点击显示 收起
1 From the Jean Mayer USDA Human Nutrition Research Center on Aging at Tufts University, Boston (PKN, NQ, and KLT), and the National Institute on Aging, National Institutes of Health, Baltimore (DM, JH, and RA).
2 Supported in part by the US Department of Agriculture, Agricultural Research Service, contract number 53-3K06-01, and by the National Institutes of Health, National Institute on Aging Intramural Program. 3 Address reprint requests to KL Tucker, Jean Mayer USDA Human Nutrition Research Center on Aging at Tufts University, 711 Washington Street, 9th Floor, Boston, MA 02111. E-mail: katherine.tucker{at}tufts.edu.
ABSTRACT
Background: Obesity has increased > 20% in the past decade in the United States, and more than one-half of US adults are overweight or obese.
Objective: Our objective was to further elucidate the nutritional etiology of changes in body mass index (BMI; in kg/m2) and waist circumference by dietary intake pattern. We hypothesized that a healthy dietary pattern would lead to smaller changes in BMI and waist circumference than would other dietary patterns.
Design: Subjects were 459 healthy men and women participating in the ongoing Baltimore Longitudinal Study of Aging. Diet was assessed with the use of 7-d dietary records, from which 41 food groups were created and entered into a cluster analysis.
Results: Five dietary patterns were derived (healthy, white bread, alcohol, sweets, and meat and potatoes). The mean annual change in BMI was 0.30 ± 0.06 for subjects in the meat-and-potatoes cluster and 0.05 ± 0.06 for those in the healthy cluster (P < 0.01). The mean annual change in waist circumference was more than 3 times as great for subjects in the white-bread cluster (1.32 ± 0.29 cm) as for those in the healthy cluster (0.43 ± 0.27 cm) (P < 0.05).
Conclusions: Consuming a diet high in fruit, vegetables, reduced-fat dairy, and whole grains and low in red and processed meat, fast food, and soda was associated with smaller gains in BMI and waist circumference. Because foods are not consumed in isolation, dietary pattern research based on natural eating behavior may be useful in understanding dietary causes of obesity and in helping individuals trying to control their weight.
Key Words: Dietary patterns • cluster analysis • obesity • body composition • diet assessment • BMI • waist circumference
INTRODUCTION
Obesity has increased > 20% in the past decade in the United States (1), and more than one-half of US adults are overweight or obese (2). Obese adults are at increased risk of the major diseases that afflict Western populations, notably cardiovascular disease, type 2 diabetes, and certain cancers (3). Body fat distribution may also be a factor significantly associated with morbidity and mortality. The ratio of waist-to-hip circumference and waist circumference alone have been associated with cardiovascular disease, premature death, stroke, type 2 diabetes, some cancers, and hypertension (4–10).
Despite considerable research effort, the nutritional etiology of obesity remains unclear and controversial, especially with regard to the roles of dietary fat (3, 11) and carbohydrate (12). Inconsistent findings may be due in part to the traditional single-nutrient approach commonly used in nutritional epidemiologic research, which is limited by colinearity among nutrients (13) and by an inability to detect small effects from single nutrients (14). A recent review of intervention trials found that mean weight loss was greater in studies in which subjects consumed a high-fiber diet compared witha low-fiber diet than in studies in which subjects consumed a low-fat diet compared with a high-fat diet (15). However, weight loss was more than 3 times as great in those studies in which subjects consumed a combination low-fat, high-fiber diet (3.4 kg over 6 mo) compared with a low-fat diet alone (1.0 kg over 6 mo), which suggests additive effects of fat and fiber in weight loss (15).
In response to the challenges of the traditional approach to understanding diet-disease relations, the measurement of dietary or eating patterns, in which various foods or nutrients (or both) are combined into a composite variable (16), has been suggested as an alternative method in nutritional epidemiologic research. A recent review of eating patterns and body mass index (BMI; in kg/m2) found that patterns defined with the use of either a diet index, factor, or cluster analysis were inconsistently related to BMI (17). All articles reviewed were cross-sectional in design. We are not aware of any prospective studies that have measured dietary patterns by using factor or cluster analysis in relation to changes in either BMI or waist circumference.
This study was designed to use cluster analysis to examine prospectively the relation between dietary patterns and body composition among women and men participating in the Baltimore Longitudinal Study of Aging (BLSA). Our objective was to further elucidate the nutritional etiology of changes in adiposity over time as measured by BMI and waist circumference by using dietary patterning methods—specifically, cluster analysis. Many studies of dietary patterns have empirically derived a "healthy" type of dietary pattern that is relatively high in fruit, vegetables, fiber, and other "healthy" foods (18–23). We hypothesized that a healthy dietary pattern would lead to smaller changes in BMI and waist circumference than would other dietary patterns.
SUBJECTS AND METHODS
Study population
The BLSA was initiated in 1963 to study the physical, mental, and emotional effects of aging among healthy, active persons; the original study design and data collection were described in detail elsewhere (24). The initial study participants were white men aged 27–88 y who were living in the Baltimore area. The study protocol was expanded in 1978 to include women and minorities. All subjects were volunteers who were recruited from Baltimore City Hospitals. Enrollment in the BLSA is open, and subjects may have entered the study as early as 1963 or as recently as today. Once enrolled, participants return approximately every 12–24 mo for repeated evaluation, eg, height and weight measurements, body-composition analysis, and dietary assessment.
Our study population was limited to those entering the study in or after 1980 (n = 921) to incorporate both men and women and to avoid biasing our study because of changing dietary trends over time. Of these candidates, we excluded those aged < 30 y (n = 46) and > 80 y (n = 92), because persons in the former group are less likely to have developed stable dietary patterns and those in the latter group are more likely to be ill or to have limited dietary intake. All subjects gave written informed consent for their participation in the study, and the institutional review boards of the Johns Hopkins Bayview Medical Research Center and the Gerontology Center approved the BLSA protocol.
Of the remaining 783 subjects who met the year of entry and age range inclusion criteria for the study, those who had not completed ≥ 4 d of dietary records were excluded (n = 9), as were those whose food group intake appeared implausible (> 6 SDs from the mean for each food group; n = 63). An additional 134 subjects were excluded because they did not have measures of height or weight at the time the dietary record was collected and at follow-up, which meant that they could not be used in a prospective study. Finally, to create a disease-free cohort, all subjects who were diagnosed with cancer, diabetes, stroke, or heart disease either before or at baseline were excluded (n = 190), which left 459 subjects available for the analysis. Of those 459 subjects, 10 were excluded from the waist circumference analysis because they did not have baseline or follow-up information on waist circumference.
Dietary assessment
Dietary intake was assessed by 7-d dietary records; reports detailing dietary collection methods and dietary intake in the BLSA population were published previously (25–27). In summary, trained dietitians instructed study participants in the procedure for completing 7-d food records. Food records were completed at home by the participant and sent back to the study center. Before 1993, subjects were given food models and a booklet of food pictures to help them assess portion size. From 1993 on, subjects were given a portable scale for weighing food portions. Participants were contacted by telephone if there were any questions about their diet records.
Dietary records from 1984–1991 were originally coded and entered into a nutrient database maintained by the BLSA, whereas diet records completed since 1994 were coded and entered into the Minnesota Nutrient Database (NDS) at Tufts University; no dietary data were collected in 1992 and 1993. Dietary data from 1984–1991 were then reentered in the NDS, and nutrient intakes were back-adjusted to correct for changes in the food supply (eg, nutrient content due to fortification of cereals) with the use of data from the USDA to correspond to appropriate intervals (28, 29).
To perform a dietary pattern analysis with the use of cluster analysis, individual foods and food ingredients from the dietary records must first be aggregated into groups. We formed 41 food groups, mainly according to macronutrient composition (eg, fat or fiber content) and culinary use; several foods (eg, pizza, eggs) composed their own groups (Appendix A). Where possible, foods were separated into full- and reduced-fat groups (eg, high-fat dairy and low-fat dairy). Cereals were divided into those that were a good source of fiber (≥ 2.5 g/serving) and those that were low in fiber (< 2.5 g/serving) (30).
For entry into the cluster analysis, food groups may be measured by absolute weight in grams, the number of servings, or the percentage of energy intake from each food group (31). We chose to consider the food group variables in terms of the percentage of energy contributed by each of the 41 groups, and we calculated those values for each subject from the average of diet records. The percentage contribution from each food group for each subject was then entered into the cluster analysis.
Cluster analysis requires a priori selection of the number of clusters to be used in the analysis. To decide which number of clusters we would specify, we ran 4–8 cluster solutions and examined each solution to see which set of clusters most meaningfully described distinct eating patterns while they also maintained adequate statistical power in each group to detect effects. From these analyses, a 5-cluster solution was selected. With the use of the PROC FASTCLUS procedure in SAS software (32), the K-means method was then used to classify subjects into 5 nonoverlapping groups in a process that iteratively compares Euclidean distances between each subject for each solution.
Anthropometric and covariate assessments
Anthropometric measurements were made by following standardized procedures (33) as fully described elsewhere (34). In summary, weight and height were measured for each subject at each visit, and BMI was calculated from these values. At each visit, waist circumference was measured with an inelastic tape used at the narrowest part of the torso at the end of expiration (34).
Demographic data were collected from each study participant at the first visit and were used to adjust for potential confounding in regression analyses. Race-ethnicity, physical activity, smoking status, education, and vitamin supplement use were determined by questionnaire when the dietary records were collected. Physical activity was measured by an adapted version of the Harvard Alumni questionnaire, which asked participants about all daily activities (eg, activities at home, at work, and during recreation or sports). The amount of time spent for each activity was summed across all activities to determine the daily energy output per kilogram of body weight (kJ/kg) and has been described previously (25, 35).
Statistical analysis
Sample means and frequencies were calculated separately for the women and the men at the baseline visit at which dietary data were collected. The percentage of subjects who were overweight (BMI: 25–29.99) and obese (BMI: ≥ 30) was calculated with the use of recommended international cutoffs (36).
To describe food and nutrient intake across the 5 clusters (dietary patterns), we calculated separately for each pattern the mean energy contribution from each food group, the mean nutrient intake from selected macronutrients and alcohol, and sample characteristic means and frequencies. We tested for differences across patterns by using PROC GLM software with Tukey-Kramer’s adjustment for multiple comparisons across groups (32). The mean annual changes in BMI and waist circumference were calculated for each dietary pattern.
We performed 2 separate regression analyses to test whether dietary patterns were associated with changes in adiposity as measured by the annual change in BMI and the annual change in waist circumference, respectively. The annual change in BMI was calculated by subtracting the BMI at visit 1 from that at visit 2, dividing that figure by the interval between visits (no. of mo), and then multiplying by 12 (representing mo). The annual change in waist circumference was calculated with the use of the same algorithm. We tested the a priori hypotheses that the healthy dietary pattern would be associated with smaller changes in BMI and waist circumference than would other dietary patterns in models adjusted for baseline anthropometry (baseline BMI and baseline waist circumference, respectively), age, sex, and sociodemographic covariates. Because total energy intake is the mechanism through which nutrients and foods may affect BMI, it is arguable whether energy should be included in the model. We fitted a final model for each analysis in which we added total energy to the multivariate model to ascertain whether adjustment for baseline energy changed our estimates. We also added terms to adjust for development of disease (stroke, cancer, diabetes, and heart disease) during the follow-up period. Because age may not be linearly related to BMI, we fitted regression models by using quadratic terms for age. We checked for effect modification by creating interaction terms for age and dietary pattern, sex and dietary pattern, and age and sex, and we then tested the interaction terms in the final models to see if the model fit improved. All analyses were performed with SAS for WINDOWS software, version 8.2 (32).
RESULTS
Sample characteristics at baseline are shown in Table 1. Most of the subjects were white. Fifty-two percent of the study participants were men, aged 60.8 y on average; the mean age of the women was 57.3 y. More of the men (42.5%) than of the women (31.1%) were overweight at baseline, whereas twice as many of the women (10.5%) than of the men (5.0%) were obese. The men had a greater mean waist circumference (90.4 cm) than did the women (77.2 cm).
View this table:
TABLE 1 . Characteristics of 459 adults participating in the Baltimore Longitudinal Study of Aging
Five clusters were derived, which we labeled healthy, white bread, alcohol, sweets, and meat and potatoes, on the basis of the food that contributed relatively greater proportions of energy to each cluster. The energy contributions from selected food groups for the 5 clusters (dietary patterns) are shown in Table 2. Differences in energy intake from food groups were seen across patterns. Energy intakes in the white-bread, alcohol, and sweets patterns were greatest from these respective food sources. The healthy pattern contained relatively greater contributions from "healthy" foods, including fruit, high-fiber cereal, and reduced-fat dairy, and relatively smaller contributions from fast food, nondiet soda, and salty snacks.
View this table:
TABLE 2 . Percentage energy contribution from selected food groups across the 5 dietary patterns identified at baseline among 459 adults participating in the Baltimore Longitudinal Study of Aging1
A significantly higher percentage of energy from carbohydrate (61.9 ± 1.5%; P < 0.0001) and a higher intake of fiber (26.6 ± 1.1 g; P < 0.001) was seen in the healthy pattern than in all other clusters (Table 3); no significant differences were seen in protein or total energy intake across patterns. The alcohol pattern contained the highest percentage of current smokers (29.3%, P < 0.01), and 62.2% of subjects in the healthy pattern used vitamins compared with 39.0% of subjects in the meat-and-potatoes pattern and 37.6% of subjects in the sweets pattern (P < 0.001). Subjects in the healthy pattern had the smallest waist circumference (P < 0.05) at baseline, but there were no significant differences in baseline BMI (P > 0.05).
View this table:
TABLE 3 . Daily nutrient intakes and sample characteristics across the 5 dietary patterns identified at baseline among 459 adults participating in the Baltimore Longitudinal Study of Aging1
Regression results associating change in BMI and change in waist circumference for all clusters compared with the healthy pattern are shown in Table 4. There was a significantly greater annual increase in BMI among subjects in the meat-and-potatoes pattern (ß = 0.25; 95% CI: 0.07, 0.43; P < 0.05) and in waist circumference among subjects in the white-bread pattern (ß = 0.90 cm; 95% CI: 0.12, 1.68; P < 0.05) than among subjects in the healthy cluster. The annual change in waist circumference for subjects in the meat-and-potatoes pattern was 0.75 cm, which was nearly significant (95% CI: -0.03, 1.53; P < 0.10) compared with changes among subjects in the healthy pattern. Estimates were similar when total energy was included in the model and when models were adjusted for development of disease during the follow-up period. Including quadratic terms for age in the model did not improve model fit. Because no significant interaction was observed between age and dietary pattern, sex and dietary pattern, or age and sex, these interaction terms were removed from the final model. For all study participants, the overall annual rate of change in BMI was 0.11, and the overall annual rate of change in waist circumference was 0.84 cm (data not shown).
View this table:
TABLE 4 . Regression coefficients (ß) and SE for dietary patterns for predicting relative change in BMI and change in waist circumference, comparing each pattern with the healthy pattern, among adults participating in the Baltimore Longitudinal Study of Aging1
The mean annual change in BMI was 0.30 ± 0.06 for subjects in the meat-and-potatoes pattern and 0.05 ± 0.06 for those in the healthy pattern (P < 0.01) (Figure 1). The mean annual change in waist circumference among subjects in the white-bread pattern (1.32 ± 0.29 cm) was more than 3 times that among subjects in the healthy pattern (0.43 ± 0.27 cm; P < 0.05) (Figure 2).
FIGURE 1. . Mean (± SE) annual change in BMI across the 5 dietary patterns identified at baseline among adults participating in the Baltimore Longitudinal Study of Aging. Healthy pattern, n = 98; white-bread pattern, n = 79; alcohol pattern, n = 60; sweets pattern, n = 140; meat-and-potatoes pattern, n = 82. *Significantly different from the healthy pattern, P < 0.05.
FIGURE 2. . Mean (± SE) annual change in waist circumference across the 5 dietary patterns identified at baseline among adults participating in the Baltimore Longitudinal Study of Aging. Healthy pattern, n = 98; white-bread pattern, n = 79; alcohol pattern, n = 60; sweets pattern, n = 140; meat-and-potatoes pattern, n = 82. *Significantly different from the healthy pattern, P < 0.05.
DISCUSSION
We derived 5 nonoverlapping dietary patterns by using cluster analysis, and the smallest gains in BMI and waist circumference were seen for subjects consuming a healthy diet. Our healthy dietary pattern is similar to the Dietary Approaches to Stop Hypertension diet, which has been shown to decrease blood pressure (37). We are not aware of any studies that have examined the relation between the Dietary Approaches to Stop Hypertension diet and changes in BMI or body composition or both. The overall annual rate of change in BMI that we observed in this study (0.11) is similar to that reported for control subjects in the Normative Aging Study with a similar baseline BMI (0.09; 38) and to that in a study among representative Canadians (0.11 for women and 0.09 for men; 39).
Our findings were strengthened by the prospective study design. Most studies of dietary patterns and weight have employed a cross-sectional design (18, 19, 40), and such studies are susceptible to reverse causation. In a cross-sectional study of 16 621 women participating in the National Health Screening Service in Oslo, Jacobsen and Thelle (41) found that high BMI was most strongly associated with lower consumption of low-fat milk and higher consumption of fish, and they concluded that this pattern may have resulted from the subjects’ attempts to lose weight, rather than reflecting the reason for weight gain.
Our study used patterning methods to assess diet, an approach that has only recently been used in observational obesity research. A recent review on the topic considered 30 cross-sectional studies that observed the relation between diet patterns defined by diet index, factor, or cluster analysis and BMI (17). The results were inconclusive, possibly because of variability in dietary assessment methods (including food-frequency questionnaires, diet diaries, single 24-h diet recalls) and inadequate control of confounding factors. However, there was limited evidence suggesting that a diet high in fruit and vegetables and low in meat and fat was associated with a lower BMI (17), as also seen in the present study. We assessed diet with the use of 7-d diet diaries, which are considered the gold standard of dietary assessment, and we were able to adjust for age, sex, physical activity, education, and smoking in our models. We were not able, however, to adjust for the role of genetics, which may modify the relation between diet and adiposity (42).
We know of no prospective studies that have considered the relation between empirically derived dietary patterns and body fat distribution. In 2 cross-sectional studies, persons with type 1 diabetes (n = 2868) who consumed a diet high in carbohydrates, cereal fiber, and low-glycemic-index foods had a significantly smaller waist circumference and BMI (43), and healthy adults (n = 2110) participating in the original and offspring US Framingham Heart Study and European SENECA studies who were in the meat and fat cluster had a larger waist circumference and BMI than did those who were in the fish and grain cluster. A recent crossover study among 11 healthy men who consumed a high-glycemic-index diet and a low-glycemic-index diet for 5 wk each found that the low-glycemic-index diet led to a decrease in trunk mass and an increase in lean body mass with no change in body weight (44). The authors suggest several possible mechanisms, including shifts in substrate utilization, decreases in the proteolytic counterregulatory hormones, and a reduction in lipoprotein lipase activity, which could explain the change in fat mass but not body weight (44). In our study, those in the white-bread pattern had significantly greater gains in waist circumference, but not in BMI, than did those in the healthy pattern. These studies and ours suggest that a diet high in fiber-rich, low-glycemic-index foods may result in lesser amounts of central adiposity.
Our study has several limitations. Dietary exposure measurements that use empirical methods such as cluster or factor analysis are data-driven and involve subjective decisions by investigators, and there is no gold standard for determining the number of clusters (17). Patterns may therefore be difficult to reproduce and compare across studies, especially between populations in whom diet differs. However, there does seem to be homogeneity of dietary patterns across populations. With the use of cluster or factor analysis, several studies have identified healthy and meat-based patterns (18–23, 45–49), a white-bread pattern (13, 50–52), a sweets pattern (18, 22, 50, 52–55), and an alcohol pattern (18, 22, 23, 48, 51, 53, 54, 56) in which food intakes were similar to those of the same-named patterns observed in this study. The consistency of patterns across studies derived from either cluster or factor analysis suggests that dietary patterns are reasonably reproducible.
A limitation of the dietary pattern approach is the inability to isolate nutrient-specific biologic effects, because dietary patterns include many foods and nutrients that act differently to affect hunger, satiety, energy metabolism, and food intake. In our study, subjects consuming the healthy dietary pattern had the highest intake of foods such as high-fiber cereal, reduced-fat dairy, fruit, nonwhite bread, whole grains, beans and legumes, and vegetables and the smallest gains in BMI and waist circumference. Several mechanisms may be responsible for this effect. These foods are high in fiber, which may affect weight by increasing satiety and satiation through decreased gastric emptying, increased colonic transit, and decreased insulin response (57). This hypothesis with regard to the role of fiber may be extended to the role of the glycemic index (58). Many of the foods in the healthy pattern are low in glycemic load, which evokes a decreased insulin response and therefore decreases hunger and energy intake (59). The healthy dietary pattern also contained foods that are low in energy density (eg, fruit and vegetables), which may be responsible for mediating energy intake rather than dietary composition per se (60–62). Compared with subjects in the healthy pattern, those in the meat-and-potatoes pattern had relatively higher intakes of meat, potatoes, fast food, pizza, and nondiet soda—foods that are low in fiber and high in energy density and glycemic load—whereas those in the white-bread pattern received almost 16% of their daily energy intake from white bread—the food with the highest glycemic index value (63).
This study cannot conclude whether one of these mechanisms or another, as yet unknown mechanism was responsible for the gains in BMI and waist circumference, but it is likely, rather, that there are multiple pathways through which food intake, including both dietary composition and total food volume, affects energy balance. Because foods are consumed not in isolation but as part of an overall dietary pattern, research based on natural eating behavior may be useful in understanding the effect of diet on weight. Such research is also more easily translated to specific dietary recommendations or advice (31), which could be helpful to persons trying to control their weight.
Our study did not address social or demographic issues that may affect dietary intake. There may be sex differences in dietary intake (13, 51, 64), and dietary patterns may also have different effects on BMI in men and women (53, 65). In our study, preliminary analysis revealed similar patterns in the men and the women, and we therefore did not stratify our analysis by sex. In addition, this study population was mainly white and highly educated, whereas dietary patterns may differ among ethnic (13) and educational (19) groups. Our population was also relatively lean (5% of the women and 10% of the men were obese) compared with a nationally representative sample of US adults, of whom 31% were obese (66). For these reasons, our findings may not be generalizeable to the overall US population.
In conclusion, our results suggest that consuming a diet high in fruit, vegetables, reduced-fat dairy, and whole grains and low in red and processed meat, fast food, and soda was associated with smaller gains in BMI and waist circumference. Additional prospective research studies are needed to further assess the relations among dietary patterns, body weight, and central fat deposition.
APPENDIX A
View this table:
Food groups (n = 41) used in the cluster analysis
ACKNOWLEDGMENTS
We gratefully acknowledge Peter Bakun and Janice Maras for assistance with data processing and preparation of dietary variables.
RA and JH contributed to the original design and to data collection for the Baltimore Longitudinal Study of Aging. PKN was responsible for the design and analysis for this report and drafted the manuscript. NQ provided statistical programming support. KT contributed to the analysis and design and oversaw Tufts’ collaboration with the BLSA. All authors made critical comments during the preparation of the manuscript, and they fully accept responsibility for the work. None of the authors had a personal or professional conflict of interest.
REFERENCES