Dietary patterns and 14-y weight gain in African American women

Abstract

Background: An inverse association between healthy dietary patterns and weight gain that has been shown in white populations is not evident in the few studies in African Americans, a population at high risk of obesity.

Objective: We prospectively assessed dietary patterns in relation to weight gain in African American women.

Design: The study included 41,351 women aged 21–54 y at enrollment in 1995 in the Black Women's Health Study. Participants reported dietary intake in 1995 and 2001 and current weight every 2 y through mailed questionnaires. By using factor analysis, 2 major dietary patterns were identified: a “vegetables/fruit” pattern and a “meat/fried foods” pattern. Multivariable mixed linear regression models were used to estimate mean weight gain in 14 y of follow-up according to each dietary pattern.

Results: Among women who maintained similar dietary patterns in 1995 and 2001, the vegetables/fruit pattern was associated with significantly less weight gain over 14 y (10.88 and 11.94 kg in the highest and lowest quintiles, respectively; P for trend = 0.003), whereas the meat/fried foods pattern was associated with significantly greater weight gain (12.02 and 10.15 kg in the highest and lowest quintiles, respectively; P for trend < 0.001). The associations were stronger among women aged <35 y, whose weight gain was greatest.

Conclusion: These findings are the first to indicate that African American women may be better able to achieve long-term weight maintenance by consuming a diet high in vegetables and fruit and low in red meat and fried foods.

INTRODUCTION

The prevalence of obesity has rapidly increased in the United States, with large disparities by age, sex, and race-ethnicity (1). Over the past 2 decades, increases in body mass index (BMI) have been highest among African American women (1). Approximately 50% of African American women were obese [BMI (in kg/m²) ≥30] as of 2007–2008 (2).

Although it is known that weight gain arises because of positive energy balance, the degree to which specific dietary factors contribute to weight gain remains unclear. Dietary pattern analysis is an important alternative to the traditional approach of examining single foods or nutrients in relation to health outcomes (3–5). A review of 30 cross-sectional studies of dietary patterns in relation to BMI and obesity reported inconsistent associations (6). In the few large follow-up studies that have assessed this question, the evidence suggests that a healthier diet is associated with less weight gain in white populations (7–11).

African Americans tend to have poorer diet quality than do whites (12, 13), but only 2 studies have prospectively examined a measure of diet quality in relation to weight gain separately among African Americans. In the Multi-Ethnic Study of Atherosclerosis (MESA), higher Healthy Eating Index scores were associated with lower BMI after 18 mo of follow-up among whites, but there was no association among African American participants (14). In the Coronary Artery Risk Development in Young Adults (CARDIA) study, a higher Diet Quality Index score was inversely associated with 20-y weight gain in white participants but was positively associated with weight gain among African American participants (15).

The objective of this study was to prospectively assess the association between dietary patterns and weight gain in 14 y of follow-up in a large cohort of African American women.

SUBJECTS AND METHODS

Study population

The Black Women's Health Study (BWHS) is an ongoing follow-up study in African American women. The study was established in 1995 when self-identified black women from all regions of the United States were enrolled through postal questionnaires. The baseline questionnaire collected information on demographic characteristics, lifestyle factors, and medical history; usual diet was assessed through a food-frequency questionnaire (FFQ). A total of 59,000 women aged 21–69 y whose addresses were judged to be valid have been followed through mailed questionnaires every 2 y. Follow-up questionnaires update exposure information and identify incident disease. Follow-up has averaged >80% of the baseline cohort over 7 questionnaire cycles. The Boston University Medical Campus institutional review board approved the protocol and reviewed the study annually.

We excluded women aged ≥55 y at baseline (n = 5716), which is the age at which women began to lose weight on average during follow-up, to isolate the effect of dietary intake on weight change from age-related weight loss. We further excluded women from this analysis for the following reasons: history of cancer at baseline or during follow-up (n = 3098), baseline weight <80 pounds (lb; 36.3 kg) or >300 lb (136.1 kg) (n = 483), history of gastric surgery (ascertained on the 1999 questionnaire; n = 128), missing weight on all follow-up questionnaires (n = 3537), >10 blank food items at baseline (n = 1228), missing or implausible total energy intake values (<400 or >3800 kcal) at baseline (n = 2195), and values beyond the 1st and 99th percentiles of dietary pattern scores (n = 1265). After all exclusions, 41,351 women remained in the present analysis.

Dietary assessment

We assessed usual diet at baseline in 1995 with a 68-item modified version of the National Cancer Institute (NCI)–Block FFQ, and in 2001 with an 85-item version (16). The 9 frequency responses ranged from never or <1/mo to ≥2/d for foods and to ≥6/d for beverages. In 1995, we asked participants to specify a small, medium, or large portion size. A medium portion size was defined for each item (eg, 0.5 cup of broccoli), and small and large servings were weighted as 0.5 and 1.5 times a medium serving size, respectively. In 2001, a super-size portion, equivalent to ≥2 times the size of medium, was added. Nutrients were calculated by using the NCI's Diet*Calc software (17). The 1995 FFQ was validated among 408 participants by using a 3-d dietary record and ≤3 telephone 24-h recalls (18). Energy-adjusted and deattenuated Pearson correlation coefficients for the FFQ compared with diet records and recalls ranged from 0.5 to 0.8 for total fat, saturated fat, protein, carbohydrate, fiber, calcium, vitamin C, folate, and β-carotene.

Assessment of body size and covariates

Participants reported height and current weight on the baseline questionnaire in 1995. Current weight was updated every 2 y by follow-up questionnaire. BMI was calculated as weight in kilograms divided by squared height in meters. In a validation study in 115 participants conducted at Howard University Cancer Center (Washington, DC), self-reported height and weight were highly correlated with measured values (r = 0.93 and 0.97, respectively) (19, 20).

Information on education was ascertained on the 1995 and 2003 questionnaires. Data on vigorous activity, smoking status, alcohol intake, parity, age at each birth, and menopausal status were obtained at baseline and have been updated on biennial follow-up questionnaires. The question on vigorous physical activity specified basketball, swimming, running, and aerobics as examples. Eight response categories ranged from none to ≥10 h/wk. In a validation study, actigraphs (activity monitors) were worn by 101 BWHS participants during their waking hours for 7 d (19). Significant positive correlations were observed between BWHS questionnaire data and actigraph measurements for vigorous activity (r = 0.40).

Change in body weight

Weight change was calculated for each participant as the difference between each participant's self-reported weight in 2-y intervals from information provided on follow-up questionnaires between 1995 and 2009.

Statistical analysis

The individual FFQ items were aggregated into 35 predefined food groups on the basis of similarity of nutrient content. For each food group, we adjusted for energy by dividing the intake of each food group by total energy intake and multiplying by 1000. Factor analysis (principal components analysis) was performed on the 35 food groups to identify dietary patterns with the use of PROC FACTOR in SAS (SAS Institute, Cary, NC). To determine the number of factors to retain, we considered an eigenvalue criterion of >2, Scree plot analysis (based on the inflection point on a plot of eigenvalues), and the interpretability of each factor (21). An orthogonal (varimax) transformation was used to rotate the correlation matrix, resulting in factors that were uncorrelated and in a simpler data structure with greater interpretability (22). Factor loadings for food groups within a pattern represent correlations between a specific food group and that pattern. Labeling of each pattern was based on food groups with a factor loading ≥0.30. Factor scores for each pattern were calculated by summing intakes of each food group weighted by its factor loading. Each participant was assigned a factor score for each identified pattern, and factor scores for each pattern were categorized into quintiles (quintiles 1 and 5 represented low and high adherence, respectively, to each dietary pattern).

To examine the pattern of BMI according to age in the cohort, we used a 3-knot restricted cubic spline model to assess the possibly nonlinear relation between BMI and age (23). We identified the number and location of knots by using a stepwise selection procedure.

Mixed linear regression models that adjusted for within-person correlation of successive weight changes over each 2-y cycle were used to calculate the multivariable-adjusted means for changes in body weight across categories of dietary pattern scores. Among 41,351 women, there were 174,857 2-y changes in weight over 14 y of follow-up before the age of 55 y. Regression coefficients, representing mean weight change in each 2-y interval, were multiplied by 7 to obtain mean weight change over the 14-y period. Dietary pattern scores at baseline were assessed in relation to weight change from 1995 to 2001, and dietary pattern scores in 2001 were assessed in relation to weight change from 2001 to 2009. Participants were censored from the analysis at age 55 y to isolate the effect of dietary intake on weight change from age-related weight loss.

Multivariable models were adjusted for factors associated with weight gain, which included age (5-y categories), questionnaire cycle, education (≤12, 13–15, ≥16 y), geographic region (Northeast, South, Midwest, West), vigorous activity (none or 0, <1, 1–2, 3–4, 5–6, ≥7 h/wk), smoking status (never; past; current: <15 cigarettes/d, ≥15 cigarettes/d), alcohol intake (<1, 1–6, ≥7 drinks/wk), age at menarche (<12, 12–13, ≥14 y), parity (0, 1, 2, ≥3 births), years since last birth (<5, 5–14, ≥15 y), menopausal status (premenopausal, postmenopausal, uncertain), and the other dietary pattern score (quintiles). Because estimated total energy intake was unrelated to weight gain, and adjustment for energy intake did not affect the results, we did not include energy intake in the models. Covariates that changed over time (eg, vigorous activity and smoking status) were treated as time-dependent variables in the analysis. Tests for linear trend were conducted by using the median of each category modeled as a continuous variable.

We conducted subgroup analyses within strata of time-varying age (<35, 35–44, and ≥45 y) and baseline BMI (<25, 25–29, and ≥30). Tests for interaction were performed by using the likelihood ratio test comparing models with and without cross-product terms between the covariate and the dietary pattern score. All statistical analyses were performed with the use of SAS version 9.1 (SAS Institute).

RESULTS

We identified 2 major dietary patterns among 41,351 women by using factor analysis; similar patterns were observed in 1995 and 2001, with some differences in factor loadings (Table 1). Vegetables, fruit, legumes, fish, and whole grains loaded highly on the first pattern (named the “vegetables/fruit” pattern), whereas red meat, processed meat, French fries, fried chicken, and added fat loaded highly on the second pattern (named the “meat/fried foods” pattern). The vegetables/fruit and meat/fried foods patterns explained 10% and 6%, respectively, of the total variance in the 35 food groups in both 1995 and 2001. The vegetables/fruit and meat/fried foods patterns are similar to the “prudent” and “Western” patterns, respectively, identified in previous studies, which have explained a similar proportion of the variance (24, 25).

TABLE 1.

Dietary pattern factor loadings in the Black Women's Health Study, 1995 and 2001¹

	1995		2001
Food or food groups	Vegetables/fruitpattern	Meat/fried foods pattern	Vegetables/fruitpattern	Meat/fried foods pattern
Noncruciferous vegetables	0.762	0.02	0.742	−0.29
Cruciferous vegetables	0.632	0.09	0.672	−0.16
Fruit	0.552	−0.15	0.532	−0.36
Beans	0.482	0.03	0.432	−0.12
Tomatoes	0.462	0.11	0.422	−0.19
Fish (not fried)	0.442	−0.10	0.482	−0.23
Rice	0.342	−0.01	0.29	−0.01
Whole grains	0.302	−0.35	0.24	−0.39
Orange or grapefruit juice	0.302	−0.02	0.21	−0.18
Potatoes (not fried)	0.29	0.22	0.372	0.23
Poultry (not fried)	0.28	−0.25	0.13	−0.28
Soup	0.28	−0.04	0.28	−0.09
Low-fat mayonnaise, salad dressing	0.24	−0.11	0.322	−0.34
Chili	0.24	0.18	0.27	0.09
Cereals	0.21	−0.12	0.302	−0.14
Pasta	0.21	−0.22	0.18	0.02
Low-fat dairy	0.14	−0.42	0.08	−0.36
Red meat	−0.27	0.532	0.14	0.612
Processed meat	−0.22	0.492	0.15	0.542
Regular mayonnaise, salad dressing	0.14	0.462	0.28	0.15
French fries	−0.17	0.452	0.08	0.502
Fried chicken	−0.19	0.442	0.07	0.502
Butter, margarine	0.13	0.362	0.05	0.25
Eggs	−0.01	0.312	0.25	0.18
Fried fish	0.06	0.29	0.28	0.29
Alcohol	0.04	0.23	0.11	0.12
Organ meats (liver)	0.08	0.20	0.23	0.17
Salty snacks	−0.20	0.15	−0.07	0.29
Refined grains	−0.16	0.13	−0.05	0.372
High-fat dairy	−0.23	0.10	−0.04	0.27
Sugar-sweetened soft drinks	−0.39	0.09	−0.19	0.342
Sweets	−0.38	−0.11	−0.17	0.19
Tea, coffee	0.08	−0.07	0.16	0.07
Nuts	0.02	−0.03	0.07	−0.11
Sugar-sweetened fruit drinks	0.01	−0.02	−0.01	0.05

¹Dietary patterns were derived by using factor analysis.

²Factor loading ≥0.30.

At baseline, women with a high vegetables/fruit pattern score were older, less likely to be obese, more educated, more physically active, and less likely to smoke (Table 2). Women with a high meat/fried foods pattern score were slightly younger, more likely to be obese, less educated, less physically active, and more likely to smoke and drink. The vegetables/fruit pattern was positively associated with dietary intakes of fiber, vitamin A, and β-carotene and inversely associated with intakes of carbohydrates, saturated fat, and cholesterol; the meat/fried foods pattern was positively associated with dietary intakes of saturated fat and cholesterol and inversely associated with intakes of fiber, vitamin A, and β-carotene.

TABLE 2.

Baseline characteristics by quintile (Q) of dietary patterns among 41,351 women in the Black Women's Health Study, 1995¹

	Vegetables/fruit pattern				Meats/fried foods pattern
Characteristic	Q1 (low)	Q3	Q5 (high)	P for trend2	Q1 (low)	Q3	Q5 (high)	P for trend2
n	8296	8269	8249		8261	8278	8273
Age (%)
<35 y	53.2	43.6	33.2	<0.001	42.7	43.2	43.1	0.04
35–44 y	34.4	37.0	38.6	<0.001	35.0	36.9	38.3	<0.001
≥45 y	12.4	19.4	28.2	<0.001	22.3	19.9	18.6	<0.001
BMI (%)
<25 kg/m2	36.8	38.7	46.5	<0.001	44.6	40.1	37.6	<0.001
25–29 kg/m2	29.4	31.6	31.3	<0.001	30.8	30.9	30.4	0.18
≥30 kg/m2	32.9	28.9	21.3	<0.001	23.7	28.1	31.0	<0.001
Region (%)
Northeast	24.2	27.1	31.7	<0.001	31.4	28.0	21.7	<0.001
South	33.9	32.7	27.3	<0.001	27.7	32.1	35.0	<0.001
Midwest	27.1	22.5	18.4	<0.001	20.9	22.6	25.7	<0.001
West	14.7	17.7	22.5	<0.001	20.0	17.3	17.6	<0.001
Education ≥16 y (%)	37.7	47.7	53.9	<0.001	59.3	46.5	36.2	<0.001
Age at menarche ≤11 y (%)	27.4	28.3	28.9	0.004	31.1	27.8	25.9	<0.001
Nulliparous (%)	35.0	38.3	43.8	<0.001	47.2	37.6	34.9	<0.001
Years since last birth <5 (%)3	19.1	21.6	19.8	0.25	22.4	21.8	19.0	<0.001
Vigorous activity ≥5 h/wk (%)	7.4	12.6	23.6	<0.001	20.7	12.4	10.2	<0.001
Current smoker (%)	21.7	15.5	11.8	<0.001	8.8	15.8	23.3	<0.001
Current drinker ≥7 drinks/wk (%)	4.3	5.1	5.8	<0.001	1.6	3.8	12.6	<0.001
Carbohydrates (g/d)	199 ± 984	153 ± 73	134 ± 64	<0.001	177 ± 82	164 ± 82	132 ± 71	<0.001
Saturated fat (g/d)	20.1 ± 10.6	16.8 ± 9.0	12.8 ± 7.2	<0.001	14.2 ± 8.2	16.9 ± 9.3	18.5 ± 9.9	<0.001
Cholesterol (mg/d)	223 ± 135	212 ± 126	176 ± 111	<0.001	161 ± 95	204 ± 118	258 ± 147	<0.001
Fiber (g/d)	7.9 ± 4.4	9.4 ± 5.3	10.6 ± 5.9	<0.001	12.0 ± 6.1	9.3 ± 5.0	7.1 ± 3.9	<0.001
Vitamin A (IU/d)	3828 ± 2663	5258 ± 3529	6279 ± 4042	<0.001	6077 ± 4033	5123 ± 3512	4378 ± 3158	<0.001
β-Carotene (μg/d)	1251 ± 1001	1790 ± 1358	2281 ± 1627	<0.001	2113 ± 1618	1793 ± 1385	1424 ± 1134	<0.001

¹With the exception of age, percentages and means were standardized to the age distribution of the cohort at baseline. Percentages do not total 100% because of missing data.

²Derived from age-adjusted tests for linear trend across quintiles of dietary pattern score.

³Among parous women.

⁴Mean ± SD (all such values).

At baseline, 31% of participants were overweight (BMI of 25–29) and 28% were obese (BMI ≥30). BMI increased most rapidly at younger ages, from a mean BMI of 25.5 at age 21 y to 28.5 at age 35 y (Figure 1). Mean 14-y weight gains were 13.12, 10.35, and 9.07 kg among women aged <35, 35–44, and ≥45 y, respectively. Mean 14-y weight gains according to baseline BMI of <25, 25–29, and ≥30 were 10.58, 12.43, and 9.93 kg, respectively.

FIGURE 1.

Mean BMI (in kg/m²) according to age in 41,351 women in the Black Women's Health Study, 1995–2009. Data were fitted by a restricted cubic spline mixed linear regression model. The 95% CIs are indicated by dashed lines.

Overall, there was no significant association between 14-y weight gain and the vegetables/fruit pattern (Table 3). However, mean weight gain was significantly lower in the highest quintile relative to the lowest quintile of the vegetables/fruit pattern among women aged <35 y (12.06 compared with 13.24 kg; P for trend = 0.004).

TABLE 3.

Mean 14-y weight gain (in kg) according to quintile (Q) of the vegetables/fruit pattern among 41,351 women in the Black Women's Health Study, 1995–2009¹

	Vegetables/fruit pattern
	Q1 (low)	Q2	Q3	Q4	Q5 (high)	P for trend	P for interaction
Overall
Model 1	10.56 ± 0.31	10.38 ± 0.31	10.33 ± 0.31	10.24 ± 0.31	10.57 ± 0.31	0.97
Model 2	10.88 ± 0.39	10.67 ± 0.39	10.64 ± 0.39	10.54 ± 0.40	10.83 ± 0.40	0.82
Age
<35 y
Model 1	13.62 ± 0.59	13.66 ± 0.60	12.63 ± 0.61	12.20 ± 0.63	12.24 ± 0.64	<0.001
Model 2	13.24 ± 0.80	13.33 ± 0.82	12.41 ± 0.83	12.01 ± 0.85	12.06 ± 0.87	0.004
35–44 y
Model 1	10.52 ± 0.45	10.28 ± 0.45	10.19 ± 0.44	10.30 ± 0.45	11.00 ± 0.46	0.19	0.007
Model 2	10.64 ± 0.59	10.35 ± 0.60	10.24 ± 0.60	10.33 ± 0.61	11.96 ± 0.62	0.37
≥45 y
Model 1	7.63 ± 0.49	7.33 ± 0.48	8.25 ± 0.46	8.08 ± 0.45	8.00 ± 0.44	0.13
Model 2	9.01 ± 0.66	8.68 ± 0.65	9.58 ± 0.64	9.38 ± 0.64	9.28 ± 0.64	0.23
Baseline BMI
<25 kg/m2
Model 1	10.81 ± 0.39	10.04 ± 0.38	9.75 ± 0.38	9.56 ± 0.37	10.00 ± 0.37	0.005
Model 2	11.17 ± 0.49	10.52 ± 0.48	10.32 ± 0.49	10.15 ± 0.48	10.62 ± 0.48	0.08
25–29 kg/m2
Model 1	11.78 ± 0.55	11.98 ± 0.55	11.47 ± 0.54	11.49 ± 0.55	11.57 ± 0.56	0.38	0.06
Model 2	12.46 ± 0.69	12.68 ± 0.71	12.20 ± 0.70	12.22 ± 0.71	12.25 ± 0.72	0.39
≥30 kg/m2
Model 1	9.48 ± 0.72	9.60 ± 0.74	10.31 ± 0.74	10.09 ± 0.76	10.26 ± 0.78	0.11
Model 2	9.87 ± 0.91	9.85 ± 0.92	10.48 ± 0.94	10.22 ± 0.96	10.25 ± 0.99	0.39

¹All values are means ± SEs estimated by using mixed linear regression models. Model 1 adjusted for age and questionnaire cycle. Model 2 adjusted for age, questionnaire cycle, education, geographic region, vigorous activity, age at menarche, parity, years since last birth, menopausal status, smoking status, alcohol intake, and the meat/fried foods pattern.

We repeated the analyses among the 12,736 women who maintained approximately the same dietary patterns over time—that is, women who remained within one quintile of each of their 1995 dietary patterns (Table 4). In this subgroup, the vegetables/fruit pattern was associated with significantly less 14-y weight gain overall (highest compared with lowest quintile: 10.88 and 11.94 kg, respectively; P for trend = 0.003). The inverse association was strongest among women aged <35 y (11.40 compared with 14.28 kg; P for trend < 0.001). The vegetables/fruit pattern was also associated with significantly less weight gain among women with normal weight at baseline (10.65 compared with 11.74 kg; P for trend = 0.02) and among women who were overweight at baseline (11.30 compared with 13.03 kg; P for trend = 0.002).

TABLE 4.

Mean 14-y weight gain (in kg) according to quintile (Q) of the vegetables/fruit pattern among 12,736 Black Women's Health Study participants who maintained the same dietary patterns over time, 1995–2009¹

	Vegetables/fruit pattern
	Q1 (low)	Q2	Q3	Q4	Q5 (high)	P for trend	P for interaction
Overall
Model 1	11.80 ± 0.49	11.16 ± 0.48	10.98 ± 0.48	10.36 ± 0.47	10.62 ± 0.49	<0.001
Model 2	11.94 ± 0.63	11.23 ± 0.62	11.15 ± 0.63	10.61 ± 0.63	10.88 ± 0.65	0.003
Age
<35 y
Model 1	14.35 ± 0.93	13.80 ± 0.91	12.72 ± 0.94	10.15 ± 0.97	10.63 ± 1.04	<0.001
Model 2	14.28 ± 1.31	13.91 ± 1.31	13.16 ± 1.31	10.81 ± 1.38	11.40 ± 1.43	<0.001
35–44 y
Model 1	11.13 ± 0.71	10.89 ± 0.70	9.89 ± 0.70	10.46 ± 0.69	10.48 ± 0.73	0.21	<0.001
Model 2	10.72 ± 0.98	10.42 ± 0.98	9.53 ± 0.98	10.18 ± 0.98	10.18 ± 1.02	0.37
≥45 y
Model 1	9.60 ± 0.78	8.10 ± 0.75	9.91 ± 0.71	8.71 ± 0.69	8.59 ± 0.70	0.25
Model 2	10.57 ± 1.03	8.93 ± 1.03	10.73 ± 1.01	9.53 ± 1.01	9.32 ± 1.02	0.16
Baseline BMI
<25 kg/m2
Model 1	11.87 ± 0.61	10.79 ± 0.59	10.36 ± 0.59	9.67 ± 0.57	10.20 ± 0.57	<0.001
Model 2	11.74 ± 0.77	10.82 ± 0.74	10.59 ± 0.76	10.05 ± 0.75	10.65 ± 0.76	0.02
25–29 kg/m2
Model 1	12.55 ± 0.90	12.65 ± 0.88	11.92 ± 0.86	10.87 ± 0.85	10.67 ± 0.90	<0.001	0.20
Model 2	13.03 ± 1.12	13.11 ± 1.16	12.47 ± 1.12	11.51 ± 1.13	11.30 ± 1.17	0.002
≥30 kg/m2
Model 1	11.47 ± 1.15	10.60 ± 1.14	11.18 ± 1.16	10.84 ± 1.21	11.20 ± 1.34	0.81
Model 2	12.29 ± 1.50	11.01 ± 1.47	11.65 ± 1.52	11.35 ± 1.58	11.64 ± 1.69	0.57

¹Values are means ± SEs estimated by using mixed linear regression models. Model 1 adjusted for age and questionnaire cycle. Model 2 adjusted for age, questionnaire cycle, education, geographic region, vigorous activity, age at menarche, parity, years since last birth, menopausal status, smoking status, alcohol intake, and the meat/fried foods pattern.

The meat/fried foods pattern was not significantly associated with 14-y weight gain overall (Table 5). However, mean weight gain was significantly greater in the highest quintile relative to the lowest quintile of the meat/fried foods pattern among women with normal weight at baseline (10.77 compared with 10.09 kg; P for trend = 0.002).

TABLE 5.

Mean 14-y weight gain (in kg) according to quintile (Q) of the meat/fried foods pattern among 41,351 women in the Black Women's Health Study, 1995–2009¹

	Meat/fried foods pattern
	Q1 (low)	Q2	Q3	Q4	Q5 (high)	P for trend	P for interaction
Overall
Model 1	10.30 ± 0.31	10.30 ± 0.31	10.46 ± 0.31	10.47 ± 0.31	10.54 ± 0.31	0.20
Model 2	10.42 ± 0.39	10.46 ± 0.39	10.64 ± 0.39	10.67 ± 0.39	10.78 ± 0.39	0.08
Age
<35 y
Model 1	12.46 ± 0.60	12.79 ± 0.61	12.74 ± 0.60	13.17 ± 0.61	13.71 ± 0.63	0.01
Model 2	12.45 ± 0.83	12.60 ± 0.84	12.41 ± 0.83	12.82 ± 0.83	13.31 ± 0.83	0.11
35–44 y
Model 1	10.26 ± 0.45	10.49 ± 0.45	10.54 ± 0.45	10.45 ± 0.45	10.50 ± 0.45	0.59	0.77
Model 2	9.76 ± 0.60	10.09 ± 0.60	10.24 ± 0.60	10.17 ± 0.60	10.29 ± 0.60	0.19
≥45 y
Model 1	8.01 ± 0.45	7.78 ± 0.46	8.04 ± 0.46	7.83 ± 0.47	7.87 ± 0.47	0.76
Model 2	9.43 ± 0.64	9.28 ± 0.64	9.58 ± 0.64	9.41 ± 0.65	9.51 ± 0.66	0.75
Baseline BMI
<25 kg/m2
Model 1	9.58 ± 0.37	9.49 ± 0.37	10.03 ± 0.38	10.39 ± 0.39	10.61 ± 0.38	<0.001
Model 2	10.09 ± 0.48	9.89 ± 0.48	10.32 ± 0.49	10.60 ± 0.49	10.77 ± 0.49	0.002
25–29 kg/m2
Model 1	11.85 ± 0.56	11.82 ± 0.55	11.62 ± 0.55	11.58 ± 0.55	11.37 ± 0.55	0.18	0.06
Model 2	12.38 ± 0.70	12.39 ± 0.70	12.20 ± 0.70	12.19 ± 0.70	12.07 ± 0.70	0.38
≥30 kg/m2
Model 1	9.79 ± 0.77	10.01 ± 0.75	10.04 ± 0.74	9.80 ± 0.74	9.92 ± 0.74	0.99
Model 2	9.84 ± 0.96	10.31 ± 0.95	10.48 ± 0.94	10.31 ± 0.94	10.57 ± 0.93	0.28

¹Values are means ± SEs estimated by using mixed linear regression models. Model 1 adjusted for age and questionnaire cycle. Model 2 adjusted for age, questionnaire cycle, education, geographic region, vigorous activity, age at menarche, parity, years since last birth, menopausal status, smoking status, alcohol intake, and the vegetables/fruit pattern.

Among women who maintained the same dietary patterns over time, the meat/fried foods pattern was associated with significantly greater 14-y weight gain overall (highest compared with lowest quintile: 12.02 and 10.15 kg, respectively; P for trend < 0.001) (Table 6). Significant positive associations were evident in each age category, with the strongest association among women aged <35 y (14.79 compared with 11.42 kg; P for trend = 0.001). The meat/fried foods pattern was also associated with significantly greater weight gain among women with normal weight at baseline (12.00 compared with 9.96 kg; P for trend < 0.001) and among women who were obese at baseline (12.86 compared with 9.73 kg; P for trend = 0.004).

TABLE 6.

Mean 14-y weight gain (in kg) according to quintile (Q) of the meat/fried foods pattern among 12,736 Black Women's Health Study participants who maintained the same dietary patterns over time, 1995–2009¹

	Meat/fried foods pattern
	Q1 (low)	Q2	Q3	Q4	Q5 (high)	P for trend	P for interaction
Overall
Model 1	9.94 ± 0.48	10.59 ± 0.48	11.12 ± 0.48	11.23 ± 0.48	11.86 ± 0.50	<0.001
Model 2	10.15 ± 0.63	10.69 ± 0.63	11.15 ± 0.63	11.25 ± 0.62	12.02 ± 0.63	<0.001
Age
<35 y
Model 1	9.96 ± 0.93	12.82 ± 0.96	12.69 ± 0.93	13.77 ± 0.93	14.35 ± 1.01	<0.001
Model 2	11.42 ± 1.34	13.65 ± 1.35	13.16 ± 1.31	14.08 ± 1.32	14.79 ± 1.35	0.001
35–44 y
Model 1	9.80 ± 0.71	10.19 ± 0.70	10.58 ± 0.71	10.71 ± 0.70	11.41 ± 0.73	0.006	0.06
Model 2	8.78 ± 0.97	9.13 ± 0.96	9.53 ± 0.98	9.63 ± 0.96	10.45 ± 0.99	0.009
≥45 y
Model 1	8.56 ± 0.71	8.51 ± 0.69	9.36 ± 0.71	8.88 ± 0.74	9.59 ± 0.75	0.06
Model 2	9.95 ± 1.01	9.90 ± 1.00	10.73 ± 1.01	10.32 ± 1.03	11.22 ± 1.05	0.04
Baseline BMI
<25 kg/m2
Model 1	9.20 ± 0.57	10.20 ± 0.57	10.21 ± 0.58	11.08 ± 0.59	11.85 ± 0.61	<0.001
Model 2	9.96 ± 0.76	10.73 ± 0.75	10.59 ± 0.76	11.24 ± 0.76	12.00 ± 0.78	<0.001
25–29 kg/m2
Model 1	10.88 ± 0.88	11.23 ± 0.86	12.19 ± 0.87	12.26 ± 0.85	11.81 ± 0.90	0.05	0.36
Model 2	11.52 ± 1.13	11.67 ± 1.13	12.47 ± 1.12	12.50 ± 1.10	12.35 ± 1.14	0.13
≥30 kg/m2
Model 1	9.84 ± 1.28	10.59 ± 1.18	11.44 ± 1.16	10.69 ± 1.15	12.23 ± 1.18	0.03
Model 2	9.73 ± 1.60	10.69 ± 1.53	11.65 ± 1.52	11.10 ± 1.49	12.86 ± 1.50	0.004

¹Values are means ± SEs estimated by using mixed linear regression models. Model 1 adjusted for age and questionnaire cycle. Model 2 adjusted for age, questionnaire cycle, education, geographic region, vigorous activity, age at menarche, parity, years since last birth, menopausal status, smoking status, alcohol intake, and the vegetables/fruit pattern.

DISCUSSION

Our results from a large prospective study are the first evidence to suggest that a healthier dietary pattern may be instrumental in reducing weight gain in African American women. We identified 2 major dietary patterns through factor analysis: a vegetables/fruit pattern and a meat/fried foods pattern. Overall, neither pattern was significantly associated with 14-y weight gain. However, associations were stronger among women who maintained similar dietary patterns over time: the vegetables/fruit pattern was associated with significantly less 14-y weight gain overall, whereas the meat/fried foods pattern was associated with significantly greater weight gain. The associations were most pronounced among women aged <35 y, who experienced the greatest amount of weight gain during 14 y of follow-up. The differences in weight gain associated with dietary patterns represented 10–20% of total weight gain.

African Americans have poorer diet quality (eg, consume fewer vegetables and have more energy-dense diets) than do whites (12, 13), but few prospective studies of the association between diet quality and weight gain have been conducted in African Americans. To our knowledge, this is the first study to observe that high diet quality is inversely associated with prospective weight gain in African Americans. The MESA, which included 1620 African Americans, found that higher Healthy Eating Index scores were associated with lower BMI after 18 mo of follow-up among white participants, whereas there was no association among African American participants (14). The CARDIA study, which included 2486 African Americans, reported that a higher Diet Quality Index score was associated with lower 20-y weight gain only among white participants, whereas African American adults with high diet quality actually gained significantly more weight than did those with low diet quality (19.4 and 17.8 kg, respectively) (15).

The results of our study are broadly consistent with the few large prospective studies largely composed of whites that have assessed dietary patterns in relation to weight gain over long-term follow-up. In the Nurses’ Health Study II, increases in a Western dietary pattern score over 8 y were associated with the greatest amount of weight gain, whereas increases in a prudent dietary pattern score were associated with the least amount of weight gain (7). In the European Prospective Investigation into Cancer and Nutrition (EPIC)–Potsdam cohort, a dietary pattern high in whole grains, fruit, and vegetables and low in meat and high-fat dairy was associated with significantly less 4-y weight gain (8); the inverse association was evident only among participants aged <50 y and among those who were not obese at baseline. An analysis from a larger number of EPIC cohorts reported that greater adherence to a Mediterranean dietary pattern, characterized by high consumption of olive oil, legumes, fruit, and vegetables and low consumption of meat, was associated with significantly less 5-y weight gain; the results were stronger among adults aged <40 y and among participants who were not obese at baseline (9). The Swedish Mammography Cohort found that increases in the Healthy pattern score over 9 y were inversely associated with changes in BMI and weight, but the strongest association was among obese women (10). In the present study, we observed stronger associations between each dietary pattern and weight gain among women aged <35 y and among women who were of normal weight at baseline. Furthermore, in the subgroup of women who maintained the same dietary patterns over time, an inverse association for the vegetables/fruit pattern was present among normal-weight and overweight women, whereas the positive association with the meat/fried foods pattern was present among normal-weight and obese women.

A vegetables/fruit dietary pattern is consistent with the Dietary Guidelines for Americans, published by the US Department of Agriculture every 5 y, which recommends consumption of a wide variety of vegetables and fruit, more whole grains than refined grains, and lean proteins such as poultry and fish (26). A meat/fried foods dietary pattern is more energy-dense than is a vegetables/fruit pattern (27). Laboratory studies indicate that individuals tend to consume the same weight of food rather than a consistent amount of energy (25, 26). Foods that characterize a Western dietary pattern, such as processed foods high in fat and sugar, are more palatable and less satiating (28), and thus can more easily lead to overconsumption of calories. Sustained positive energy balance will result in weight gain over time; this could explain our findings that weight gain was greatest among women who consumed a meat/fried foods pattern in both 1995 and 2001. Because total energy intake is largely determined by body size, metabolic efficiency, and physical activity, it is difficult to assess energy balance directly (29). Moreover, FFQs >contain questions about a limited number of foods and do not estimate absolute energy intake with precision (29). It is therefore not surprising that estimated total energy intake was unrelated to weight gain in our study, whereas it is plausible that a dietary pattern high in energy density would be associated with long-term positive energy balance and subsequent weight gain.

Strengths of our study include its prospective design, length of follow-up, and focus on African American women. The study is larger than most previous prospective studies (8, 10, 14, 15) and included >15 times the number of African Americans as did each of the 2 previous studies that reported on this population group (14, 15). We used measures of diet at 2 time points, which allowed us to update dietary changes midway through follow-up. We were also able to use repeated reports of body weight every 2 y, and we updated data on covariates, including vigorous physical activity, smoking, and parity for each questionnaire cycle. Limitations of the study were the use of FFQs to estimate dietary intake and the use of self-reported measures of body size. Whereas a validation study found satisfactory correlation for nutrient intakes with the use of 24-h recalls and 3-d diet records (18), misclassification of dietary pattern on weight gain would have tended to result in underestimation of the effects on weight gain (30). Although data from a validation study indicated high correlation between self-reported and measured anthropometric variables (19), weight tends to be underreported, particularly among obese women (31). It is possible that weight gain was underestimated in our study, especially at higher BMIs. We were able to control for important potential confounders, but we cannot rule out residual confounding by unmeasured lifestyle factors.

In recent decades, the greatest increases in BMI among US adults have occurred in African American women (1). Among both black and white women, rates of weight gain have been reported to be highest in young adulthood (1, 32, 33), which is consistent with our finding that the greatest amount of weight gain occurred among women aged <35 y. The present findings suggest that African American women may be better able to achieve long-term weight maintenance by consuming a diet rich in vegetables and fruit and low in red and processed meats and fried foods. Because the association between a healthier diet and weight gain appears to be greatest at younger ages and among those with lower BMI, promotion of the substitution of fruit and vegetables for meats and fried foods could have important implications for the prevention of obesity as young African American women grow older.