μmol/L) by total iron binding capacity (μmol/L) as assessed by a modification of the automated AAII-25 method that used ferrozine as the chromogen and dialysis to remove protein.29 Both iron assays were conducted at the Central Laboratory for NHANES, National Center for Environmental Health, NCEH, CDC.

Genotyping methods

Specimens were genotyped in the Molecular Biology Branch of the Division of Laboratory Sciences, NCEH, CDC, using genomic DNA extracted from EBV-transformed cell lines. The wild type (designated as “+”) and the C282Y and H63D mutations were genotyped using TaqMan technology,3033 in which amplification and genotyping are simultaneously performed using the ABI PRISM 7700 (Applied Biosystems, Foster City, California). These methods are described in detail elsewhere.12

Statistical analysis

As a result of the requirements to maintain anonymity, the analysis was broken into several steps. First, we used a weighted multiple linear regression with TS as the dependent variable and age, sex, race and ethnicity, fasting status, and time of day (and all interactions of age, sex, race, and ethnicity, up to and including three variables) as independent variables. The weights used were the NHANES III sampling weights for the examination in phase II. To test the validity of the model, we plotted the residuals against the predicted values and examined the distribution of residuals. We found that the variance did increase slightly with an increase in the predicted value, but transforming transferrin saturation using a square root or log function did not substantially improve the variance. The residuals were normally distributed with a mean of −0.10% and a standard deviation of 10.4%. The purpose of this preliminary analysis was to attempt to account for the variability in TS34 (except that due to genetic factors) prior to linking race/ethnic group, sex, age group, mean weight, and decile of residual TS to genotype information. This analysis was done using the full NHANES III phase II database of 9041 persons who were 12 years and older, nonpregnant, with data on transferrin saturation, and who were non-Hispanic black, non-Hispanic white, or Mexican-American.

We focused our attention on the residuals from this model (i.e., the variability in TS levels that was unexplained by the variables in our model, which we denote “residual TS” or rTS). We ranked all persons by their residual TS value and categorized a person as having “high rTS” if their value of residual TS was in the highest decile. We next determined that our definition of high rTS corresponded well to high values of raw TS. The sensitivity of our high rTS variable for actual TS values > 45% ranged from 98.8 to 100% across sex, race/ethnic, and age groups indicating that in most cases our threshold for high rTS was below 45%. Specificity ranged from 92 to 95.6% (Table 1). Sensitivity was lowest for white males, and specificity was lowest for non-Hispanic black females.

Table 1 Sensitivity and specificity of high residual transferrin saturationa for transferrin saturation >45% in the US population, N = 9041b

Sampling cells were created that contained no fewer than five persons who shared the same set of background characteristics (i.e., sex, race and ethnicity, age group, and decile of rTS). Of the persons within a sampling cell, 20% or 2 of the subjects, whichever was larger, were eliminated randomly. Thus the probability of selection of data from these cells varied from 60% (3/5) for cells n = 5 to 80% for cells n = 8 or greater. At this stage the link to individual identifiers was removed. The results of the genetic testing were then linked to the dataset using the random numbers that had been attached to both the DNA samples and the analytic dataset. Individual sample weights were replaced by average sample weights in each analytic cell. We adjusted the NHANES sample weights to account for the probability of selection of data from the cells given the number of people in each cell (see above). The final data set included the random number, age group, sex, race/ethnicity, residual TS decile, and adjusted sample weight.

We computed the weighted prevalence estimates of high rTS by genotype in the population.27 Because we analyzed residuals that resulted after adjusting for age, race/ethnicity, and sex, we expect that the prevalence of high rTS is similar across these demographic variables. The associations between HFE genotypes and high rTS, as well as the interactions between each of the demographic variables (age, racial and ethnic group, and sex) and HFE genotype with respect to risk of high rTS, are separate and distinct from the associations between the demographic variables and high TS based on raw TS values.3537 Thus, the second step was to examine the interactions between HFE genotype and each demographic characteristic with respect to residual TS higher than the 90th percentile. Hence, we stratified our analysis of HFE genotype and high TS separately by each demographic characteristic (i.e., age, racial and ethnic group, and sex).

We used logistic regression to calculate the odds of high rTS by HFE genotype compared to that of the homozygous wild type in our final dataset as well as separately for each race/ethnic group, each age group (12–49, 50+) and each sex. As explained in Appendix 138 and above, it is still may be necessary to account for the interactions between demographic variables and the HFE mutations when computing the odds of high rTS even though values of rTS were calculated from a regression model that included these variables. The logistic regression was weighted using the sample weights described above to reproduce the odds of high rTS in the target (U.S.) population. For these analyses, the at-risk genotypes were C282Y/ C282Y, C282Y/ H63D, H63D/ H63D, C282Y/+, and H63D/+, where + denotes the wild type. All logistic regressions were rescaled to the actual sample size using the NORMALIZED option after the weight statement in SAS.39,40 To test potential interactions of HFE genotypes and demographic characteristics with the risk of high rTS, we calculated the difference in log likelihood ratios for a logistic regression model with the HFE genotypes and the specified demographic characteristic (e.g., age Group 50 or more years vs. age Group 12–49 years) compared with a model including the HFE genotypes, the demographic characteristic, and the interaction between the HFE genotypes and the demographic characteristic. We compared the difference in log-likelihood ratios with the tables for chi-square,35 and we examined the significance of individual interaction terms. We corrected the odds ratios and 95% confidence intervals obtained from the logistic regression equations to better approximate prevalence ratios using the following equation:41

OR equals the odds ratio, and P0 is the prevalence of the outcome of interest (e.g., high rTS) in the +/+ group.

We were unable to calculate standard errors accounting for the complex sample design because anonymity requirements prevented access to cluster variables. Standard errors that do not account for the cluster variables would underestimate the true error of the estimates. Instead we used the binomial distribution42 to construct approximate confidence intervals for the weighted prevalence estimates using an assumed design effect of 1.5, although in reality, the design effect for this analysis may be lower or higher than 1.5.40 Random selection within strata tends to lower the design effect.43 We also multiplied the errors from the coefficients of the logistic regression models by the square root of 1.5 in order to calculate confidence intervals corrected for the assumed design effect of 1.5. We used SAS, Version 8.1, for all analyses.39

RESULTS

Although the confidence intervals were wide, we observed that more persons with C282Y homozygosity had high rTS (69.1%, 95% CI, 17.3, 95.1) than persons without either mutation (8.8%, 95% CI, 7.8, 10.0) (Table 2). Persons with H63D homozygosity and compound heterozygosity were also more likely to have high rTS then persons without either mutation. The prevalence of high rTS appeared similar among males and females, but the number of persons with C282Y homozygosity was too small to determine significant differences in the associations for any of the demographic subgroups.

Table 2 Estimated prevalence of high rTS in the US populationa by hemochromatosis (HFE) genotype and by sex

The associations between high rTS and simple heterozygosity for H63D and C282Y in comparison with the wild type (+/+) were not statistically different by racial and ethnic groups (Table 3), except the association of simple H63D heterozygosity with high rTS was statistically significant only among Mexican-Americans. Among Mexican-Americans, those with simple H63D heterozygosity were 1.8 times (95% CI, 1.2, 2.6) as likely to have high rTS as those without either mutation (i.e., +/+). In comparison, non-Hispanic whites with simple H63D heterozygosity were 1.1 times (95% CI, 0.7, 1.6) as likely to have high rTS as those without either mutation.

Table 3 Estimated prevalence of high rTS in the US populationa by hemochromatosis (HFE) genotype and by race/ethnic group

Compared with persons aged 12 to 49 years with the +/+ genotype, persons aged 50 years and older with the same genotype were 0.6 times (95% CI, 0.4, 0.8) as likely to have high rTS (5.9% vs. 10.1%, see Table 4). Compound heterozygosity was associated with increased prevalence of high rTS in both age groups. The difference in the log-likelihood ratios for the models with and without the interactions with age group was statistically significant (P < 0.025) indicating a different association between HFE genotype in persons aged 50 years and older than in those < 50 years of age. Only the interactions for H63D homozygosity or simple H63D heterozygosity and age were statistically significant. The prevalence ratio for high rTS among persons with H63D homozygosity compared with the homozygous wild type (i.e., persons without either mutation) was 3.1 times (95% CI, 1.2, 9.9) greater among persons 50 years and older than the prevalence ratio among those < 50 years of age. The prevalence ratio for high rTS among persons with H63D heterozygosity compared with the homozygous wild type was 1.9 times (95% CI, 1.1, 8.8) greater among persons 50 years and older than the prevalence ratio among those < 50 years of age. The prevalence ratio among persons with C282Y heterozygosity compared with the homozygous wild type was the same for persons 50 years and older and those < 50 years of age (prevalence ratio = 1.0, 95% CI, 0.5, 5.5).

Table 4 Estimated prevalence of high rTS in the US populationa by hemochromatosis (HFE) genotype and by age group, y

DISCUSSION

We evaluated the association between TS, an indicator of circulating iron, and the C282Y and H63D mutations in the HFE gene in a representative sample of the U.S. population. Although the confidence intervals were wide, our estimate of about two-thirds of persons with C282Y homozygosity having high rTS is consistent with previous studies,1315,1920,23,24 in which 50 to 100% of persons with C282Y homozygosity had high TS. Koziol and colleagues13 reported that among 10,198 adults attending a health appraisal clinic in California, 48% of women and 81% of men with C282Y homozygosity (N = 42) had TS > 45%. Similarly, among whites in two different studies, 7523 to 9419 percent of persons with C282Y homozygosity had a TS > 45%. Distante and colleagues15 studied a group of 505 unrelated hospital employees and found that both of the two employees with C282Y homozygosity had a TS > 50%. Deugnier and colleagues44 found among people attending a health appraisal clinic in France that 41% of women (n = 44) and 80% of men (n = 10) with C282Y homozygosity had elevated TS compared with 5% of nonhomozygotes. Phatak et al.24 found among 4,865 unselected primary care patients that 75% of men (n = 4) and 100% of women (n = 8) had a TS >45%. Although the majority of male C282Y homozygotes have elevated transferrin saturation in our study and most other studies to date, in the larger studies, including ours, usually only about half of the women do. The exception is the study conducted among primary care patients where 8 of out 8 women with C282Y homozygosity had high TS levels.24 In Phatak’s study,24 the mean age of the patients was 52 years. The lower proportion of high TS levels among females than males in the majority of studies may be explained by loss of iron due to menstruation and pregnancy among women of childbearing age.

Our study had the largest sample of Mexican Americans and blacks thus far with HFE genotype and TS values. Our data suggested that simple heterozygosity for H63D was significantly associated with elevated rTS in Mexican-Americans. However, the lower prevalence of C282Y mutations in Mexican-Americans and non-Hispanic blacks,12 and the resulting small sample size of exposed persons, make it difficult to address the associations between C282Y mutations and high rTS in these groups. Because of the low prevalence of C282Y and the strong association between H63D and high rTS in Mexican-Americans, the attributable fraction of clinical hemochromatosis arising from H63D mutations may be higher than that arising from C282Y mutations in this group. This possibility requires further investigation in screening studies that use genetic and iron testing for hemochromatosis among larger samples of Mexican-Americans.

Our study indicates that the associations between the H63D homozygosity and high rTS strengthen with increasing age. In screening studies for hemochromatosis using elevated serum iron measures and HLA-typing, symptoms and disease complications increased with age among persons considered homozygous for hemochromatosis.1 The small sample of people with C282Y homozygosity and compound heterozygosity decreased our ability to examine these groups. The results by age are unclear for simple H63D or C282Y heterozygosity. The association with high rTS was slightly greater among older persons with simple H63D heterozygosity, but not among those with simple C282Y heterozygosity. The finding was consistent with a previous study20 that found no difference in the association between simple C282Y heterozygosity and serum ferritin distribution by age.

The small number of variables and the inability to use continuous outcomes limited our analysis. We could not use the common thresholds for TS (e.g., 45%–70%) because sample cell sizes became too small in some subgroups to meet the requirements for protection of the identity of human subjects. There is no consensus about the optimal TS threshold to identify people for further evaluation for hemochromatosis. Our TS threshold (> 90th percentile of rTS) needed to maintain anonymity was slightly < 45% in some population sub groups, but had comparable sensitivity and specificity to a threshold of 45%, the low end of phenotypic expression of hemochromatosis.45 Because 45% is at the lower end of phenotypic expression, it generally has a high sensitivity, but low specificity for diagnosis of hemochromatosis.45 Similarly, because the sample size would become too small in some subgroups to preserve anonymity, we could not include many variables that may affect transferrin saturation values such as anemia of various causes, inflammation, megaloblastic conditions, or hepatocellular damage attributable to concurrent viral disease, alcohol, or drugs. Thus we cannot rule out that high TS was due to these factors. The sample size limited our ability to precisely estimate the association of C282Y homozygosity with high TS and the associations of H63D homozygosity and compound heterozygosity with high TS for some of the demographic subgroups. Finally, we only had one measure of TS, and although we were able to adjust for fasting status and time of day of specimen collection, because of the biological variability in TS, repeated measures are needed to better establish risk of hemochromatosis.1,45

The results of our study and others together indicate that although genotype and TS are correlated, a substantial proportion of people who are homozygous for C282Y, and the majority of people who are heterozygous for HFE genotypes, do not have high TS. In addition, our study suggests that the association between HFE genotype and high TS may vary with age. In Mexican-Americans, the presence of H63D may be more important than C282Y as a risk factor for hemochromatosis because of its higher prevalence and its positive association with transferrin saturation. Our study reaffirms the need for additional information about the risk of iron overload and chronic disease associated with HFE genotypes, as well as other genetic and environmental factors that modify this risk, to make informed decisions regarding genetic screening for hemochromatosis. Much of this information will be gained through the National Institutes of Health sponsored Hemochromatosis and Iron Overload Screening (HEIRS) Study, a multicenter screening study including 10,000 Hispanic and 27,000 African American individuals.46 Because iron overload has a variable age of onset, may be influenced by dietary, alcohol, and other factors, and may be affected by modifier genes, a longitudinal study is required to determine the true penetrance of the HFE genotypes.