1 From the Bandim Health Project, INDEPTH Network, Statens Serum Institut, Bissau, Guinea-Bissau (CW, RO, PR, PK, PG, PA, and MS); the Infectious Disease Research Unit, Skejby (CW, RO, and PLA) and the Department of Internal Medicine, Silkeborg (HG), Aarhus University Hospital, Aarhus, Denmark; the Department of Human Nutrition, Faculty of Life Science (PK) and the Department of Infectious Diseases (MS), University of Copenhagen, Copenhagen, Denmark; and the Infectious Diseases Research Group, Department of Clinical Sciences, Lund University, Malmö, Sweden (PG)
2 Supported mainly by The Danish Research Council for Developmental Research and by the Segels, Beckett, SSAC, Jakob Madsen, Lily Benthine Lund, and Skejby University Hospital research foundations. CW was supported by a PhD scholarship from the University of Aarhus. 3 Reprints not available. Address correspondence to C Wejse, Department of Infectious Diseases, Aarhus University Hospital, Brendstrupgaardsvej, 8200 Aarhus N, Denmark. E-mail: wejse{at}dadlnet.dk.
ABSTRACT
Background: Little is known regarding vitamin D deficiency (VDD) in African populations and in tuberculosis (TB) patients. VDD has been shown to be associated with TB.
Objective: We aimed to compare the degree of vitamin D insufficiency (VDI) and VDD in TB patients and healthy adult controls in a West African population.
Design: An unmatched case-control study was performed at a Demographic Surveillance Site in Guinea-Bissau. Serum 25-hydroxyvitamin D3 [25(OH)D3] concentrations were measured in 362 TB patients and in 494 controls.
Results: Hypovitaminosis D [25(OH)D3 75 nmol/L] was more common in TB patients, but VDD [25(OH)D3 50 nmol/L] was more common and more severe in controls. We observed hypovitaminosis D in 46% (167/362) of the TB patients and in 39% (193/494) of the controls; the relative risk (RR) of hypovitaminosis D was 1.18 (95% CI: 1.01, 1.38). VDD was observed in 8.5% (31/362) of the TB patients and in 13.2% (65/494) of the controls. The RR was 0.65 (95% CI: 0.43, 0.98), mainly because severe VDD [25(OH)D3 25 nmol/L] was observed in only 1 of 362 TB patients (0.2%) and in 24 of 494 controls (4.9%). After adjustment for background factors, hypovitaminosis D was not more frequent in TB patients than in healthy controls, but the mean serum 25(OH)D3 concentration remained lower.
Conclusions: Hypovitaminosis D was highly prevalent in TB patients and in healthy controls living at 12 °N; severe VDD was rare in TB patients. The finding indicates that the serum 25(OH)D3 concentration is associated with TB infection, but whether this role is a symptom or is causal was not established.
Key Words: Hypovitaminosis D • 25-hydroxyvitamin D • tuberculosis • Guinea-Bissau
INTRODUCTION
Tuberculosis (TB) constitutes a major health problem in sub-Saharan Africa (1). Vitamin D deficiency (VDD) was shown to be associated with TB in small studies from Indonesia, India, and Kenya (2–4) and in studies of foreign-born persons in Britain (5–7). African Americans have significantly lower serum 25-hydroxyvitamin D [25(OH)D] concentrations than do whites (8) and have an increased susceptibility to Mycobacterium tuberculosis infection (9). In addition, some studies suggest that certain vitamin D–receptor polymorphisms may be involved in the susceptibility to TB (10). We hypothesized that VDD is associated with TB. Hence, we carried out a population-based study in Guinea-Bissau, where the incidence of TB is high (470/100,000) (11).
SUBJECTS AND METHODS
Study area
We conducted the study at the Bandim Health Project, a Demographic Surveillance Site with a current population of 92 000 in the capital of Guinea-Bissau (12° N) on the West African coastline. The staple foods are rice, small amounts of fresh water, and sea fish; soybean oil is consumed frequently. Red palm oil, fruit, vegetables, and nuts are consumed seasonally (P Kaestel, personal communication, 2006).
Study population
We described the vitamin D status in an unmatched case-control study in TB patients and healthy adult controls from the same area. From November 2003 to February 2006 we included 362 TB patients in a treatment trial for TB (ISRCTN35212132). Inclusion criteria for cases were as follows: diagnosis of TB according to World Health Organization guidelines (12), residence in study area, and age >15 y. Field assistants visited the 3 health centers and the TB hospital in the study area daily and invited new incident TB patients starting treatment to come to the inclusion site the following day. We assessed demographic variables in a baseline questionnaire and collected nonfasting blood samples at inclusion. Mean lag time from the start of treatment to inclusion and time of blood sampling was 7 d, 311 of 362 patients were included within 2 wk after treatment initiation.
We enrolled a random population sample from the study area between April 2005 and February 2006 and obtained blood samples from 494 adults for a study of genetic risk factors for TB (13). We included trios consisting of mother, father, and child (regardless of age) from each house and collected blood samples from all. Only the adults were included in the present study. We drew a random list of houses from the study database; houses with a case of TB during the past 2 y were excluded, as were individuals who had experienced a cough for >2 wk or who had previously had TB. In case of refusal or if the household did not have a relevant trio, residents of the neighboring house were solicited. A demographic questionnaire was completed, and anthropometric measures were made.
We conducted the study in accordance with the Helsinki Declaration, and the procedures followed were in accordance with the ethical standards of the Bandim Health Project. Ethical approval was obtained from the ethics committee within the Ministry of Public Health in Guinea-Bissau and by the Central Ethical Committee of Denmark.
Anthropometric measures
Height was measured with a meter scale; the weights of the TB patients and controls were measured with the same weight scale. Body mass index was calculated as weight (kg)/height squared (m).
Seasonality
In Bissau the rainy season lasts from June to November. During the rainy season it is cloudy and sunlight exposure is diminished, but the days are slightly longer. We coded samples taken from December through May as being from the dry season (mean sunshine: 224 h/mo) (14); samples collected from June through November were coded as being from the rainy season (mean sunshine: 147 h/mo).
Tuberculin skin test
Laboratory technicians performed the tuberculin skin test (TST) using purified protein derivative (PPD) as a measure of tuberculin reaction. We applied Tuberculin (PPD, 0.1 mL SSI RT23 2T.U.) intradermally in the ventral aspect of the forearm. We read TST reactions by measuring 2 diameters of the area with skininflammation with a ruler and ballpoint technique after 48–72 h (15). We used 10 mm as the cutoff for a positive reaction, referred to as latent TB infection (LTBI) (16, 17).
Socioeconomic index
Socioeconomic status was drawn from the Bandim Health Project database on the 750 individuals with a valid identification number. This index divides the population into the poorest, less poor, and richest according to household information on type of roof, indoor toilet, electricity, and TV (18, 19).
Laboratory measurements
Serum was harvested and stored at –20 °C. Samples were transported to Denmark every 3 mo and stored at –80 °C. Samples were analyzed in batches at the Department of Clinical Biochemistry, Aarhus University Hospital, in February, June, and October 2006. We measured serum 25-hydroxyvitamin D2 [25(OH)D2, ergocalciferol] and serum 25-hydroxyvitamin D2 [25(OH)D3, cholecalciferol] by isotope-dilution liquid chromatography–tandem mass spectrometry on an API3000 mass spectrometer (Applied Biosystems, Foster City, CA) using a method adapted from Maunsell et al (20): routine isotope-dilution liquid chromatography–tandem mass spectrometry assay for simultaneous measurement of the 25-hydroxy metabolites of vitamins D2 and D3. The method was calibrated by using Serum Calibration Standards from an external supplier (ChromSystems, Munich, Germany). The quality control was performed by daily analysis of internal control samples and participation in the DEQAS Vitamin D External Quality Assessment Scheme. The interassay and intraassay CVs were 9.4% and 9.7%.
We defined vitamin D insufficiency (VDI) as a serum 25(OH)D3 concentration of 51–75 nmol/L, mild VDD (mVDD) as a serum 25(OH)D3 concentration of 26–50 nmol/L, and severe VDD (sVDD) as a 25(OH)D3 concentration of 25 nmol/L according to Vieth (21) and Holick (22). We refer to hypovitaminosis D as any of the above and VDD as all with a serum 25(OH)D3 concentration 50 nmol/L. We analyzed all samples for 25(OH)D2 and 25(OH)D3 (23).
Serum calcium and albumin were measured by absorbance (Corba Integra; Roche Diagnostics, Mannheim, Germany). We corrected total serum calcium for individual variations in albumin by using the following equation: adjusted serum calcium (mmol/L) = total serum calcium (mmol/L) x 0.00086 x [650 – serum albumin (µmol/L)]. The reference range according to Roche Diagnostics (24) is 2.10–2.75 mmol/L.
Statistical analysis
The study had 99% and 74% power, respectively, to detect a 10% and a 5% difference in prevalence of VDD (<50 nmol/L) among TB patients and controls. Categorical variables with missing information were given a separate category, thereby preserving the power of the study. Analyses were adjusted for family relation by clustering. Pearson chi-square was used to assess statistical differences in proportions between groups (P < 0.05), Student's t test was used to assess differences in means between 2 groups when there was a normal distribution, and Wilcoxon's rank-sum test was used when nonparametric analysis was needed (25). Logistic regression analysis was used to adjust for categorical differences between cases and controls; linear regression analysis was used to adjust for differences in mean serum concentrations. Spearman's rank correlation coefficient () was used for correlation analysis. A 2-sided P < 0.05 was considered significant. Statistical analyses were performed with STATA software (version 9; StataCorp, College Station, TX).
RESULTS
Only 7 samples (2 TB patients, 5 controls) had detectable concentrations of 25(OH)D2; 25(OH)D2 concentrations ranged from 25 to 44 nmol/L and were found in subjects with 25(OH)D3 values in the range 36–137 nmol/L. Only 25(OH)D3 concentrations are discussed below.
Prevalence of hypovitaminosis D in TB patients and healthy controls
Characteristics of TB cases and healthy adult controls are described in Table 1. Mean and median 25(OH)D3 concentration were significantly lower in TB patients than in healthy controls.
View this table:
TABLE 1. Characteristics of 362 tuberculosis (TB) patients and 494 healthy adult controls1
We observed hypovitaminosis D in 46% (167/362) of the TB patients and in 39% (193/494) of the healthy controls; the relative risk (RR) of hypovitaminosis D was 1.18 (95% CI: 1.01, 1.38) in TB patients compared with controls. We observed sVDD [25(OH)D3 25 nmol/L] in only 1 of 362 TB patients (0.2%) and in 24 of 494 controls (4.9%). VDD + sVDD (all with 25(OH)D3 50nmol/L) was observed in 8.5% (31/362) of the TB patients and in 13.2% (65/494) of the healthy controls. Hence, the RR of VDD was 0.65 (95% CI: 0.43, 0.98) for TB patients compared with controls. The proportions of various degrees of lack of vitamin D in the 2 groups are displayed in Figure 1.
View larger version (13K):
FIGURE 1.. Distribution of the degree of hypovitaminosis D among the tuberculosis (TB) patients and controls.
We investigated whether the duration of TB treatment was important to vitamin D status, but found only a weak correlation between 25(OH)D3 concentration and days passed since the start of the 4-drug treatment regimen (Spearman's = 0.08, P = 0.12).
The controls were not matched, and the men were overrepresented in the TB group. Mandingas were overrepresented in the healthy control group, which possibly reflected differences in family structure, because Mandinga adults are more likely to be married and to have been present in the trios.
We found significant differences in schooling and nutritional variables; TB patients had less formal schooling and significantly lower BMI and albumin values. The total 25(OH)D3 concentrations were also significantly lower in TB patients than in controls.
Multivariate analysis
To assess whether the difference in vitamin D status between TB patients and healthy controls was due to background factors, we conducted univariate and multivariate analyses of hypovitaminosis D and deficiency (mVDD + sVDD), respectively, controlling for sex, season, ethnic group, religion, schooling, socioeconomic index, age-group, and BMI group. Only variables that affected the vitamin D estimate for TB by >10% were entered in the final multivariate analysis.
In the analysis for hypovitaminosis D, the univariate estimate was an OR of 1.33 (95% CI: 1.01, 1.76); when adjusted for clustering the OR was 1.33 (95% CI: 0.99, 1.78). Only BMI group affected the estimate and, when adjusted for BMI, TB was no longer associated with hypovitaminosis D (OR = 1.19; 95% CI: 0.87, 1.62), but low BMI was significantly associated with hypovitaminosis in this model (OR = 1.38; 95% CI: 1.0, 1.9).
The univariate estimate for the association between TB diagnosis and VDD was 0.62 (95% CI: 0.39, 0.97); when adjusted for clustering the OR was 0.62 (95% CI: 0.38, 1.0). Of the variables in Table 1, only the socioeconomic index affected this estimate. In a model including the 750 subjects with an available socioeconomic index, TB was significantly negatively associated with risk of VDD (OR = 0.53; 95% CI: 0.32, 0.90). In this model, the most poor socioeconomic group was insignificantly associated with risk of deficiency (OR = 2.1; 95% CI: 0.50, 9.0).
In a linear regression analysis, we assessed the influence of confounding factors on mean differences in 25(OH)D3 concentrations between TB patients and controls for the same background factors as in Table 1. None of these variables changed the significant difference shown in Table 2 between TB patients and controls by >5%. In a subgroup analysis of the 735 individuals with all background variables, the mean difference between TB patients and controls remained highly significant; 25(OH)D3 concentrations were lower in TB patients (8.1 nmol/L; 95% CI: 2.3; 13.9 nmol/L) than in controls. In this model, only lack of formal schooling was significant and raised mean 25(OH)D3 concentrations by 6.4 nmol/L (95% CI: 0.9, 11.8 nmol/L).
View this table:
TABLE 2. Analysis of predictor variables by vitamin D concentration1
Vitamin D status in healthy controls
A significant interaction of sex was observed on the association between TB status and vitamin D status (Table 2). With a cutoff at 50 nmol/mL (VDD), further significant interactions of ethnic and religious groups and a tendency for an interaction with rainy season (P = 0.09) were seen on the association between TB status and vitamin D status. For comparison with other populations, we therefore present 25(OH)D3 concentrations in the healthy population sample and risk factors for suboptimal vitamin D status.
Hypovitaminosis D was found in 42% (77/182) of the samples taken in the dry season and in 37% (116/312) of the samples taken in the rainy season (NS). However, for sVDD there was a difference because nearly all samples with sVDD were taken in the rainy season; the prevalence was 1% (1/182) in samples taken in the dry season and 7% (23/312) in those taken in the rainy season (P = 0.001). Mean 25(OH)D3 concentrations did not differ by season, as shown in Table 1.
Hypovitaminosis D was observed in 49% (77/157) of Moslems compared with 34% (82/243) of Christians and 36% (34/94) of animists (P = 0.007). The Fula ethnic group was highly associated with hypovitaminosis D and VDD; 63% (32/51) had 25(OH)D3 concentration 75 nmol/L (P < 0.001) and 22% (11/51) had 25(OH)D3 concentrations 50 nmol/L (P = 0.06), which was also observed in 21% (27/131) of the Pepel ethnic group (P = 0.003). Sex, BMI group, and no formal schooling were not found to be significantly associated with hypovitaminosis D or VDD.
The TST reaction was measured in only 426 adult controls, because not all of the controls could be located for test reading 2–3 d after application. We found a tendency toward a higher frequency of LTBI for those with VDD: 35% (18/51) with VDD and 25% (92/375) without VDD had a 25(OH)D3 concentration 50 nmol/L (P = 0.09).
In a logistic regression analysis, we examined the following background factors: age group (15–35, 35–50, and 50–87 y), sex, BMI group (13–20, 20–25, and 25–42), season, lack of formal schooling, ethnic groups (Balanta, Fula, Mandinga, Pepel, and others), and religious groups (Animist, Christian, and Moslem) as potential determinants of VDD and hypovitaminosis D. Female sex, Fula ethnic group, and Moslem religion were significantly associated with VDD, whereas only the Fula ethnic group was significantly associated with hypovitaminosis in this model (Table 2).
In a subgroup of 422 healthy controls with TST results, the ORs were 1.2 (95% CI: 0.8, 2.0) and 1.8 (95% CI: 0.9, 3.5), respectively, for hypovitaminosis or deficiency among individuals with LTBI when we controlled for significant background factors.
DISCUSSION
VDD and VDI occurred frequently in a large sample of healthy individuals and TB patients from a West African capital, as studies in other sub-Saharan populations also have shown (26–29). Among healthy controls, there was an insignificant tendency for individuals with LTBI to have VDD in both crude and adjusted analyses, which suggests a causal role of VDD in acquiring TB. Associations were also seen in the Fula and Pepel ethnic groups, but, after adjustment for background factors, only the association with Fula ethnicity and hypovitaminosis remained. The Fula group is the most light-skinned group in Guinea-Bissau, and we have no explanation for why they have a greater prevalence of hypovitaminosis because they are not known to be less exposed to the sun than the other ethnic groups in the area.
We hypothesized a relation between hypovitaminosis D and TB, and we found that hypovitaminosis D was indeed more frequent because of more frequent VDI in this group. However, we also found much less severe VDD among TB patients, which was unexpected. The lower prevalence of sVDD in active TB patients contrasts with other studies of vitamin D status in TB patients (2–7). The absence of sVDD among TB patients was surprising because TB patients were in a much worse nutritional condition than were the healthy controls with considerably lower BMI and albumin concentrations. The worse nutritional status of TB patients explains the higher degree of VDI among TB patients to some extent, because hypovitaminosis was not significantly more frequent among TB patients after the adjustment for BMI, which suggests that hypovitaminosis is a feature of bad nutritional status. However, mean 25(OH)D3 concentrations remained lower after adjustment for BMI, and the dietary intake of vitamin D is usually not considered sufficient to maintain good vitamin D status if sunshine is avoided (30, 31). Dietary differences are also unlikely to explain the contradictory finding of the absence of sVDD but significantly more prevalent VDI among TB patients.
The present study was, however, limited by the lack of detailed diet information. Fish is consumed regularly in the region of study, both freshwater and saltwater fish, but intake varied considerably within the different groups of the population (P Kaestel, personal communication, 2006). We have no data on the vitamin D content of these fish or individualized information on the intakes, but only fatty fish have significant amounts of vitamin D (32). All TB patients reported having eaten fish during the past week and may have, because of their disease, been allowed larger portions of the available meat in the family.
This study was further limited by the lack of information on sun exposure in the individuals, but we found more cases of hypovitaminosis D in the rainy season and virtually all cases of severe VDD were found in the rainy season. However, we found no clear seasonal difference, as was also reported in a study from Puerto Rico at the 18th latitude (33). Different exposures to sunlight may, to some extent, explain our findings, because controls were more often included in the study in the rainy season, which should be associated with a higher risk of sVDD. We would, however, expect this difference in timing of sampling to bias toward a lower mean 25(OH)D3 concentration in controls, which we did not find, and adjustment for season did not change the mean difference between TB patients and controls. The fact that formal schooling was less frequent among TB patients may have been the reason why more individuals in this group than in the control group engaged in outdoor manual labor, which could have led to more sun exposure in the TB group. This may explain why VDD was absent, but would also be expected to lead to higher mean 25(OH)D3 concentrations.
The absence of VDD in the TB patients may also be explained by the long diagnostic delay during which they may have been prescribed multivitamins, likely to be ergocalciferol. We sampled 6 different multivitamin brands from 5 pharmacies in Bissau. The 3 major brands are sold cheaply in small plastic bags and are the most common; they all contained ergocalciferol. Much more expensive brands that contain cholecalciferol are also available, but they are rarely sold. Hence, the rare occurrence of 25(OH)D2 in our measurements makes it unlikely that frequent multivitamin supplementation accounted for the differences found.
The simultaneous increased frequency of VDI and the complete absence of sVDD among the TB patients possibly explained the different vitamin D metabolism of the TB patients and controls. Liu et al (34) recently showed an effect on host vitamin D metabolism induced by infection with Mycobacterium tuberculosis via stimulation of toll-like receptors and induction of the 1,25-hydroxylating enzyme. Whether such an influence on vitamin D metabolism is of importance to serum 25(OH)D3 concentrations or whether serum 25(OH)D3 concentrations are important to host defense against TB remains to be shown, but a possible mechanism may be that VDD predisposes to the acquisition of LTBI or to the progression toward active TB disease, which then leads to increased production of vitamin D metabolites by granulomas (35, 36).
Perhaps simple random variation is the most obvious reason for our findings. Twenty-three of the 24 controls with sVDD were sampled during June-July 2005 within the same area of the city, and this group of 23 account for the differences in sVDD among TB patients and controls; hence, we may have encountered a cluster with prevalent sVDD. TB patients and controls had similar proportions of mVDD (Figure 1). It is also possible that TB patients with sVDD were the first to die; some identified TB patients died before inclusion in the study, and their vitamin D status is unknown. Mortality rates are very high in the study area, even among TB patients receiving treatment (11, 37–40). Furthermore, when TB patients overall have lower mean 25(OH)D3 concentrations but are absent in the group with lowest concentrations, selection bias is a possibility. As we follow this population prospectively, we will be able to assess the mortality risk associated with hypovitaminosis D in future studies.
There was, however, an important interaction of sex on the association between TB status and vitamin D status that modulated the risk of hypovitaminosis significantly. Sex has also been shown to modify the association between mortality risk and vitamin A supplementation in the study area (42), and we interpret this finding to possibly indicate that TB disease plays a role in vitamin D metabolism.
We found a lower albumin-corrected calcium concentration in the TB patients than in the healthy controls, which was also present when calcium concentrations were not corrected for albumin. This finding was likely explained by the fact that hypovitaminosis was more frequent among TB patients than among controls; calcium absorption is known to be impaired when 25(OH)D3 concentrations are <75 nmol/L (41).
A limitation of this study was the unmatched case-control design, which impedes strong conclusions when comparing TB patients and the random population sample. A prospective study following individuals with vitamin D insufficiency for the development of TB and changes in vitamin D status during the course of disease and treatment would, however, be difficult and costly.
In conclusion, hypovitaminosis D was highly prevalent among TB patients and healthy controls in a West African country; hypovitaminosis D was more frequent among the TB patients, but sVDD was very rare in this group. After adjustment for socioeconomic and demographic factors, hypovitaminosis D was not more frequent among TB patients than among healthy controls, but the mean differences in serum 25(OH)D3 concentrations remained lower. Furthermore, we reported a contradictory finding of less sVDD among the TB patients. The findings support the conclusion that the serum 25(OH)D3 concentration plays a role in TB infection, whether this role is a symptom or is causal was not established.
ACKNOWLEDGMENTS
We thank the dedicated field staff in Bissau, the hard working laboratory staff in Aarhus, Jens Nielsen for statistical consultancy, and Lene Heickendorf and Holger Jon Møller for advice about the vitamin D measurements.
The authors' responsibilities were as follows—CW (primary investigator): initiated the study and drafted the first version of the manuscript; RO: collected samples from the healthy control cohort; PR: collected clinical and demographic data; PG: provided advise on the field study design; PLA and HG: primarily responsible for the conception of the study and the data interpretation. PA and MS: helped draft the protocol and supervised the study conduct. All authors took part in the interpretation of the data and revision of the manuscript and participated intellectually and practically in the study. The authors solemnly declared that they had no personal or financial support or involvement with organizations with financial interest in the subject matter and had no conflicts of interest to disclose.
REFERENCES
1 From the Laboratory for Human Nutrition, Swiss Federal Institute of Technology, Zürich, Switzerland (MBZ and RB); the Division of Human Nutrition, Wageningen University, Wageningen, Netherlands (MBZ); the Medical Research Council, Cape Town, South Africa (PLJ and SS); and the University of Venda, Thohoyandou, South Africa (NSM, LFM, and XM)
2 Supported by the Swiss National Science Foundation (Bern, Switzerland), the Medical Research Council (Cape Town, South Africa), and ETH Zürich (Switzerland). Guerbet (Roissy CdG Cedex, France) provided the iodized oil capsules, and The Foundation for Micronutrients in Medicine (Rapperswil, Switzerland) provided the placebo capsules. 3 Reprints not available. Address correspondence to MB Zimmermann, Laboratory for Human Nutrition, ETH Zürich, LFV E19, Schmelzbergstrasse 7, CH-8092, Zürich, Switzerland. E-mail: michael.zimmermann{at}ilw.agrl.ethz.ch.
ABSTRACT
Background: Vitamin A (VA) deficiency (VAD) and iodine deficiency (ID) often coexist in children in Africa. VAD may affect thyroid function and the response to iodine prophylaxis.
Objective: The aim was to investigate the effects of supplementation with iodine or VA alone, and in combination, in children with concurrent VAD and ID.
Design: A 6-mo randomized, double-blind, 2 x 2 intervention trial was conducted in 5–14 y-old South African children (n = 404), who, on average, had mild-to-moderate VAD and ID. At baseline and after 3 mo, children received 1) iodine (191 mg I as oral iodized oil) + placebo (IS group), 2) VA (200000 IU VA as retinyl palmitate) + placebo (VAS group), 3) both iodine and VA (IS+VAS group), or 4) placebo. At baseline, 3 mo, and 6 mo, urinary iodine (UI), thyroid volume, thyrotropin (thyroid-stimulating hormone; TSH), total thyroxine (TT4), thyroglobulin, serum retinol (SR), and retinol-binding protein (RBP) were measured.
Results: SR and RBP increased significantly with VA supplementation (P < 0.05). For UI, SR, and RBP, there were no significant treatment interactions between iodine and vitamin A. The 3-factor and all three 2-factor interactions were significant for thyroid volume, TSH, and thyroglobulin (P < 0.001), whereas none of these interactions were significant for TT4. There was a clear effect of VAS without IS on TSH, thyroglobulin, and thyroid volume; all 3 variables decreased significantly (P < 0.05).
Conclusions: Iodine prophylaxis is effective in controlling ID in areas of poor vitamin A status. VA supplements are effective in treating VAD in areas of mild ID and have an additional benefit—through suppression of the pituitary TSHß gene, VAS can decrease excess TSH stimulation of the thyroid and thereby reduce the risk of goiter and its sequelae.
Key Words: Vitamin A • iodine • supplementation • deficiency • Africa • children
INTRODUCTION
More than one-third of the global population is affected by either vitamin A (VA) deficiency (VAD) or iodine deficiency (ID) (1–4). These deficiencies often coexist, and 32–50% of school-age children in rural West and North Africa have both VAD and goiter (5–7). High-dose oral VA supplementation (VAS) is a recommended strategy to control VAD in affected populations (2), many of whom are also iodine deficient. Conversely, many VA-deficient children in the developing world are consuming iodized salt.
Thyroid metabolism and the response to iodine prophylaxis in areas of ID are influenced by multiple nutritional factors (8–13), including VA status (14, 15). VAD has multiple effects on thyroid function in animals (14): 1) in the thyroid, VAD decreases thyroidal iodine uptake and iodine incorporation into thyroglobulin and increases thyroid size (16–20); 2) in the periphery, VAD increases circulating thyroid hormone concentrations (21); and 3) in the pituitary, VA status modulates thyrotropin (thyroid-stimulating hormone; TSH) production by retinoid X receptor (RXR)–mediated expression of pituitary TSHß mRNA (22–27), and VAD in rats increases pituitary TSHß mRNA, TSH, and circulating thyroid hormone (21). In a recent study (28), rats with concurrent ID and VAD were supplemented with iodine and VA, alone and in combination. Primary hypothyroidism in animals with concurrent VAD and ID did not reduce the efficacy of VAS, nor did VAD reduce the efficacy of iodine to correct thyroid dysfunction due to ID. Moreover, VAS given alone without iodine supplementation reduced pituitary TSHß mRNA expression, circulating TSH, and thyroid weight (28).
In a cross-sectional study of African children, VAD in children with severe ID was associated with an increase in TSH stimulation and thyroid size and a reduced risk of hypothyroidism (15). In the same population, an intervention trial compared the efficacy of iodized salt alone to iodized salt given with VAS and found greater decreases in TSH and thyroid volume in the IS+VAS group (15). However, in that study, there was no control group and VAS was given only with iodized salt. Therefore, the aim of the present study was to investigate the safety and efficacy of repletion with iodine or VA alone, and in combination, in school-age children with concurrent VAD and ID.
SUBJECTS AND METHODS
The study was conducted in Limpopo Province in South Africa. The subjects were 5–14-y-old children from rural primary schools. Power calculations indicated that a sample size of 85 children per group was required to yield 80% power at 5% significance to detect a 20 nmol/L difference in mean total serum thyroxine concentration, which allowed for 10% loss to follow-up. Ethical approval for the study was obtained from ETH Zürich, the University of Venda, and the Provincial Department of Education in South Africa. Informed written consent was obtained from the parents, and oral assent was obtained from the participating children. All children in the schools were invited to participate; the only exclusion criteria were major chronic medical illnesses and recent consumption of iodine, VA supplements, or both. None of the consenting children were excluded on the basis of these criteria. At baseline, weight was measured with a TANITA Digital Scale 1631 (Itin Scale, Brooklyn, NY) and height with a rigid stadiometer. A spot urine sample was collected for measurement of urinary iodine concentration (UI). Whole blood was collected by venipuncture for the measurement of total thyroxine (TT4), thyrotropin (TSH), thyroglobulin, serum retinol (SR), retinol-binding protein (RBP), and C-reactive protein (CRP) concentrations. Thyroid volume was measured with a portable Aloka SSD-500 Echocamera (Aloka, Mure, Japan) or a Toshiba Justvision 200 (Toshiba, Tokyo, Japan) with high-resolution 7.5-MHz linear transducers (29).
The 6-mo study was a double-blind trial that used a 2 x 2 factorial design. Children from the screening were randomly assigned to receive, at baseline and at 3 mo, one of the following: 1) iodine (191 mg I as oral iodized oil; Lipiodol, Guerbet, Paris) + placebo (sunflower oil) (IS group), 2) vitamin A (200000 IU as retinyl palmitate; RpScherer, Aprilia, Italy) + placebo (VA group), 3) both iodine and VA (IS+VAS group), or 4) placebo. The capsules were swallowed with water under direct supervision by the investigators. Baseline measurements were repeated at 3 and 6 mo. At completion, all study children who had not received supplementation were given VAS, IS, or both.
Laboratory analyses
Serum and urine samples were portioned and frozen at –20°C until analyzed. UI was measured by using the Pino modification of the Sandell-Kolthoff reaction (30). At UI concentrations of 47 and 79 µg/L, the CVs of this assay in our laboratory are 10.3% and 12.7%, respectively. Whole-blood TSH and serum TT4 were measured by using immunoassays (31); normal reference values are <3.7 mU/L for TSH and 65–165 nmol/L for TT4. Whole blood was spotted onto filter paper, and thyroglobulin was measured by using an immunoassay; the normal reference value is 4–40 µg/L thyroglobulin (32). SR was measured by HPLC (33) and RBP by an immunoassay (Immundiagnostik AG, Bensheim, Germany). VAD was defined as an SR concentration <0.70 µmol/L (2), and an SR concentration <1.05 µmol/L indicated low VA status (4). CRP was measured by nephelometry (TURBOX; Orion Diagnostica, Espoo, Finland); the normal reference value is a CRP concentration <10 mg/L.
Data and statistical analyses
Data processing and statistical analyses were done by using SPLUS (2000; Insightful Corporation, Seattle, WA), and PRISM (version 3; GraphPad, San Diego, CA). References for thyroid volume in school-age children according to sex and body surface area were used to define goiter (29). EPINFO (version 3.3.2; Centers for Disease Control and Prevention, Boston, MA) was used to calculate height-for-age z scores and weight-for-age z scores using World Health Organization (WHO) references. If data were not normally distributed, statistical analyses were done after log transformation. We studied the effects of the 2 treatments and of their interactions by analysis of covariance. If the interaction was significant, the differences were given for both groups, those also receiving the other treatment and those not receiving it. Because of the potential confounding effects of inflammation on SR and RBP, SR and RBP values from children with an elevated CRP (>10 mg/L) were not included in the data analyses. Significance was set at P < 0.05.
RESULTS
The results of the baseline screening are shown in Table 1. Overall, the children were reasonably well-nourished, as reflected by median height-for-age and weight-for-age z scores of –0.56 and –0.59, respectively. However, 12% of the children had an SR concentration <0.7 µmol/L, which indicated that VAD in this region is a moderate public health problem according to the WHO (34). The median UI was 74 µg/L, which indicated mild ID; 31% of the children had a UI <50 µg/L, and 27% of the children were goitrous. Thyroid hormone concentrations were generally in the normal range; only 4.5% of the children were hypothyroxinemic. At baseline, 3 mo, and 6 mo, 3–6% of children had an elevated CRP value, but no differences in the prevalence of elevated CRP values between the 4 groups was found at any time point (data not shown). At 6 mo, because subjects moved away from the study area or were absent on the days of data collection, the dropout rates in the placebo, IS, VAS, and IS+VAS groups were 7%, 8%, 5%, and 7%, respectively.
View this table:
TABLE 1. Baseline characteristics of children from primary schools in Limpopo Province, South Africa
There were no significant differences in any of the baseline variables in Table 2, Table 3, or Table 4 between the 4 groups after randomization. As shown in Table 3, both iodine and VA supplementation were effective: the iodine-by-time interaction was significant for UI, with an effect size of 117.9 µg/L (P < 0.0001); the vitamin A–by-time interaction was significant for SR and RBP, with effect sizes of 0.029 µmol/L and 9.47 mg/L, respectively (P < 0.001). However, for UI, SR, and RBP, there were no significant treatment interactions between iodine and vitamin A. Changes in the thyroid variables are shown in Table 4. The 3-factor and all three 2-factor interactions were significant for thyroid volume, TSH, and thyroglobulin (P < 0.001), whereas none of these interactions were significant for TT4.
View this table:
TABLE 2. Age, sex ratio, weight, and height in children who received, at baseline and at 3 mo, 1 of 4 treatments1
View this table:
TABLE 4. Thyroid volume (Tvol), whole-blood thyrotropin (TSH) and thyroglobulin (Tg), and serum total thyroxine (TT4) in children who received, at baseline and after 3 mo, 1 of 4 treatments1
View this table:
TABLE 3. Urinary iodine (UI), serum retinol (SR), and retinol-binding protein (RBP) concentrations in children who received, at baseline and at 3 mo, 1 of 4 treatments1
DISCUSSION
Our findings showed that mild ID does not impair the SR and RBP response to VAS in children with concurrent ID and VAD. Conversely, the data also indicate that mild VAD does not reduce the efficacy of IS to correct thyroid dysfunction in children with concurrent ID and VAD. These latter findings differ somewhat from those of several animal studies in which severe VAD impaired the pituitary-thyroid axis, even when the iodine supply was adequate. The adverse effects in these studies included reduced thyroidal iodine uptake (19), impaired synthesis of thyroglobulin and coupling of iodotyrosine residues to form thyroid hormone, and reduced hepatic conversion of T4 to triiodothyronine (T3) (20). However, a recent study in rats with only mild-to-moderate VAD and ID (28) reported findings similar to those in the present study; that is, provision of an iodine-sufficient diet entirely reversed the abnormalities of the pituitary-thyroid axis produced by VA and iodine depletion, regardless of VA status.
Our data indicate that VAS in children receiving IS had minimal effects on the thyroid axis compared with the effects of IS alone (Table 4). These findings contrast with those of several previous studies in iodine-sufficient animals, in which pharmacologic doses of VA did affect the pituitary-thyroid axis and decreased thyroid size, pituitary TSH content, and circulating TT3 and TT4 (16, 35). A similar effect was reported in lymphoma patients who developed hypothyroidism after treatment with a synthetic retinoid that specifically binds to the RXR (36). In contrast, a recent study in iodine-sufficient rats with mild-to-moderate VAD given high-dose VAS ( 50 mg/kg body wt), reported that VA treatment had no discernible effects on the pituitary-thyroid axis (28). Together with the findings of the present study, these data suggest that high-dose VAS in iodine-sufficient areas is unlikely to affect thyroid function.
The major new finding of this study was that VAS alone in iodine-deficient children with mild VAD reduces circulating TSH, serum thyroglobulin, and thyroid size without significantly affecting thyroid hormone concentrations (Table 4). This finding is consistent with that of a previous report in hypothyroid rats, in which pharmacologic doses of VA reduced basal TSH secretion and the TSH response to thyroid-releasing hormones (27). This finding is also consistent with that of a recent study in iodine- and vitamin A–deficient rats (28), in which VAS reduced TSHß mRNA expression, serum TSH concentrations, and thyroid weights. Control of TSH production by the pituitary depends on 2 main factors: the binding of the thyroid hormone receptor, which is activated by T3 and T4, and the binding of the RXR, which is activated by retinoic acid (23). Both receptors suppress the transcription of the pituitary TSHß gene by occupying half-sites on the promoter DNA of the gene; thus, VA status modulates TSH production (23–25). In the present study, reduced TSH stimulation might have been expected to reduce thyroid hormone production, but circulating concentrations of TT4 did not decrease. This implies that either the sensitivity of the thyroid to TSH improved with VA repletion or that the metabolism of circulating thyroid hormone was altered to maintain their concentrations. The findings of previous animal studies support both of these mechanisms. For example, Ingenbleek (20) reported that a combination of VAD and ID in rats impaired thyroid hormone synthesis by reducing iodine incorporation into thyroglobulin; these adverse effects were reversed by VA treatment. On the other hand, Morley et al (35) found that high-dose VAS in iodine-sufficient rats altered peripheral thyroid hormone metabolism and increased hepatic conversion of T4 to T3.
These results are important in the context of the findings from a previous cross-sectional study in African children (15), in which increasing VAD severity was a predictor of greater thyroid volume and higher concentrations of TSH, thyroglobulin, and TT4; in children with VAD, the odds ratio for goiter was 6.51, whereas the odds ratio for hypothyroidism in VAD was 0.06. A concern raised by that study (15) was that VAS given alone in areas of ID might reduce pituitary TSH secretion and thereby impair thyroid hormone production in the face of marginal iodine status. The findings of the present study do not support that contention; there was no significant decrease in thyroid hormone concentrations in the children receiving VAS, despite the reduction in circulating TSH.
In a previous study (15), the efficacy of iodized salt alone was compared with that of iodized salt given with VAS (200000 IU at 0 and 5 mo) in a 10-mo trial. In contrast with the results in the present study, in which IS+VAS had no measurable effect on thyroid status compared with IS alone, in the study by Zimmermann et al (15), mean thyroglobulin, median TSH, and the goiter rate significantly decreased in the IS+VAS group compared with iodized salt alone. The varying results of these 2 studies may have been due to differences in the severity of ID, study length, the route of iodine repletion (iodized oil compared with salt), or a combination thereof.
It is possible that only in the case of severe ID, when TSH stimulation of the thyroid is very high, does the additive effect of concurrent VA repletion become apparent. Thus, the results in this population with moderate VAD and mild ID may not be generalizable to populations with more severe deficiencies.
The findings of the present study may at least partially explain the inverse correlation between VA status and goiter reported in previous cross-sectional studies in developing countries. In Senegalese adults, there was a strong negative correlation between increasing severity of goiter and SR and RBP concentrations (37, 38). In Ethiopian children, those with visible goiters had significantly lower SR and RBP concentrations than did children without goiters or with smaller goiters (39). However, in these studies it was not possible to distinguish the effects of VAD from protein malnutrition, which also can reduce SR and RBP concentrations. Among adequately nourished 7–14-y-old Philippino children, the prevalence of goiter was 1.8% in children without VAD and 5.3% in those with VAD (40).
In summary, our data indicate that iodine prophylaxis will be effective in controlling ID, even in areas of poor VA status, and high-dose VAS is likely safe and effective in areas of mild ID. The findings suggest that high-dose VAS in a population can modify indicators of ID, such as thyroglobulin and goiter, independent of a change in iodine nutrition. In areas of endemic goiter, through suppression of transcription of the pituitary TSHß gene, VAS may decrease excess TSH stimulation of the thyroid and thereby reduce the risk of goiter and its sequelae.
ACKNOWLEDGMENTS
We thank the teachers and children in the schools for their participation. We also thank TI Mbhenyane, AP Netshivhulana, CN Nesamvuni, HV Mbhatsani, and TC Mandiwana (University of Venda, Thohoyandou, South Africa); E Strydom (Medical Research Council, Cape Town, South Africa); and F Hilty (ETH Zürich, Switzerland) for their technical assistance.
Each of the authors made substantial contributions to the study design, data collection, and data analyses and to the writing, editing, or writing and editing of the manuscript. None of the authors has a personal or financial interest in the companies or organizations sponsoring this research, including advisory board affiliations.
REFERENCES
1 From the Johns Hopkins Bloomberg School of Public Health (JNN, KOO, S-CC, and LEC), the Johns Hopkins Hospital (FRW, JM, and MSN), and the Johns Hopkins School of Medicine, Department of Obstetrics and Gynecology (FRW), Baltimore, MD
2 Reprints not available. Address correspondence to KO O'Brien, Cornell University, Division of Nutritional Sciences, MVR, Room 340, Ithaca, NY 14853-6301. E-mail: koo4{at}cornell.edu.
ABSTRACT
Background: Because pregnant African American women and teens are at risk of low birth weight, they are frequently counseled to strive for gestational weight gains at the upper limits of the Institute of Medicine's recommended ranges.
Objective: The objective was to examine whether such weight gains improve birth outcomes in a cohort of disadvantaged African American adolescents of low (<19.8), average (19.8 to 26.0), or high (>26) prepregnancy body mass index (BMI; in kg/m2).
Design: Data were extracted from the medical charts of 1120 African American adolescents who received prenatal care at an inner-city maternity clinic between 1990 and 2000 and analyzed by using analysis of covariance and multivariate regression methods.
Results: Data were available for 815 adolescents, 711 of whom delivered at term (37 wk). Fifty-eight percent (n = 409) of all term deliveries and 74% of the high-BMI adolescents (n = 126) had gains in the upper half of or above the recommended ranges. For all BMI groups, the most significant differences in birth outcomes were found in comparisons of teens who gained below the recommended ranges with those who gained in the lower half of the recommendation range. Further gains were not clearly beneficial, particularly for infants of high-BMI mothers.
Conclusions: African American adolescents entering pregnancy underweight or at average weight should be counseled to gain within the recommended ranges, whereas overweight adolescents need support to avoid excessive gestational weight gain. Such advice would be prudent in light of the known associations between obesity and the increased likelihood of chronic diseases.
Key Words: Obesity • body mass index • disadvantaged African American adolescents • pregnancy • birth outcomes
INTRODUCTION
Pregnant adolescents in the United States are at high risk of adverse birth outcomes, including preterm delivery, low birth weight, and increased perinatal mortality (1). These elevated risks are thought to be due in part to the greater prevalence of such biological risk factors as low prepregnancy weight, the heightened nutritional demands of growth, and the structural and hormonal consequences of physiologic immaturity (2). The risks are also due to socioeconomic factors: adolescent mothers are more often poor, are more often members of racial minorities, have lower educational levels, and receive suboptimal prenatal care (3, 4).
Appropriate nutrient intake and weight gain during pregnancy are considered 2 of the most important modifiable behaviors for improved maternal and infant outcomes (5). In its 1990 report, the Institute of Medicine (IOM) suggested that because of their higher risk of lower birth weight for a given weight gain, young adolescents and African American women should be advised to gain at the upper limits of the IOM's recommended weight gain ranges for women of low and average prepregnancy body mass index (BMI: weight in kg/height squared in m) (5). However, adolescents have also been shown to be at increased risk of excessive gestational weight gain (2, 6, 7), and African American women have been found to have an increased risk of postpartum weight retention (8, 9) and overweight and obesity (10, 11). For African American women in particular, there is increasing evidence that higher weight gains during pregnancy do not improve infant outcomes and instead may elevate the mothers' long-term risk of chronic disease (12, 13). In 1997 an expert panel proposed an update of the 1990 IOM recommendations, suggesting that adolescents and African American adults be counseled to stay within the IOM-recommended BMI-specific weight gain ranges (14). Further research on the relation between observed weight gain and birth outcomes in African American adolescents is needed to provide a more solid basis for recommendations about optimal weight gain. The purpose of this study was to evaluate associations between levels of gestational weight gain and birth outcomes in a large sample of African American adolescents who delivered singleton infants in Baltimore between 1990 and 2000.
SUBJECTS AND METHODS
Subject characteristics
The data set used for this analysis was compiled from a retrospective medical chart review of all adolescents aged 17 y at conception who received prenatal care at the Maternity Center East clinic. This inner-city clinic was affiliated with the Johns Hopkins Hospital and served a predominantly disadvantaged African American community. Methods used to extract the medical data have been described elsewhere (15). Combined prepregnancy, weight gain, and birth weight measures were available for 815 (73%) of the 1120 African American pregnancies in the database. Examination showed no significant differences in maternal age, height, parity, smoking, or incidence of preeclampsia between subjects with and without birth weight or weight gain data (data not shown). Data on the characteristics of this population and on relations between hematologic status and birth outcomes, diet, and fetal growth were previously reported (15–17). The study was approved by the Joint Committee of Clinical Investigation at the Johns Hopkins Hospital.
Weight gain determinations
Prepregnancy BMI was calculated by using measured height and self-reported prepregnancy weight. Although the tendency for overweight subjects to underreport weight and for underweight subjects to overreport weight may introduce bias, previous research suggests that self-reported weights correlate well with actual weight among adults (18) and among teens (19). In these populations the correlation between reported prepregnancy weight and the first measured weight was 0.93; however, a detailed inspection of our data set showed that 7% of adolescents had large differences between the recalled prepregnancy weight and a first measured weight in the first trimester (gains of 1 kg/wk). The data were therefore adjusted according to the following rules: 1) accept reported prepregnancy weight if the resulting rate of gain was <1 kg/wk during the first trimester; otherwise 2) extrapolate prepregnancy weight if 2 weight gain measures were recorded before 15 wk gestation, from which a first trimester rate of gain could be estimated; 3) replace reported prepregnancy weight with the first measured weight if a measure was recorded by 8 wk gestation but the second measure was recorded after 15 wk; 4) drop observation if the rate of gain was implausible (1 kg/wk) and the correction was not feasible.
Subjects were then classified according to the adult references used by the IOM into 4 categories: underweight (BMI < 19.8), average weight (BMI = 19.8–26.0), overweight (BMI = 26.1–29.0), and obese (BMI > 29). Because few overweight and obese adolescents delivered small-for-gestational-age infants (n = 12 and n = 14, respectively), and because weight gain recommendations are similar for these 2 groups, the data for these 2 groups were combined for the analysis. In addition, a preliminary analysis indicated that differences in total weight gain and birth weight between these 2 groups were not statistically significant.
The subjects were weighed at each prenatal visit in street clothes and no shoes following standard clinical practice, and total weight gain was calculated as the difference between the final recorded weight within 4 wk of delivery and the prepregnancy weight; 86% of these final weight measures were made within 2 wk of delivery. The number of weeks between the final weight measure and delivery was included in all multivariate analyses. In addition to being examined as a continuous variable, total weight gain at term (37 wk) was examined in 4 categories: below the IOM recommendation for BMI, in the lower half of the recommended range (12.5–15.2 for low-, 11.5–13.8 for average-, and 7.0–9.2 kg for high-BMI adolescents), in the upper half of the recommended range (15.3–18 kg for low-, 13.9–16 kg for average-, and 9.3–11.5 kg for high-BMI adolescents), and above the recommendation for BMI. Date of last menstrual period was estimated on the basis of a best obstetrical estimate algorithm using all available data (self-reported last menstrual period, physical examination, and earliest available ultrasound), and length of gestation was calculated as the distance between birth and last menstrual period (15).
Infant birth weight classification and outcome variables
The outcome variables examined included average birth weight and small-for-gestational-age (SGA) size, which was defined as birth weight less than the 10th percentile for gestational age according to a national sex-specific reference for fetal growth based on 1991 natality data (20). SGA birth weight was selected as an outcome variable rather than low birth weight because it adjusts for duration of pregnancy. Large-for-gestational-age (LGA) size according to the same national reference was also examined, but the incidence was too low (2%) to allow for multivariate analysis. Birth weight at term was categorized as suboptimal (<3000 g), optimal (3000–4000 g), and above optimal (>4000 g) on the basis of previous research findings that adverse outcomes are minimized for infants of African American adult and adolescent mothers when birth weights range between 3000 and 4000 g (21).
Other variables in the data set were examined for their association with these outcomes and with gestational weight gain on the basis of findings previously reported in the literature. These variables included young maternal chronologic age (<15 y at infant birth), low gynecologic age (2 y since the self-reported date of the onset of menses), parity (0 versus 1), height, self-reported history of smoking or drug use (any versus none), medical diagnosis of preeclampsia (blood pressure > 140/90 mm Hg accompanied by abnormally high urinary protein and symptoms of edema), health insurance status (private, Medicare, or none), iron status (results of hemoglobin and hematocrit screens, typically conducted at entry into prenatal care and at 28 wk gestation), gynecologic infections diagnosed at any time during pregnancy (including chlamydia, gonorrhea, bacterial vaginosis, vaginal infections, and urinary tract infections), adequacy of prenatal care as measured by the Kotelchuck index (based on timing of entry, total number of prenatal visits, and gestational age at delivery) (22), and infant sex.
Statistical analyses
Descriptive analyses included visual examination of distributions, means, and SDs for continuous variables and frequency distributions for categorical variables. Chi-square analysis was applied to test relations between pairs of categorical variables and one-way analysis of variance to test for significant relations between categorical and continuous variables. Significance for bivariate relations was set at P < 0.05. Analysis of covariance (ANCOVA) was used to examine differences in mean birth weight by prenatal weight-gain category (below, in the lower half of, in the upper half of, and above recommended ranges) with adjustment for variables that proved significant in the bivariate analyses and remained significant in the ANCOVA equation. Logistic regression models were developed to examine the effect of weight gain (as a continuous and as indicator variables for the 4 weight-gain categories) on the risk of being born SGA and suboptimal birth weight (<3000 g) after control for covariates and potential confounding factors. Linear regression models were used to test these effects on birth weight. Associations with SGA and suboptimal and mean birth weights were examined for the group as a whole and stratified by prepregnancy BMI. Because each BMI stratum included a wide range of BMI values, BMI was also included in the multivariate analyses as a continuous variable. All variables for which there was a significant difference for any BMI group or which are known confounders of the relation between weight gain and birth weight were tested in multivariate models. Significance for retention in the multivariate models was defined as P 0.10. We found a significant interaction between BMI and weight gain for both birth weight and suboptimal birth weight outcomes, which provided further evidence of the importance of stratifying our analysis by BMI and weight-gain categories. This interaction effect was not significant for SGA birth weight, possibly because of its lower incidence; thus, only main effects of this outcome are presented. Statistical analyses were performed by using STATA release 8.2 (Stata Corp LP, College Station, TX).
RESULTS
The distribution of selected anthropometric and sociodemographic variables in the group as a whole, and stratified by prepregnancy BMI, is presented in Table 1. Underweight adolescents were more likely to be taller (P < 0.01) and of low gynecologic age (P < 0.001); overweight adolescents were more likely to be older (P = 0.02) and multiparous (P < 0.01). Overweight adolescents were less likely to undergain weight and about twice as likely to overgain weight (P < 0.001), similar to the patterns found in adults (23). However, unlike many adult African American populations (14), our teens were more likely to overgain than to undergain weight.
View this table:
TABLE 1. Selected characteristics of African American adolescents by prepregnancy BMI (in kg/m2) category1
Mean (±SD) birth weight in this population was 3062 ± 647 g, and 37% of the newborns had suboptimal birth weights (<3000 g) (Table 2). The incidence of SGA birth weight (16%) was elevated and of LGA size (2%) was low compared with the incidence in adult African Americans from the same community (12). Risk of SGA and suboptimal birth weight were significantly and negatively associated with prepregnancy BMI (P < 0.02 and P < 0.001, respectively), gestational weight gain (P < 0.001 for both), and maternal height (P = 0.03 for both). The rate of smoking was low in this cohort; nevertheless, it was significantly associated with suboptimal birth weight (P = 0.05). Preeclampsia more frequently coincided with SGA birth weight (P < 0.01) but not with suboptimal birth weight (P = 0.08).
View this table:
TABLE 2. Unadjusted birth weight outcomes for African American adolescents by prepregnancy BMI (in kg/m2)1
The actual mean birth weights of infants born at term (37 wk) to mothers gaining below, within the lower or upper half, or above the BMI-specific recommended weight gain ranges and birth weights adjusted for potentially confounding factors are shown in Table 3. Also presented are the proportions of infants in each weight-gain category that were of suboptimal birth weight. In general, the most biologically and statistically significant reductions in birth weight were found in comparisons of gains below the recommended ranges with those in the lower half of the recommended range. The significance of further improvements in birth weight from gains in the upper half of the recommended ranges was inconsistent; the data adjusted for confounders suggested biologically modest additional increases in birth weight, with the highest increases (88.4 g) seen in the average-BMI group. Further increases in birth weight were achieved from weight gains above the recommended ranges, but the mean gestational weight gain in this category for all 3 BMI groups was >20 kg.
View this table:
TABLE 3. Actual and adjusted birth weights (g) for term deliveries of African American adolescents (n = 705) by low, average, and high prepregnancy BMI (in kg/m2) and gestational weight gain category1
The incidence of SGA and suboptimal birth weight in mothers who gained less than the recommended range for their BMI compared with those who gained within or above their recommended ranges were significantly higher in the low-BMI (P < 0.0001) and average-BMI (P < 0.01) groups but not in the high-BMI group (P > 0.30), although it should be noted that only 16.5% of these high-BMI teens gained below the recommended range. Differences in the rates of these outcomes between those who gained in the lower half and those who gained in the upper half of the recommended ranges were also inconsistent; however, the number of subjects in some categories was relatively small. For example, in the high-BMI teens, the incidence of SGA birth weight did not differ significantly between those who gained in the lower half of the recommended range and those who gained in the upper half of the recommended range, whereas the incidence of suboptimal birth weight increased. In low-BMI teens, the incidence of SGA birth weight increased, whereas the incidence of suboptimal birth weight decreased. In the average-BMI teens, for whom the sample size was largest, the incidence of both outcomes decreased. For all BMI groups, LGA birth weight and above-optimal birth weights (>4000 g) occurred almost exclusively in those who gained above their recommended range (data not shown).
For the group as a whole, the risk of SGA birth weight after control for potentially confounding variables was found to be significant (Table 4), which indicated that the largest benefits were achieved with gains from below to within the lower half of the recommended ranges. Regression analyses of birth weight outcomes examined separately by BMI category suggested that the effect on birth weight of increasing gains from below to within the recommended range was most significant for low- and for high-BMI teens (birth weight increased by 260 g in both groups).
View this table:
TABLE 4. Effect of weight-gain category on risk of small-for-gestational-age (SGA) birth in African American adolescents who delivered at term1
DISCUSSION
In this cohort of pregnant African American adolescents, birth weight outcomes markedly improved in teens in all BMI groups when weight gain increased from below to within the lower half of the IOM recommended range. The additional benefits from gains in the upper half of the recommended range were modest or equivocal. Additional weight gain conferred the greatest benefit on infants of low- and average-BMI teens and had the least effect on overweight teens, most of whom had infants of healthy birth weight regardless of the level of weight gain. Furthermore, there appeared to be a tendency for higher weight gains to be associated with a higher proportion of adverse outcomes in this BMI group. The vast majority of the heavier teens (64%) gained above the recommended range, which likely increased their risk of obesity and its associated chronic diseases. Moreover, whereas 23% of the study sample entered pregnancy overweight (n = 191), an additional 18.5% (n = 151) entered pregnancy with low or average BMIs but gained >40 kg; 22.5% (n = 183) entered pregnancy with low or average BMIs and gained more than the IOM recommended range and therefore were at increased risk of overweight after pregnancy. These risks may be considerable in light of research indicating that African American women have elevated rates of postpartum weight retention (9, 24).
These findings are consistent with a number of other recent studies that recommend a reconsideration of the IOM weight recommendations for adolescents and African Americans (12, 14, 25, 26). A study by Schieve et al (25), which examined a much larger population of African American adult women (n = 33 101), reported adjusted birth weights that were slightly higher than ours and a considerably lower incidence of low birth weight but very similar trends. With greater power than the current study, Schieve et al also found that increases in gestational gains had a somewhat contradictory effect on the birth weight outcomes examined (mean birth weight and risk of low birth weight) in African Americans and concluded that gains in the upper half of the recommended range were of questionable benefit. Hickey et al (26) came to similar conclusions after examining a cohort of 2219 African American adults. Another study, which examined adult gravidas in Baltimore (12), calculated that among black women of average and high BMI, the gains required to reduce their risk of SGA birth weight to levels observed among white women would be excessive (25 kg for women of average prepregnancy BMI) and thus ill-advised. All 3 studies, as well as an expert panel convened in 1996 by the Federal Maternal and Child Health Bureau (14), cited the lack of clear evidence of benefits and the increased risk of postpartum weight retention and obesity among black women as reason to be cautious about recommending higher gestational weight gains. Our study was the first to examine the effect of weight gain below, within, and above the IOM recommendations in a sample of adolescents.
The African American teens in our study showed less tendency to undergain than did African American women in other studies (14, 26), but these findings were consistent with studies of adolescents (2, 14). In our study population, 29% of the adolescents who delivered at term gained excessive amounts of weight (>18 kg, or gains above the highest upper limit recommended by the IOM), a similar proportion to that reported for other adolescent cohorts (2, 6). Despite their higher weight gains, the median birth weight of 3151 g for this cohort was below the 1991 national reference of 3495 g for infants of similar gestational age (20), below the reference birth weight for African Americans of 3330 g (27), but similar to the birth weight reported for other adolescents (28). Thus, as did other researchers who studied pregnancy during adolescence, we observed that higher maternal weight gains did not necessarily contribute to higher infant birth weights.
The 1990 IOM report noted that the limited data available suggest that young adolescents (<2 y postmenarche) may give birth to smaller infants for a given weight gain than do older women (5). We found this to be the case in our African American cohort. In the subjects who delivered at term for whom we had data on age at menarche (n = 607), those of low gynecologic age had infants with significantly lower mean birth weights than did those of higher gynecologic age (3079 ± 347 compared with 3237 ± 437 g; P = 0.02), despite similar mean weight gains (15.1 ± 7.4 compared with 14.8 ± 7.1 kg; P > 0.10). This reduction in birth weight remained significant after control for smoking, prepregnancy BMI, parity, preeclampsia, gestation duration, and weight gain. However, total gestational gain and birth weight of infants born to mothers of young chronologic age (<15 y at infant birth) who delivered at term were not significantly different from those of older mothers (16.1 ± 6.5 compared with 14.8 ± 7.1 kg; P > 0.20) and (3269 ± 464 compared with 3218 ± 427 g; P > 0.40), respectively. The incidence of SGA and suboptimal birth weight did not differ significantly by either gynecologic or chronologic age.
Our adolescents were recruited from a disadvantaged community. According to the 2000 census, the median income in their neighborhood was <$27 000; 30.8% of the households with children aged <18 y were below the poverty level, 96.4% of the population was African American, and violent crime rates were high (29). A growing body of literature shows that neighborhood poverty and violence, racial discrimination, and associated psychological stress are associated with the risk of poor birth outcomes, particularly among African Americans (30, 31). It is possible that such adversity contributed, in ways that we did not measure, to the elevated rates of growth restriction and suboptimal birth weight observed.
Some limitations to our study should be considered when interpreting these results. Prepregnancy weight was self-reported, and we do not know whether the adolescents were willing and able to report their prepregnancy weight accurately. We adjusted 6.6% of the reported weights, and it is possible that additional inaccurate weights were undetected, which could have led to misclassification by both prepregnancy BMI and weight-gain category. Our adjustments tended to increase estimates of prepregnancy weight and decrease estimates of gestational weight gain but did not affect our overall findings or conclusions. Missing data on prepregnancy weight, weight gain, or birth weight reduced our power to detect differences in these variables when subjects were stratified by BMI and weight-gain category.
Our data did not allow us to determine which adolescents may have gained fat, lean mass, and height as a necessary part of their own maturation in addition to the gain supporting fetal growth. Also, we could not identify those adolescents that possibly retained excess maternal fat postpartum. Longitudinal research suggests that most female teens continue to grow in stature to age 18 y and beyond (32) and to gain weight as part of this growth until age 18 y (33), although there is evidence that black girls may mature earlier and gain less in height after age 14 y (34). In our cohort we noted no biologically or statistically significant differences in gestational weight gain among those with early (<11 y) or late (>12.9 y) onset of menses, which are potential markers of different rates of growth. It is noteworthy that teens who had given birth previous were significantly more likely to be overweight than were primiparas. Moreover, preliminary evidence from our research suggests that dietary quality among these teens is low (16) and, thus, poor eating habits may contribute to inappropriate weight gain. In conclusion, our data suggest no clear benefit from gestational weight gains at the upper limits of the IOM recommendations, particularly in teens who enter pregnancy overweight. The long-term costs to these mothers of additional weight gain may overshadow the modest improvements in birth weight. Other risk factors—such as poverty, racial discrimination, or dietary quality—may play a more important role than do higher levels of weight gain in optimizing birth weight in this population.
ACKNOWLEDGMENTS
JNN was responsible for the statistical analyses and manuscript preparation with the guidance of KOO and LEC. S-CC compiled the original data set, conducted extensive analyses of the data, and reviewed these analyses for consistency. JM and MSN provided prenatal care and nutrition counseling, respectively, to the subjects included in the data set and were responsible for much of the data entered into the medical charts. FRW reviewed all medical aspects of the analyses, including calculations of adjusted last menstrual period, adjustments of prepregnancy weights, and interpreting birth weight differences, and assisted with manuscript preparation and interpretation.
REFERENCES
1 From the Departments of Chemical Pathology (IK and ZARG) and Pediatrics (KJN and PM), University of Zimbabwe College of Health Sciences, Harare, Zimbabwe; the Department of Hematology, Parirenyatwa Group of Hospitals, Harare, Zimbabwe (BM); and the Center for Sickle Cell Disease, Howard University, Washington, DC (IK, ML, and VRG)
2 Supported by a grant from the Research Board of the University of Zimbabwe, by the International Federation of Clinical Chemistry and Laboratory Medicine–Roche National Award 2002 (to IK), by NIH research grant no. UH1 HL02679 funded by National Heart, Lung, and Blood Institute and the Office of Research on Minority Health, and by Howard University General Clinical Research Center grant no. MO1-RR10284. IK was a Fulbright Research Fellow at Howard University. 3 Reprints not available. Address correspondence to I Kasvosve, Department of Chemical Pathology, University of Zimbabwe College of Health Sciences, PO Box A178, Avondale, Harare, Zimbabwe. E-mail: ikasvosve{at}medsch.uz.ac.zw.
ABSTRACT
Background: Iron deficiency is common in African children, but genetic variations affecting susceptibility have not been identified. The Q248H mutation in ferroportin, a cellular iron exporter regulated by iron status and inflammation, may be associated with high iron stores in African adults.
Objective: The study examined the prevalence of iron deficiency in African children in an area where malaria transmission is low to absent and investigated whether ferroportin Q248H provides protection from iron deficiency.
Design: Complete blood counts, serum markers of iron status and inflammation, and ferroportin Q247H were measured in 208 preschool children in Harare, Zimbabwe. Iron deficiency was defined by serum ferritin and C-reactive protein (CRP) concentrations (definition 1) or by ferritin and the ratio of transferrin receptor to log10 ferritin (definition 2).
Results: Q248H was present in 40 children (38 heterozygotes, 2 homozygotes), elevated CRP was present in 26 (12.5%), and iron deficiency was present in 50 (24.0%) (definition 1) or 55 (26.4%) (definition 2). The interaction between ferroportin Q248H and CRP was significant for ferritin concentrations (P = 0.027) in a 2-factor analysis of variance model. With elevated CRP, the estimated geometric 
(SE range) ferritin concentration was 74 (52–106) µg/L for Q248H heterozygotes but 24 (20–30) µg/L for wild-type subjects (P = 0.016). With normal CRP, the ferritin concentration was 16 (14–
Conclusions: Any effect of Q248H in protecting against iron deficiency may be observable in children exposed to repeated inflammatory conditions. Further studies of iron status and ferroportin Q248H in African children are needed.
Key Words: Ferroportin mutation • African children • iron deficiency • inflammation
INTRODUCTION
Iron deficiency is a common nutritional disorder in children in developing countries (1). Several studies have suggested that iron deficiency in infancy may be associated with impaired cognitive function during the school years (2, 3). In sub-Saharan Africa, low dietary iron and chronic gastrointestinal blood loss due to hookworm infestation are major causes of iron deficiency (4). Hereditary hemorrhagic telangiectasia (Osler-Weber-Rendu disease) may, rarely, cause iron deficiency through gastrointestinal blood loss (5), but no cases have been reported in which genetic disorders caused childhood iron deficiency in the absence of gastrointestinal blood loss.
The SLC40A1 gene encodes a multiple transmembrane domain protein, ferroportin, which is responsible for the efflux of iron from mature enterocytes of the duodenum and from macrophages of the spleen and bone marrow to plasma (6–9). Macrophages are responsible for recycling iron that is recovered from the catabolism of aged erythrocytes (10). Cellular iron export by ferroportin is regulated by hepcidin, which in turn is regulated by iron stores and inflammation (11, 12).
The cDNA 744GT substitution in exon 6 of the ferroportin gene, which results in the replacement of glutamine with histidine at position 248 (Q248H), is a common polymorphism in Africans and African Americans that may be associated with a tendency in adults to iron loading (13–15). Cellular studies indicate that, unlike the ferroportin mutations that are associated with macrophage iron overload in white families—eg, A77D (16) and V162del (17)—the Q248H mutation has not been reported to decrease the macrophage membrane expression of ferroportin or to influence the cellular expression of transferrin receptors or the cellular ferritin content in the absence of inflammation (18).
In the current study, we examined the prevalence of iron deficiency in children in Harare, Zimbabwe, an area where malaria transmission is low to absent (19) and where there is a very low prevalence of hookworm infestation (PK Nathoo, personal communication, 2003). We also investigated whether ferroportin Q248H might provide protection from iron deficiency.
SUBJECTS AND METHODS
Study participants
Two hundred eight infants and preschool children attending well-child clinics in Harare, Zimbabwe, were studied. Body weight and height (length) of the children were measured, and the mother or guardian of each child was asked whether the child had a history of malaria. This was an exploratory study, and the sample size was determined by the limited funds available.
Written informed consent was obtained from the mother or guardian of each child. The institutional review boards of the Medical Research Council of Zimbabwe and of Howard University (Washington, DC) approved the protocol.
Study samples and laboratory measurements
Morning peripheral blood samples (5 mL) were collected into a 2-mL evacuated tube containing K3-EDTA and a 5-mL evacuated tube without the anticoagulant. Complete blood counts were performed by using an automated analyzer (Sysmex, Norderstedt, Germany). The analyzer was calibrated every morning with the use of standards provided by the manufacturer, and the performance of the machine was evaluated by using commercial blood as a quality control. Thin blood smears stained by using Giemsa solution were assessed microscopically for malaria. Serum ferritin, transferrin receptor, and C-reactive protein (CRP) concentrations were measured by using enzyme immunoassays (serum ferritin and transferrin receptors: Ramco Laboratories, Stafford, TX; CRP concentrations: ALPCO Diagnostics, Windham, NH).
Definitions of iron deficiency
Serum ferritin is a sensitive indirect measure of iron stores in healthy persons (20). In the presence of acute and chronic inflammation, the serum ferritin concentration increases independently of iron stores, which reduces its usefulness for a diagnosis of iron deficiency (21). The ratio of transferrin receptor to log10 ferritin was shown to identify iron deficiency in the setting of inflammation in a large series that documented iron status by using bone marrow examination (22). We employed 2 models to categorize iron status.
Definition 1
In definition 1, iron deficiency was defined as a serum ferritin concentration < 10 µg/L (23). If the serum ferritin concentration was 10 µg/L and the CRP concentration increased, the subject was considered to have indeterminate iron status. CRP concentrations > 8.2 mg/L were taken to be elevated, as specified by the manufacturer of the kit. The drawback of the use of definition 1 is that, if the child has both elevated CRP and a serum ferritin concentration 10 µg/L, a determination as to whether the child is iron deficient is not possible.
Definition 2
In definition 2, iron deficiency was defined as a serum ferritin concentration < 10 µg/L or, if the serum ferritin concentration was 10 µg/L, a transferrin receptor:log10 ferritin > 10.8. If the serum ferritin concentration was 10 µg/L and the transferrin receptor:log10 ferritin was 6.7–10.8, then the subject was considered to have indeterminate iron status. (In the current data set, among 46 subjects with serum ferritin concentration < 10 µg/L, the 2.5 percentile value for transferrin receptor:log10 ferritin was 6.7. Among 136 participants with serum ferritin concentrations 10 µg/L and without elevated CRP concentrations, the 97.5 percentile value for transferrin receptor:log10 ferritin was 10.8). An advantage of the use of definition 2 is that some children with both elevated CRP and serum ferritin concentrations 10 µg/L could be classified as to their iron-deficiency status.
Detection of the ferroportin 744GT mutation
DNA was isolated from leukocytes that were separated from whole blood by using lymphocyte separation medium (MediaTech, Sterling, VA). Exon 6 of ferroportin was amplified by using a set of primers encompassing portions of the introns that flank the exon (forward primer: 5'-CAT CGC CTG TGG CTT TAT TT-3'; reverse primer: 5'-GCT CAC ATC AAG GAA GAG GG-3'). After an initial denaturation at 94 °C for 3 min, a polymerase chain reaction was performed for 5 cycles of heating at 94 °C for 45 s, cooling at 56 °C for 45 s, and heating at 68 °C for 45 s, which were followed by 25 cycles of heating at 94 °C for 45 s, cooling at 52 °C for 45 s, and heating at 68 °C for 45 s and a final cycle of 15 min at 68 °C in a thermocycler (PTC-100; MJ Research Inc, Waltham, MA). The 392-base pair (bp) product was digested with PvuII enzyme (MBI Fermentas, Hanover, MD), and the resulting DNA fragments (252 and 140 bp) were fractionated on 3% agarose gel and detected with ethidium bromide.
Statistical analysis
Weight-for-height z scores were calculated by using EPI-INFO software (version 6.04; Centers for Disease Control and Prevention, Atlanta, GA). Statistical analysis was conducted with SYSTAT software (version 11; SYSTAT Software, Inc, Point Richmond, CA). Proportions were compared by using Fisher's exact test. Continous variables that followed a skewed distribution were analyzed with the Kruskal-Wallis test or log transformed for analysis of variance (ANOVA) and logistic regression models. The effect of the ferroportin Q248H polymorphism on serum ferritin concentrations was examined by using a 2-factor ANOVA model that included elevated or nonelevated C-reactive protein concentration as the second factor and that adjusted for age and weight-for-height z score. Because there was a significant C-reactive protein x ferroportin Q248H interaction for serum ferritin concentrations (P = 0.027), we also compared serum ferritin concentrations according to ferroportin Q248H status in separate ANOVAs for participants with and without elevated CRP. Each ANOVA model included age and weight-for-height z score as covariates. Logistic regression analysis was used to estimate the association between iron deficiency and Q248H heterozygosity after control for age and weight-for-height z score.
RESULTS
Two hundred eight apparently healthy children aged 3–60 mo who were visiting well-baby clinics in Harare, Zimbabwe, were studied. Eleven children (5.3%) had a history of malaria infection, but none had a positive malaria slide at the time of the study. The clinical and laboratory characteristics of the study population are shown in Table 1. The prevalence of obesity (weight-for-height z score > 2.0; 24) was 3% and that of underweight (weight-for-height z score <–2.0) was 7.2%. The ferroportin Q248H mutation was present in 40 children (38 heterozygotes and 2 homozygotes), absent in 157 individuals and we could not determine the presence of the mutation in 11 subjects due to the poor quality of the DNA recovered. The ferroportin Q248H allele frequency in the population studied was 0.107. Twenty-six (12.5%) children had elevated CRP concentrations. The median CRP concentration was 0.8 mg/L (interquartile range: 0.3, 2.8 mg/L) in Q248H heterozygotes and 0.6 mg/L (interquartile range: 0.1, 2.5 mg/L) in ferroportin wild-type children (P = 0.1).
View this table:
TABLE 1. Demographic and laboratory features of 208 children attending well-child clinics in Harare, Zimbabwe1
In a 2-factor ANOVA model, there was a significant ferroportin Q248H x CRP category interaction for serum ferritin concentrations (P = 0.027). After adjustment for age, weight-for-height z score, and that interaction, the estimated (SE range) serum ferritin concentration was 32 (27–39) µg/L in 38 Q248H heterozygotes and 21 (19–23) µg/L in 157 ferroportin wild-type children (P = 0.033) (Table 2
View this table:
TABLE 2. Serum ferritin concentration according to ferroportin Q248H1
On the basis of definition 1, 50 (24.0%) of the children had iron deficiency and 22 (10.5%) had indeterminate iron status. On the basis of definition 2, 55 (26.4%) of the children had iron deficiency and 21 (10.1%) had indeterminate iron status (Table 1). Under definition 1, 6 of 38 (15.8%) ferroportin heterozygotes had iron deficiency and 42 of 157 (26.8%) ferroportin wild-type subjects had iron deficiency (P = 0.259). Under definition 2, 6 of 38 (15.8%) ferroportin heterozygotes had iron deficiency and 47 of 157 (29.9%) ferroportin wild-type participants had iron deficiency (P = 0.204). In logistic regression modeling, younger age was significantly (P 0.0006) associated with iron deficiency under both definitions; a higher weight-for-height z score was significantly (P = 0.031) associated with iron deficiency only under definition 2 (Table 3). After adjustment for age and weight-for-height z score, the odds ratio for iron deficiency in ferroportin Q248H heterozygotes compared with that in ferroportin wild-type participants was 0.53 (95% CI: 0.18, 1.40; P = 0.222) according to definition 1 and 0.39 (95% CI: 0.14, 1.07; P = 0.068) according to definition 2 (Table 3).
View this table:
TABLE 3. Logistic regression models of the association between iron deficiency and Q248H heterozygosity adjusted for age and weight-for-height z score1
DISCUSSION
In the current study, we found iron deficiency in 25% of children aged < 5 y from an area of Zimbabwe where malaria is not transmitted and hookworm infection is not endemic, but, in an additional 10% of the children, iron status could not be described with confidence. The prevalence in the study population of subjects with a ferroportin Q248H mutation was 20%, which is similar to our earlier findings in African adults (13, 15). Overall, the statistically adjusted serum ferritin concentration was significantly higher in ferroportin Q248H heterozygotes than in wild-type subjects. This finding was fully accounted for by the markedly higher serum ferritin concentration in Q248H heterozygotes with elevated CRP than in wild-type subjects with elevated CRP. There was a trend to a lower prevalence of iron deficiency in Q248H heterozygotes, but the difference was not statistically significant. In keeping with studies in other countries (25–27), iron deficiency under both definitions was associated with younger age. Iron deficiency under definition 2 was associated with a higher weight-for-height z score (Table 3), which is consistent with the higher prevalence of iron deficiency among overweight or obese children and adolescents reported by others (25–27).
The ferroportin polymorphism by CRP concentration observed for the serum ferritin concentration is of interest in view of a recent report on the role of hepcidin in regulating ferroportin expression on the membrane of macrophages (12). Hepcidin, which is produced by hepatocytes stimulated by inflammatory cytokines (28), directly interacts with ferroportin on the membrane of macrophages, causing internalization of ferroportin and subsequent degradation of the protein by lysosomes. This process is associated with less export of iron from macrophages; because of this iron retention, ferritin synthesis increases (12). In the current study, the greater elevation in serum ferritin concentration with inflammation (as indicated by increased CRP concentrations) in Q248H heterozygotes than in wild-type subjects is consistent with the possibility that this mutation somehow enhances the interaction of hepcidin with macrophage membrane ferroportin, which results in greater retention of iron. However, the origin of serum ferritin in inflammatory states and the relation of serum ferritin to ferroportin function are not known.
In this study with a small sample size, no firm conclusion regarding a possible protective effect of the ferroportin Q248H variant against iron deficiency can be drawn. The observation of elevated serum ferritin concentrations only in Q248H heterozygotes with elevated CRP raises the possibility that any effect of Q248H in protecting from iron deficiency might be observable mostly in children exposed to repeated or prolonged inflammatory conditions. Given the importance of iron deficiency and the need to understand genetic factors in nutritional disorders, the relation between iron deficiency and ferroportin Q248H should be investigated in a study with a larger sample size and a power analysis that is based on the results of this report.
ACKNOWLEDGMENTS
We thank Lynne McNamara (Witswatersrand University, Johannesburg, South Africa) for her suggestions in developing primers and Anup Madan (Neurogenomics Research Laboratory, University of Iowa) for confirming the ferroportin gene sequence.
IK, ZARG, KJN, and VRG developed the study protocol and the experimental design and obtained funding. IK, KJN, and PM were responsible for subject recruitment and drawing of blood. Serum assays and ferroportin Q248H determinations were carried out by IK and ML, and BM performed complete blood counts. Statistical analysis of the data was performed by VRG and IK. All authors contributed to the preparation of the manuscript. None of the authors had a personal or financial conflict of interest.
REFERENCES
1 From the Laboratory of Membrane Biochemistry and Biophysics, Division of Intramural Clinical and Biological Research, National Institute on Alcohol Abuse and Alcoholism, National Institutes of Health, Bethesda, MD (KDS, RJP, SB, MM, and NS); the Food Composition Laboratory, Beltsville Human Nutrition Research Center, US Department of Agriculture, Beltsville, MD (RJP and VPF); the Department of Medicine, Brigham and Women's Hospital, Harvard Medical School, Boston, MA (HR); and the Departments of Obstetrics & Gynecology, School of Medicine (JJ, MB-A, JEW, RJS, and JHH) and Psychology (JHH), Wayne State University, Detroit, MI
2 Supported by the National Institute on Alcohol Abuse and Alcoholism research grant N01-AA83019 to JHH. 3 Address reprint requests to N Salem Jr, LMBB, NIAAA, 5625 Fishers Lane, Room 3N-07, Rockville, MD 20852. E-mail: nsalem{at}niaaa.nih.gov.
ABSTRACT
Background:African American women and socioeconomically challenged women are at risk of compromised folate status and, thus, of folate-related birth defects. Data are limited on circulating folate concentrations in pregnant African American women after folic acid fortification of the food supply was implemented.
Objective:The objective was to determine the influence of smoking and alcohol consumption on plasma 5-methyltetrahydrofolic acid (5-MTHFA) concentrations in pregnant African American women.
Design:Alcohol consumption, smoking exposure, and other characteristics of pregnant African American women reporting to an inner-city antenatal clinic were assessed. At 24 wk of gestation, blood samples and food-frequency intake data were collected. Plasma 5-MTHFA concentrations were determined by liquid chromatography–mass spectrometry for 116 subjects and examined in a correlational study design.
Results:Dietary folate and markers of alcohol consumption were positively associated, whereas exposure to smoke was negatively associated with plasma 5-MTHFA. More than one-half of the participants in this population failed to meet the recommended dietary allowance for dietary folate equivalents of 600 µg/d during pregnancy.
Conclusions:Most inner-city African American women are not meeting the recommended dietary allowance for dietary folate during pregnancy, and smoking may further compromise their folate status. Programs to reduce smoking and raise awareness about the importance of folate and multivitamin supplementation during pregnancy need to target this population.
Key Words: 5-Methyltetrahydrofolic acid • folate • folic acid • fortification • African American women • pregnancy • nutrition • smoking • alcohol • electrospray mass spectrometry • human plasma • polyunsaturated fatty acids
INTRODUCTION
In 1996 the federal government mandated that grain-based foods manufactured in the United States be fortified with folic acid (140 µg/100 g). This fortification was considered the most effective method of increasing folate intake in women of childbearing age (1). Fortification of grain products with folic acid increases folate concentrations in middle-aged and older populations (2, 3) and in women of childbearing age (4, 5). The implementation of folic acid fortification has lowered the incidence of neural tube defects (NTDs) in the United States by 20% on the basis of birth certificate data (6) and by >50% when the incidence of NTDs in stillbirths and terminated pregnancies were included in the analysis (7). Congenital anomalies of the central nervous system, including spina bifida and anencephaly, have not decreased in Detroit African Americans after folic acid fortification (3.0 per 1000 live births in 1995 compared with 3.1 per 1000 live births in 2000) (8).
In women of childbearing age, African American women have lower folate concentrations in serum and erythrocytes than do white, Hispanic (4, 5), and Asian (4) women in the United States. These lower values coincide with lower intakes of dietary folate equivalent (DFE), after energy adjustment, in pregnant African American women than in white women (9). Socioeconomically disadvantaged women also have lower folate status than advantaged women (4). Low socioeconomic status has been associated with a cluster of health-detrimental behaviors, including poor diet, smoking, and lack of physical exercise (10). Other methods of increasing folate status such as multivitamin supplementation by inner-city African American women are difficult and require substantial effort because awareness about folate and NTDs is low (11).
Active and passive exposures to tobacco smoke are associated with lower serum and erythrocyte folate concentrations (12), and smoking during pregnancy decreases maternal plasma folate concentrations (13). Exposure to smoke is associated with both lower intake of dietary folate and lower blood folate status. The association of smoke exposure to folate status persists after adjusting for differences in intake, suggesting nondietary as well as dietary influences (12). Smoking and alcohol consumption have been positively correlated in African American women (14). Halsted et al (15) reviewed the negative effects of chronic alcohol intake on folate homeostasis in alcoholics and in animal models, but the effects may be complicated by the folate content of some alcoholic beverages, specifically beer (16). The effects of alcohol on folate metabolism during pregnancy in humans remain unclear (17).
The primary purpose of this study was to examine the influence of diet, smoking, and alcohol consumption in pregnant African American women at risk of alcohol-related birth defects on plasma concentrations of 5-methyltetrahydrofolic acid (5-MTHFA), the predominant circulating folate metabolite. We expected these results to provide insight into the effectiveness of folic acid fortification on the intakes of DFE and the concentrations of plasma 5-MTHFA in a population known to be at risk of both folate deficiency and of giving birth to infants with complex disorders of the nervous system. In addition, this study provides highly accurate and specific determinations of plasma 5-MTHFA by a stable-isotope liquid chromatography–mass spectrometry method, the advantages of which have been discussed previously (18).
SUBJECTS AND METHODS
Subjects and study design
Pregnant African American women (n = 116) presenting between February 1999 and January 2001 at the antenatal clinic at Wayne State University in Detroit were recruited on the basis of reported alcohol intake (described in "Alcohol and smoking exposure"). Women with high-risk pregnancies were excluded from the study. All procedures and protocols received prior approval by the Wayne State University Human Investigations Committee, and informed consent was obtained during the initial clinical visit.
A structured interview at the first antenatal visit determined eligibility and assessed demographic characteristics, alcohol intake, and smoking exposure (19). Socioeconomic status was measured with the use of a modified Hollingshead index (20). At the regular obstetrical visit at 24 wk of gestation, a 15-mL fasting blood sample was collected by venipuncture. Specimens were collected into heparinized tubes, kept cold (4 °C) until centrifuged (2000 x g for 5 min at 4 °C) to separate plasma and erythrocytes, and frozen at –75 °C until analyzed. Nutritional status was assessed with the use of a validated food-frequency questionnaire (21) modified to quantify selected dietary fats. Quantification was based on the US Department of Agriculture National Nutrient Database for Standard Reference, release 14 (22). Intakes of individual nutrients were adjusted for total energy intake by the nutrient residual method (23) to reduce measurement error (24). The energy adjusted nutrient intakes were used for statistical analyses and are herein described as "adjusted" throughout. DFEs were calculated according to the method recommended by the Food and Nutrition Board (25). Briefly, for natural food sources of folate, 1 µg folate = 1 µg, DFE; for synthetic vitamin preparations, 1 µg folic acid = 2 µg DFEs; and, for mixed food and supplement products, µg DFE is calculated by food folate (in µg) x 1 plus folic acid (in µg) x 1.7. All participants were advised about nutrient supplementation during pregnancy and received a prescription for a prenatal vitamin.
Alcohol and smoking exposure
At-risk drinking was determined by several proven screening tests, including the Michigan Alcoholism Screening Test (MAST) and the Tolerance, Annoyed or Angry, Cut down or quit, Eye opener (T-ACE) questionnaire (26-28). Quantitative alcohol intakes were determined by 14-d recalls from the time of conception and at the first prenatal visit. Recall information generated estimates of alcohol intakes as grams of absolute alcohol per day around the time of conception (AAD0) and at the time of the first prenatal visit (AAD1), grams of absolute alcohol per drinking day (AADD0; AADD1), and proportion of drinking days over those 14 d (PROPDD0; PROPDD1). Cigarette smoking by both the mother and the father were determined by maternal recall of the number of cigarettes smoked per day around the time of conception and at the first prenatal visit. All women with AAD0 14.2 g (0.5 oz) were recruited into the study, plus a random 8% sample of the remaining patients were recruited. This selection strategy oversampled the high-risk drinking women (19, 29).
Sample analyses
Liquid chromatography–mass spectrometry was used to determine the plasma 5-MTHFA concentrations as described previously (18). Before extraction, 13C5-5-MTHFA (10 ng) was added to 0.5 mL plasma as an internal standard. The analyte was isolated with the use of solid-phase extraction (Strata phenyl column 100 mg/mL; Phenomenex, Torrance, CA), washed with 0.03 M K2HPO4, and eluted with 0.5 mL HPLC mobile phase (acetonitrile:methanol:water, 26:14:60). Extract (40 µL) was injected onto a C18 HPLC column (150 x 4.6 mm; Phenomenex) with the use of a binary pumped Agilent 1100 HPLC (Palo Alto, CA) interfaced to an ion trap mass spectrometer (Finnigan LCQ, San Jose, CA), and samples were analyzed by electrospray ionization in the positive ion mode.
Statistical analyses
Pearson's correlations (two-tailed) were used to determine the bivariate associations of plasma 5-MTHFA with energy-adjusted DFEs and selected maternal variables. As the accuracy of DFE estimates are questionable, bivariate correlations to adjusted DFEs were also determined according to rank by Spearman's correlation coefficients. Adjusted DFE, food folate, fortified folate, and plasma 5-MTHFA were grouped according to adjusted DFE quartiles for trend analyses and mean comparisons by one-way analysis of variance (ANOVA) with a priori comparisons between the lowest and highest quartiles. Associations with 5-MTHFA were also evaluated with the use of multiple linear regression analyses. A parsimonious model with all included independent variables having P values < 0.10 and a controlled model with potential confounders included as independent variables were generated. Variables were included on the basis of information from the literature, influence on the model R2 value, and degree of collinearity with other variables. Adjusted DFE was entered as both continuous data and after being assigned the appropriate rank. Associations with specific types of alcoholic beverages were also examined by linearregression. In addition, participants were grouped according to levels of periconceptional smoking (>0 or 0) and drinking (AAD0 > 0 or AAD0 = 0) to examine demographic characteristics. The subgroups were analyzed by a two-factor ANOVA with interaction for drinking and smoking with Holm's (30) post hoc comparisons of individual means. Data are presented as the mean ± SD with P value < 0.05 accepted as significant. Ad hoc bivariate correlations between maternal characteristics with father's smoking were also determined. The Mantel-Haenszel statistical procedure was used to compare percentages of drinking fathers. All statistical analyses were completed with SPSS for WINDOWS statistical software (release 11.5.1; SPSS Inc, Chicago).
RESULTS
Maternal characteristics
The analyses of plasma 5-MTHFA were completed for samples from 116 subjects with an age range of 16–38 y. Mean gestational age at the first prenatal visit was 16.4 ± 6.5 wk. The means ± SDs for selected demographic characteristics, alcohol intake variables, smoking exposure variables, and selected dietary intakes are shown in Table 1. Demographic characteristics were examined by two-factor ANOVA for categorized drinking and smoking status with interaction (data not shown). No significant effects of categorized drinking and smoking were observed on plasma 5-MTHFA. A significant interaction (P < 0.05) between smoking and drinking was detected for education with women who smoke and drink having the lowest acquired education (11.2 ± 1.3 grade acquired) and the nonsmoking drinkers having significantly higher acquired education (12.4 ± 2.0 grade acquired). Drinking women scored significantly higher (4.2 ± 1.0) than nondrinking women (3.6 ± 1.1) on the Hollingshead index, indicating lower socioeconomic status. The percentage of drinking fathers was higher for drinking women (77%) than for nondrinking women (27%), and the amount of father's smoking was almost 3 times greater for women who smoked (9.4 ± 10.4 cigarettes/d) than for nonsmoking women (3.2 ± 6.1 cigarettes/d). Smoking women also ate fewer carbohydrates (45.8 ± 8.0%) and more fat (38.9 ± 5.4% of energy) than nonsmoking women (carbohydrates: 48.9 ± 5.9% of energy; fat: 36.3 ± 4.4% of energy). Several significant associations of these various maternal characteristics and exposures with adjusted DFE intakes and plasma 5-MTHFA were determined by Pearson's correlations (along with Spearman's correlations for adjusted DFE intakes) and are presented in Table 1. Pre-pregnancy maternal smoking and maternal smoking during pregnancy were negatively associated with adjusted DFE as assessed by Spearman's correlations ( = –0.22, P = 0.017 for both). The association was slightly weaker when assessed by Pearson's correlations (r = –0.18, P = 0.049 for pre-pregnancy smoking; r = –0.18, P = 0.051 for smoking during pregnancy). Adjusted DFE was also negatively associated with dietary fat intake and positively associated with dietary vitamin B-12 and carbohydrate intake in both correlation analyses. Vitamin B-6 intake was also positively associated with adjusted DFE by the Spearman's correlational test only. Plasma 5-MTHFA was positively correlated to various screening test scores that indicate at-risk drinking and negatively correlated to the amount of father's smoking.
View this table:
TABLE 1. Characteristics and dietary intakes of pregnant African American women and correlations with adjusted dietary folate equivalents (DFE) and plasma 5-methyltetrahydrofolic acid (5-MTHFA)1
Associations with plasma 5-MTHFA
Plasma 5-MTHFA was positively associated with adjusted DFE (r = 0.29, R2 = 0.09, P = 0.001) by linear regression (Figure 1). This association was lower but remained significant when adjusted DFE was ranked and regressed against plasma 5-MTHFA (r = 0.25, R2 = 0.06, P = 0.007) or when adjusted DFE and plasma 5-MTHFA were examined by Spearman's correlation test ( = 0.19, P = 0.045). On the basis of the initial linear regression relation, the plasma 5-MTHFA concentration of 18.2 ng/mL (95% CI: 4.23, 32.28 ng/mL) would be predicted from a DFE intake of 600 µg/d, the current recommended dietary allowance (RDA) for DFE for pregnant women (25). The means, ranges, and CIs for adjusted DFE intakes, food folate, fortified folate, and plasma 5-MTHFA according to quartiles of adjusted DFE are shown in Table 2. About 57% of the participants reported adjusted DFE < 600 µg/d. The 57th percentile of plasma 5-MTHFA was 19.4 ng/mL. Mean differences between the lowest and highest quartile of adjusted DFE, food folate, and fortified folate were significant (P < 0.001) by a priori contrasts after significant one-way ANOVA result. Trend analyses confirmed that the adjusted folate intakes and plasma 5-MTHFA increased from the lowest to highest quartile (Table 2). Without fortification, almost the entire sample (98th percentile) would fail to consume the recommended 600 µg DFE/d, as indicated by adjusted food folate intake (Table 2). Fortification contributed an estimated 198 µg DFE/d as indicated by adjusted fortified folate intake. Contributions from daily multivitamin use are minimal because compliance by urban African American women educated about folic acid and NTDs has been previously reported to be only 9% (31).
View larger version (20K):
FIGURE 1.. Relation between adjusted dietary folate equivalent intakes and plasma 5-methyltetrahydrofolic acid in pregnant African American women. Dietary folate intakes were adjusted with the use of the nutrient residual model and were linearly correlated to plasma folate by linear regression: r = 0.29, R2 = 0.09, P = 0.001 (n = 116). RDA, recommended dietary allowance; EAR, estimated average requirement.
View this table:
TABLE 2. Various folate intake variables and plasma 5-methyltetrahydrofolic acid (5-MTHFA) concentrations of pregnant African American women by quartiles of adjusted dietary folate equivalents (DFE)1
Plasma 5-MTHFA was positively correlated with adjusted DFE, MAST score, and total energy intake and negatively correlated with father's smoking (Table 3). The multiple regression analyses were restricted to 77% of the sample (n = 89), largely because of missing data about father's smoking. These associations persisted after control for maternal education level, total pregnancies, and body mass index. Mother's age, socioeconomic status, maternal smoking, alcohol intakes, and other screening tests were also considered. The effect of alcohol consumption around the time of conception (AAD0) approached significance (standardized ß = 0.17, P = 0.08) when substituted for MAST score in the unadjusted model from Table 3 when a single outlier was excluded from the analysis (standardized ß = 0.11, P = 0.25 with outlier included). This substitution did not significantly change the associations of adjusted DFE, total energy, and father's smoking. No significant changes were observed in these results when the adjusted DFE intakes were ranked and substituted into either model.
View this table:
TABLE 3. Multiple linear regression results with plasma 5-methyltetrahydrofolic acid concentrations set as the dependent variable and dietary folate equivalents, smoking, drinking, and other maternal characteristics set as independent variables1
Maternal smoking was negatively, but not significantly, associated with changes in plasma 5-MTHFA (standardized ß = –0.12, P = 0.23) when entered in place of father's smoking. The stronger association between father's smoking and maternal plasma 5-MTHFA, compared with the association between mother's smoking and maternal plasma 5-MTHFA, was unexpected. Ad hoc Pearson's correlation coefficients between various maternal characteristics and father's smoking were determined. Father's smoking was positively correlated with mother's smoking (r = 0.27, P = 0.009) and frequency of maternal drinking (PROPDD0; r = 0.27, P = 0.009) and negatively correlated with maternal plasma 5-MTHFA (r = –0.21, P = 0.043).
Dietary sources of food folate and folic acid as determined by food-frequency questionnaire
The percentage contribution of individual foods to food folate intake, folic acid intake, and DFEs from food as determined by food-frequency questionnaires are presented in Table 4. Cold breakfast cereal contributed 26.9 ± 12.0% of the DFE intake from food, 45.7 ± 20.5% of the folic acid intake, and 4.2 ± 2.8% of the food folate intake. With the exception of orange juice, the foods providing the greatest contribution to folic acid intake were also the greatest contributors to the DFE intake. Orange juice provided the greatest percentage of food folate at 26.3 ± 18.0% that translated into 11.9 ± 9.6% of the DFE intake from food sources. After orange juice, the next 3 biggest contributors to food folate intake were milk, French fries, and breads, respectively.
View this table:
TABLE 4. Percentage contribution of individual food sources on the food-frequency questionnaire to food folate intake, folic acid intake, and dietary folate equivalents (DFE) from food1
Types of alcoholic beverages
After age, smoking, body mass index, total energy, and adjusted DFE were controlled for, alcohol consumption from alcoholic coolers at the first prenatal visit was negatively associated with plasma 5-MTHFA as a separate variable (ß = –1.90, P = 0.042) and with other alcohol beverages included as independent variables (ß = –1.86, P = 0.050). Beer was positively related to plasma 5-MTHFA in the various models but failed to reach statistical significance. Around the time of conception, the percentages of the total alcohol consumed for the beverage categories were 28% (beer), 13% (wine), 7% (coolers), and 51% (liquor). On average, women reported drinking 93% less alcohol at their first prenatal visit than they reported drinking around the time of conception. The percentages of distribution of each individual alcohol type consumed at the time of the first prenatal visit were 26% (beer), 37% (wine), 7% (coolers), and 30% (liquor).
DISCUSSION
In pregnant African American women at risk of alcohol-related birth defects, adjusted DFE intake was a positive and the most significant indicator of plasma folate status. The adjusted DFEs in this sample suggest that daily folate intake for at-risk pregnant women in this community is below the current RDA of 600 µg/d for more than 57% of the women and below the estimated average requirement (EAR) of 520 µg/d for 34% of the women. The estimated contribution of fortified folate in this population was 198 ± 98 µg/d (
± SD) that is similar to, but below, recent estimates of 215–240 µg/d of higher folate intake from fortified foods (32). Without folate fortification, the results suggest that 98% of the women would fail to reach the EAR for pregnant women, and 76% would fail to meet the EAR of 320 µg/d for women of childbearing age. With fortification, only 2% were below the EAR for women of childbearing age, and 3% had adjusted DFE intakes greater than the tolerable upper intake level of 1000 µg/d. Food fortified with folic acid contribute more than 50% of the total DFE intakes in this population (Table 4). However, these observations must be considered in light of the limitations in estimating dietary folate intakes (1,
The mean plasma concentration of 5-MTHFA in these subjects was 18.0 ± 7.1 ng/mL and is similar to serum folate concentrations in other socioeconomically disadvantaged women (4). The lowest plasma concentration of 5-MTHFA in this study was 6.6 ng/mL and is still higher than the mean serum folate concentrations (4.0 ng/mL) reported for non-Hispanic black women before fortification in the National Health and Nutrition Examination surveys (5). To our knowledge, this is the first study to assess the relations between dietary folate intake and blood folate concentrations in pregnant women and to estimate that a DFE intake of 600 µg/d results in plasma 5-MTHA concentrations of 18.2 ng/mL (95% CI: 4.23, 32.28 ng/mL). However, the applicability of this estimate is limited for several reasons. Because the present study determined plasma concentrations of 5-MTHFA by a highly accurate and specific liquid chromatography–mass spectrometry method, it is unlikely that plasma values are inaccurate. However, plasma folate concentrations decrease throughout pregnancy as a result of plasma volume expansion (34). Previous work by Ward et al (35) suggested that in middle-aged men a similar concentration of folate intake through the use of supplements rather than fortification results in a plasma concentration of = 17.4 ng/mL) (36). Also, it was shown that food-frequency questionnaires tend to overestimate folate intakes compared with folate-focused dietary recalls (37). Previous correlations between adjusted DFE and plasma folate reported in the literature of smaller studies in men with r = 0.26 (38) and nonpregnant women with r = 0.35 (
The effects of alcohol consumption and cigarette smoking on folate status are complex. It is difficult to isolate the effects of smoking and alcohol consumption because of their strong positive correlation. This strong association between smoking and drinking, and their apparently opposite effects on folate status, may be masking associations of maternal smoking and particular measures of alcohol consumption. The significant correlation between father's smoking and plasma 5-MTHFA is difficult to interpret. Father's smoking may be interpreted as an indicator of any of the following: passive smoke exposure, maternal smoking without the strong bias toward underreporting smoking during pregnancy, and combined tobacco smoke exposure. The latter interpretation may account for the stronger correlation between folate status and father's smoking than mother's smoking.
Lower blood folate concentrations in smokers is often attributed to lower dietary folate intakes (25), and in this study, too, a negative correlation was observed between pre-pregnancy maternal smoking and adjusted DFE. However, the negative association between father's smoking and maternal plasma 5-MTHFA persisted when DFE intake was included in the model, suggesting a nondietary mechanism, such as higher folate turnover, may be involved. A persistent negative association of smoke exposure after adjusting for dietary folate has been reported previously (12).
Alcohol consumption has been associated with lower blood folate status in chronic alcoholics (15), but it has been associated also with higher erythrocyte folate concentrations in drinking, pregnant women (16). In the present study, at-risk drinking, as indicated by screening tools, was associated with higher plasma 5-MTHFA both in bivariate models and in multivariable linear regression models with adjustments for potential confounds. Positive correlations between plasma folate concentrations and quantitative alcohol variables (ie, AAD0, AADD0, and PROPDD0) were not significant. The positive association between alcohol intake and blood folate concentrations may be a result of folate content in beer (16), although the consumption of spirits has also been positively associated with erythrocyte folate in women (39). Neither beer nor liquor consumption in the current study was significantly correlated with plasma concentrations of 5-MTHFA, although the coefficients were positive in all the regression models examined. The consumption of alcoholic coolers at the first prenatal visit was negatively correlated to plasma folate. The negative association of alcoholic coolers suggests a negative effect of ethanol on plasma folate status.
In African Americans, the prevalence of the C677T and A1298C mutation in methylene tetrahydrofolate reductase is low (4, 40). Therefore, folate status is largely determined by diet and environmental factors. In the current study, dietary folate, exposure to smoke, and indexes of at-risk drinking are significant predictors of plasma concentrations of 5-MTHFA in pregnant African American women. Dietary folate fortification contributed 200 µg DFE/d, but more than one-half of the pregnant women in this study failed to meet the RDA for folate of 600 µg DFE/d. Only 18% of inner-city African American women of childbearing age have heard of NTDs and only 9% know folate could prevent NTDs (11). For mothers in Michigan, folic acid awareness is lower among black women, women with unplanned pregnancies, and women with no high school education (41). Education programs can augment knowledge about NTDs and folate and can raise daily multivitamin use in pregnant African American women (31).
On the basis of the current results, fortification of food with folic acid alone may not be sufficient to provide the RDA for urban African American women during pregnancy; indeed, it does not appear that fortification has been capable of reducing the number of congenital central nervous system defects in African Americans living in the Detroit area (8). The results of this study suggest that targeted programs for education on folate use, as well as compliant, daily multivitamin supplementation during pregnancy, are required in this population and that the assessment of both supplementation compliance and concentrations of folate fortification of foods should continue (32). This study also highlights the need to control for lifestyle factors, including amount and type of alcohol consumption and active and passive tobacco exposure, when examining dietary folate and circulating folate concentrations in humans.
ACKNOWLEDGMENTS
We express our appreciation to Rudolf Moser of Merck Eprova AG for supplying a pure standard of 5-MTHFA. We are indebted to the Population and Provide Data Unit at the Vital Records and Health Data Development Section, Epidemiology Services Division, Bureau of Epidemiology, Michigan Department of Community Health for providing data on birth defects for African Americans in Detroit from 1993 to 2000. We thank the technical staff (E Russell, T Martin, L DiCerbo, et al) and the participants for their commitment to this study.
KDS was responsible for data management, interpretation, and statistical analysis and was the primary writer of the manuscript. RJP and VPF were responsible for the 5-MTHFA analysis. RJP made significant contributions to data interpretation and manuscript preparation. SB and MM contributed to data management and interpretation. MM was responsible for the study design. MB-A was responsible for study coordination and management and analyses of the data. JJ provided statistical advice and analyses. HR provided quantitative interpretation of the nutrition surveys. JEW made decisions for clinical exclusion and delivered the infants. RJS was the director of antenatal research and assisted in the study design, data interpretation, and manuscript preparation. JHH was a coprincipal investigator, secured partial funding, was responsible for the study design and management, and contributed to manuscript preparation. NS was a coprincipal investigator, provided funding, and contributed to the study design and management, data interpretation, and manuscript preparation. None of the authors had a conflict of interest.
REFERENCES
1 From the Department of Human Sciences, Loughborough University (NC, PLG, MMW, CB, NCD), Loughborough, United Kingdom, and the Mineral Metabolism Research Unit, University of the Witwatersrand (JMP, SAN), Johannesburg, South Africa.
2 The Birth to Twenty birth cohort study receives financial and logistic support from the Urbanisation and Health Programme of the Medical Research Council of South Africa; the Anglo-American Chairman's Fund; Child, Youth, and Family Development of the Human Sciences Research Council of South Africa; and the University of the Witwatersrand. The Bone Health study is financially supported by the Wellcome Trust (United Kingdom). 3 Reprints not available. Address correspondence to N Cameron, Department of Human Sciences, Loughborough University, Loughborough, LE11 3TU, United Kingdom. E-mail: n.cameron{at}lboro.ac.uk.
ABSTRACT
Background: The regression equations of Slaughter and Dezenberg, which are based on mixed ethnic samples, are currently recommended for predicting body fat from skinfold-thickness measures in prepubescent children of African ancestry. These equations contain methodologic problems that could make them inappropriate for African children.
Objective: The objective was to apply the Slaughter and Dezenberg equations to predict body fat in African prepubertal children and to compare the results with body fat measured by dual-energy X-ray absorptiometry (DXA). If significantly different outcomes were observed, then the objective was to develop new prediction equations and validate them on African children.
Design: The Slaughter and Dezenberg equations were applied to a cross-sectional sample of 214 prepubescent (Tanner stage 1) African children (118 boys). Body fat was determined by DXA, and subcutaneous fat at triceps, biceps, subscapular, suprailiac, thigh, and calf sites was measured with use of Holtain calipers. A randomly selected sample of 134 participants (78 boys) was used to generate new prediction equations that were validated on the remaining 80 participants (40 boys).
Results: The Slaughter and Dezenberg equations significantly underestimated (P < 0.001) body fat compared with DXA in both boys and girls. The best combination of skinfold thicknesses to predict body fat in African prepubertal boys, controlling for chronologic age, was triceps, biceps, subscapular, suprailiac, and thigh (SEE = 2.87), and for girls it was biceps, subscapular, suprailiac, thigh, and calf (SEE = 3.51).
Conclusion: The Slaughter and Dezenberg equations are unsuitable for predicting body fat in 9-y-old African prepubertal children. New equations that are based on skinfold-thickness combinations from African children provide more accurate estimates.
Key Words: Body fat • children • prepubertal children • African children • skinfold thickness • DXA • birth cohort
INTRODUCTION
The prediction of total body fat from subcutaneous fat assessed by anthropometry has become an accepted indirect method of determining body composition. Of the prediction equations that exist for children only those developed by Slaughter et al (1) and Dezenberg et al (2) were recommended for children of African (specifically African American) ancestry. Both sources used mixed Caucasoid and African American samples to derive prediction equations for prepubertal children. Although the former (1) developed equations specifically for black children, the latter (2) produced equations for ethnically mixed samples.
There are, however, several problems with the Slaughter et al (1) and, to a lesser extent the Dezenberg et al (2), methodologies. First, the Tanner staging technique for pubertal development was misapplied. Tanner (3) assigned a 5-point scale to pubertal development as seen in the development of breasts and pubic hair in girls and genitalia and pubic hair in boys. A further stage 6 could be applied to pubic hair when "... the pubic hair spreads further ... in the mid-twenties or later..." (3, p 33). However, Slaughter et al (4) and Dezenberg et al (2) assigned children with Tanner stage 2 into the "prepubescent" groups when, by definition, they were pubertal. Similarly, children in Tanner stage 4 were assigned by Slaughter et al (1) to a "postpubescent" group when puberty is not finished until Tanner stage 5. Thus, according to Slaughter and colleagues puberty only occurred in Tanner stage 3.
Second, it is not possible to have a single rating in the Tanner scale because the staging technique involves assessment of 2 aspects of pubertal development: pubic hair in both sexes, breast development in girls, and genitalia development in boys. No indication is provided by Slaughter et al (1, 4) or Dezenberg et al (2) of exactly how the Tanner scale was applied to result in a single rating figure, although a group described as "adult" by Slaughter et al (1), because they were in "Tanner stage 6," implies that pubic hair development was the criterion measure of puberty. In addition, Dezenberg et al (2) report that not all of their participants were actually assigned a Tanner stage and were simply assumed to be prepubertal. Although such assumptions could be valid for the younger end of the sample, they are certainly not valid for the older end, a fact acknowledged by Dezenberg et al (2) by describing this lack of Tanner staging as a limitation to their study.
Third, although Slaughter et al (1, 4) provided detailed descriptions of sample sizes, they did not provide the sample size of the black participants. Sample sizes of mixed ethnic groups were, in any case, small (50 prepubescent males and 16 females) and apparently not determined by power analysis. Dezenberg et al (2) also do not appear to use power analysis, although their African American samples are statistically more reasonable at 31 boys and 38 girls.
To test the applicability of the Slaughter and Dezenberg equations to African children, we applied them to prepubertal black participants, ie, children in Tanner stage 1 for breasts and genitalia and for pubic hair, from a birth cohort study set in South Africa. This paper reports on that application and the subsequent use of the anthropometric data to develop prediction equations for total body fat that are applicable to African children.
SUBJECTS AND METHODS
Subjects
The Birth to Twenty (BT20) study is a longitudinal study that tracked a cohort of 3274 children of all ethnic groups born between April and June 1990 living in Soweto and Johannesburg, South Africa, and was described in detail elsewhere (5, 6). A subsample of children from this study was used to undertake a separate bone health study from the age of 9 y (n = 369). A sample of prepubertal African children from the bone health study was used in the current analysis (n = 225). Nine children were excluded from the analysis because of missing anthropometric data, and a further 2 were excluded because sum of skinfold data for the triceps and subscapular regions was >35 mm, hence making the Slaughter equation used inappropriate for these individuals. Therefore, data from 214 African children (boys = 118; girls = 96) aged 9.51 ± 0.28 y were available for this analysis. All children were prepubescent, ie, in Tanner stage 1, for both breast and genitalia and for pubic hair.
Measurements
Anthropometric measurements were taken for all of the children with use of standard techniques (7). These measurements include height, weight, and skinfold-thickness measurements (with use of a Holtain Tanner/Whitehouse skinfold caliper; Holtain Ltd, Crymych, Wales) at the triceps, biceps, subscapular, suprailiac, midthigh, and medial calf sites. Each skinfold-thickness measurement was taken 3 times and a mean was calculated. A trained anthropometrist made all measurements, and the intraobserver variation for the skinfold measurements ranged between 2.0% and 2.7%. Pubertal development was assessed by trained same-sexed observers with use of the Tanner scaling technique on breasts and genitalia and pubic hair. A fan-beam densitometer (model QDR 4500A; Hologic Inc, Bedford, MA) was used to obtain dual-energy X-ray absorptiometry (DXA) readings of bone and body-composition components in the entire skeleton. Body composition [fat mass (kg), lean tissue mass (kg), and bone mineral content (g)] was assessed by using software version 8.21 (Hologic Inc) under standardized patient positioning and scan analysis. A spine phantom was scanned daily to determine the intrinsic CV of the machine. During the course of the study CVs for bone mineral content, bone area, and bone mineral density were 0.48%, 0.39%, and 0.35%, respectively. A trained technician for DXA performed all scans, and intraobserver variation for our study was found to be below 1% for all skeletal sites.
In recent publications [Tylavsky et al (8, 9)] the effectiveness of DXA as the criterion measure of body composition in adults was questioned. In a sample of overweight and obese adults Tylavsky et al (8) found small absolute differences in the loss of whole body lean soft-tissue mass and fat mass between the fan-beam and pencil-beam DXA system. Both units showed the same relation between changes in whole body lean soft-tissue mass and changes in total body water. In a sample of 58 adults aged 70–79 y Tylavsky et al (9) found systematically higher estimates of fat-free mass and lower estimates of fat mass from DXA as opposed to a 4-compartment model of body composition. However, those studies were on samples of adults who were either overweight or obese (8) or aged (9). In children and adolescents Gately et al (10) and Roemmich et al (11) showed that estimates of percentage body fat from DXA are significantly comparable to the 4-compartment model.
Ethics
All procedures were approved by the Committees for Research on Human Subjects of the Faculty of Health Sciences of the University of the Witwatersrand, South Africa, and Loughborough University, United Kingdom.
Analysis
Slaughter equations
An estimate of percentage body fat was calculated with the use of sex-specific Slaughter equations (1) for sum of skinfolds for triceps + subscapular, and triceps + calf, developed for prepubertal children with a triceps + subscapular skinfold <35 mm. The equations for boys are the following: percentage fat = 1.21(triceps + subscapular) – 0.008(triceps + subscapular)2 – 3.2 and percentage fat = 0.735(triceps + calf) + 1.0. The equations for girls are the following: percentage fat = 1.33(triceps + subscapular) – 0.013(triceps + subscapular)2 – 2.5 and percentage fat = 0.610(triceps + calf) +5.1.
The percentage fat values obtained from the Slaughter equations were regressed against percentage body fat from DXA in the South African sample with use of linear regression analysis with sex included as a covariate. The resulting SEE and R2 were then compared with those given by Slaughter (1) on the basis of the US sample.
Dezenberg equations
Fat mass was determined with use of the Dezenberg et al (2) equations for their first 4 stepwise variable combinations (weight, weight + triceps skinfold, weight + triceps + sex, weight + triceps + sex + ethnicity). Abdominal skinfolds (step 5) were not assessed in the BT20 study, and Dezenberg et al (2) stipulate that this skinfold is not essential. In addition to comparisons of descriptive statistics, the fat mass values obtained from the Dezenberg equations were regressed against percentage body fat found by DXA with use of linear regression analysis.
New prediction equations
The sum of skinfold combinations (from 2 to 6 skinfolds) from the BT20 data was also used in regression to determine whether they were more accurate than the Slaughter and Dezenberg equations in predicting body fat from DXA. Each regression model controlled for child's age and included a squared term for the sum of skinfolds to account for the nonlinear association between percentage body fat and the sum of skinfolds. In addition, an interaction term between age and the sum of skinfolds was also tested for significance.
New prediction equations were generated with use of subsamples of 78 boys and 56 girls who were randomly selected from the main sample. These prediction equations were subsequently validated with use of the remaining 40 boys and 40 girls in the BT20 sample. (A power analysis, with Z = 1.96 and Zß = 0.84, demonstrated that >31 subjects were needed to detect a difference of 5% body fat between the prediction equations and the DXA values in a validation sample.) The best predictors of percentage body fat on the basis of the highest R2 value and the lowest SEE for each combination of 2–6 skinfold measures were identified. All analyses were undertaken with use of SPSS statistical software, version 11 (SPSS Inc, Chicago).
Bland-Altman plots were used to determine bias and to investigate limits of agreement between the prediction equations and DXA in the validation groups for boys and girls. Bias was measured by considering predicted percentage body fat (from our own models or from Slaughter's models) minus percentage body fat from DXA on the y-axis and the mean of predicted body fat and DXA percentage body fat on the x-axis. Limits of agreement were assessed with use of 95% CIs defined by the mean differences ± 2 SDs. In addition, for the best combination for each number of skinfolds, DXA percentage body fat was regressed against predicted fat from the validation study, separately for boys and girls. Predicted values were deemed to be valid if Student t test revealed no significant deviation from the line of identity (intercept = 0, slope = 1).
RESULTS
Fat mass and percentage body fat determined by DXA were significantly greater (independent t test; P < 0.01, P < 0.001, respectively) in girls compared with boys, whereas SDs and interquartile ranges were similar (Table 1). Mean values for age, height, and weight were not different for boys and girls. Percentage body fat calculated by the 2 Slaughter equations (triceps + calf and triceps + subscapular) are shown in Table 2 and mean fat mass from the Dezenberg equations in Table 3.
View this table:
TABLE 1. Means and medians for age, height, weight, percentage body fat, and total body fat in boys and girls1
View this table:
TABLE 2. Predicted mean and median percentage body fat by using Slaughter's equation on the sample of African prepubertal children
View this table:
TABLE 3. Mean and SD of fat mass from dual-energy X-ray absorptiometry (DXA) and Dezenberg regression equations
Slaughter's triceps and calf skinfold combinations were significantly greater (paired t test; P < 0.001) than the values calculated from triceps and subscapular combinations (Table 2). The values for percentage body fat predicted from the Slaughter equations were significantly less (paired t test; P < 0.001) than those from DXA illustrated in Table 1 with differences between the medians varying between 6.1% and 14.1%.
Dezenberg predictions of fat mass were significantly different from DXA estimates (paired t test; P < 0.001) for regression steps 2 through 4 when the sexes were combined. When sexes were tested separately, predictions were significantly different for all steps in both sexes (paired t test; P < 0.001).
To compare how well Slaughter's 2-skinfold combinations predicted body fat in the US and African samples, the R2 and SEE values from predicting fat as measured by density, water, and bone and reported by Slaughter et al [(1) Table 3, p 715] were compared with the values that resulted from predicting percentage fat determined by DXA in our African sample. In short, we were determining whether the same equations when applied to the best estimate we have of fat in our sample were actually good predictor combinations (Table 4). The skinfold combinations produced higher R2 values and lower SEE values with the US sample than those produced from the BT20 sample. Similarly, Dezenberg's equation 4 produced an R2 = 59% and an SEE = 1.87 on the African sample in comparison to an R2 of 92% on the African American data. Thus, Slaughter's and Dezenberg's predictions apparently related better to body fat when applied to the African American sample than to the African sample.
View this table:
TABLE 4. Comparison of prediction quality of percentage body fat using Slaughter's body density, water, and bone mineral calculation on the American sample and dual-energy X-ray absorptiometry (DXA) values from the Birth to Twenty sample1
The new prediction equations and their associated R2 and SEE values separately for boys (n = 78) and girls (n = 56) are shown in Table 5 for each numerical combination of skinfolds (2, 3, ... 6) that produced the best prediction of body fat on the basis of the highest R2 and lowest SEE values. Note that in general the R2 values are not different or are slightly lower than either Slaughter's or Dezenberg's results, but our SEEs are lower, indicating a better estimation. The best-performing combinations of 2 skinfolds were triceps + suprailiac for boys and thigh + suprailiac for girls. These combinations produced lower SEE values (3.2 and 3.7) than those that used the Slaughter combination of triceps + calf (5.2) or triceps + subscapular (4.9) to predict DXA percentage body fat on these African children (Table 4). Indeed, they also produced similar or lower SEE values than the Slaughter combination did in predicting body fat in African American children (Table 4).
View this table:
TABLE 5. Prediction equations for percentage body fat from dual-energy X-ray absorptiometry (DXA) for various skinfold-thickness combinations by using the Birth to Twenty sample1
Bland-Altman plots were applied to assess the accuracy of the new prediction equations and the equations of Slaughter for the validation samples of boys and girls (plots not shown here). Plots that used our own equations were restricted to the best-performing equations (defined by the highest R2) for boys (5 skinfolds; triceps, biceps, subscapular, suprailiac, and thigh) and girls (5 skinfolds; biceps, subscapular, suprailiac, thigh, and calf). The limits of agreement (mean difference ± 2 SD) for girls were –7.21% to 8.35%, with a mean difference close to zero (0.57%). For boys the mean difference ± 2 SD was –8.17% to 10.10%, with a mean difference that was, again, close to zero (0.97%). The Slaughter equations for triceps and calf for both boys and girls show mean differences that indicate a bias toward underestimation for the Slaughter equations in comparison to DXA (–6.41% for girls and –5.70% for boys). The limits of the difference for the Slaughter triceps + calf equations ± 2 SD were –17.09% to 4.27% and –17.39% to 6.00% for girls and boys, respectively. The Slaughter equations for triceps and subscapular show mean differences that indicate an even stronger bias toward underestimation for both boys and girls in comparison to DXA than the triceps + calf equations (–11.4% for girls and –9.14% for boys). The limits of the difference for the Slaughter triceps and subscapular equations ± 2 SD were –22.27% to –0.56% and –18.71% to 0.42% for girls and boys, respectively. Pitman test for difference in variance also showed P < 0.01 for comparisons between DXA and the Slaughter equations, indicating significant differences in variance with average percentage body fat. In contrast P > 0.01 was observed from Pitman test for difference in variance for comparisons between our own equations and DXA, indicating no significant difference in variance with average percentage body fat. Regression demonstrated that, for the best combination of each number of skinfolds, the intercept and the slope of the validation sample were not significantly different from the line of identity.
The average percentage of body fat was slightly overpredicted in both boys and girls by <1.0%. There were no significant differences between the means of predicted body fat for any of the combinations of skinfolds and the mean DXA percentage body fat for both sexes. Clearly, in any prediction situation there are going to be individuals who do not predict well. In a fat prediction situation these individuals will typically be the children who are at the extremes of body fat values or who could have subcutaneous fat distributions and patterning that differ from those of the average child. Statistically, they are identified as those children falling outside the ±1.96 SEE values of the prediction equations. In this study 8 boys and 5 girls fell into this category. Six of the 8 boys had DXA fat values greater than (n = 3) or less than (n = 3) the mean DXA values ±2 SD values for boys. The other 2 boys predicted well for combinations of skinfolds <5 and were only identified at the 5 and 6 skinfold combination levels. All 5 girls were greater (n = 2) or less (n = 3) than the DXA mean ± 2 SD values for girls. Of the 13 poor predictors, 8 predicted poorly for all combinations of skinfolds, and all were outside the sex-specific DXA ± 2 SD range. Hence, they were either very fat or thin in comparison to their peers.
DISCUSSION
The prediction of body fat in prepubertal children of black African ancestry (ie, either African Americans or actual African children) has previously only been possible through the use of equations of Slaughter et al (1) to predict percentage fat on the basis of combinations of triceps, calf, and subscapular skinfolds or the equations of Dezenberg et al (2) to predict fat mass. For the first time we were able to test the accuracy of these predictions against DXA-derived values for percentage body fat in African prepubertal children. There is conflicting evidence that DXA can both overestimate and underestimate body fat, depending on the age and sex of the children being assessed. For example, Goran et al (12) found underestimation by DXA in relation to skinfolds in white children aged 5–8 y, whereas Gutin et al (13) recorded a systematic tendency for DXA values to be higher in relation to skinfolds in 9–11-y-old children. In our study it is clear that the Slaughter equations underestimate percentage body fat by 30–50% of the DXA estimate in such children, and Dezenberg equations significantly underestimated fat mass by 15% of the mean values. It is important to try to understand why this is the case.
The current analysis is based on a sample of black African children, whereas Slaughter et al (1, 4) and Dezenberg et al (2) used mixed samples of children of European and African American ancestry and controlled for "race" or ethnicity in their analyses. In addition, the age range of the current sample is narrow (9.1–9.9 y) compared with Slaughter et al's (1, 4) sample (7–11 y) and Dezenberg et al's (2) sample (4–10 y) and strictly prepubertal according to Tanner scaling. Both Slaughter et al (1) and Dezenberg et al (2) chose to use a more liberal interpretation of prepubertal by including children who were already displaying pubertal changes and were rated 2 on the Tanner scales. Finally, the techniques to determine body fat were different between this study and Slaughter's study. DXA was used in Dezenberg's study (2), and the current analysis compared with combinations of body density (hydrodensitometry), body water (deuterium oxide dilution), and bone mineral (photon absorptiometry) in the Slaughter et al (1) study.
The narrow age range and strictly prepubertal nature of the current sample provides a more homogeneous sample from which to investigate body composition. The mixed-pubertal nature of Slaughter's sample is likely to include a large proportion (close to 50%) of children in early pubertal development with consequently greater body fat than strictly prepubertal children. This effect could be balanced by the higher proportion of younger children, but there is no way of knowing how effective this balance was. The use of a multicompartmental model by Slaughter goes some way toward reducing the effect of differing pubertal status by accounting for the chemical changes in the fat-free body known to occur during puberty. In Dezenberg's study it is not possible to determine how many children were actually pubertal, because pubertal assessments were not made in all cases. Clearly, however, Slaughter's and Dezenberg's equations are aimed at samples with a greater range of age and prepubertal development and, in the case of the latter equations, a greater range of ethnicity. However, the African children within this study, although being within a narrow age range and strictly prepubertal, were all within the chronologic and morphologic range of the source samples and, thus, ought not to have produced significantly different results.
There are some interesting implications of the Slaughter and Dezenberg equations that underestimate body fat in African children and in the different combinations of skinfolds that were the best predictor variables. First, it could be that more fat is situated subcutaneously in African American children than in African children. In other words, the same skinfold thickness or combination of predictor variables in African American and African children will reflect less total body fat in African Americans. Second, it would appear that the patterning of fat in African children could be different to that of the African American sample in that particular combinations of skinfolds did not present as best predictor combinations in both samples. Total body fat is apparently better represented by a combination of suprailiac + triceps skinfolds in African boys and subscapular + calf in African girls, compared with triceps + calf or triceps + subscapular in the African American sample.
The current African sample is shorter and lighter than the American sample but appeared to have similar or greater body fat, depending on the method of determination in the American samples. The American boys in Slaughter's study have means of percentage body fat ranging from 19.0% to 23.5% and girls from 23.2% to 27.8% compared with the median African values of 21.8% and 25.8% for boys and girls, respectively. However, the African children tend to deposit less fat in subcutaneous stores, thus justifying the use of ethnic-specific regression equations when calculating body fat. Clearly, there is a need to widen the age range of this prepubertal sample to provide greater applicability, but the current analysis provides a useful check point on body fat at the beginning of adolescence that makes its dissemination desirable at this time.
ACKNOWLEDGMENTS
NC, JMP, and SAN were involved in the initiation of the study, financial support, the logistical organization, and collection of data. NC, PLG, and MMW were responsible for strategies of data analysis. MMW, CB, and NCB were responsible for data analysis and initial drafts of the paper. NC was responsible for the final manuscript and preparation for publication. All authors reviewed and commented on drafts of the paper. None of the authors has any financial or personal interest in any company or organization sponsoring the research, including advisory board affiliations.
REFERENCES
1 From the Department of Medicine, Obesity Research Center, St Luke’s—Roosevelt Hospital, Columbia University, College of Physicians and Surgeons, New York
2 Supported by National Institutes of Health Grant RO1-NIDDK 42618 and by a National Institutes of Health Minority Fellowship Award (to AJ). 3 Reprints not available. Address correspondence to SB Heymsfield, St Luke’s—Roosevelt Hospital, Weight Control Unit, 1090 Amsterdam Avenue, 14th Floor, New York, NY 10025. E-mail: sbh2{at}columbia.edu.
ABSTRACT
Background: Body composition differs between African American (AA) and white women, and the resting metabolic rate (RMR) is likely to be lower in AA women than in white women.
Objective: We tested 2 hypotheses: that AA women have a greater proportion of low-metabolic-rate skeletal muscle (SM) and bone than do white women and that between-race musculoskeletal differences are a function of body weight.
Design: Hypothesis 1 was tested by comparing SM, bone, adipose tissue, and high-metabolic-rate residual mass across 22 pairs of matched AA and white women. Magnetic resonance imaging and dual-energy X-ray absorptiometry were used to partition weight into 4 components, and RMR was both calculated from tissue-organ mass and measured. Hypothesis 2 was evaluated by measuring SM, bone, fat, and residual mass in 521 AA and white women with the use of dual-energy X-ray absorptiometry alone.
Results: Hypothesis 1: AA women had greater SM (
± SD group difference: 1.52 ± 2.48 kg; P < 0.01) and musculoskeletal mass (1.72 ± 2.66 kg; P < 0.01) than did white women. RMR calculated from body composition and measured RMR did not differ; RMR estimated by both approaches tended to be lower (160 kJ/d) in AA women than in white women. Hypothesis 2: SM was significantly correlated with weight, height, age, and race
Conclusions: Lower RMRs in AA women than in white women are related to corresponding differences in the proportions of heat-producing tissues and organs, and these race-related body-composition differences increase as a function of body weight.
Key Words: Energy requirements • resting metabolic rate • obesity • nutritional assessment
INTRODUCTION
Of the energy that humans and other mammals expend over time, the largest fraction is resting metabolic rate (RMR; 1), which is reflective of the collective ongoing biological processes involved in cellular and tissue maintenance and repair (2). Interindividual variation in RMR, after adjustment for body size, is of intense research interest with respect to human energy requirements (2). A consistent observation has been that RMR is lower in African American (AA) women than in white women of comparable weight, height, age, and fat-free mass (FFM; 3-7). Some investigators suggest that a relatively low RMR in AA females may be a predisposing risk factor for long-term weight gain and obesity (7).
A new approach to the exploration of between-group RMR differences is the modeling of energy exchange in the context of tissue-organ body composition (1, 8). The RMR of each tissue and organ is derived as the product of organ mass and tissue-specific metabolic rate. Tissue and organ mass content are derived by whole-body magnetic resonance imaging (MRI; 8, 9). The specific metabolic rates are known and validated for adults aged <50 y (8). Earlier reports support the validity of this RMR estimation approach for MRI models ranging from 4 to 8 tissue-organ components (8-10).
The 4-compartment MRI model partitions body mass into adipose tissue, skeletal muscle (SM), bone, and residual mass (9-11). The residual component includes brain, liver, kidneys, heart, gastrointestinal tract, and other organs and tissues. Brain and visceral organs are high-metabolic-rate compartments that account for a disproportionately large fraction of RMR relative to their mass (12). Earlier reports suggest that AA women have a greater SM and bone mass than do white women of similar weight, height, and age (13, 14). SM and bone are low-metabolic-rate tissues (8, 12, 15, 16), and the extent to which the greater musculoskeletal mass in AA women is offset by a lower residual mass and adipose tissue mass than is seen in their equivalent-weight white counterparts remains unknown. Similarly, a lower residual mass in AA women correspondingly leads to a lower RMR than is seen in white women. Alternatively, a greater musculoskeletal mass accompanied by a lower adipose tissue mass would produce small effects on RMR.
The present study consisted of 2 linked experiments. The first experiment, based on earlier observations, was formulated on the basis of the hypothesis that adult AA women have a greater amount of SM (13, 14) and bone (13, 17) than do matched white women. In this framework, we examined corresponding differences in other body components and RMR with the ultimate aim of establishing whether and to what extent body-composition effects might account for observed differences in RMR between AA and white women. The initial results led us to propose a second hypothesis—that body-composition differences between the races are a function of body mass—and to conduct a second experiment to test that hypothesis.
SUBJECTS AND METHODS
Experimental design and protocol
Experiment 1
Healthy adult AA and white women were included in the first phase of the study. Race was determined by each subject’s self-report that each parent and all 4 grandparents were of the same race. Age was limited to <50 y for RMR modeling purposes because the applied formulas are accurate in younger subjects but less reliable in older subjects (8, 10). The source database consisted of 220 women aged >18 y who were evaluated as part of a long-term body-composition study (18). Each AA woman was matched to a white woman by age (±10 y), weight (±4 kg), and height (±5 cm). Twenty-two matches were completed, for a total of 44 subjects. Women were all premenopausal and were studied independent of menstrual cycle activity.
Four tissue-organ compartments were evaluated: adipose tissue, SM, bone, and residual mass. Adipose tissue and SM mass were estimated by whole-body MRI scanning as previously reported (8-10). Bone mineral mass was measured by dual-energy X-ray absorptiometry (DXA), and bone mass was calculated as 1.8 x bone mineral mass (9, 11). Residual mass was then calculated as the difference between body mass and the sum of the 3 measured compartments.
RMR was measured after an overnight fast using a ventilated hood indirect calorimetry system (Delta-Trac II metabolic monitor; SensorMedics, Yorba Linda, CA). RMR was also calculated from body composition as the summed products of compartment mass and known specific metabolic rate. The specific metabolic rates, as previously estimated for adults aged <50 y, are for adipose tissue, SM, bone, and residual mass 18.8, 54.3, 9.6, and 225.7 kJ · kg–1 · d–1, respectively (9). We assume in this study that there are no race differences in these specific metabolic rates.
We also examined the relations between RMR estimates and FFM in the matched women and, for consistency, we calculated FFM from MRI estimates, rather than DXA, as the sum of SM, bone, residual, and fat-free adipose tissue mass. We assumed that fat-free adipose tissue mass is 15% of adipose tissue mass (9, 11). In an earlier study we found FFM, as analyzed by MRI, to be almost identical to FFM measured by DXA (9, 11).
All measurements were made within one day of each other. The subject’s body weight was measured to the nearest 0.01 kg using a digital scale (Weight-Tronix; Scale Electronics Development, New York). Standing barefoot height was measured to the nearest 0.1 cm with a wall-mounted Holtain stadiometer (Holtain Limited, Crosswell, United Kingdom).
Experiment 2
Once analyzed, the initial database confirmed a significant but small difference in SM mass between AA and white women. On the basis of a review of composite earlier studies (3-7) and our own new data, we advanced the second hypothesis: that the magnitude of race differences in SM mass is a function of body mass. Accordingly, we assembled a second data set of 521 healthy adult AA and white women from the center’s archives (14, 18). Each subject had completed a DXA scan, and we then evaluated the measured appendicular lean soft tissue mass by using Kim’s equation (19) to estimate total-body SM mass:
RESULTS
Experiment 1
Subjects
The characteristics of the subjects in the first experiment are presented in Table 1. There were 22 subject pairs with no significant between-group differences in age, height, or body mass index.
View this table:
TABLE 1. Baseline characteristics and body composition of subjects in experiment 11
Body composition
The results of experiment 1 body-composition studies are summarized in Table 1. Body weight was not significantly different in the matched groups: AA women weighed 67.0 ± 13.5 kg, and white women weighed 66.9 ± 12.5 kg. Adipose tissue, residual mass, and bone mass did not differ significantly between AA and white women. AA women, however, had greater SM mass than did white women (between-group difference: 1.52 ± 2.48 kg; P < 0.01).
The calculated fat-free component of adipose tissue comprised the smallest fraction of FFM in both groups, and the fractions were increasingly larger in bone, residual mass, and SM (observations not shown). The fraction of FFM as SM was larger (P < 0.05) in AA women than in white women, but the fat-free component of bone, residual mass, and adipose tissue as a percentage of FFM did not differ significantly between AA women and white women.
Musculoskeletal mass was 1.72 ± 2.66 kg larger (P < 0.01) in the AA women (25.9 ± 2.8 kg) than in the white women (24.2 ± 2.9 kg; Figure 1). Expressed as a fraction of FFM, musculoskeletal mass was significantly greater (P = 0.01) in the AA women (0.56 ± 0.05) than in the white women (0.53 ± 0.04), for a between-race fractional of 0.032 ± 0.060.
View larger version (36K):
FIGURE 1.. Mean (±SEM) musculoskeletal mass in African American and white women (n = 44) expressed as an absolute mass (upper panel) and as a fraction of the fat-free mass (FFM; lower panel). *Significantly different from the white women, P < 0.01 (paired t test).
Resting metabolic rate
The calculated and measured RMR results are summarized in Table 2. Measured and calculated RMR did not differ significantly between the AA women and the white women. The RMR was 160 kJ/d lower in AA women by both estimation methods, but this difference was not statistically significant. Analysis of covariance also showed no significant effect (P = 0.29) of race on RMR values after control for age, weight, height, and body composition. As expected, age, body weight, and height were significant predictors of RMR (P < 0.01; data not shown).
View this table:
TABLE 2. Resting metabolic rate (RMR) results1
Experiment 2
The second evaluated cohort consisted of 521 women (171 AA and 350 white), and the group’s demographic and body-composition characteristics are summarized in Table 3. The regression model results are presented in Table 4.
View this table:
TABLE 3. Baseline characteristics and body composition of subjects in experiment 21
View this table:
TABLE 4. Multiple regression models for body composition1
SM mass was predicted by weight, height, age, and race as separate independent variables, each of which was statistically significant (SM model 1 in Table 4: P < 0.05). SM mass was also predicted by weight, height, age, and the race x weight interaction term (model 2), although the model correlation coefficient and SE of estimate were similar to those in model 1 with race alone as a predictor. Race was not a significant predictor variable for SM mass when added separately to the interaction model including the race x weight term (model 3).
Similarly, the race x weight interaction term was a significant predictor of bone mineral mass (models 2 and 3) and residual mass (RM models 2 and 3). In contrast, race alone and race x weight failed to be a significant predictor of fat mass (model 1).
The magnitude of the race x weight term in body-composition prediction can be shown by using an example of 2 pairs of female subjects, in which one pair is AA and the other pair is white, both subjects in each pair are 165 cm in height and 25 y of age, but one subject has a body weight of 50 kg and the other has a body weight of 100 kg. The models in Table 4 can be used to calculate the mass of each component for the 4 women. In the pair of 50-kg women, SM and bone mineral mass would be larger by 1.0 and 0.3 kg, respectively, and residual mass smaller by 0.9 kg in the AA woman than in the white woman, and there would be no predicted differences in fat mass. We assume the small residual mass difference of 0.4 kg represents either model prediction error or a nonsignificant race difference in fat mass. In the 100-kg pair, the AA woman would have 2.0 kg and 0.5 kg more SM and bone mineral mass, respectively, and 1.8 kg less residual mass than would the white woman, and there would be no predicted differences in fat mass.
DISCUSSION
The present study was prompted by the finding in earlier studies of a lower RMR in AA women than in white women after adjustment for conventional measures, such as weight, height, and age; body composition as fat; and FFM (3-7, 21-29). In this study we explored additional body-composition and related energy expenditure measures under the general working theory that previously observed RMR differences may be accounted for by race variation in the proportions of tissue-organ components.
In the first experiment we used whole-body MRI and DXA to partition body mass into 4 compartments differing in metabolic activity, ie, adipose tissue, SM, bone, and residual mass. Our observations support the view that AA women have a greater musculoskeletal mass than do white women who are similar in age, weight, and height (13, 14, 17), and this difference persists even when musculoskeletal mass is expressed as a fraction of FFM. Similarly, AA girls matched to white girls for age, Tanner stage, and body mass index had greater limb lean body mass than did the white girls, as assessed by DXA (30). However, the absolute body-composition differences by race that were observed in the present study were not large—only 1.7 kg for the sum of SM and bone mass.
Because the between-group subject weights in the first experiment were matched, by necessity the AA women had 1.7 kg less of other components—0.6 kg of adipose tissue and 1.1 kg of residual mass—than did their white counterparts. The net result is that in AA women there was a slight shift in the proportion of heat-producing tissues favoring lower RMR SM and bone over higher RMR organ-containing residual mass; estimated heat production differences secondary to adipose tissue were negligible. Because we assumed that no race difference exists in tissue-organ–specific metabolic rates, this shift in tissue-organ distribution was reflected in a small, nonsignificantly lower predicted RMR of 167.2 kJ/d in the AA women. A small difference in the same direction and magnitude was observed in measured RMR. Thus, largely the combined effects of a greater musculoskeletal mass and lower residual mass could account for the difference between RMR in AA women and that in their white counterparts. This small metabolic rate effect combined with a relatively small sample size might be one reason for the nonsignificant between-group RMR difference.
The completion of the first experiment validated our body-composition hypothesis, but the magnitude of calculated and measured RMR differences was relatively small and statistically nonsignificant. Accordingly, we revisited earlier body-composition and RMR studies exploring these issues and noted 2 findings: either AA women were heavier than were their white counterparts, or both AA and white women were heavier than were the women evaluated in the current study. Most of the earlier studies reported absolute body-composition and RMR differences between AA and white subjects rather than differences expressed as a function of body mass. We therefore postulated a second hypothesis linking differences in body composition and, by inference, in RMR to body mass. Our observations in the second experiment support this hypothesis and show, in a large sample of women, increasing SM, bone, and residual mass differences between AA and white women as a function of body mass. According to the developed prediction models, RMR would differ by 167.2 kJ/d between AA women and white women weighing 50 kg, but this difference would be doubled between women weighing 100 kg. These differences would reflect a 3–5% lower RMR in AA women than in white women.
Previous studies in both children and adults reported either no race-related RMR differences or statistically significant RMR differences of 630–840 kJ/d (3-7, 21-29). The extent to which measurement error, small sample sizes, and subject characteristics contribute to these observed differences is unknown. The results of our study suggest that RMR differences between AA women and white women may be a function of body mass, and this phenomenon may account for some of the variation observed across studies. That is, there are between-race weight differences both within and between earlier studies. Our findings predict larger AA-white RMR differences in subject groups with greater body mass index.
In a study relevant to the present investigation, Hunter et al (4) examined RMR in AA and white women. Fat-free mass, fat mass, and regional lean tissue were also determined by DXA. The AA women had a lower RMR (506 kJ/d) than did the white women, and this significant difference persisted after adjustment for FFM and limb lean tissue mass, which is mainly SM, but it disappeared after adjustment for trunk lean tissue. The investigators suggest that relatively low volumes of metabolically active organs mediate the low RMR of AA women; this hypothesis is consistent with the findings of the present study. Similar results were observed in an earlier study of prepubertal girls (31). In that study, black girls had an RMR of 385 kJ/d, which was significantly lower than the RMR of white girls matched for age, Tanner stage, and body mass index, after adjusting for FFM. Specifically, after matching or control for body weight, our own findings show greater musculoskeletal muscle mass and correspondingly less higher-metabolic-rate residual mass in the AA women than in the white women.
Rather than providing definitive results, our findings suggest the need for future studies in a larger sample with well-defined characteristics along with direct MRI measurements of the high-metabolic-rate organs and tissues rather than of the less specific residual mass component. Other critical limitations of our study are that we assumed that no race differences exist in tissue- and organ-specific metabolic rates, and that we limited our study to subjects aged <50 y. Whereas the similar calculated and measured RMRs in both groups of women support our assumed specific metabolic rates, a need exists to move forward to direct measurements of organ metabolic rates in vivo. In vivo measurement of organ consumption of O2 is now possible with positron emission tomography by using 15O inhalation (32), and future studies should examine whether differences in tissue- and organ-specific metabolic rates exist between race groups.
Conclusion
The present study results suggest that AA and white women differ in relative body composition as a function of body mass and that there are larger race differences at greater weights. These differences appear to be small overall, but quantifiable with appropriate methods and adequate sample size. Our analysis also supports the view that these differences in body composition translate to small differences in RMR and, potentially, in energy requirements. These observations help to reconcile the variable findings in earlier studies by showing the small magnitude of body-composition and, potentially, RMR effects that vary as a function of body mass. Finally, on the basis of the observed lower residual mass, our study also suggests a lower mass of higher-metabolic-rate visceral organs in AA women than in white women.
ACKNOWLEDGMENTS
AJ, M-PS-O, and SBH were responsible for the study design; AJ, WS, DG, and ZW were responsible for data collection; AJ, WS, SH, ZW, and SBH were responsible for data analysis; and AJ, WS, DG, SH, M-PS-O, ZW, and SBH were responsible for writing the manuscript. None of the authors had any financial or personal conflict of interest.
REFERENCES
1 From the Georgia Prevention Institute, Department of Pediatrics (JCD, FAT, PB, BG, and HS) and the Office of Biostatistics and Bioinformatics (RHP and PB), Medical College of Georgia, Augusta; the EMGO Institute, Department of Social Medicine and Research Centre Body@Work TNO VU, Amsterdam (JCD); and the Twin Research & Genetic Epidemiology Unit, St Thomas’ Hospital, London (HS).
2 Supported in part by grants HL 69999, HL 35073, and HL 41781 from the National Heart, Lung, and Blood Institute and by a State of Georgia Biomedical Initiative grant to the Georgia Center for the Prevention of Obesity and Related Disorders. 3 Reprints not available. Address correspondence to H Snieder, Georgia Prevention Institute, Medical College of Georgia, Building HS 1640, Augusta, GA 30912-3710. E-mail: hsnieder{at}mcg.edu.
ABSTRACT
Background: Obesity is associated with multiple health problems, often originating in childhood.
Objective: The objective was to investigate differences in the development of adiposity from childhood to adulthood as related to race, sex, and socioeconomic status (SES).
Design: Individual growth curve modeling for waist circumference, body mass index, and sum of skinfold thicknesses (triceps, subscapular, and suprailiac) was performed in an 11-y cohort study of 622 African Americans and European Americans aged 4.2–27.5 y. We examined the development of adiposity in 2 ways: 1) differences related to race, sex, and parents’ education (SES), and 2) differences between obese, overweight, and normal-weight persons at the end of their childhood (> 17 y of age).
Results: The sum of skinfold thicknesses was greater in females than in males, with a larger increase with age. Race, sex, and SES showed a complex relation with body mass index and the sum of skinfold thicknesses. The low-SES group showed the fastest increase in waist circumference with age. The obese group showed the most rapid increase in the 3 measures of adiposity. Growth curves for the obese group were distinguishable from those for the normal-weight persons at an earlier age for African Americans than for European Americans.
Conclusions: The development rate of adiposity from childhood into early adulthood is influenced by sex and SES but not by race. However, race, sex, and SES had joint effects on adiposity levels. The development of obesity can begin to be distinguished in midchildhood, but the age at which this distinction becomes apparent depends on race.
Key Words: Body mass index • sum of skinfold thicknesses • waist circumference • race • sex • growth curve modeling
INTRODUCTION
The prevalence rates of obesity in both adults and children in the United States have significantly increased over the past decade and continue to increase (1), particularly among minorities such as black Americans (2). Obesity is associated with complications such as dyslipidemia, hypertension, and insulin resistance and is an important risk factor for cardiovascular disease (CVD), type 2 diabetes, and cancer (3–5). Because obesity in childhood is a major risk factor for adult obesity (6), greater insight into the development of adiposity from childhood into adulthood in different demographic groups and early identification of children at risk would be beneficial to prevention efforts.
A limited number of pediatric longitudinal studies on the development of obesity have been performed (7–13), and only a few of these studies have evaluated race and sex effects. Moreover, only one of these longitudinal studies explored the development of measures of both general and central adiposity (11). Studies from the early 1980s (14, 15) observed that central obesity was more strongly associated with cardiovascular morbidity and mortality than was general obesity, and later studies identified visceral adipose tissue as the culprit (16).
In addition to race and sex, the prevalence of obesity also varies with socioeconomic status (SES), although this relation is complex and poorly understood (17, 18). Only 2 studies addressed the effect of SES on the development of obesity from childhood into early adulthood and showed that youth with a lower SES had the largest increase in body mass index (BMI; in kg/m2) 6 or 7 y later (13, 18).
To the best of our knowledge, this is the first longitudinal study to report the development of both general and central adiposity from childhood to early adulthood within the context of race, sex, and SES. We approached this issue from 2 different angles. We tested whether the development from childhood into early adulthood of BMI and sum of skinfold thicknesses as measures of general adiposity and of waist circumference as a measure of central adiposity were influenced by race, sex, and SES. We also investigated what distinguished the development of general and central adiposity between obese (BMI =" BORDER="0"> 30), overweight (BMI = 25.0–29.9), or normal-weight (BMI < 25.0) subjects at the end of their childhood (> 17 y). To this end, we used growth curve modeling, which is particularly suited for the analysis of longitudinal data (19), to explore interindividual differences in the development of general and central obesity over time in a sample of 622 European American (EA) and African American (AA) males and females aged 4.2–27.5 y.
SUBJECTS AND METHODS
Subjects
A total of 748 subjects (166 AA males, 186 AA females, 205 EA males, and 191 EA females) participated in this study. The subjects were participants in an ongoing longitudinal study of the development of CVD risk factors in which annual evaluations were conducted from 1987 to 1998 (20–22).
Participants had a verified positive family history of CVD, including essential hypertension, premature myocardial infarction (< 55 y of age), or both in one or both biological parents or in one or more grandparents (20). A family history of CVD was verified by the subjects’ physician or medical records. On the baseline evaluation, the subjects were normotensive for age and sex and were apparently healthy on the basis of parental report of the child’s medical history. The subjects were classified as AA or EA according to the criteria described previously (20). Informed consent was obtained from one of the parents and from the children in accordance with procedures approved by the Institutional Review Board at the Medical College of Georgia.
Recruitment and evaluation of subjects began in 1987 and is described elsewhere (23, 24). The annualized attrition rate has been < 4%/y, which has been primarily due to some of the subjects moving out of the region. There have been no significant differences in age, race, or sex distributions between the dropouts and the subjects that remained in the study.
Anthropometric measures
Anthropometric evaluations were conducted at each annual laboratory visit over an 11-y period. Height was measured to the nearest 0.1 cm while the subjects were shoeless and weight was measured to the nearest 0.1 kg while the subjects were wearing shorts and a shirt with a medical scale that was calibrated daily. Skinfold thicknesses (ie, triceps, subscapular, and suprailiac) were measured on the right side of the body with Lange calipers according to established protocols (25). Three sets of readings were recorded and averaged. Waist circumference (in cm) was measured twice at the center of the umbilicus and the values were averaged. From these primary measures, BMI (wt/ht2) and the sum of 3 skinfold thicknesses were calculated as measures of general adiposity. Waist circumference was used as the measure of central adiposity (26).
Socioeconomic status
SES was represented by parental education level, ie, the mother’s or father’s education level, because these measures remained highly stable across the years of the study. Thus, the parental education level at the midpoint of the study was considered representative of the entire study period. Parental education level was measured in years on a 7-point scale that ranged from less than high school to postgraduate education and was subsequently divided into 3 categories: low (< 12 y), medium (=" BORDER="0"> 12 and < 16 y), and high (=" BORDER="0"> 16 y).
Statistical analyses
The aims of our study were twofold: 1) to test whether the development of general and central adiposity from childhood into early adulthood is influenced by race, sex, and SES; and 2) to investigate the development of general and central adiposity in persons who were obese (BMI =" BORDER="0"> 30), overweight (BMI = 25.0–29.9), or of normal weight (BMI < 25.0) after age 17 y, at which age the prediction of adult overweight is reported to be very good (27). To achieve these 2 aims, we used individual growth curve modeling (28, 29). This statistical technique is particularly suited for the analysis of longitudinal data and has several advantages over traditional methods for analyzing longitudinal data (19, 22).
The analysis of individual growth curves was implemented by using mixed linear models in PROC MIXED of the SAS/STAT software package (release 8.02, 1999; SAS Institute Inc, Cary, NC). Three dependent variables were analyzed separately: BMI, waist circumference, and sum of skinfold thicknesses. BMI was log transformed, and waist circumference and sum of skinfold thicknesses were square root transformed to eliminate convergence problems. Only subjects with =" BORDER="0"> 3 observations were included because the analyses depend on fitting polynomial regression curves to each subject’s data. Six hundred forty-nine subjects (144 AA males, 168 AA females, 173 EA males, and 164 EA females) met this criterion.
For the analysis of the first aim, each dependent variable was analyzed by using a mixed linear model that included age, age2, age3, race, sex, SES, and the interactions among these factors as fixed effects. To avoid computational problems, age was rescaled and centered by using age = (age/mean age) - 1, where mean age is the average age of the entire group. Age2 and age3 were calculated from this rescaled age. Rescaled age, age2, and age3 were included as continuous variables, whereas race, sex, and SES were included as categorical variables. Effects of race, sex, SES, and their interactions represent effects on the growth curve level. Effects on the rate of change of central and general adiposity measures were modeled as interactions with age, age2, and age3. Additionally, the intercept, age, age2, and age3 were included as random effects, with separate coefficients for each of these independent variables being fit to each subject’s data. Significant variability in these coefficients indicated that growth rates differed between the subjects. Mother’s education was never found to have a significant effect, so this variable was not included in any further analyses. Only 622 subjects (136 AA males, 156 AA females, 169 EA males, and 161 EA females) were used in this analysis because 27 subjects had no record of father’s education. Mean (± SD) values for characteristics at the subjects’ first visit (mean age: 11.4 y; range: 4.2–23.9 y) are shown in Table 1. The data set is complicated because not all subjects had the same number of visits and because the subjects were recruited into the study at different ages and in different years. However, > 80% of all 622 subjects had =" BORDER="0"> 5 visits, which made this data set very informative for the study of adiposity changes over time.
View this table:
TABLE 1. Characteristics at the subjects’ first visit1
To compare the development of general and central adiposity among subjects who were obese, overweight, or of normal weight after age 17 y, the 3 dependent variables were included in a mixed linear model that included age, age2, age3, race, sex, BMI category, and the interactions among these variables as fixed effects. The average BMI for all visits at which a subject was =" BORDER="0"> 17 y of age was calculated for each subject, and this average BMI was then used to categorize the subjects as obese, overweight, or of normal weight. Father’s education was never significant in these analyses and was therefore removed from the model. Age, age2, and age3 were rescaled and centered as above. A total of 510 subjects were included in this analysis because 139 subjects were not =" BORDER="0"> 17 y of age by the end of the study.
The significance of fixed effects was determined by using an F test based on the method of Kenward and Roger (30) to calculate the appropriate df. All random effects (intercept, age, age2, or age3) were tested by using a likelihood ratio test, which asymptotically has a chi-squared distribution.
In our study, 123 of the 622 subjects were siblings. Siblings share genes and environment and consequently will be more alike than will subjects from different families. Although this dependency between siblings does not lead to biased estimates, it may result in an overestimate of the significance of observed effects (31). However, when siblings were excluded from the analyses, the pattern of significant results was virtually identical, so the results for the entire sample are reported here.
RESULTS
Effects of race, sex, and socioeconomic status on growth curves
BMI ranged from 12 to 20 in children < 7 y of age and increased to a range of 15 to > 50 in subjects in their late teens. Both waist circumference and the sum of skinfold thicknesses showed similar general patterns of increase. Analyses of individual growth curves indicated that the intercept and the linear, quadratic, and cubic components of the growth curves for all variables differed between subjects (the random components were significant for all variables). The pattern of significant results of the fixed effects on the growth curves was similar for the 3 individual skinfold-thickness measures that made up the sum of skinfold thicknesses (Table 2). As such, the 3 skinfold-thickness measures are well summarized by the results for the sum of skinfold thicknesses.
View this table:
TABLE 2. Summary of growth curve modeling for BMI; waist circumference; triceps, subscapular, and suprailiac skinfold thicknesses; and sum of skinfold thicknesses with a model that included race, sex, and socioeconomic status (SES), as categorical variables1
Growth curves for both waist circumference and the sum of skinfold thicknesses, but not for BMI, differed between males and females (Figure 1). For waist circumference, the linear and cubic components (significant sex x age and sex x age3 interactions) were significantly different between the sexes (Table 2). The linear, quadratic, and cubic coefficients reported below pertain to rescaled age, age2, age3, and the transformed dependent variables. The linear increase for males (slope = 1.681, SE = 0.1066) was greater than that for females (slope = 1.368, SE = 0.1021), whereas the cubic change did not differ significantly from zero for males (coefficient = 0.006, SE = 0.4116) and was much larger for females (coefficient = 1.340, SE = 0.0402). For the sum of skinfold thicknesses, the linear and quadratic components for the growth curve differed between the sexes (significant sex x age and sex x age2 interactions; Table 2). The linear increase of females (slope = 2.3394, SE = 0.2968) was greater than that of males (slope = 0.7773, SE = 0.3108). The curve for females leveled off with age (quadratic coefficient = -1.6026, SE = 0.5033), whereas the curve for males showed greater increases with age (quadratic coefficient = 0.5749, SE = 0.4828).
View larger version (12K):
FIGURE 1.. Estimated growth curves for waist circumference and sum of skinfold thicknesses by sex. Males and females differed in their linear and cubic components of growth curves for waist circumference (significant sex x age and sex x age3 interactions; Table 2) and in the linear and quadratic components for sum of skinfold thicknesses (significant sex x age and sex x age2 interactions; Table 2). Age distribution of measurements: 10 y, 11.6%; 11–15 y, 34.2%; 16–20 y, 43.2%; > 20 y, 11%.
The growth curves for waist circumference also differed among the SES categories (significant SES x age interaction; Table 2), with subjects in the lowest SES category having the greatest linear increase in waist circumference (slope = 8.9533, SE = 0.07228) and the other 2 SES categories having similar increases (mid-SES: slope = 8.7305, SE = 0.03504; high-SES: slope = 8.6648, SE = 0.07226). The difference in slopes between the low-SES and the other 2 SES categories was highly significant (F[1,460] = 11.12, P = 0.0009). The growth curves for BMI showed a similar pattern, with marginally significant differences among the SES categories (SES x age interaction; Table 2). For BMI, the difference between the low-SES slope and the other 2 SES categories was also significant (F[1,501] = 5.26, P = 0.022).
The mean for each variable reflects the relative level of each growth curve. As such, the positioning of the growth curves was affected by the specific combination of race, sex, and SES for both BMI and sum of skinfold thicknesses (SES x race x sex interaction, Table 2). BMI was lowest in the high-SES group for EA males and EA females, but was highest in the medium-SES group for AA males and lowest in the medium-SES group for AA females (Figure 2). The sum of skinfold thicknesses showed a similar pattern (Figure 2). Waist circumference differed among the 4 race and sex categories (race x sex interaction, P < 0.05; Table 2
View larger version (18K):
FIGURE 2.. Mean (± SE) BMI values and sum of skinfold thicknesses by race, sex, and socioeconomic status (SES). AA, African American; EA, European American.
Effects of body mass index category on growth curves
Subjects who were of normal weight after reaching adulthood (average BMI of < 25.0 after 17 y of age) began with average BMIs of 15 (children aged < 7 y), and their mean BMI increased to 24 by early adulthood (Figure 3). Subjects who were overweight after reaching adulthood (average BMI of 25.0–29.9 after 17 y of age) began with average BMIs of 16, which increased to 27 by early adulthood. Subjects who were obese once they reached adulthood (average BMI of =" BORDER="0"> 30 after 17 y of age) began with an average BMI of 18, which increased to 37 by early adulthood. The growth curves for BMI differed in a complex way between the race, sex, and final BMI categories: BMI category x race x sex x age2 and BMI category x race x sex x age3 interactions were significant (Table 3 and Figure 3). Regardless of race or sex, the increase in BMI was directly related to the final BMI categories; subjects who were obese as adults showed the fastest increase in BMI, subjects who were overweight showed a moderate increase in BMI, and subjects who were of normal weight showed the slowest increase in BMI (Figure 3). Thus, the general pattern was that differences in BMI between the 3 final BMI categories increased with age. Race and sex together with final BMI category also affected the growth curves. The growth curves for EAs were not significantly different between the final BMI categories at early ages, whereas the growth curves for AAs showed differences in BMI at early ages between those who were obese and those who were of normal weight at the end of the study; the initial BMI values for the middle (overweight) group, relative to the other 2 groups, differed between AA males and females.
View larger version (22K):
FIGURE 3.. Estimated growth curves for BMI by race, sex, and average BMI after age 17 y. AA, African American; EA, European American.
View this table:
TABLE 3. Summary of growth curve modeling for BMI; waist circumference; triceps, subscapular, and suprailiac skinfold thicknesses; and sum of skinfold thicknesses with a model that included race, sex, and BMI category after age 17 y as categorical variables1
Waist circumference and sum of skinfold thicknesses showed similar patterns, with growth curves differing in a complex way between the race, sex, and final BMI categories: BMI category x race x sex and BMI category x race x sex x age2 interactions were significant (Table 3). The growth curves for both waist circumference and sum of skinfold thicknesses showed increasing differences between the 3 final BMI categories with age and showed discrimination between BMI categories at earlier ages in AAs than in EAs, as was observed for BMI. For sum of skinfold thicknesses, males of a normal weight showed no increase on average, whereas females did (data not shown).
DISCUSSION
The main objective of this study was to assess the effects of race, sex, and SES on the development of BMI and sum of skinfold thicknesses (as measures of general adiposity) and of waist circumference (as a measure of central adiposity) from childhood into early adulthood. Our results showed that both general and central adiposity measures increased from childhood through early adolescence, with males and females showing different trajectories for sum of skinfold thicknesses and waist circumference but not for BMI. Increases in general adiposity (7, 10, 12) as well as in central adiposity (11) were reported previously in childhood and adolescence. Our finding of different growth curves for sum of skinfold thicknesses for males and females is consistent with the different male and female trajectories for percentage body fat observed by Labarthe et al (7). In contrast with our findings for waist circumference, Huang et al (11) did not find sex differences in the growth rates of visceral and subcutaneous abdominal fat during childhood and early adolescence, possibly because of their relatively small sample size (n = 138) and short follow-up period of 3–5 y.
The shape of the growth curve for general and central adiposity was unaffected by race in our study, although race, sex, and SES had joint effects on general adiposity levels. Huang et al (11) also observed no race differences for growth rate in subcutaneous abdominal fat but did find such differences for growth in visceral fat. Partly in line with our findings, Kimm et al (12) found that racial differences in sum of skinfold thicknesses and BMI were relatively stable during early adolescence, with black girls showing higher levels than white girls from age 12 y onward; however, BMI and sum of skinfold thicknesses increased at a greater rate in black girls in their late teens.
One unique feature of this study was the evaluation of the influence of SES on the development of general and central adiposity from childhood into early adulthood. SES is usually strongly confounded with race, and our study was no exception in that it overrepresented EAs in the high-SES category and AAs in the low-SES category. However, inclusion of both variables in our models allowed us to assess their independent effects as well as their potential interactions. Our results showed that SES did affect the level of general adiposity (BMI and sum of skinfold thicknesses) in combination with race and sex. As expected, BMI was lowest in the highest SES group in EA males and females. This result agreed with the generally observed inverse relation between SES and obesity risk in white men and women (32). Unexpectedly, we found general adiposity to be highest and lowest in the medium-SES group for AA males and females, respectively. The differential relation of BMI and sum of skinfold thicknesses with SES between AA and EA females is particularly striking and similar to the findings of Burke et al (33), who noted a negative association between education and body size in white but not in black women aged 18–30 y. Body image (or satisfaction) is known to be different in black than in white females and has been shown to vary with SES (34). The complex relation between race, SES, and body image (or self-perceived body weight) may, therefore, offer a possible explanation for the difference in the relation of SES with general adiposity between black and white women (17, 34).
The association of SES with central adiposity was more straightforward, with waist circumference inversely related to SES irrespective of race and sex. Not only did subjects in the low-SES category have the largest waist circumference, they also showed the strongest increase in this measure over time (as did BMI to a lesser extent). Interestingly, EA females had smaller waist circumferences than did EA males, but this pattern was the reverse in AAs.
A second aim of our study was to investigate the differences in development of general and central obesity between persons who were obese, overweight, or of normal weight after age 17 y. We chose this cutoff point because Guo and Chumlea (27) showed that the prediction of adult overweight is excellent at 18 y of age, and later cutoff points would have resulted in progressively smaller sample sizes because subjects would not have reached the cutoff age by the end of our study. Only small differences in growth curves between subjects in the 3 final weight categories were found before age 10 y, especially in EAs. However, differences became progressively larger as the subjects aged; the obese subjects showed the most rapid increase in all 3 adiposity measures. Growth curves in the obese subjects could be distinguished from those of the normal-weight subjects at an earlier age in AAs than in EAs. Although not explicitly addressed in our study, this observation implies that the prediction of adult obesity is likely to be more accurate in early childhood in AAs than in EAs. This result is promising given the higher prevalence of overweight and obesity and its related morbidity and mortality in AAs in general, particularly females, than in EAs (2, 12). Early identification of children who are at risk of becoming obese is important for many reasons. Preventive lifestyle changes, such as increases in physical exercise, reductions in television viewing, and the consumption of healthier and smaller portions of food, are easier to implement in childhood. Furthermore, prevention efforts must be started in childhood before obesity-related health problems have had the chance to become established.
Participants in our study had a verified positive family history of CVD, which may have increased their risk of developing obesity (35). This seems to be supported by the observation that a large percentage of AA (> 45%) and EA (> 30%) subjects was either overweight or obese when they reached adulthood in our study. However, we have no reason to believe that the relative effects of race, sex, SES, and final BMI category in our high-risk cohort were different from those in the general population.
Although the prevalence of obesity in American children has sharply increased over the past few decades, evidence for a concomitant rise in food intake is scarce (36). This apparent paradoxical finding suggests that the high prevalence of adiposity is at least partly due to a decrease in physical activity (ie, a sedentary lifestyle). Indeed, the amount of physical activity in American youth is lower than the recommended level, and a significant decrease in reported physical activity has been found in the high school years (37). Kimm et al (38) reported an even more dramatic decrease in physical activity during the teen years among American girls, especially AAs. Thus, the observed increase in central and general obesity from childhood into early adulthood in our study may have been due, at least in part, to the strong decrease in physical activity observed in American children. Unfortunately, reliable measures of physical activity and television viewing were not available for the entire duration of this longitudinal study, so such influences could not be assessed. In a recent cross-sectional analysis of the same subjects, we found only a small effect of self-reported physical activity on adiposity (39). However, careful experimental manipulation of the physical activity level in randomized clinical trials, as done in many studies conducted by our research group, has been shown to reduce adiposity in children (40, 41).
In conclusion, our results suggest that the rate of development of adiposity from childhood into early adulthood is influenced by sex and SES but not by race. However, race, sex, and SES have joint effects on adiposity levels. AA females and persons from low-SES backgrounds have the highest risk of becoming obese in adulthood. The development of obesity can begin to be distinguished in midchildhood and is often established by early adulthood, but the age at which the distinction becomes apparent depends on race. Therefore, prevention efforts should start in childhood. AAs should be targeted at an earlier age than should EAs, and the focus should be on the most vulnerable groups—such as those from a low-SES background—to yield the highest benefit.
ACKNOWLEDGMENTS
JCD participated in the design of the study and drafted the manuscript. RHP performed the statistical analysis and participated in the design of the study and in the drafting of the manuscript. FAT conceived of the original longitudinal study, participated in its design and coordination, and edited the manuscript. PB and BG provided significant advice and edited the manuscript. HS developed the original idea for the study and participated in the design of the study and in the drafting of the manuscript. None of the authors had any conflict of interest to report.
REFERENCES
