Obesity and depressed mood associations differ by race/ethnicity in adolescent girls

Abstract

Objectives

To evaluate bidirectional associations between obesity and depressed mood in adolescent girls, and assess whether these associations differed by racial/ethnic group.

Methods

We analyzed data collected from 918 adolescent girls studied in 6^th and 8^th grades as part of the Trial of Activity for Adolescent Girls (TAAG). Racial/ethnic group was defined as non-Hispanic white, non-Hispanic black, and Hispanic. Height and weight were measured and obesity was defined as a BMI-for-age at or above the 95^th percentile. The Center for Epidemiologic Studies Depression Scale (CES-D) was used to measure depressive symptoms. Generalized estimating equations were used to examine associations between 6^th grade obesity and 8^th grade depressed mood, as well as 6^th grade depressed mood and 8^th grade obesity.

Results

Racial/ethnic group was a statistically significant effect modifier in both directions of association (p < 0.02). Among white girls, 6^th grade obesity was associated with greater likelihood of depressed mood in 8^th grade (odds ratio (OR) = 2.47, 95% confidence interval (CI): 1.85, 3.30); for black and Hispanic girls OR= 1.16 and 0.82, respectively. Also for white girls, 6^th grade depressed mood was associated with greater likelihood of obesity in 8^th grade (OR = 4.47, CI: 1.96, 10.24); whereas for black and Hispanic girls OR= 0.83 and 1.89, respectively.

Conclusions

These analyses suggest that associations between obesity and depressed mood may be most problematic among adolescent girls in the white racial/ethnic group, and our results are consistent both with depressed mood contributing to obesity and obesity contributing to depressed mood.

INTRODUCTION

Independently, obesity and depression each have a large negative impact on public health (1). Understanding the psychological antecedents and consequences of adolescent obesity is important in the context of the population increases in obesity prevalence that have occurred during the past three decades (2). Evidence indicates that obesity can contribute to depressed mood (3–5) and also that depressed mood can contribute to obesity and excess weight gain (6–11). These bidirectional associations may be more evident in females compared to males (12). In the Children in the Community Study (3, 6), and the Dunedin Multidisciplinary Health and Development Study (9) males and females were studied, but associations between depressed mood and obesity were only present among females. However, Mustillo et al. found that for boys but not girls, chronic obesity between the ages of 9 and 16 years in the Great Smoky Mountains Study was associated with depression (13), and other studies have not found consistent associations in either boys or girls (14).

Cultural differences in preference for and/or acceptance of a higher body weight may contribute to lower levels of body dissatisfaction among African-American adolescent girls compared to white girls (15, 16). As a result, a weaker linkage between obesity and depression for African-American adolescent girls compared to white girls has been hypothesized (17). However, the few prospective analyses of United States adolescents or young adults that have examined differences in associations between depression and obesity by racial/ethnic group have not found evidence for such interactions (7, 11).

Using prospective data collected from a sample of adolescent girls studied in 6^th and 8^th grades, our objective was to investigate whether bidirectional associations between obesity and depressed mood exist in this sample of middle school girls and whether these associations were different among white, black, and Hispanic girls. We sought to understand the nature of the relationship between depressive symptoms and obesity, and particularly to determine whether patterns varied by racial/ethnic group.

METHODS

Study sample

Data were collected as part of the Trial of Activity for Adolescent Girls (TAAG). TAAG was a multi-center group-randomized trial designed to test an intervention to reduce the usual decline in physical activity in middle-school girls (18, 19). TAAG was a collaboration consisting of six field centers (San Diego State University, Tulane University, University of Arizona, University of Maryland, University of Minnesota, and University of South Carolina) the coordinating center (University of North Carolina Chapel Hill), and the project office (National Heart, Lung, and Blood Institute). Public middle schools in which a majority of students lived in the surrounding community were eligible to participate. Six schools were selected at each of the six field centers for a total of 36 schools. TAAG schools represented the demographic and socioeconomic makeup of their school districts, with preference given to schools with greater racial/ethnic and socioeconomic diversity. Details of recruitment and participation are presented elsewhere (20, 21). Parental consent and student assent were obtained prior to each measurement period. The protocol was approved by the institutional review boards of all participating institutions in fall 2002 and annually thereafter.

Girls were selected within schools by random sampling of all eligible girls. Reasons for ineligibility were unable to read and understand English, told by a doctor to avoid exercise, or other medical contraindication. In spring 2003, a sample of 45 to 60 sixth-grade girls per school was selected; parental consent and student assent were obtained for 1721 of the 2160 eligible girls for an average recruitment rate of 80%. Two years later, in spring 2005, a new cross-sectional sample of 90 to 120 eighth-grade girls was selected from the same schools and to maximize the number of girls available for longitudinal studies, all previously-measured 8^th grade girls who remained in their original school were recruited, even if they were not in the new cross-sectional sample. 1214 girls were assessed both in 6^th and 8^th grade.

Our analyses include those girls assessed in 6^th and 8^th grades, who provided complete data on age, depressive symptoms, and weight status. Seven girls were excluded because they did not have information on depressive symptoms at 8^th grade, two girls were excluded because they were over age 16 at 6^th grade, and 27 girls were excluded because they were missing information on 1 or more covariates (we conducted analyses to determine the extent to which inclusion of these girls impacted our findings). The objective of our analyses was to understand differences among racial/ethnic groups in associations between depressive symptoms and obesity; thus we limited our analyses to the three most prevalent racial/ethnic groups (Hispanic, non-Hispanic black, non-Hispanic white). We excluded 121 girls whose racial/ethnic group was Asian, Native Hawaiian or other Pacific Islander, American Indian or Alaska Native, or other, leaving 918 girls for our analyses. Girls were not excluded from our analyses based on their 6^th grade BMI or level of depressive symptoms.

Measurements

Measurements were taken during spring 2003 and 2005 by trained and certified staff. Periodic recertification ensured that performance standards were met continuously.

Obesity

Height (without shoes) was measured to the nearest 0.1 cm using a portable stadiometer (Shorr Productions). Weight (light clothing without shoes) was measured to the nearest 0.1 kg using a digital scale (Seca 880). BMI was calculated as weight (kg)/height² (m). The Centers for Disease Control and Prevention (CDC) 2000 sex-specific BMI-for-age growth charts were used to determine BMI z-scores (22). Following recommendations for defining obesity in children and adolescents in the US (23), we defined obesity as a BMI-for-age at or above the 95^th percentile of the CDC growth reference.

Depressive symptoms

The Center for Epidemiologic Studies Depression Scale (CES-D) was used to measure depressive symptoms. This 20-item self-administered instrument is widely used to assess depression in community studies and has been validated with adolescents (24). Consistent with previous analyses of TAAG data (25), we defined depressed mood as a CES-D score at or above 24. This cut point was determined to maximize sensitivity and specificity to detect Major Depressive Disorder in adolescent girls (26).

Racial/Ethnic Group

Each girl responded to two questions about her racial/ethnic group. The first asked whether the girl thought of herself as Hispanic, Mexican American, or of Spanish origin. The second asked whether the girl thought of herself as white, black or African-American, Asian, Native Hawaiian or other Pacific Islander, American Indian or Alaska Native, or other. We categorized racial/ethnic group based on these questions as non-Hispanic white, non-Hispanic black, and Hispanic; we will refer to these groups as white, black, and Hispanic. Too few girls identified with other racial/ethnic groups to allow for meaningful analyses, and thus we restricted our analysis to these three major race/ethnicity groups.

Covariates

Exact age was calculated from reported birth date and the date of assessments. As an indicator of low family income we relied on whether a girl qualified for free/reduced school lunch at 6^th grade. Participation in the free/reduced lunch program was assessed via self-report; girls were asked, “Do you get free or low-cost lunches at school?” We also included as a covariate a measure of the amount of time typically spent home alone after school on weekdays because this variable was related to depressive symptoms in previous analyses of TAAG (25). Girls reported the number of hours they typically spent at home alone after school in 6^th grade, and we categorized this as none, less than 1 hour per day, or 1 or more hours per day. Self-esteem and physical activity are related to obesity and to depressed mood, but we did not consider these variables to be potential confounders of associations between obesity and depressed mood; both of these variables may be intermediates between obesity and depressed mood, and it is also likely that associations of self-esteem and physical activity with depressed mood and obesity differ by racial/ethnic group. Future research could delineate these relationships, which were not the focus of our current analyses.

Statistical analyses

Differences by racial/ethnic group in characteristics of girls were compared by chi-square, with Bonferoni adjustment for multiple comparisons, and by analysis of variance, with Tukey-Kramer adjustment for multiple comparisons. Pearson correlations between CES-D score and BMI z-score were determined overall and by racial/ethnic group.

We present cross-tabulations, by racial/ethnic group, of cross-sectional associations between 6^th grade obesity and depressed mood, as well as cross-tabulations of each direction (6^th grade obesity predicting 8^th grade depressed mood; and 6^th grade depressed mood predicting 8^th grade obesity) of the prospective associations. We present crude odds ratios and 95% confidence intervals for each direction of the prospective association overall, as well as stratified by 6^th grade level of the 8^th grade outcome. We tested for homogeneity of the odds ratios across these strata using the Breslow-Day test, and calculated the Mantel-Haenszel summary odds ratio and 95% confidence-interval. All analyses were conducted using SAS Version 9.1.3.

We used generalized estimating equations (GEE), implemented via SAS PROC GENMOD, to adjust for covariates and to determine whether prospective associations differed by racial/ethnic group. These models account for the clustering of girls within schools. Analyses included the main effect for the predictor of interest, a three-level indicator for racial/ethnic group, and the interactions between racial/ethnic group and the predictor. Our focus is on obesity and depressed mood as dichotomous variables, so we employed a logit link and binomial error distribution; we also evaluated associations in each direction of association using CES-D score and BMI z-score as continuous variables using an identity link and normal error distribution. Preliminary analyses were conducted to determine whether quadratic or cubic terms should be included with linear terms as continuous predictors; nested models were compared with log-likelihood tests and linear models were found to be adequate. We adjusted for 6^th grade age, free or reduced lunch, and time home alone after school; we assessed the need to adjust for whether the school was a TAAG intervention or control school, and this did not affect the associations.

To assess the potential for confounding by 6^th grade levels of the 8^th grade outcome we evaluated models that were adjusted for 6^th grade obesity or depressed mood as appropriate; we also evaluated the extent to which results differed with adjustment for 6^th grade CES-D or BMI z-score as continuous variables. We present models both with and without adjustment for 6^th grade obesity or depressed mood. To assess effect modification we present models stratified by obesity or depressed mood at 6^th grade.

RESULTS

In 6^th grade, girls were on average 12 years old; 15.6% were obese, and 14.3% had depressed mood. Two years later, 15.1% of girls were obese and 17.9% had depressed mood (Table 1). Differences in characteristics were apparent by racial/ethnic group. At 6^th and 8^th grades, black and Hispanic girls had higher BMI z-scores and were more likely to be obese than white girls. At 6^th grade, white girls had lower average CES-D scores than black or Hispanic girls, and white and Hispanic girls were less likely than black girls to have depressed mood. At 8^th grade, the percentage of white and Hispanic girls with depressed mood had increased and the percentage of black girls with depressed mood had decreased; thus differences by race/ethnicity in depressive symptoms at 8^th grade diminished.

Table 1.

Selected characteristics of girls by racial/ethnic group at 6^th and 8^th grades

	White (n=538)	Black (n=179)	Hispanic (n=201)	P valuea	Total (n=918)
6th grade
Age (years)	11.9 (0.4)b	12.0 (0.5)	11.9 (0.4)	.008ce	11.9 (0.4)
Free/reduced lunch (%)	16.5	62.0	48.3	<.0001cde	32.4
Time alone after school (%)				<.0001cd
None	39.6	36.3	41.8		39.4
<1 hour per day	39.2	24.0	27.4		33.7
≥1 hour per day	21.2	39.7	30.9		26.9
BMI (kg/m2)	19.6 (3.9)	22.3 (5.4)	21.5 (4.8)	<.0001cd	20.5 (4.5)
BMI z-scoref	0.2 (1.0)	0.8 (1.1)	0.7 (1.0)	<.0001cd	0.5 (1.1)
Obesity (%)g	9.3	26.8	22.4	<.0001cd	15.6
CES-D	12.6 (8.5)	16.5 (10.4)	14.4 (8.2)	<.0001cd	13.7 (8.9)
Depressed mood (%)h	10.4	25.1	14.9	<.0001ce	14.3
8th grade
Age (years)	13.9 (0.4)	14.0 (0.5)	13.9 (0.5)	.19	13.9 (0.4)
BMI (kg/m2)	21.4 (4.3)	24.3 (6.0)	23.3 (5.1)	<.0001cd	22.4 (5.0)
BMI z-score	0.4 (1.0)	0.9 (1.0)	0.8 (1.0)	<.0001cd	0.6 (1.0)
Obesity (%)	8.6	27.4	21.9	<.0001cd	15.1
CES-D	14.0 (9.5)	15.2 (8.9)	16.0 (10.0)	.029d	14.7 (9.5)
Depressed mood (%)	16.2	18.4	21.9	.19	17.9

BMI = body mass index; CES-D = Centers for Epidemiologic Studies Depression Scale

^aTest for differences by racial/ethnic group: Chi-square for percentages, analysis of variance for means (Tukey-Kramer adjustment for multiple comparisons)

^bMean (standard deviation) tabulated unless % indicated

^cWhite different from black

^dWhite different from Hispanic

^eBlack different from Hispanic

^fBMI z-score from CDC 2000 BMI-for-age growth reference

^gObesity = BMI z-score ≥95^th percentile

^hDepressed mood = CES-D ≥24

The correlation between BMI z-score at 6^th and 8^th grades (r = 0.91) was greater than the correlation between CES-D score at 6^th and 8^th grades (r=0.39), and this pattern was consistent across racial/ethnic groups (results not shown). Most girls (79%) who were obese at 6^th grade remained obese at 8^th grade, and the number of girls not obese at 6^th grade who became so by 8^th grade was small (n=26). Compared to obesity, there was less persistence in depressed mood between 6^th and 8^th grades; fewer than half of girls (37%) who had depressed mood at 6^th grade remained depressed at 8^th grade, and 115 girls who did not have depressed mood at 6^th grade had depressed mood at 8^th grade.

At 6^th grade, 30 girls (12 white, 9 black, 9 Hispanic) were both obese and had depressed mood; the odds of having depressed mood in 6^th grade for white girls who were obese in 6^th grade were 3.2 times (95% confidence interval (CI): 1.6, 6.5) that of white girls who were not obese (Table 2a); whereas for black and Hispanic girls this cross-sectional odds ratio was not statistically significant and was of smaller magnitude (black girls, OR=0.6 (95% CI: 0.3, 1.4); Hispanic girls, OR=1.6 (95% CI: 0.7, 3.8)). In the bottom of Table 2a, we present cross-tabulations by racial/ethnic group of prospective associations between 6^th grade obesity and 8^th grade depressed mood; in Table 2b we present these cross-tabulations stratified by 6th grade depressed mood. Among white girls, overall, and for those who did not have depressed mood in 6^th grade, obesity in 6^th grade was associated with higher odds for depressed mood in 8^th grade. The Mantel-Haenszel summary odds ratio (95% CI) across strata of 6^th grade depressed mood was 2.0 (1.0, 3.9) for white girls, 1.3 (0.5, 3.1) for black girls, and 0.8 (0.4, 1.9) for Hispanic girls (Table 2b). In Table 3, we present cross-tabulations by racial/ethnic group of prospective associations between 6^th grade depressed mood and 8^th grade obesity overall and with stratification by 6th grade obesity. The Mantel-Haenszel summary odds ratio (95% CI) across strata of 6^th grade obesity was 3.9 (1.1, 13.8) for white girls, 1.6 (0.4, 5.5) for black girls, and 1.5 (0.4, 6.2) for Hispanic girls (Table 3b).

Table 2a.

Cross-tabulations of obesity at 6^th grade and depressed mood at 6^th grade and at 8^th grade by racial/ethnic group

Racial/ethnic group	White/		Black		Hispanic
Cross-sectional	6th grade mood a		6th grade mood		6th grade mood
6th grade weight status	CES-D ≥24	CES-D < 24	CES-D ≥24	CES-D < 24	CES-D ≥24	CES-D < 24
BMI ≥ 95th percentileb	12 (24.0%)c	38 (76.0%)	9 (18.8%)	39 (81.3%)	9 (20.0%)	36 (80.0%)
BMI < 95th percentile	44 (9.0%)	444 (91.0%)	36 (27.5%)	95 (72.5%)	21 (13.5%)	135 (86.5%)
OR (95%CI)		3.2 (1.6, 6.5)		0.6 (0.3, 1.4)		1.6 (0.7, 3.8)
Prospective	8th grade mood		8th grade mood		8th grade mood
6th grade weight status	CES-D ≥24	CES-D < 24	CES-D ≥24	CES-D < 24	CES-D≥24	CES-D < 24
BMI ≥ 95th percentile	15 (30.0%)	35 (70.0%)	9 (18.8%)	39 (81.2%)	9 (20.0%)	36 (80.0%)
BMI < 95th percentile	72 (14.8%)	416 (85.2%)	24 (18.3%)	107 (81.7%)	35 (22.4%)	121 (76.6%)
OR (95%CI)		2.5 (1.3, 4.8)		1.0 (0.4, 2.4)		0.9 (0.4, 2.0)

OR = odds ratio, CI = confidence interval, BMI = body mass index, CES-D = Centers for Epidemiologic Studies Depression Scale

^aDepressed mood = CES-D ≥24

^bBMI z-score from CDC 2000 BMI-for-age growth reference; Obesity = BMI z-score ≥95^th percentile

^cn (row percent) tabulated

^dBreslow-Day test for homogeneity of the odds ratios; P value <0.05 would suggest that the odds ratios across strata are not homogeneous.

Table 2b.

Stratified by 6^th grade depressed mood – cross-tabulation of 6^th grade Obesity and 8^th grade depressed mood

Prospective, stratified by 6th grade depressed mood
Girls with depressed mood at 6th grade
	White n=56		Black n=45		Hispanic n=30
	8th grade mood		8th grade mood		8th grade mood
6th grade weight status	CES-D ≥24	CES-D < 24	CES-D ≥24	CES-D < 24	CES-D ≥24	CES-D < 24
BMI ≥ 95th percentile	5 (41.7%)	7 (58.3%)	5 (55.6%)	4 (44.4%)	3 (33.3%)	6 (66.7%)
BMI < 95th percentile	16 (36.7%)	28 (63.6%)	13 (36.1%)	23 (63.9%)	7 (33.3%)	14 (66.7%)
OR (95%CI)		1.3 (0.3, 4.6)		2.2 (0.5, 9.7)		1.0 (0.2, 5.2)
Girls without depressed mood at 6 th grade
	White n=482		Black n=134		Hispanic n=171
	8th grade mood		8th grade mood		8th grade mood
6th grade weight status	CES-D ≥24	CES-D < 24	CES-D ≥24	CES-D < 24	CES-D ≥24	CES-D < 24
BMI ≥ 95th percentile	10 (26.3%)	28 (73.7%)	4 (10.3%)	35 (89.7%)	6 (16.7%)	30 (83.3%)
BMI < 95th percentile	56 (12.6%)	388 (87.4%)	11 (11.6%)	84 (88.4%)	28 (20.7%)	107 (79.3%)
OR (95% CI)		2.5 (1.1, 5.4)		0.9 (0.3, 2.9)		0.8 (0.3, 2.0)
Breslow-Day P valued		.37		.34		.78
Mantel-Haenszel OR (95% CI)		2.0 (1.0, 3.9)		1.3 (0.5, 3.1)		0.8 (0.4, 1.9)

OR = odds ratio, CI = confidence interval, BMI = body mass index, CES-D = Centers for Epidemiologic Studies Depression Scale

^aDepressed mood = CES-D ≥24

^bBMI z-score from CDC 2000 BMI-for-age growth reference; Obesity = BMI z-score ≥95^th percentile

^cn (row percent) tabulated

^dBreslow-Day test for homogeneity of the odds ratios; P value <0.05 would suggest that the odds ratios across strata are not homogeneous.

Table 3a.

Cross-tabulations of depressed mood at 6^th grade and obesity at 6^th grade and at 8^th grade by racial/ethnic group

Racial/ethnic group	White		Black		Hispanic
Cross-sectional	6th grade weight status		6th grade weight status		6th grade weight status
6th grade mooda	BMI ≥ 95th %tileb	BMI < 95th %tile	BMI ≥ 95th %tile	BMI < 95th %tile	BMI ≥ 95th %tile	BMI < 95th %tile
CES-D ≥24	12 (21.4%)c	44 (78.6%)	9 (20.0%)	36 (80.0%)	9 (30.0%)	21 (70.0%)
CES-D < 24	38 (7.9%)	444 (92.1%)	39 (29.1%)	95 (70.9%)	36 (21.1%)	135 (78.9%)
OR (95% CI)		3.2 (1.6, 6.5)		0.6 (0.3, 1.4)		1.6 (0.7, 3.8)
Prospective	8th grade weight status		8th grade weight status		8th grade weight status
6th grade mood	BMI ≥ 95th %tile	BMI < 95th %tile	BMI ≥ 95th %tile	BMI < 95th %tile	BMI ≥ 95th %tile	BMI < 95th %tile
CES-D ≥24	13 (23.2%)	43 (76.8%)	11 (24.4%)	34 (75.6%)	9 (30.0%)	21 (70.0%)
CES-D < 24	33 (6.8%)	449 (93.2%)	38 (28.4%)	96 (71.6%)	35 (20.5%)	136 (79.5%)
OR (95% CI)		4.1 (2.0, 8.4)		0.8 (0.4, 1.8)		1.7 (0.7, 4.0)

OR = odds ratio, CI = confidence interval, CES-D = Centers for Epidemiologic Studies Depression Scale, BMI=body mass index

^aDepressed mood = CES-D ≥2

^bBMI z-score from CDC 2000 BMI-for-age growth reference; Obesity = BMI z-score ≥95^th percentile

^cn (row percent) tabulated

^dBreslow-Day test for homogeneity of the odds ratios; P value <0.05 would suggest that the odds ratios across strata are not homogeneous.

Table 3b.

Stratified by 6^th grade obesity - cross-tabulations of 6^th grade depressed mood and 8^th grade obesity

Prospective, stratified by obesity at 6th grade
Girls obese at 6th grade
	White n=50		Black n=48		Hispanic n=45
	8th grade weight status		8th grade weight status		8th grade weight status
6th grade mood	BMI ≥ 95th %tile	BMI < 95th %tile	BMI ≥ 95th %tile	BMI < 95th %tile	BMI ≥ 95th %tile	BMI < 95th %tile
CES-D ≥24	11 (91.7%)	1 (8.3%)	8 (88.9%)	1 (11.1%)	8 (88.9%)	1 (11.1%)
CES-D < 24	25 (65.8%)	13 (34.2%)	33 (84.6%)	6 (15.4%)	28 (77.8%)	8 (22.2%)
OR (95% CI)		5.7 (0.7, 49.3)		1.5 (0.2, 13.8)		2.3 (0.2, 21.1)
Girls not obese at 6th grade
	White n=488		Black n=131		Hispanic n=156
	8th grade obesity		8th grade obesity		8th grade obesity
6th grade mood	BMI ≥ 95th %tile	BMI < 95th %tile	BMI ≥ 95th %tile	BMI < 95th %tile	BMI ≥ 95th %tile	BMI < 95th %tile
CES-D ≥24	2 (4.6%)	42 (95.5%)	3 (8.3%)	33 (91.7%)	1 (4.8%)	20 (95.2%)
CES-D < 24	8 (1.8%)	436 (98.2%)	5 (5.3%)	90 (94.7%)	7 (5.2%)	128 (94.8%)
OR (95% CI)		2.6 (0.5, 12.6)		1.6 (0.4, 7.2)		0.9 (0.1, 7.8)
Breslow-Day P valued		.54		.93		.56
Mantel-Haenszel OR (95% CI)		3.9 (1.1, 13.8)		1.6 (0.4, 5.5)		1.5 (0.4, 6.2)

OR = odds ratio, CI = confidence interval, CES-D = Centers for Epidemiologic Studies Depression Scale, BMI=body mass index

^aDepressed mood = CES-D ≥2

^bBMI z-score from CDC 2000 BMI-for-age growth reference; Obesity = BMI z-score ≥95^th percentile

^cn (row percent) tabulated

^dBreslow-Day test for homogeneity of the odds ratios; P value <0.05 would suggest that the odds ratios across strata are not homogeneous.

Multivariate models: obesity in 6^th grade as predictor of depressed mood two years later

The association between 6^th grade obesity and 8^th grade depressed mood depended upon racial/ethnic group (p value for interaction <0.001) (Table 4, top). For white girls, but not black or Hispanic girls, obesity in 6^th grade was associated with higher odds for depressed mood in 8^th grade and this association was not substantively changed with adjustment for 6^th grade depressed mood (Table 4), 6^th grade CES-D score (results not shown), or in the strata of girls who did not have depressed mood at 6^th grade (Table 4). Odds ratios for white girls were statistically significantly different from Hispanic girls; white girls were significantly different from black girls in the unadjusted model, but with adjustment for covariates and 6^th grade depressed mood the precision of estimates for black girls was reduced (Table 4).

Table 4.

Bidirectional prospective associations between obesity and depressed mood

	White	Black	Hispanic	Interaction P valuea
6th grade obesityb as predictor of 8th grade depressed moodc
Unadjusted	2.48 (1.93, 3.19)d	1.03 (0.46, 2.31)	0.86 (0.38, 1.94)	<.0001ef
Adjusted for covariates	2.47 (1.85, 3.30)	1.16 (0.50, 2.72)	0.82 (0.35, 1.92)	<.0001f
Adjusted for covariates and 6th grade depressed moodg	2.09 (1.44, 3.02)	1.29 (0.57, 2.88)	0.76 (0.31, 1.85)	<.0001f
Stratified by 6th grade depressed mood
Among those with depressed mood at 6th grade	1.26 (0.37, 4.31)	2.11 (0.78, 5.72)	0.96 (0.29, 3.24)	.50
Among those without depressed mood at 6th grade	2.50 (1.57, 3.98)	0.98 (0.16, 5.97)	0.72 (0.26, 1.95)	.0002f
6th grade depressed mood as predictor of 8th grade obesity
Unadjusted	4.13 (1.77, 9.64)h	0.82 (0.40, 1.67)	1.66 (0.58, 4.73)	.022e
Adjusted for covariates	4.47 (1.96, 10.24)	0.83 (0.40, 1.71)	1.89 (0.66, 5.40)	.014e
Adjusted for covariates and 6th grade obesityi	3.68 (1.72, 7.87)	1.58 (0.65, 3.83)	1.64 (0.36, 7.49)	.22
Stratified by 6th grade obesity
Among those obese at 6th grade	6.27 (0.49, 80.04)	1.35 (0.23, 7.96)	2.13 (0.17, 26.30)	.59
Among those not obese at 6th grade	3.11 (0.85, 11.32)	1.56 (0.51, 4.72)	1.34 (0.12, 14.81)	.56

BMI = body mass index; CES-D = Centers for Epidemiologic Studies Depression Scale

^aP value (Wald χ²) for interaction by race/ethnicity

^bObesity = BMI z-score ≥95^th percentile of the 2000 CDC BMI-for-age growth charts

^cDepressed mood = CES-D≥24

^dOdds ratios (95% confidence interval) for 8^th grade depressed mood from 6^th grade obesity calculated with generalized estimating equations; all models (except unadjusted) are adjusted for 6^th grade age, free lunch, and time home alone. Reference category is girls of specified racial/ethnic group who were not obese in 6^th grade.

^e= white different from black;

^f= white different from Hispanic.

^g3-way interaction between 6^th grade depressed mood, 6^th grade obesity, and race not statistically significant (p=0.28)

^hOdds ratios (95% confidence interval) for 8^th grade obesity from 6^th grade depressed mood calculated with generalized estimating equations; all models (except unadjusted) are adjusted for 6^th grade age, free lunch, and time home alone. Reference category is girls of specified racial/ethnic group who did not have depressed mood at 6^th grade.

ⁱ3-way interaction between 6^th grade obesity, 6^th grade depressed mood, and race not statistically significant (p=0.91)

Multivariate models: depressed mood in 6^th grade as a predictor of obesity two years later

There was some evidence that the prospective association between 6^th grade depressed mood and obesity in 8^th grade differed by racial/ethnic group (Table 4, bottom). The interaction between 6^th grade depressed mood and racial/ethnic group was statistically significant (p<.05) in the unadjusted model as well as with adjustment for covariates. Among white girls with depressed mood in 6^th grade, the odds of obesity in 8^th grade were 4.47 times (95% CI: 1.96, 10.24) the odds for white girls without depressed mood; this was statistically significantly different from the association among black girls (OR=0.83). Although with adjustment for 6^th grade obesity, the interaction with racial/ethnic group was no longer statistically significant (p=0.22), the odds ratio for white girls remained statistically significant (OR=3.68, 95% CI: 1.72, 7.87). For black and Hispanic girls, odds ratios were smaller and confidence intervals were wide (Table 4). Very few girls were obese in 8^th grade that had not been obese in 6^th grade, and stratification by 6^th grade obesity status resulted in odds ratios with extremely wide confidence intervals.

Results of models in which BMI z-score and CES-D score were tested as continuous variables were consistent with results using dichotomous variables (results not shown). Results were also not substantively different with inclusion of girls who were excluded because they were missing information on one or more covariates (results not shown).

DISCUSSION

In this prospective study of white, black, and Hispanic middle-school girls in the United States, bidirectional associations between obesity and depressed mood differed by racial/ethnic group. The strongest associations between depressed mood and obesity were seen for white girls. Among white girls studied over a two-year follow-up period, obesity predicted higher odds for depressed mood, and depressed mood predicted higher odds for obesity. These prospective associations for white girls were consistent with cross-sectional associations at 6^th grade, and remained with adjustment for covariates, and with adjustment for the 6^th grade level of the outcome. Among black and Hispanic girls, obesity was not strongly associated with depressed mood in either direction of association. However, confidence intervals for black and Hispanic girls were wide.

It is difficult to compare our results to those of other studies of adolescents because of differences in how obesity and depression are assessed and whether results were stratified by gender. In a large nationally representative sample of US adolescents, BeLue et al.(27) observed that obese white and Hispanic youth were more likely than their non-obese peers to be reported to have mental health problems, whereas this was not found for black youth. However in two large multi-ethnic cross-sectional samples of adolescents in the United Kingdom, obesity and depressive symptoms were not strongly related and no differences were observed by gender or ethnicity (14). Our results differ somewhat from those found in the NHLBI Growth and Health Study (NGHS), in which no interaction with race was observed in models predicting young adult obesity from adolescent depression (11). In the NGHS, Hispanic girls were not studied, results are not presented stratified by racial/ethnic group, and their participants were older than the girls in TAAG (11). In addition, girls in TAAG were assessed approximately 5 to 10 years after NGHS, and it is possible that cohort differences in prevalence and perception of obesity could be responsible for the different findings. Bidirectional associations between depressive symptoms and obesity were examined in the National Longitudinal Study of Adolescent Health (7). In that study, for all racial/ethnic groups, depressed mood predicted obesity one year later. The study differed from ours in that it included boys, data were collected in 1995 and 1996, and obesity was based on self-reported height and weight. Associations between obesity and depressive symptoms may be stronger in females than males (3, 6, 9, 12), and misclassification of obesity based on self-reported height and weight may be related to depression (28).

Our results are consistent with other studies that have shown cultural differences in body shape preferences (15, 29). Whereas African-American and Hispanic cultures may have more tolerance for, or even prefer, a larger female figure, in the US the white culture has adopted a preference for a slimmer female body type (29). These cultural differences, whether internalized by girls themselves or contributing to stigmatization, could be hypothesized as one potential explanation for the differences by racial/ethnic group we observed in associations between obesity and depressed mood.

Glymour reported that when an exposure is related to baseline levels of a dependent variable in an observational study, bias can be introduced by including the baseline value of the dependent variable as a covariate or by restricting the analysis to those who did not have the dependent variable (e.g., obesity) at baseline (30). To address this potential for bias, as well as evaluate confounding and effect modification, we present our results with and without adjustment for the 6^th grade level of the 8^th grade outcome, as well as stratified by the 6^th grade level of the 8^th grade outcome.

Tracking of weight status in adolescents and adults is high (31). Consistent with this, we observed a correlation of 0.9 between 6^th and 8^th grade BMI z-scores that was not related to racial/ethnic group; of the 918 girls we studied, only 26 girls were obese in 8^th grade who were not also obese in 6^th grade, and only 30 girls were not obese in 8^th grade who had been obese in 6^th grade. Thus we have limited power to determine the extent to which depressed mood in 6^th grade predicts 8^th grade obesity in analyses stratified by 6^th grade obesity. Persistence of depressed mood between 6^th and 8^th grades was less striking, but still substantial. Given the tracking of weight status and depressive symptoms, it is difficult to determine temporal relationships. Nevertheless, in our analyses these relationships appear to be different by racial/ethnic group.

Our research has some limitations. We were not able to account for maturational timing in our analyses. Differences in pubertal timing among racial/ethnic groups have been established (32), obesity is associated with earlier maturation in females (33, 34), and earlier maturation may increase risk for depression (35, 36). Other potential confounding variables that we were unable to adjust for are cigarette smoking and use of anti-depressant medication. Although we do not think that a large proportion of girls in 6^th grade were smokers, cigarette smoking has been shown in other studies to be related to depression (37) and reduced body weight (38). Although the overall level of depressed mood observed in this cohort of adolescent girls is consistent with estimates from community studies of US adolescents (39, 40), the decline in the prevalence of depressed mood among black girls between 6^th and 8^th grades (25.1% to 18.4%) was unexpected.

In summary, in prospective analyses of US middle-school girls, we found that bidirectional associations between obesity and depressed mood differed by racial/ethnic groups and were strongest among white girls. Further studies are needed to understand why these associations differ by racial/ethnic group and to identify possible protective factors among black and Hispanic youth that could be conveyed to all adolescents. Although the association between obesity and depressed mood was strongest among white girls, the prevalence of obesity and of depressed mood among black and Hispanic girls was higher than for white girls. Thus, these conditions are likely to often co-occur in adolescent girls irrespective of racial/ethnic group.