ELSEVIER

Race, Socioeconomic Status, and Obesity in 90 to lO#Year-Old Girls: The NHLBI Growth and Health Study

SUE Y. S. KIMM, MD, EVA OBARZANEK, PHD, BRUCE A. BARTON, PHD,

CHRISTOPHER E. ASTON, PHD, SHARI L. SIMILO, MS, JOHN A. MORRISON, PHD,

Z. I. SABRY, PHD, GEORGE B. SCHREIBER, DSc, AND ROBERT P. McMAHON, PHD

The purpose of this investigation was to detennine whether measures of socioeconomic status (SES) are inversely associated with obesity in 9- to IO-year-old black and white girls and their parents. Subjects were participants in the Growth and Health Study (NGHS) of the National Heart, Lung, and Blood Institute. Extensive SES, anehropometric, and dietary data were colkcted at baseline on 2379 NGHS participants. The prevalence of obesity was examined in the NGHS girls and parents in relation to SES and selected environmental factors. Less obesity was observed at higher kveels of household income and parental education in white girls but not in black girls. Among the mothers of the NGHS participants who were seen, lower prevalence of obesity was observed with higher levels of income and education for white mothers, but no consistent patterns were seen in black mothers. Univariate logistic models indicated that the prevaknce of obesity was signijicandy and inversely associated with parenti income and education and number of parents in the household in white girls whereas caloric intake and TV viewing were significantly and positively associated with obesity. Among black girls, only TV viewing was significantly and positively associated with ~Jw prevalence of obesity. M&variate logistic regression models revealed that lower parental educational attainment, one-parent household, and increased caloric intake were significantly associated with the prevalence of obesity in white girls; for black girls, only increased hours of TV viewing were significant in these models. It is concluded that socioeconomic status, as measured by education and income, was related to the prevalence of obesity in girls, with racial variation in these associations. A lower prevaience of obesity was seen at higher levels of socioeconomic status in white girls, whereas no clear relationship

was detected in black girls. These findings raise new questions regarding the correlates of obesity in black girls. Ann Epidemiol 1996;266-275.

KEY WORDS: Obesity, socioeconomic status, race, girls, poverty, African-Americans, television, caloric intake.

INTRODUCTION

Obesity in humans is observed to aggregate in families (l), presumably due to both genetic influences and shared life- style within the family unit (2, 3). Genetic transmission is said to account for approximately 25% of the variation of obesity in the general population (4). Cultural factors are also believed to be transmitted within family units and have

From the Department of Family Medicine and Clinical Epidemiology, School of Medicine, University of Pittsburgh, Pittsburgh, PA (S.Y.S.K.); Division of Epidemiology and Clinical Applications, National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, MD (E.O.); Maryland Medical Research Institute, Baltimore, MD (B.A.B., S.L.S., R.P.McM.); Department of Human Genetics, Graduate School of Public Health, University of Pittsbureh, Pittsburgh. PA (C.E.A.): Children’s Hos- pital Medical Center, Cincinnati, OH (JyA:M.);‘Univer.ty of California, Berkeley, CA (Z.I.S.); and Westat, Rockville, MD (G.B.S.).

Please address reprint requests to: S. Y. S. Kimm, MD, Department of Family Medicine and Clinical Epidemiology, School of Medicine, Univer- sity of Pittsburgh, M-200 Scaife Hall, Pittsburgh, PA 15261.

Received January 8, 1996; accepted April 30, 1996.

0 1996 by Eisevier Science Inc. 655 Avenue of the Americas, New York, NY 10010

been estimated to account for 30% of the variation in obesity (4). Hence, a sizeable contribution to the familial aggrega- tion observed in human obesity probably stems from shared environmental and behavioral factors within the household as well as inherited factors.

Among environmental factors, socioeconomic status (SES) has been reported to be inversely related to obesity, particularly in white women (5-8). The high prevalence of obesity seen in black women in the United States (9,lO) has been attributed to greater poverty among black Americans. Because young children are dependent on their adult care- takers for food availability and these caretakers also deter- mine the overall family lifestyle, SES may affect childhood obesity much as it affects adult obesity (11). A comprehen- sive literature review by Sobal and Stunkard (5) as well as other reports (12-15) reveal mixed findings on the relation- ship between obesity and SES in children. These observa- tions suggest that the inverse relationship may not be a clear one, but perhaps comprise a combination of or interaction

1047s2797/96/$15.00 PI1 SlO47-2797(96)00056-7

between several factors, including those not related to SES. For example, Fuchs and Reklis included childhood obesity as one of the indicators of the declining social conditions of U.S. children (12). They noted that childhood obesity increased in the United States during the 196Os, when material conditions improved and cultural changes oc- curred. In the 198Os, when material conditions declined, childhood obesity continued to increase. Therefore, the authors conlectured that the increasing trend in childhood cjbesity over the past several decades probably involved both economic as well as cultural factors.

One study reported an inverse association between the prevalence of obesity among white adults of European origin and the number of generations that their families had lived in the United States (16). The number of generations lived in the United States and SES appeared to have independent associations with the prevalence ofobesity. Attitudes toward slimness appeared to he linked to the extent of acculturation, measured as the number of generations in the United States, rather than to socioeconomic status. The authors speculated thar there might be an “acculturation phenomenon” op- erating and that a person’s adult weight may be partly a product of social intluences operating in his/her childhood.

At present, the manner in which low levels of income and educ;mon are associated with obesity is not well understood. Mo$t reports deal with the observed inverse relationship between SES and body mass index (BMI) and generally do not provide further insight into the nature of this relation- ship. For instance, it is not clear how SES affects behavior (e.g., through overeating or physical inactivity), which then may lead to obesity.

The purpose of this research was to examine the associa- tion between race, SES, and the prevalence of obesity at baseline in ;I large cohort of 9- to IO-year-old black and white girls and their mothers or female guardians participat- ing m the Growth and Health Study (NGHS) of the Na- tional Heart, Lung, and Blood Institute. In this report, obe- sity in black and white girls is assessed in relation to various measures of socloeconomlc status and selected environ- mentnl factors that included caloric intake and television viewing.

METHODS

The NGHS is a multicenter longitudinal study designed to assebs factors associated with the development of obesity in black and white girls during adolescence and to examine the effects of the development of obesity on major risk factcjrs for cardiovascular disease. The three field centers are at the University of California-Berkeley, Children’s HOS- pita1 Medical Center (Cincinnati, OH), and Westat (Rock- ville, MD). The coordinating center is at the Maryland Medical Research Institure (Baltimore, MD). A total of

2379 girls (1213 black and 1166 whitt.), aged 9 and 10 years, were enrolled in the NGHS for long-term follow-up. Parents or guardians of the participant-s were also enrolled and are being followed. This report 1s based on observations made on the NGHS cohort at the time ot entry into the study in 1987. The detailed NGHS study design and baseline characteristics of the cohort were reported elsewhere ( 17). The NGHS protocol was approved by the institutional re- view boards vf the participating clinical c:znters and by NHLBI.

Study Population

The NGHS participants were recruiteci trom .&ools in Richmond, CA, and Cincinnati, OH, ;md from families enrolled in Humana Group Health Plan, Inc., a health maintenance organization (HMO) in the Washington, D.C. area. Sampling was designed to achieve :I brc& spectrum in measures of SES between the black and white NGHS participants to allow for comparison amcmp different SES strata by race. Selection of potential pu& am! parochial schools for subject recruitment was based (XI ct.%sus tract data that showed approximately equal percentages ctf black and white children and the least disparity m income and education between black and white rcsiclents. Participant recruitment strategy in the school setring involved ap- proaching a whole class of potentially age-rlrgiblr girls. Even those girls who were not eligible (i.e., .4+n. Hispanic, or otherwise not meeting the eligibility criteria) were allowed to participate in clinic activities; they tvcre not, however, officially included as NGHS partickpanfs. Th!a inclusive strategy (which yielded a very low refusal r;+t<: coverall) was to help assemble a study population that would reflect, as closely as possible within volunteer parcicipant~:, normal free-living 9- to lo-year-old black and whit-e girls living in the community where the clinical sites ,~rt’ I( ILateii. Subject recruitment in the HMO was base<1 on rilndorn sampling from a membership list of families with qgc-eligible girls. Recruitment of white participants was augmented by Girl Scouts from the same geographic areas a.~ ?he predc~minantly white HMO clinic>.

Eligibility was based on the girl \ ', age ;md rilie anti 011

the parents’ provision of household information. C;irls had to he aged 9 or IO years at the time ot [lie hrst clinic vi.5it. Participation was restricted to girls who self-declared as being either white or black and who were iiving in a racially concordant household. For the data analysts, etther mothers or female guardians were included since: the :riln of this study was to assess environmental rather &an genetic influ- ences. For brevity, future references to tnotber will include female guardian

Baseline Information

At entry in 1987 detailed household, ~lemographic. and

268 Klmm et al. RACE,SES, ANDORESITYIN GIRLS

AEP Vol. 6, NC,. 4 July 1996. 266-275

psychosocial information was obtained for the NGHS child participants and their parents who consented to participate. The measures of SES included in this analysis were parental education, total household income, and number of parents living at home. Level of education for the family was taken as the maximum level of education achieved by either parent. Maximum parental education was originally measured in 13 categories, ranging from zero to 6 years of school to graduate school. Total annual household income was originally mea- sured in nine levels, ranging from -=z $5,000 to > $75,000. Number of parents living at home was included in the analysis as an adjustment variable for the total household income as well as an indicator variable for family social structure.

Measurements of height and weight were obtained for children and mothers. BMI (kg/m2) was calculated from the baseline measures of body weight and height. All informa- tion was collected and measurements made by certified in- terviewers and examiners who followed a common protocol. Details of the methods of measurement have been described (17). Caloric intake was assessed by 3-day food records, which were completed by the girls, reviewed via in-depth interview by trained staff, and coded centrally for nutrient analysis at the University of Minnesota Nutrition Coordi- nating Center (18). Th e mean daily caloric intake was calculated by averaging the daily calories obtained from the 3-day food record. Information on the amount of television viewing was obtained by a questionnaire administered by a trained interviewer with the aid of prompts such as the local TV guides. The total number of hours of TV and videos viewed per week was derived by adding the number of hours of specific programs or videos that the child had reported as having viewed during a typical week. Television viewing was included in the analysis because of a previous report linking TV viewing to BMI in the NGHS cohort (19).

Obesity was defined as the highest quartile (i.e., > 75th percentile) of BMI of respective NGHS population distribu- tions (child, > 20.5 kg/m2; mother, > 30.7 kg/m2) of the two races combined. BMI rather than skinfold measurements was used to assess obesity because BMI is the most frequently used measure in published reports. Overweight is commonly defined as being above the cut point of the 85th percentile of the second National Health and Examination Survey (NHANES II) data (20). However, the NHANES II popula- tion sample is mostly white since it is based on a probability sampling to reflect the overall U.S. population. Further, when stratified by age and gender, the actual sample size for children becomes quite small. The NGHS data base provides more robust information for the derivation of a BMI cut point for studying a biracial cohort. The NHANES II 85th BMI percentile for 9-year-old girls is 19.6 kg/m2, and for lo-year-old girls it is 20.9 kg/m2 (21). The 75th percentile cutpoint for the NGHS cohort (9- and lo-year-

old girls) used in this report is 20.5 kg/m’, a figure very similar to the NHANES II reference criterion.

Statistical Analysis

Prevalence of child and mother’s obesity, total hours of television viewing per week, and mean daily caloric intakes were examined in relation to levels of family income and the mother’s education. Bartholomew’s test for monotonic trend was used to assess statistical significance in these vari- ables across income and education categories for each race (22, 23). For the trend test, the original 13 categories of education were collapsed into six categories (did not com- plete high school (HS); high school graduate or equivalent (HS Grad); course work past high school (post-HS); college l-3 years; college 4’ years; and graduate school). Total annual household income was also collapsed from the original nine to seven categories (< $10,000; $lO,OOO- 19,999; $20,000-29,999; $30,000-39,999; $40,000-49,999; $50,000-74,999; 3 $75,000). Associations among ordered categorical and continuous data were examined with Spear- man’s rank order correlation coefficient. Tests for significant differences in correlation coefficients between racial groups were calculated by use of Fisher’s z-transformation of the correlations (24). To estimate the association of indepen- dent variables with obesity, we used the logistic regression model within each race (25). The education and income categories were further collapsed from above for the logistic regression to have sufficient sample size in each race for each stratum of education and income. Education was col- lapsed from the above to three (HS or less, some college including post-HS, and college 4’ years) categories. Income was further collapsed to four categories for logistic regression (< $10,000; $10,00&19,999; $20,000-$39,999; 2 $40,000). All variables considered in univariate logistic models were included in a multivariate model without using stepwise selection procedures. Regression diagnostics (condition in- dices, variance inflation factors) were calculated to investi- gate whether sufficient collinearity existed among the inde- pendent variables to make model results unreliable (26).

The significance of any single predictor variable on risk was assessed using Wald’s test, approximating the sampling distribution of the ratio of the square of the estimated coef- ficient to its variance with a x2 distribution. These calcula- tions used the SAS logistic regression procedure (27).

RESULTS

Although there was a broad distribution in SES measures among black and white participants, the NGHS cohort contained a greater proportion of black girls in lower family income and parental education categories as compared to whites. For example, 27.9% (n = 317) of black families had

TABLE 1. Age, anthropometric measures, total caloric intake, and television viewing by race --.--- .-_._.-_-

Vari;dde No.

White

Mean SD No.

Rlack

MeZl SD

family incomes of < $10,000, while only 7.8% (n = 87) of white families fell into this category; 19.2% (n = 218) of

black families had family incomes between $10,000 and $19,999, while only 9.5% (n = 105) of white families were in this group. For the upper income strata, although 15% (n = 171) of black families had incomes of > $50,000 per

year, 33.6% (n = 373) ofwhite families were in this stratum. Thia imbalance may be expected to affect the power of the analyses, but- not to bias estimation of trends seen across levels of income and education. However, bias may occur

if the mean values of income within the categories differ between rhe races. Due to the categorical nature of the NGHS income data, such mean values could not be cal-

culated. At baseline, the mean age of the NGHS cohort was 10.0

years, with black girls 1 month older than white girls (Table 1). Black girls had significantly higher mean BMI than did white girls (19.2 kg/m’ vs. 17.9 kg/m’). Average caloric in-

take of blacks was approximately 53 calories per day higher than that of white girls (I’ = 0.02). The most noticeable black-white difference was seen in the TV viewing, with

black girls viewing an average of 11.4 hours more per week than white girls. Heights and weights were obtained on 52% of the mc&rrs (5 1% black and 64% white). Obesity was

more prevalent among black girls (30.6% > 20.5 kg/m’) and their mothers (35% > 30.7 kg/m’) than among the

white girls (19.3%) and their mothers (16.8%). The preva- lence of obesity was significantly lower at higher household income levels for white girls (I’ < 0.001 for the trend) but not for black girls (P = 0.512) (Figure 1A). Likewise, the

prevalence of obesity was significantly lower at higher paren- tal education for white girls (P < 0.001) but not for black girls (I-’ = 0.282) (Figure 1B). Although the prevalence of obesity in black girls with the highest level of parental edu- cation appeared to he the lowest, a x2 test failed to reveal

a significant difference in the obesity prevalence between the highest and lowest educational categories (I’ = 0.318).

In general, results were similar among mothers by race (data not shown). There was a significantly lower prevalence

of obesity at higher income and educatmn levels among white mothers (P < 0.001; P = 0.019, respectively); this

trend was similar to that seen in their daughters. The preva- lence of obesity was not associated with income among black mothers (I’ = 0.332). There was, however, a lower

prevalence of obesity at higher levels of education among black mothers (I’ = 0.003). Note that rhc mothers may represent a self-selected group, since analysis was limited to those who came for a clinical assessment.

Television viewing (hours per week) was examined by

income and education categories (Figure 2). Television viewing was negatively associated with higher household income and parental education in white girl5 (P < 0.001

for both income and education). The asrociations were sig- nificantly weaker in black girls (P = 0.114 and P = 0.067, respectively). N o associations of caloric rntake across in-

come and education categories were obser~,ed CX either race (data not shown; P \:alues of the Bartholc )tnt’w tt*sth for monotonic trends ranged from 0.24 to 1).@?.

Table 2 displays the results ot correl,itron ;malysis of education, income, number of parents m household, total

caloric intake, and weekly hours of TV viewing. In both t-aces parental education was significanth ~orrelatcd with

both income and number of parents in tile household and negatively correlated with TV viewing. However, the nega- tive correlation between education and Ti’ viewing was

significantly stronger in whites than in bin&. Income was significantly correlated with number ofparents in the house- hold as well as negatively correlated with TV viewing in both races. There was a significant negative correlation between the number of parents in the household and the weekly hours of TV viewing among the NGHS girls of both

races. There was no significant correlation between caloric intake and income, education, or TV viewing in rither race.

Table 3 shows the odds ratios (OR) from the univariate logistic models for the association of income, cducaticm, number of parents in household, TV viewing, and caloric

intake with the prevalence of obesity. These OR are consis- tent with the lower prevalence of obesity ~\t liigher income

270 Kimm ct al.

RACE, SES, AND OBESITY IN GIRLS

AEP Vol. 6, /-Jr>. 4 July 199h: X16-275

(4

r

Trend, p = ,512

Trend, p < ,001

W

p = ,282

p < ,001

FIGURE 1. Prevalence of obe- sity in NGHS girls across income levels (A), and education levels (B). Error bars indicate 95% con- fidence intervals for the percent- age obese.

and education levels seen in white girls (Figure 1). The odds ratio for obesity prevalence in white girls (OR,) are significantly < 1.00 for the highest level of education (col- lege 4’ years) and for the higher levels of income ($20,000-

39,999 and 2 $40,000; these OR, correspond to the inverse associations seen in Figure 1). The odds ratios in the black girls (OR,) generally were > 1.00 for all higher levels of income, although the increases were not statistically signifi- cant except for the $20,000-39,999 income level. (P values

are not shown for individual income categories in Table 3.) Both higher total caloric intake and living in a single- parent household were significantly associated with greater prevalence of obesity in white girls but not in black girls. Number of hours of TV viewing was significantly associated with the prevalence of obesity in both black and white girls,

with greater hours of TV associated with higher prevalence to a similar extent among both racial groups.

To account for the correlations observed between inde- pendent variables, we applied multiple logistic regression

(Table 4). For white girls, lower education levels remained significantly associated with the prevalence of obesity, as did having one parent in the household and higher caloric

intake. The OR, for “some college” was not significant but that for “college 4’ years” was. However, income and hours of TV viewing were no longer significantly associated with the prevalence of obesity (P = 0.718 and 0.201, respec-

tively) in the multivariate model for white girls. In contrast, there was essentially no change in the results for the black girls from the results of the univariate regression model (Table 3). In the multivariate model, TV viewing remained

~<$x~~~---~ Trend, p = 0.067

\ Trend, p 4 0.001

$ White + Black

significantly associated with prevalence of obesity. Though number of hours of TV viewing was no longer significantly associated with obesity prevalence in white girls, the OR, was virtually identical to that for black girls (adjusted ORW = 1.08 vs. adjusted ORa = 1.10). Income remained unasso-

ciated with obesity prevalence in black girls (P = 0.113; Table 4) as in the univariate analysis; however, the $20,000- $39,999 annual income category was associated with a higher prevalence of obesity (ORB = 1.6; 95% confidence interval = 1.06-2.41).

DISCUSSION

The results of these analyses provide further evidence of an important association between income and education in the

FIGURE 2. Hours of TV view- ing per week hy NGHS girls across income levels (A), and ed- ucation levels (B). Error bars in- dicate 95% conhdence intervals for the percentage obese.

prevalence of obesity. However, there appears to be racial variation in the association of obesity with rhese measures.

For white girls, there was an inverse relationship between parental income and education and the occurrence of obe- sity. In contrast, for black girls, neither rncome nor educa-

tion appeared to be inversely related to obesity.

What is it about low SES that accounts for its association with obesity in white girls? In this group, both TV viewing

and obesity were strongly inversely associated with house- hold income as well as with parental education. Similarity in these associations with different measures of SES is not unexpected given the high correlation between some of these measures (Table 2). Thus, we see concordance in associations with income and education. However, in white

girls, when other environmental factors were included in

272 Klmm et al. AEI’ VIJI 6, NI>. 4

RACE, SES, AND OBESITY IN GIRLS ]uly 199h: 266-275

TABLE 2. Spearman correlation coefficients among independent variables

No. of parents

Variable Income ,n household

Education”

White 0.51 0.26

(0.0001)” (0.0001)

Black 0.44 0.25

(0.0001) (0.0001)

Black-white difference’ 0.0324 0.7949

Income* White 0.44

(0.0001)

Black 0.50 (0.0001)

Black-white differenceb 0.0688

No. of parents in househo White

Black

Black-white difference*

Total calories’ White

Black

Black-white difference”

‘Education: s high school, some college, college 4+ years. hNumber in parentheses is P value for test if correlation equals zero. ‘Black-white difference p-value: Black-white comparison of correlation coefficients. “Annual income < $lO,ooO, $1O,ooo$19,999, $2O,ooO-$39,999, 2 $40,000. ‘No. of parents: 1 or 2 (includes guardians). ‘Total calories (k&/day) and TV viewing (hr): continuous.

Total calories

0.04

(0.2008) 0.02

(0.4886)

0.6455

0.01

(0.6446) -0.01

(0.6906)

0.6527

0.01

(0.8274) -0.01

(0.7206)

0.6455

TV viewmg

-0.37 (0.0001)

-0.07 (0.0158)

0.0001

-0.26

(O.c001)

-0.11 (0.0005)

0.0003

-0.12

(0.0001) -0.08

(0.0068)

0.3371

0.04

(0.1498) 0.06

(0.0708)

0.6455

a multivariate model, lower education and a one-parent household were found to be independently associated with obesity. Caloric intake was also significantly associated, but to a lesser extent. The association with income, however, was no longer significant. This absence of association may be due in part to its high correlation with education (r = 0.51; Table 2) and the number of parents in the household (r = 0.44, Table 2). In the regression model, parental educa- tion may be better than income as an indicator of environ- mental influences linked to obesity in white girls. For white girls there also appeared to be less TV viewing in better- educated or two-parent households (Table 2, Figure 2).

The NGHS data suggest that the prevalence of obesity in black girls was higher at the higher income levels. This pattern contrasts with the inverse association seen in white girls. A greater prevalence of obesity is observed with higher socioeconomic status in traditional societies where access to food supply is directly correlated with SES (28, 29). Examination of data from the Health and Examination Survey conducted approximately 20 years prior to NGHS revealed a positive association between the mean BMI (BMI

data derived from the population mean values of body weight and height for 1 l-year-old girls) and income for white girls but no variation in BMI with income for ll- year-old black girls (30). One might conjecture that the prevailing cultural milieu of the white household in the early 1960s (at the time of the survey from which the above data were extracted) was more analogous to that seen in traditional societies. This pattern was reversed by 1987, when white girls were seen at NGHS baseline visit. Interest- ingly, neither in the 1960s HES survey nor in the 1980s was income associated (directly or inversely) with obesity among black girls. However, though not statistically signifi- cant, there was a trend toward a higher prevalence of obesity at the higher income levels in NGHS black girls, possibly suggesting that the relationship between income and obesity prevalence appears to be more akin to that of white girls in the 1960s and even possibly to that of traditional societies. Hence, what we observe in the NGHS cohort may be a reflection of differential social development in our society, where a certain lag period may need to elapse between an era when food availability is a concern to an era of affluence

TABLE 3. Odds ratios (OR) and 950/ o confidence intervals (95% CI) of predictors for obesity risk by race (univarcarr iogiwc models) __- Variable ORw” (95% Cl) Pb OR< (95% (:I) P” --. ---_-

i 0.001 d.hC% 0.76 1.14

(0.52-1.12) (O.86.-I.'11

0.40 1 .a1 (0.28-0.58) (@.7Sm I.=?!

Income lev<+ i 0.001 $1 o,ooo-8 19 99’1 0.86 l.ZL’

(0.47-1.59) (0.81--i.;;! $20,000-$39.999 0.47 1.52

(0.28-0.79) (1.08-2.13) -s $40,005 0.36 1.3"

(0.22-0.60) (0.96-2.00)

0.397

-rw, partmi’ 0.40 < 0.001 1 .Oh 0.528 (0.28-0.55) (0.X4-i ;q

TV wewtng/wk 1.18 0.001 1 .oli 0.02 3 (1.06-1.30) (1.01-1.16)

Total calow intake/Jay” 1.04 0.015 O.YO c.225 (1.01-1.08) (0.96-1.011 -̂

,‘ORw = ~xfd, rats> for white girl\. ‘I' v&~e based on Wald’s test for OR = 1.0. ORR = odda ratio tor black girls. ‘Reference level IS s high school. Rdrrence level 1s c $10,000 annual mc<,me.

‘Reference level 1s one parent. Odds for IO-h mcrrment in hours of TV wewing per week; the OR for blacks and whites cumhmed was 1 .I2 (95% (3 = I .05-l .lP), wrth FI P v due tnf -: 3.CO1 ‘Odds for IOO-kcal increment in calorws

with no such concern before an inverse relationship between SES and obesity prevalence can be seen.

On the other hand, some environmental influences may be SE.5independent but may be more pervasive in a specific cultural setting. For instance, there may be greater social tolerance of obesity among African Americans (10, 31). There is some evidence that black girls may feel less peer pressure to be thin or have a greater sense of social accept- ability of obesity. Black NGHS participants showed a prefer- ence for a less-thin body shape than did white girls and also scored higher in the Harter social acceptance subscale as compared with white girls of the same range of adiposity (32, 33).

Finally, those attributes of lifestyle subsumed under “TV viewing” may be SES independent among black girls. After adjustment for mdicators of SES, the magnitude of the OR for TV viewing in white girls was reduced to 1.08 from the unadjusted OR of 1.18. The OR for TV viewing among black girls was virtually unchanged after adjustment for SES indicators (adjusted ORB = 1.10 vs. unadjusted ORB = 1.08). It is of interest that the adjusted OR for TV viewing among black and white girls were almost identical, although the effect of TV viewing was not significant in the multi- variate model for whites. Black girls watched an average of 365 hours of TV per week as compared with 25.1 hours for white girls; therefore the average level of TV viewing by black girls exceeded that of all but the lowest income

groups of white girls. There may have been less ability to detect a similar association of TV viewing with obesity prevalence in white girls as compared with black girls be- cause of the clustering of TV viewing at Iower values among white girls. In addition, because of the stronger negative correlation of TV viewing with both income and education in white girls in contrast to black girls, the relationship between TV viewing and the prevalence of obesity among white girls might have been masked.

There has been much debate regarding the role of TV viewing in childhood obesity (34-39). Television viewing is often regarded as a surrogate measure of relative physical inactivity, since television viewers are not engaged in active outdoor play during TV viewing. A previous report showed a direct association between TV viewing and BMI in the NGHS cohort (19). The factor accounting for the relation- ship between television viewing and obesity prevalence has not been ‘defined. Television viewing haa also been impli- cated with a lower resting metabolic rate in children (36, 40). The popular notion attributing the association between obesity and TV viewing to the influence of increased calories from snacking to the association of obesirp and TV viewing (41) is not supported by the very tow correlation seen be- tween total daily caloric intake and TV viewing in white and black girls (r = 0.04 and 0.06 respectively, Table 2).

The results of our study suggest that the two most com- monly used indicators of SES, namely, education and in-

274 Kimm et al. AEI’ Vol. 6, Nu. 4

RACE, SES, AND OBESlTY IN GIRLS Jul!, 1996: 266-275

TABLE 4. Odds ratios (OR) and 95% confidence intervals (95% CI) of predictors for obesity risk by race (multivariate logistic models)

Variable OR,,,<’ (95% CI) P OR$ (95% Cl) Ph

Maximum educationd 0.018 0.822

Some college 0.86 0.90

(0.56-1.32) (0.64-1.27)

College 4’ years 0.53 0.90

(0.33wl.85) (0.59-1.39)

Income 1eveP $10,000-$19,999

$20,000-$39,999

3 $40,000

0.84

(0.43-1.64) 0.70

(0.38-1.29) 0.76

(0.40-1.45)

0.718 0.113

1.14 (0.73-1.79)

1.60 (1.06-2.41)

1.59

(0.98-2.57)

Two parents’ 0.54 0.004 0.90 0.519

(0.3&0.82) (0.65-1.25)

TV viewing/wkK

Total caloric intake/da$

“ORs = odds ratio for white girls.

1.08 0.201 1.10 0.016

(0.961.21) (1.02-1.19)

1.04 0.020 0.98 0.163

(1.01-1.08) (0.961.01)

hP value based on Wald’s test for OR = 1.0. ‘ORB = odds ratio for black girls. dReference level is < high school. ‘Reference level is < $lO,ooO annual income. ‘Reference level is one parent. Qdds for 10-h increment in hours of TV viewing per week. Qdds for 10%kcal increment in calories.

come, may play a complex role in the prevalence of obesity. On the other hand, measurements of income and education, captured largely through ordered categorical responses, are imprecise measures of socioeconomic status, and their ascer- tainment and significance may vary among different races. Correlations between income and education may introduce further complexity. Thus, regression analyses that include income and education may result in one variable assuming primacy over the other in multivariate models. Neverthe- less, the univariate results show inverse associations between the prevalence of obesity and education and income in white girls; these results are consistent with the published reports for adults. In black girls no inverse association was demonstrable.

Our findings raise new questions regarding correlates of obesity in black girls and women. The role of poverty, which has been traditionally viewed as a major determinant of obesity in minority populations remains to be defined. Obe- sity in black women appears not to be linked to simple associations with socioeconomic status. This lack of associa- tion with SES in blacks suggests that some other aspect of racial difference might be involved. Yet, to ascribe genetic susceptibility as a major causal factor for obesity among black women presupposes the scientific validity of race as rational biologic taxonomy (42). Modem genetic tech- niques have undermined the scientific validity of such cate- gorization (43). Hence, the concept of race should be viewed

as primarily social and cultural in origin (44). Nevertheless, the possibility that genetic factors linked to obesity may be differentially distributed between these two racial groups cannot be ruled out.

In conclusion, these findings from the NGHS should stimulate further research on the complex interplay of social forces that lead to racial differences in the development of obesity. Addressing this issue is particularly important since obesity has been assumed to be associated with high cardio- vascular disease mortality in U.S. black women. The absence of an inverse association of obesity prevalence with parental income and education among black girls raises intriguing questions about possible racial differences in social and cul- tural factors associated with obesity in childhood.

This research was performed under contracts NO-HC-55023-26 of the National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, MD. We gratefully acknowledge the following individuals for

their assistance: Andrea McAllister, B.S., in data analysis; Susan Hojnacki, M.S., in editorial tasks; Paul Hoppe, M.&i, in manuscript preparation. We

also acknowledge the long-term commitment of all NGHS participants and their families.

Participating NGHS Centers were as follows: Clinical Centers: Children’s

Medical Center, Cincinnati, Ohio; University of California, Berkeley, California; Westat, Inc., Rockville. MD. Coordinnring Cenrer: Maryland Medical Research htitute, Baltimore, MD. NIH Program Office: Division of Epidemiology and Clinical Applications, National Heart, Lung, and Blood Institute, Bethesda, MD.

REFERENCES

1.

2.

3.

4.

5

6.

7

6.

0

10.

11.

12.

Ii.

14.

15.

16.

17.

IS.

I9

20

21

22

23

R,lvussin E, Swmhum BA. Pathophysiology of obesity. Lancet. 1992; 34@:404-408.

C;arn ,i;M. Family-lme and s(xx~econnmic fxtorc in fatness and oheslty. Nutr lit-v. 1986;44:381-386.

Borahard C, Perussc 1.. Grnetlc aspects of obesity. Ann NY Acad SC,. lW3;t>99:?6-35.

%~hal J. Stunkard AJ. Socioeconomic status and obesity: a review of rhc hteraturc. Psycho1 Bull. 1989;105:260-275.

National (:cnter fur Health Statistics. Health promotion and disease preventton, Unrted States, 1985. Vital Health Stat [lo], No. 163. l)HHS puhlicatlon no. (PHS) 88-1591. Washington, D.C.: U.S. De- partment <If Health and Human Services; 1987:20, table 3.

i brn SM, Barley SM. Cole PE, Higgins ITT. Level of education, level of ~nc~xne, and It~vel of fatness in adults. Am J Clin Nutr. 1977; 30:7?1-725.

1x1$ II?, Fries JF, Huhert HR. Gender and race differences in the cixrelatton hetwecn body mass and education m the 1971-75 VHANES I. J Epldemk~l Community Health. 1992;46:191-196.

Najlx MF, Rc)wland M. Anthropometrtc reference dara and prevalence ;jf overwclght. Llnited Scare\, 1976-80. DHHS publication no. (PHS) 37- lh88. Washmgton. D.C I!.S. Department c)f Health and Human -;ct\mY 1387.

Kum,tn) iL;l S. Ohelry m black women. Epidemiol Rev. 1987;9:31-50.

i)l~vvtt~r S/i, Ellts<m RC, MoOrc LL, (?illman MW, Garrhie EJ, Singer MR. I’xenr-child rel;monsh[p in nutrient intake: The Framingham < :hlltlrcv~‘~ Srudv. Am J Clin Nutr. 1992;56:593-598.

Fuch\ \‘R, Reklii l)M. America’s children: Economic perspectives and !xh\ ~q?tt~)n~. %ience. l992;255:4l-46.

(;arn SM, tltlpkins PK, Ryan AS. Differential fatness gam of low mcomc’ boo\ and gtrls. Am J Clin Nutr. 1981;34:1465-1468.

<;oIdcn Ml’. Saltzer EB, Dcl’aul-Snyder L, Reiff MI. Ohestty and ~~OCIOC~ ~rmrmi< clas> In children and therr mothers. JDBP 1983:4: 1 I i -1 I?.

Stunkxd AJ. D‘.4qudi E, Fox S, Filion RDL. The influence of social ( I,~Q on olxhlty ,md thmness m children. JAMA. 1972;221:579-584.

G)ldhlstt PB, Moore ME, Stunkard AJ. Social factors in obesity. IAMA. 1’~1~5;192:1@~~-1044.

The National Heart, Lung, and Blood Institute Growth and Health Stud\, Reae.&~ (;roup. Oheaq and cardiovascular disease risk bctors in black and white girls: The NHLBI C Jrowth and Health Study. Am J Public H&h. 1992;82:161 3--1620.

( :rawt<lrd PB, Oharzanek E, Morrison J, Sahry Zl. Comparative advan- race t>f i-day food records ovct 24-hour recall and 5-day fond frequency \xi&tcd by <+servation of 9- and IO-year-old girls. J Am Diet A\s,K 19YJ;Y4:h26-030.

Oharxnek E. Schtelher GB, Crawford PB, et al. Energy Intake and I’hyslcal XI ivory in relatron to Indexes uf hody fat: The National Heart, I.uny, ,uxl BLx>J Instlrute (%owth anJ Health Study. Am J Clm Nutr. 1’)94:6Oli-22.

I lxlan WR. Ep&miolo~~ of childhood obesity: A national perspec- IIW. .4nn NY Acad Sci. 1993;699:1-5.

N,ctlonal Center for Health Statistics. Anthropometnc reference data ,md prevalence c)foverweight: United States, 1976-80. DHEW puhlica- r Ion no. (PHS) 87-1688. Hyattsville, MD: U.S. Department of Health ;mJ Human Servlccs; 1987.

liartholomew t>J. A test of hc>mogeneity for ordered alternatives. Biomernk,l 1 Q59;46:36-48.

Bartholr~mcw I )J. A test <If homogeneity for ordered alternatlves. If. Riomt~trlka 1<)59;46: 328-335.

24.

25.

26.

27.

28.

29.

30.

31.

32.

13.

34.

15.

16.

17.

18.

39.

40.

41.

42.

43.

44.