Consumption of restaurant foods and incidence of type 2 diabetes in African American women

Abstract

Background: Type 2 diabetes is a major problem in Western nations. Profound secular changes in the food environment and eating habits may play a role. In particular, consumption of foods prepared outside the home has greatly increased.

Objective: We investigated the relation of restaurant meal consumption to incidence of type 2 diabetes among African American women with the use of data from the prospective Black Women's Health Study.

Design: The participants have completed mailed follow-up questionnaires every 2 y since 1995, including food-frequency questionnaires that asked about the frequency of eating restaurant meals of various types. Cox proportional hazards models were used to calculate incidence rate ratios and 95% CIs for the association of type 2 diabetes incidence with various categories of consumption of each restaurant food relative to the lowest category, with adjustment for diabetes risk factors.

Results: Among 44,072 participants aged 30–69 y and free of diabetes at baseline, 2873 incident cases of type 2 diabetes occurred during 10 y of follow-up. Consumption of restaurant meals of hamburgers, fried chicken, fried fish, and Chinese food were independently associated with an increased risk of type 2 diabetes. Incidence rate ratios for ≥2 such meals per week relative to none were 1.40 (95% CI: 1.14, 1.73) for hamburgers and 1.68 (95% CI: 1.36, 2.08) for fried chicken. Control for body mass index greatly reduced the estimates, which suggests that the associations are mediated through weight gain and obesity.

Conclusion: The present study has identified a risk factor for type 2 diabetes that may be readily modifiable by dietary changes.

INTRODUCTION

The rapidly growing epidemic of type 2 diabetes in the United States is most notable in African American women. The prevalence of type 2 diabetes among African American women is twice that among white women and 1.4 times that among African American men (1). Obesity is a powerful determinant of diabetes risk but is not easily modified (2, 3). In the search for risk factors that might be more amenable to change, we and others have identified several dietary factors that increase the risk of diabetes, such as a high intake of sugar-sweetened sodas and fruit drinks (4–6), a low intake of cereal fiber (7–12), and a high intake of foods with a high glycemic load (7, 10–12).

The consumption of restaurant foods, a relatively unexplored factor, is a potential contributor to the increase in diabetes occurrence among African American women. Profound changes in the food environment and in individual nutrition have taken place in the United States in the past few decades. Between 1977–1978 and 1994–1996, consumption of foods prepared away from home increased from 18% to 32% of total calories (13). In 2006, 42% of the expenditure on food in the United States was for food consumed away from home (14). About half of restaurant meals are now from fast-food restaurants (15, 16). In urban areas, the ratio of fast-food to full-service restaurants is greatest in areas that are poorer or have a higher proportion of black residents (17–20). Foods eaten at fast-food restaurants have a higher energy density than do foods eaten at home (21), portion sizes tend to be larger at restaurants (16), and frequent consumption of fast food has been associated with an increased risk of obesity (22–26). We used data from a prospective cohort study of 59,000 African American women to examine the relation of consumption of restaurant foods, including the types of foods most often eaten at fast-food restaurants, to incidence of type 2 diabetes in African American women.

SUBJECTS AND METHODS

Study population

The Black Women's Health Study (BWHS) is an ongoing prospective follow-up study of African American women from across the United States (27). The study began in 1995 when women aged 21–69 y were enrolled through postal questionnaires mailed to subscribers of Essence magazine, members of several professional organizations, and friends and relatives of early respondents. The baseline questionnaire collected information on demographics, medical and reproductive history, height, weight, physical activity, usual diet, and cigarette and alcohol use, among other factors. The procedures followed were in accordance with the ethical standards of the Institutional Review Board of Boston University.

After the exclusion of women whose addresses were judged to be invalid, 59,000 women made up the cohort that has been followed. Biennial follow-up questionnaires collect updated information on incident disease, weight, smoking, physical activity, and other factors. Follow-up has averaged >80% of the baseline cohort over 5 completed questionnaire cycles.

The present analyses are based on follow-up from 1995 through 2005, with follow-up beginning at age 30 y to exclude possible cases of type 1 diabetes. We excluded women who at baseline reported a history of diabetes (n = 2934), gestational diabetes (n = 635), stroke (n = 360), myocardial infarction (315), or cancer (n = 1143), who were pregnant at baseline (n = 958), had missing data on weight or height at baseline (n = 884), who left >than 10 dietary questions blank (n = 2736), who had implausible energy intake values (<500 or >3800 kcal; n = 3031) as estimated by dietary questionnaires, or who did not answer the questions on restaurant meals (n = 544). The final analytic cohort consisted of 44,072 women.

Case definition

Each follow-up questionnaire asked about new diagnoses of diabetes during the previous 2-y period. We assessed the accuracy of self-reported diabetes among a sample of 227 participants whose physicians provided data from their medical records. The diagnosis of type 2 diabetes was confirmed for 218 (96%) of the women. Of the remaining 9 women, 3 did not have diabetes, 2 had type 1 diabetes, 2 had gestational diabetes, 1 had steroid-induced diabetes, and 1 was classified as having metabolic syndrome.

Assessment of dietary factors

Diet was assessed at baseline in 1995 and 6 y later in 2001 by the short-form National Cancer Institute (NCI)–Block food-frequency questionnaire (FFQ) (28). We added to the FFQ several food items commonly eaten by African American women. Participants were asked to mark how often they had consumed each food in the past year and the portion size. In 1995, respondents could chose medium size (defined on the questionnaire for each item), one-half that size (small), or 1.5 times that size (large). In 2001, a “super” portion size was added that was twice the designated medium size. Nutrient estimates from the 1995 FFQ were calculated by using version 3.7 of NCI DIETSYS software (29), and estimates from the 2001 FFQ were calculated with NCI DIET*A software (DHQ-Diet*Calc Analysis Software, version 1.4.3; National Cancer Institute, Bethesda, MD).

The FFQ was validated in a sample of 408 BWHS participants who completed both a 3-d food diary and up to three 24-h dietary recalls (30). Comparisons of the FFQ data with the diaries and recalls indicated satisfactory agreement, of about the same magnitude as in studies of other populations (31–33) for fat, protein, carbohydrate, dietary fiber, calcium, iron, vitamin C, folate, and β-carotene: the correlation coefficients (energy adjusted and deattenuated) ranged from 0.4 to 0.8.

In both 1995 and 2001, participants were asked how often they had eaten breakfast, lunch, dinner, or takeout of the following specific types of foods at restaurants: burgers, fried chicken, pizza, Chinese food, Mexican food, fried fish (with separate responses for each type). Response options ranged from “never in past year” to “about everyday.” Among the 462 women who inadvertently completed duplicate questionnaires, the reproducibility of responses was good to excellent, with κ statistics ranging from 0.63 to 0.74.

Assessment of nondietary factors

Information on education, height, and family history of diabetes was obtained at baseline in 1995. Data on current weight, vigorous physical activity, television watching, smoking status, and alcohol intake were obtained at baseline and have been updated on biennial follow-up questionnaires. With regard to physical activity, participants were asked how many hours per week in the past year they spent in vigorous activity such as running, swimming, basketball, and aerobics. In a validation study within the BWHS, questionnaire reports of vigorous physical activity were significantly associated with physical activity as measured by actigraphs and daily activity diaries (34). Body mass index (BMI) was calculated as weight in kilograms divided by squared height in meters.

Statistical analysis

Person-years were calculated from baseline in 1995 to the year of diagnosis of type 2 diabetes, loss to follow-up, death, or end of follow-up (May 2005), whichever came first. Time-varying covariates were reassigned every 2 y by using the Anderson-Gill data structure (35). The Anderson-Gill data structure creates a new record for every follow-up cycle at which the participant is at risk and assigns covariate values reported for that specific questionnaire cycle. Cox proportional hazards models were used to calculate incidence rate ratios (IRRs, also known as hazard ratios) and 95% CIs for the relation of restaurant food consumption to type 2 diabetes risk. IRRs were calculated for the highest category of restaurant food consumption relative to the lowest category. FFQ data from 1995 were used for follow-up through 2001, and 2001 FFQ data were used for follow-up from 2001 through 2005. Three multivariable models were constructed. Model 1 included terms for age, time period, family history of diabetes, years of education, television watching, physical activity, and cigarette smoking. Model 2 further controlled for dietary factors, including coffee consumption, cereal fiber intake, intake of sugar-sweetened soda, glycemic index of the diet, alcohol consumption, calcium intake, vitamin D, and total energy intake. Other dietary factors such as consumption of French fries and consumption of fruit and vegetables, which have not been consistently associated with type 2 diabetes but might be related to consumption of restaurant foods, were considered as potential confounders. The IRRs were unchanged with the addition of these variables and thus they were not included in model 2. Model 3 added a term for BMI, which may be an intermediate in the causal pathway between eating restaurant foods and risk of diabetes. All statistical analyses were conducted by using SAS 9.1 (SAS Institute Inc, Cary, NC).

RESULTS

Among the various types of restaurant foods, burgers were eaten most often (at least once per week in 1995 by 18% of participants) followed by fried chicken (13% at least once per week) (Table 1). In 2001, burgers remained the most commonly eaten type of restaurant food, whereas consumption of food from Mexican restaurants showed the greatest increase (from 4% to 9% for at least once per week) and pizza the greatest decrease (from 10% to 4%). At baseline, women in the highest category of each type of restaurant food consumption were younger and consumed a diet characterized by higher intakes of total energy, fat, protein, soda, and alcohol and a lower glycemic index than women in the lowest categories of consumption of those restaurant foods. For other factors, there were differing associations depending on the type of restaurant food. Consumption of burgers, fried chicken, and fried fish were consistently associated with poorer health behaviors (more smoking and less physical activity) and with a lower level of educational attainment, whereas some of the other types of food were associated with less smoking, more physical activity, and higher levels of educational attainment.

TABLE 1.

Baseline characteristics by frequency of restaurant food intake: Black Women's Health Study

	Burgers		Fried chicken		Fried fish		Chinese food		Pizza		Mexican food
	Never	≥1 time/wk	Never	≥1 time/wk	Never	≥1 time/wk	Never	≥1 time/wk	Never	≥1 time/wk	Never	≥1 time/wk
Subjects
n	5453	7790	3949	5811	9026	3277	3178	5076	2556	4350	15,428	1600
(%)	12	18	9	13	20	7	7	12	6	10	35	4
Age (y)	41.4 ± 11.31	34.4 ± 8.32	40.3 ± 11.1	36.7 ± 9.22	38.1 ± 10.7	38.3 ± 9.53	41.1 ± 11.7	36.4 ± 8.92	46.5 ± 11.1	34.6 ± 8.02	40.1 ± 10.7	35.1 ± 8.72
Education (y)	15.3 ± 1.7	14.8 ± 1.72	15.3 ± 1.8	14.6 ± 1.82	15.0 ± 1.8	14.5 ± 1.82	14.5 ± 1.9	14.9 ± 1.72	14.7 ± 1.9	14.9 ± 1.72	14.6 ± 1.9	15.0 ± 1.72
Energy (kcal/d)	1426 ± 624	1948 ± 7302	1442 ± 617	1958 ± 7432	1495 ± 654	1988 ± 7612	1524 ± 692	1921 ± 7332	1419 ± 646	1992 ± 7332	1546 ± 694	2023 ± 732
Fat (g/d)	45.2 ± 24.8	72.4 ± 27.92	44.9 ± 24.4	73.2 ± 27.82	50.0 ± 26.2	72.6 ± 28.22	52.5 ± 27.6	69.3 ± 28.22	47.1 ± 26.3	72.2 ± 28.12	54.1 ± 27.8	72.5 ± 28.32
Protein (g/d)	54.2 ± 24.5	69.2 ± 25.82	54.6 ± 24.5	70.4 ± 26.12	56.1 ± 25.0	71.1 ± 26.52	54.2 ± 25.2	72.6 ± 25.92	53.9 ± 25.3	71.7 ± 25.82	56.5 ± 25.6	74.4 ± 25.72
Cereal fiber (g/d)	4.4 ± 3.2	3.9 ± 2.72	4.7 ± 3.4	3.7 ± 2.82	4.1 ± 3.0	4.0 ± 2.92	4.1 ± 3.1	3.9 ± 2.83	4.1 ± 3.2	4.1 ± 2.93	3.8 ± 2.9	4.0 ± 2.92
Glycemic index	50.9 ± 8.1	47.2 ±7.32	50.6 ± 8.0	47.8 ± 7.22	50.7 ± 7.9	47.7 ± 7.72	51.2 ± 8.5	47.7 ± 7.22	51.9 ± 8.9	46.7 ± 7.02	51.1 ± 8.0	44.9 ± 7.52
Family history of diabetes (%)	26.3	27.8	26.3	29.22	26.4	30.02	28.2	28.0	30.0	27.02	29.0	26.32
Vigorous activity, ≥5 h/wk (%)	20.6	10.22	22.0	10.82	16.8	11.92	13.3	13.5	14.6	12.72	12.4	14.62
Cigarettes, ≥15/d (%)	3.6	6.02	4.2	6.32	4.7	7.62	5.8	5.6	6.7	5.32	6.4	5.32
Soda, ≥1 can/d (%)	7.0	29.72	7.9	29.42	13.5	26.62	18.1	23.72	11.7	27.72	18.3	24.13
Coffee, ≥2 cups/d (%)4	11.3	11.12	12.5	11.33	12.1	12.2	10.9	12.4	13.2	10.92	11.3	14.42
Alcohol, ≥7 drinks/wk (%)	5.0	6.1	5.1	7.92	4.9	8.82	4.5	7.42	6.2	7.0	5.2	7.72

¹Mean ± SD (all such values).

²P ≤ 0.001 (derived by using trend tests across ordinal categories of intake of each type of food: never, 1–4/y, 5–11/y, 1–3/mo, 1/wk, ≥2/wk).

³P < 0.05 and >0.001 (derived by using trend tests across ordinal categories of intake of each type of food: never, 1–4/y, 5–11/y, 1–3/mo, 1/wk, ≥2/wk).

⁴1 cup = 237 mL.

We examined the relation of BMI to consumption of restaurant foods in greater detail because BMI is such a strong risk factor for type 2 diabetes. The mean BMIs at baseline for low frequency of consumption in the past year (not at all) and high frequency of consumption in the past year (at least once per week) of each type of restaurant food are shown in Table 2. Mean BMIs were markedly higher for women in the high-frequency than in the low-frequency categories for burgers (difference in means: 2.2), fried chicken (difference in means: 2.9), and fried fish (difference in means: 1.9). The difference in mean BMI by frequency of consumption was smaller (0.8) for Chinese food but was still statistically significant. For pizza and Mexican food, however, mean BMIs were essentially the same for women in the low- and high-consumption categories.

TABLE 2.

Mean BMI (in kg/m²) according to frequency of consumption of types of restaurant food at baseline in 1995

	Never		≥1 time/wk
Type of restaurant food	n	BMI	n	BMI	Difference	P value1
Burgers	5453	26.1 ± 5.42	7790	8.3 ± 7.2	2.2	<0.0001
Fried chicken	3949	25.9 ± 5.4	5811	28.8 ± 7.3	2.9	<0.0001
Fried fish	9026	26.9 ± 6.2	3277	28.8 ± 6.9	1.9	<0.0001
Chinese food	3178	27.6 ± 6.5	5076	28.4 ± 7.0	0.8	<0.0001
Pizza	2556	27.8 ± 6.2	4350	27.6 ± 6.8	0.2	0.28
Mexican food	15,428	28.1 ± 6.6	1600	27.7 ± 6.8	0.4	0.07

¹Derived by using a t test for differences in means.

²Mean ± SD (all such values).

During 10 y of follow-up, we identified 2873 incident cases of type 2 diabetes. A higher frequency of eating restaurant burgers was associated with an increased risk of type 2 diabetes (Table 3). After adjustment for age, family history of diabetes, television watching, vigorous activity, and education (model 1), the IRR for consuming burgers ≥2 times/wk week relative to never in the past year was 1.58 (95% CI: 1.29, 1.94). The association was reduced (IRR: 1.40; 95% CI: 1.14, 1.73) after adjustment for other dietary factors and energy intake (model 2). Further adjustment for BMI markedly reduced the IRR to 1.15 (95% CI: 0.93, 1.42). The IRRs for eating fried chicken from restaurants ≥2 times/wk in the past year relative to not at all were 1.84 (95% CI: 1.50, 2.27) in model 1, 1.68 (95% CI: 1.36, 2.08) after adjustment for dietary factors and energy intake, and 1.27 (95% CI: 1.02, 1.57) after control for BMI . Although there was some correlation between consumption of burgers and fried chicken, the associations of each of these types of restaurant meal with incidence of diabetes persisted after control for the other.

TABLE 3.

Incidence rate ratios (IRRs) and 95% CIs for the association of restaurant food intake with the risk of type 2 diabetes¹

		IRR (95% CI)
Type of food and frequency of intake	Diabetes cases	Model 12	Model 23	Model 34
Burgers
Never	320	Reference	Reference	Reference
1–4 times/y	752	1.31 (1.15, 1.49)	1.33 (1.17, 1.52)	1.23 (1.07, 1.40)
5–11 times/y	587	1.40 (1.22, 1.61)	1.41 (1.22, 1.62)	1.27 (1.10, 1.46)
1–3 times/mo	656	1.49 (1.30, 1.70)	1.45 (1.26, 1.67)	1.26 (1.10, 1.46)
1 time/wk	321	1.60 (1.36, 1.87)	1.50 (1.28, 1.77)	1.28 (1.09, 1.52)
≥2 times/wk	145	1.58 (1.29, 1.94)	1.40 (1.14, 1.73)	1.15 (0.93, 1.42)
P for trend5		<0.0001	<0.0001	0.05
Fried chicken
Never	202	Reference	Reference	Reference
1–4 times/y	807	1.22 (1.05, 1.43)	1.23 (1.05, 1.45)	1.11 (0.95, 1.30)
5–11 times/y	724	1.35 (1.16, 1.58)	1.33 (1.14, 1.56)	1.14 (0.97, 1.34)
1–3 times/mo	631	1.27 (1.09, 1.50)	1.23 (1.04, 1.45)	1.04 (0.88, 1.22)
1 time/wk	309	1.53 (1.28, 1.83)	1.43 (1.19, 1.72)	1.16 (0.97, 1.40)
≥2 times/wk	165	1.84 (1.50, 2.27)	1.68 (1.36, 2.08)	1.27 (1.02, 1.57)
P for trend5		<0.0001	<0.0001	0.23
Fried fish
Never	482	Reference	Reference	Reference
1–4 times/y	879	1.05 (0.94, 1.17)	1.05 (0.94, 1.17)	1.01 (0.90, 1.13)
5–11 times/y	558	1.14 (1.01, 1.29)	1.11 (0.98, 1.26)	1.05 (0.92, 1.18)
1–3 times/mo	574	1.27 (1.13, 1.44)	1.21 (1.07, 1.37)	1.15 (1.01, 1.30)
1 time/wk	244	1.26 (1.08, 1.48)	1.17 (1.00, 1.37)	1.07 (0.91, 1.25)
≥2 times/wk	40	1.10 (0.79, 1.52)	0.97 (0.70, 1.35)	0.89 (0.64, 1.24)
P for trend5		<0.0001	0.005	0.10
Chinese
Never	252	Reference	Reference	Reference
1–4 times/y	742	0.94 (0.81, 1.08)	0.96 (0.83, 1.10)	0.93 (0.81, 1.08)
5–11 times/y	703	0.96 (0.83, 1.11)	0.97 (0.84, 1.13)	0.94 (0.82, 1.09)
1–3 times/mo	743	1.07 (0.92, 1.24)	1.06 (0.91, 1.22)	1.00 (0.86, 1.15)
1 time/wk	271	1.18 (0.99, 1.41)	1.14 (0.95, 1.36)	1.02 (0.86, 1.22)
≥2 times/wk	92	1.37 (1.07, 1.74)	1.29 (1.01, 1.65)	1.19 (0.94, 1.52)
P for trend5		<0.0001	0.005	0.12
Pizza
Never	245	Reference	Reference	Reference
1–4 times/y	1037	1.11 (0.97, 1.28)	1.12 (0.97, 1.29)	1.08 (0.94, 1.24)
5–11 times/y	632	1.00 (0.86, 1.16)	0.98 (0.84, 1.14)	0.97 (0.83, 1.13)
1–3 times/mo	643	1.03 (0.89, 1.21)	0.98 (0.84, 1.14)	0.95 (0.81, 1.11)
1 time/wk	202	1.02 (0.84, 1.24)	0.93 (0.76, 1.13)	0.92 (0.76, 1.12)
≥2 times/wk	29	0.90 (0.61, 1.32)	0.76 (0.51, 1.12)	0.76 (0.51, 1.13)
P for trend5		0.29	0.01	0.01
Mexican
Never	1191	Reference	Reference	Reference
1–4 times/y	829	0.94 (0.86, 1.03)	0.96 (0.88, 1.05)	0.98 (0.89, 1.07)
5–11 times/y	334	0.93 (0.82, 1.05)	0.95 (0.84, 1.08)	0.96 (0.85, 1.08)
1–3 times/mo	284	0.93 (0.82, 1.06)	0.94 (0.82, 1.07)	0.96 (0.84, 1.09)
1 time/wk	89	1.09 (0.88, 1.36)	1.06 (0.85, 1.32)	1.05 (0.84, 1.31)
≥2 times/wk	22	1.06 (0.69, 1.61)	0.98 (0.64, 1.51)	1.01 (0.66, 1.55)
P for trend5		0.65	0.55	0.68

¹IRRs were derived by using proportional hazards models.

²Adjusted for age, time period, education, family history of diabetes, television watching, vigorous activity, and smoking.

³Adjusted for variables in model 1 plus smoking, coffee consumption, glycemic index of diet, and intake of cereal fiber, alcohol, sugar-sweetened soda, calcium, vitamin D, and energy.

⁴Adjusted for variables in model 2 plus BMI.

⁵Derived by using Wald chi-square tests.

Frequent consumption of restaurant meals of Chinese food and fried fish was associated with an increased incidence of diabetes in models 1 and 2. However, there was no significant association with either Chinese food or fried fish after control for BMI. Pizza and Mexican food intake were not associated with an increased incidence of type 2 diabetes. For pizza, there was a statistically significant inverse trend with consumption but none of the individual estimates was significant.

Burgers, fried chicken, and fried fish are cooked and eaten at home as well as in restaurants. In an effort to determine whether the observed associations with burgers, fried chicken, and fried fish in restaurant meals were due to restaurant factors (eg, mode of preparation and accompanying foods) or the food itself, we carried out a subanalysis restricted to women who reported eating these foods from restaurants infrequently. For these analyses, the consumption of burgers, fried chicken, and fried fish reported by participants on the FFQs would largely have reflected meals prepared at home. Among women who ate restaurant meals of burgers less than once per week, increasing consumption of burgers was associated with an increased risk of type 2 diabetes (P for trend <0.0001). In contrast, among women who ate restaurant meals of fried chicken less than once per week or of fried fish less than once per week, there was no association of diabetes risk and frequency of consumption of fried chicken or fried fish as assessed in the FFQ (P for trend = 0.30 for fried chicken; P for trend = 0.87 for fried fish). We did not assess home consumption of Chinese and Mexican foods because such foods are types of restaurant food and were not included as items in the regular FFQ. In addition, pizza consumption in the United States is almost entirely from pizzas prepared in restaurants. Analyses were repeated within age strata (<45 and ≥45 y) and results were generally consistent in the 2 age groups (data not shown).

DISCUSSION

In this large prospective study of African American women, we found positive associations between frequent consumption of certain types of restaurant meals and risk of type 2 diabetes. The associations were strongest for restaurant meals of burgers and fried chicken and weaker for meals of fried fish and Chinese food. Restaurant meals of pizza or Mexican food were not associated with diabetes risk. Controlling for BMI markedly reduced the magnitude of the associations of frequent consumption of restaurant foods with diabetes risk, which suggested that a major explanation for the association is the effect of restaurant food consumption on weight gain and obesity. Indeed, BMI values at baseline were considerably higher in women who frequently ate burgers, fried chicken, and fried fish than in women who seldom ate such meals. In contrast, BMI values differed very little according to frequency of consumption of pizza and Mexican food.

Burgers, fried chicken, and fried fish are also prepared and eaten at home. Analyses designed to distinguish the effects of eating the specific food from the effects of eating restaurant meals containing those foods indicated that for fried chicken and fried fish the observed associations with increased risk were likely due to the restaurant meal rather than the food itself. Burgers, however, were associated with an increased risk of diabetes even if prepared at home.

To our knowledge, this is the first study to examine the relation between the consumption of food from restaurants and risk of type 2 diabetes. The results are supported by earlier findings from the CARDIA study, in which a 15-y follow-up of 1444 black men and women and 1587 white men and women showed a significant association between frequency of visits to fast-food restaurants and increases in body weight and insulin resistance (25). Several cross-sectional studies have also found an association between fast-food consumption and body weight (22–24, 26).

Data from 4 studies in the United States suggest that, in urban areas, restaurants serving the least healthy foods—“fast food” restaurants—predominate in black neighborhoods (17–20). For example, in a study in New Orleans, neighborhoods in which 80% of residents were black had 2.4 fast-food restaurants per square mile compared with 1.5 per square mile in neighborhoods in which 20% of the residents were black (17). Findings were similar in Los Angeles, where 58% of the restaurants in a predominantly white, higher-income area were full-service compared with 27% in an area of lower income with a higher proportion of black residents (18). We did not have information on whether the restaurants meals consumed by BWHS participants were from full-serve restaurants or from fast-food restaurants. However, we linked the participants’ addresses to US census block group data on racial composition and the percentage of households living below the federal poverty level. More than a third of BWHS participants live in neighborhoods in which ≥20% of households are below the federal poverty level and 46% live in neighborhoods in which ≥50% of residents are black. Thus, it is likely that a considerable proportion of the food consumed at restaurants by BWHS participants was from fast-food restaurants.

Fast-food consumption has been associated with higher intakes of energy, fat, saturated fat, sodium, and carbonated soft drinks and lower intakes of vitamins A and C, fruit and vegetables, and milk (16, 23). In the present study, participants who reported frequent consumption of food from restaurants also had higher energy and fat intakes and drank more sugar-sweetened soda. Portion sizes tend to be larger in restaurant meals than in meals eaten at home (23). Large portion sizes can lead to overconsumption of calories, thereby leading to weight gain and obesity. The energy density of restaurant foods such as burgers, fried chicken, and French fries is very high (21), which can lead to “passive overconsumption,” the tendency of humans to consume a similar bulk of food with little reference to its energy density (21, 36), which in turn can cause obesity. Fried foods also contain large amounts of partially hydrogenated oils, which may lead to insulin resistance (37).

Among the strengths of our study was the large sample size. High statistical power is important in dietary studies because the intractable problem of imprecision of dietary data tends to dilute associations. The prospective design diminished the potential for recall bias. FFQs have been used successfully to measure diet in prospective studies (38). A validation study of the FFQ used in the present study indicated that dietary intake measured by the FFQ was significantly correlated with diet measured by using diet recalls and diaries. Follow-up rates in the BWHS were sufficiently high to reduce the likelihood of material bias resulting from differential loss related to both exposure and outcome. Important confounding factors such as physical activity were taken into account in the analysis. It was not possible, however, to control for consumption of diet soda because the baseline FFQ asked only about sugar-sweetened soda.

A potential limitation was the identification of diabetes cases by self-report. However, a validation study indicated that >95% of women who reported type 2 diabetes did indeed have that condition. We cannot rule out the possibility that some women with undiagnosed diabetes were misclassified as noncases, but the prevalence of undiagnosed disease was likely to be low. Recent estimates from the National Health and Nutrition Examination Survey (NHANES) indicate that the prevalence of undiagnosed diabetes among African American women in the United States is 4.1% (1). This degree of misclassification would have had a minimal effect on the IRRs in the present study and, if random, would have diluted the IRR estimates.

BWHS participants are from across the United States, with approximately equal numbers from the northeast, south, west, and midwest. Ninety-seven percent of the participants have completed high school or a higher level of education. Among the US black female population of the same ages, 83% have at least a high school education (39). Our results should therefore be applicable to most US black women and possibly to other women and to men as well, because it is plausible that the same underlying mechanism for the association between restaurant foods and risk of type 2 diabetes would be at work in all these groups.

The present study has identified a new and readily modifiable risk factor for type 2 diabetes—the frequent consumption of certain types of restaurant meals. This finding is relevant to African Americans in particular because of the higher prevalence of fast-food restaurants in predominantly black neighborhoods. Fast-food restaurants typically serve the types of foods that are most strongly associated with diabetes incidence in our study: burgers and fried chicken. The findings are also relevant to other populations, especially persons living in countries in which fast food restaurants have made appreciable inroads.