Are prevailing patterns of diet behavior in a population, assessed using factor or cluster analysis, related to body weight or risk of obesity?
ConclusionLimited and inconsistent evidence from epidemiological studies examining dietary patterns derived using factor or cluster analysis in adults found that consumption of a dietary pattern characterized by vegetables, fruits, whole grains, and reduced-fat dairy products tends to be associated with more favorable body weight status over time than consumption of a dietary pattern characterized by red meat, processed meats, sugar-sweetened foods and drinks, and refined grains.
GradeIII – Limited
- Cluster and factor analyses are data-driven approaches that describe the dietary patterns consumed by the study population. Variability in the studies included in this review, including populations considered, dietary assessment methods used, the number and type of food groupings included in the analyses, and the statistical techniques employed, made comparisons among studies challenging.
- The number of patterns identified in the studies ranged from 2 to 6 and some similarities emerged among them. The patterns were not consistently defined by specific foods, but rather by a range of foods with overlap among the patterns. What differentiated the patterns was the amount or frequency of each food consumed.
- Dietary patterns that emerged in factor or cluster analysis that were associated with lower risk of obesity were characterized by the presence of vegetables, fruit, whole grains, and reduced-fat dairy. In adults, results pointed toward a more favorable weight status, lower weight/waist circumference (WC) gain, and lower body mass index (BMI) over time.
- Dietary patterns derived from factor or cluster analysis associated with a higher risk of obesity were characterized by the presence of red meat and processed meats, sugar-sweetened foods and drinks, and refined grains. Results related to consumption of these patterns pointed toward increased body weight and waist circumference measures over time.
- Ethnicity and socioeconomic status were often not reported or included in analyses. Insufficient evidence was available to support conclusions related to children and adolescents.
Evidence Summary OverviewDescription of the Evidence
Factor and cluster analyses are data-driven approaches that empirically derive food intake patterns (appendix A). A total of 11 studies met the inclusion criteria for this systematic review (table 4-A-II-1). All studies were prospective cohort studies, and 10 of 11 received a positive quality rating. Eight of the eleven studies were conducted in the United States, with additional studies from the United Kingdom. (McNaughton, 2007), Iran (Hosseini-Esfahani, 2011), and Sweden (Newby, 2006). The sample sizes for the studies were from 206 to 51,670 participants (3 studies <500, 4 studies <2,500, 1 study <4500, and 3 >30,000). The follow-up times for these observational studies ranged from 3 to 20 years. The majority of the studies were conducted with generally healthy adult men and women (Duffey, 2012; Hosseini-Esfahani, 2011; McNaughton, 2007; Newby, 2003; Newby, 2004; Togo, 2003), while five studies focused on dietary patterns in women only (Boggs, 2011; Bewby, 2006; Quatromoni, 2002; Ritchie, 2007; Schulze, 2006). Finally, one study (Ritchie, 2007) was conducted in children to examine weight gain in adolescence over the period of follow-up. In general, the strengths of these studies include their prospective design and length of follow-up.
Dietary intake in these prospective cohort studies was assessed using various methods. Six studies used food frequency questionnaires ranging from 26 to 168 items (Boggs, 2011; Hosseini-Esfahani, 2011; Newby, 2006; Quatromoni, 2002; Schulze, 2006; Togo, 2003). The dietary assessment of three cohorts was by diet records of 3, 5, or 7 days (McNaughton, 2007; Newby, 2003; Newby, 2004). One study assessed intake using a diet history questionnaire (Duffey, 2012). To derive dietary patterns, seven studies used factor analysis (Boggs, 2011; Hosseini-Esfahani, 2011; McNaughton, 2007; Newby, 2004; Newby, 2006; Schulze, 2006; Togo, 2003) and four studies used cluster analysis (Duffey, 2012; Newby, 2003; Quatromoni, 2002; Ritchie, 2007).
The endpoint outcomes of interest in this review were measures of weight status. Three studies examined change in body weight; seven studies reported on BMI; and six studies reported on waist circumference.One study each examined percent body fat and incidence of overweight/obesity. In eight of the studies, the outcomes were measured by study personnel; in three studies, body weight was self-reported (Boggs, 2011; Newby, 2006; Schulze, 2006).
Evidence Summary ParagraphsBoggs et al., 2011 (positive quality) conducted a prospective cohort study in the U.S.to assess dietary patterns in relation to weight gain using data from the Black Women's Health Study (BWHS) in 1995, 2001, and 2009. Participants (n = 41,351), ranging in age from 21 to 54 years old, self-administered a 68-item FFQ in 1995 and an 85-item FFQ in 2001. FFQ items were aggregated into 35 predefined food groups on the basis of similarity of nutrient content. Two major dietary patterns were identified using factor analysis (principal components analysis):
- “Vegetables/Fruit” (vegetables, fruit, legumes, fish, and whole grains)
- “Meat/Fried Foods” (red meat, meat, fries, fried chicken, and added fat)
Duffey et al., 2012 (positive quality) conducted a prospective cohort study in the United States to examine the association of different dietary patterns with or without diet beverage consumption and the risk of cardiometabolic outcomes, using data from the Coronary Artery Risk Development in Young Adults (CARDIA) study. Participants (n = 4,161), ranging in age from 18 to 30 years old, responded to a validated, interviewer-administered CARDIA diet history questionnaire, followed by a quantitative diet history. Foods and beverages from the baseline diet history were categorized into 43 food groups by using the Nutrition Coordinating Center algorithm, measured as energy per day from food group. The “diet beverages” food group (measured by servings per day) was used to identify baseline consumers of diet beverages, and these groups were labeled “consumers” and “nonconsumers.” Two distinct baseline dietary patterns were identified using cluster analysis:
- “Prudent diet” (fruit, milk, yogurt, cheese, nuts, seeds, fish, and whole grains)
- “Western diet” (meats, poultry, refined grains, soda, fast food, fruit drinks, egg and egg dishes, legumes, and snacks)
Hosseini-Esfahani et al., 2011 (positive quality) extracted a subset of data from the Tehran Lipid and Glucose Study (TLGS), a prospective cohort study conducted in Iran, and used factor analysis to determine whether changes in food patterns were related to obesity. The data was from two periods of the TLGS survey: 1999-2001 and 2005-2007. Participants (n = 206), mean age 42 years old, answered a 168-item FFQ. FFQ items were then aggregated into 21 food groups according to macronutrient composition. During the two stages of study, three dietary patterns were identified:
- “Healthful” (high intake of fruit, vegetables, dairy, oil, whole grains, poultry, and fish)
- “Western” (high intake of processed meats, fat, salty snacks, fatty sauces, and sweet beverages)
- “Mix” (high-fat red meats, legumes, nuts and seeds, sweets, tea, and coffee)
McNaughton et al., 2007 (positive) analyzed data from the Medical Research Council (MRC) National Survey of Health and Development (NSHD), also known as the 1946 British Birth Cohort, to assess the relationship between dietary patterns during adult life (at ages 36, 43, and 53 years) and risk factors for chronic disease at age 53. The dietary data was collected in 1982, 1989, and 1999 and risk factors were measured in 1999, when subjects were age 53 years. Dietary intake was assessed at each time point using a 5-consecutive-day food diary. Foods and beverages consumed in 1982, 1989, or 1999 were allocated to one of 126 food groups. The distribution of foods and beverage item consumption was highly skewed, so a binary variable was created for each item (consumers or nonconsumers). Exploratory factor analysis was conducted and dietary patterns consistent across the three time points were identified (described further in Mishra, 2006).
For women, three dietary patterns were identified:
- “Ethnic foods and alcohol” (Indian and Chinese meals, rice and pasta, oily fish and shellfish, olive oil, some vegetables, and alcoholic beverages (particularly red and white wine)
- “Meat, potatoes, and sweet foods” (red meat, bacon and ham, all types of potato and potato dishes, sweet pies, cakes, puddings, and desserts with negative loadings for pasta and skimmed milk)
- “Fruit, vegetables, and dairy” (low-fat/reduced-fat dairy products, fruit, some vegetables, and whole meal bread, with negative loadings for meat, meat products, and white bread)
For men, two dietary patterns were identified:
- “Ethnic foods and alcohol” (Indian and Chinese meals, rice and pasta, shellfish, olives, some vegetables and legumes, and alcoholic beverages [particularly red and white wine], with negative loadings for meat pies, fried chips, and animal fats)
- “Mixed” (many fruits and vegetables, low-fat/low-calorie yogurt and soya milk and a range of sweet foods including cakes, sweet biscuits, sweet pies, puddings, desserts, confectionery, and ice cream)
Newby et al., 2003 (positive quality) and Newby et al., 2004 (positive quality) conducted a prospective cohort study in the United States to evaluate the nutritional etiology of changes in BMI and WC by dietary intake pattern, using data from the Baltimore Longitudinal Study of Aging. Participants (n = 459), age range 30 to 80 years old in 1980, completed 7-day dietary records. Then, cluster and factorial analysis were used to define dietary patterns:
In 2003, the food intake data were aggregated into 41 food groups based on macronutrient composition and culinary use, and five food patterns were identified using cluster analysis:
- “Healthy" (high-fiber cereal, fruit, and reduced-fat dairy)
- “White bread” (refined grains, poultry, meat, and high-fat dairy)
- “Alcohol” (refined grains and alcohol)
- “Sweets” (refined grains, fruits, high-fat baked goods, meat, and high-fat dairy)
- “Meat and potatoes” (refined grains, fruit, meat, and high-fat dairy)
In 2004, the food intake data were aggregated into 40 food groups according to macronutrient composition and culinary use, and six factors or patterns were identified using principal component analysis, and then patterns were treated categorically (i.e., divided into quintiles):
- “Factor 1: Reduced-fat dairy products, fruits, and fiber” (nuts, seeds, legumes, white bread, and refined grains)
- “Factor 2: Protein and alcohol” (seafood and poultry)
- “Factor 3: Sweets” (sweetened juices, dairy desserts, and fast food)
- “Factor 4: Vegetable fats and vegetables” (starchy vegetables, white bread, and refined grains)
- “Factor 5: Fatty meats” (organ meats, nonwhite bread, vegetables, fruits, and processed meats)
- “Factor 6: Eggs, bread, and soup” (whole grains)
Newby et al., 2006 (positive quality) conducted a prospective cohort study in Sweden to examine whether changes in food patterns were associated with changes in BMI using data of the Swedish Mammography Cohort in 1987 and 1997. Participants (n = 33,840 women), mean age 52.5 years old, self-administered a 65-item FFQ in 1987 and a 97-item FFQ in 1997. Food groups were created according to fat and fiber content, culinary use, and previous research on the study of food patterns and body composition. Due to minor differences in dietary assessment at each time, 29 groups were created in 1987, while 32 groups were created in 1997. Confirmatory factor analysis was used to derive food patterns in each set of food groups:
- “Healthy” (vegetables, fruit, whole grains, fruit juice, and cereals)
- “Western/Swedish” (meat, processed meat, liver, potatoes, and refined grains)
- “Alcohol” (wine, spirits, snacks, beer, and chocolate)
- “Sweets” (sugary foods, sweet baked goods, soda, chocolate, fruit soup, refined grains, and dairy desserts)
Quatromoni et al., 2002 (positive quality) conducted a prospective cohort study in the United States to explore relationships between dietary patterns and the development of overweight (BMI ≥25) over a period of 12 years using data from female participants in the Framingham Offspring-Spouse study (FOS). Participants (n = 737), age range 30 to 89 years old, answered a 145-item FFQ. FFQ items were classified into 42 foods groups based on similar levels of macronutrients and key micronutrients, and then 13 food groupings were identified, each containing multiple food categories. In a second step, women were separated into groups with no overlap groups based on similarities in their frequency of consumption of the 13 food groupings. Five clusters were identified through this process:
- “Heart healthy" (vegetables, fruits, low-fat milk, other low-fat and fiber-rich foods, whole grains, fish, low-fat cheeses, lean poultry, legumes, and fewer servings of diet beverages)
- “Light eating” (more moderate eating patterns and higher consumption of beer and poultry with skin)
- “Wine and moderate” (more moderate eating patterns and higher consumption of wine)
- “High fat” (high amounts of animal and vegetables fats, sweet desserts, meat, and mixed dishes)
- “Empty-calorie” (high amounts of animal and vegetables fats, sweet desserts, meat, mixed dishes, and sweetened beverages)
Ritchie et al., 2007 (positive quality) analyzed data from a prospective cohort study conducted in the United States to determine the relation of dietary patterns with nutrient intakes and measures of adiposity over a period of 10 years using data from the National Heart, Lung, and Blood Institute Growth and Health Study (NGHS) Cohort. Participants were Black and White girls (n = 2,371), age range 9 to 10 years old at baseline, who recorded their dietary intake for 3 days (2 weekdays, 1 weekend) on an annual basis. Food intakes were combined into 40 food groupings based on frequency of usage, contribution to total energy intake, and customary use in the diet. Cluster analysis was used to define four separate patterns for Black girls and White girls.
For Black girls, the patterns were:
- “Customary” (low intakes of diet drinks, coffee/tea, yogurt, cheese, plain grains, crackers, fish/poultry [not fried], red meat, other soups, and most vegetables)
- “Snack-type foods” (high intakes of diet drinks, coffee/tea, yogurt, crackers, pretzels, other soups, and green salads, and low intakes of flavored milks, several other grain groupings, and processed meats/sandwiches)
- “Meal-type foods” (high intakes of plain breads and grains, other breakfast grains, and most types of sandwiches and protein sources, including legumes, fried and not fried potatoes)
- “Sweets and cheese pattern” (large amounts of sweets and flavored milk and cheese, with relatively small amounts of many other foods, for example, eggs, fried fish/poultry, and fried potatoes)
- “Convenience” (high intakes of pizza, fried fish/poultry, and ramen, with relatively low intakes of juice, plain milk, many of the grain-type groupings, eggs, unfried fish/poultry, cheese/spread sandwiches, other soups, fruit, and most of the vegetable-rich groupings)
- “Sweets and snack-type foods” (high intakes of sweetened and diet drinks, juice, cheese, other desserts, candy, crackers, pretzels, nuts/popcorn, peanut butter and cheese-spread sandwiches, and mixed dishes, with low intakes of flavored milk, processed meats/sandwiches, and mixed dishes)
- “Fast-food pattern” (high intakes of flavored milk, burgers, fried potatoes, eggs, red meat, processed meats/sandwiches, chips, legumes, and baked desserts, with low intakes of diet drinks, yogurt, cheese, candy, other desserts, crackers, pretzels, and peanut butter sandwiches)
- “Healthy” (low intakes of sweetened drinks, baked desserts, chips, fried fish/poultry, red meat, burgers, pizza, and fried potatoes, with relatively high consumption of plain milk, yogurt, plain breads and grains, cereal, other breakfast grains, mixed dishes, other soups, fruit, green salads, unfried potatoes, and other vegetables)
Schulze et al., 2006 (positive quality) conducted a prospective cohort study in the United States to examine the association between adherence to dietary patterns and weight change using data from the Nurse’s Health Study II from 1991 to 1999. Participants, age range 26 to 46 years old, self-administered a 133-food item FFQ in 1991, 1995, and 1999. Foods from the FFQ were then classified into 39 food groups based on nutrient profiles or culinary usage. The final sample included 51,670 women. Eating patterns defined by principal component analysis for each time-point were described as:
- “Prudent pattern” (higher intakes of fruits, vegetables, whole grains, fish, poultry, and salad dressing)
- “Western pattern” (higher intakes of red and processed meats, refined grains, sweets and desserts, and potatoes)
Togo et al., 2003 (positive quality) conducted a prospective cohort study in Denmark to assess associations between food-intake factor scores and BMI changes, as well as longitudinal associations between changes in food-intake patterns and subsequent changes in BMI using data from the Monitoring of Trends and Determinants in Cardiovascular Diseases (MONICA) study from 1982 to1994. Participants, age range 30 to 60 years old, answered a 26-item FFQ, and factor analysis was used to identify dietary patterns separated by gender.
For women, two patterns were found and named as:
- “Green” (high intakes of fruits, vegetables, whole grains, fish, and cheese)
- “Sweet-Traditional” (more meat and foods of higher energy density, with fewer vegetables)
- “Green” (high intakes of fruit, vegetables, whole grains, fish, and cheese)
- “Sweet” (baked goods, candy, soft drinks, ice cream, honey, or jam)
- “Traditional” (more meat and foods of higher energy density, with fewer vegetables)
Table 4-A-II-1 Summary of Findings Dietary patterns identified using factor analysis or cluster analysis (shaded rows) and association with body weight status and obesity
Qualitative Assessment of the Collected Evidence
Quality and Quantity
In terms of quantity, an exhaustive search in four electronic databases, supplemented with a hand search, identified 11 prospective cohort studies that met the inclusion criteria for evaluating dietary patterns with regard to body weight status. In total, over 135,000 adults and 2,300 adolescents were examined by studies in this review. Quality assessment of these studies involved an examination of their methodological rigor in order to minimize bias and random error in the systematic review findings. Components of the evaluation included the number and selection of data samples, use of statistical tools, and detection or reporting bias. Ten of eleven of the included studies were found to be of positive quality, indicating a low risk of bias and random error.
ConsistencyThe studies in this review did not use a consistent approach for examining the association between dietary patterns and body weight. Both factor and cluster analysis methodologies were used. Factor analysis summarizes a number of original variables into smaller composite factors, while cluster analysis groups individuals into clusters, so that individuals in the same cluster are homogeneous and there is heterogeneity across clusters. In addition, there are several methodological differences within the study designs, including different methods of grouping foods and selecting patterns. Nevertheless, some similarities among the findings can be seen. Most of the studies found at least two generic food patterns: a “healthy/prudent” food pattern and an “unhealthy/western” pattern. Generally, healthier patterns were associated with more favorable body weight outcomes, while the opposite was seen for unhealthy patterns. However, not all studies reported significant associations.Furthermore, because the patterns are data-driven, they represent what was consumed by the study population, and thus it is difficult to compare across the disparate patterns.
This body of evidence directly addressed the diet exposures and examined health outcomes of interest for this systematic review. Overall, the results observed were clinically meaningful from a public health perspective, particularly related to BMI. However the association between dietary patterns and body weight status may be mitigated by the presence of other important lifestyle factors that influence body weight but are difficult to measure.
The total number of studies in this review is not large, but several of them were conducted in large prospective cohort studies. Ethnicity, SES, sex, age, and BMI are among the variables considered when examining generalizability of the results to the general U.S. population. Ethnicity and socioeconomic status were often not reported or included in analyses, which makes generalizing the results difficult. There was a potential difference in associations found by gender: of the three studies that analyzed men and women separately, men tended to have null results. However, there were insufficient data to draw conclusions about population subgroups. In addition, nearly all the studies were in adults, only one was in adolescents, and none in young children. For these reasons, caution should be observed when extrapolating the findings to other populations.
- Factor and cluster analyses are data-driven approaches that describe the dietary patterns in a particular population. The studies describe preexisting dietary patterns within the population and the dietary patterns are not based on a hypothesized association to health. The patterns derived through analyses may not represent the most beneficial or detrimental patterns relative to the health outcome of interest.
- Among the studies reviewed, the dietary pattern analyses varied with regard to the dietary assessment methods, the number and type of food groupings, and the statistical analysis techniques, which make comparisons challenging.
- In factor and cluster analysis, the consolidation of food items into food groups, the number of factors or clusters to extract, and even the labeling of components are subjective. Furthermore, patterns derived from either factor or cluster analysis may not be reproducible across studies because elements of dietary patterns and analytic decisions differ.
- Dietary pattern analysis using factor or cluster methods may not be very informative in determining which elements of the diet or which biological relationships between these elements are responsible for the health outcome.
- Some studies completed over long periods of time did not account for changes to subjects’ diets or seasonal variations in food supplies, which may have influenced the food components of patterns.
- One study analyzed the dietary patterns of pre-pubescent children transitioning into adolescence. In general, the results show that patterns vary widely at this age and caution should be observed when analyzing these data because the diet of children changes rapidly, as well as their weight.
- Insufficient evidence was available in population subgroups to examine the relationship between dietary patterns derived using factor and cluster analyses and body weight status. Future studies using this methodology should examine variables such as ethnicity, SES, sex, baseline weight status, and age. In additon, it is important to incorporate environmental and behavioral factors, such as physical activity, non-leisure physical activity, eating practices (eating out, cooking at home), indulgence over the weekend, among others, as potential confounders These variables may be moderators that in the long term will define the association between a particular pattern and weight status. There is a need for more research into specific ethnic groups and how cultural practices may influence dietary patterns and their repercussions for body weight.
- Research is needed to further examine if various dietary patterns influence body weight status differently among participants who are normal weight, overweight, or obese. There is some indication that obese versus normal weight individuals respond differently to changes of food patterns on body weight measures. Research in this area may help uncover better approaches to body weight management practices.
- There is a need to examine the most common unhealthy/western pattern components, variations, and amounts of food consumed by those who have such a diet. Rationale: If a pre-existing pattern is found to be detrimental to health, there is an impetus for dietary pattern modification.
- Boggs DA, Palmer JR, Spiegelman D, Stampfer MJ, Adams-Campbell LL, Rosenberg L. Dietary patterns and 14-y weight gain in African American women. Am J Clin Nutr. 2011 Jul;94(1):86-94. Epub 2011 May 18. PMID: 21593501.
- Duffey KJ, Steffen LM, Van Horn L, Jacobs DR Jr, Popkin BM. Dietary patterns matter: diet beverages and cardiometabolic risks in the longitudinal Coronary Artery Risk Development in Young Adults (CARDIA) Study. Am J Clin Nutr. 2012 Apr;95(4):909-15. Epub 2012 Feb 29. PMID: 22378729.
- Hosseini-Esfahani F, Djazaieri SA, Mirmiran P, Mehrabi Y, Azizi F. Which Food Patterns Are Predictors of Obesity in Tehranian Adults? J Nutr Educ Behav. 2011 Jun 7. [Epub ahead of print] PMID: 21652267.
- McNaughton SA, Mishra GD, Stephen AM, Wadsworth ME. Dietary patterns throughout adult life are associated with body mass index, waist circumference, blood pressure, and red cell folate. J Nutr. 2007 Jan;137(1):99-105. PubMed PMID: 17182808.
- Newby PK, Weismayer C, Akesson A, Tucker KL, Wolk A. Longitudinal changes in food patterns predict changes in weight and body mass index and the effects are greatest in obese women. J Nutr. 2006 Oct;136(10):2580-7. PMID: 16988130.
- Newby PK, Muller D, Hallfrisch J, Andres R, Tucker KL. Food patterns measured by factor analysis and anthropometric changes in adults. Am J Clin Nutr. 2004 Aug;80(2):504-13. PMID: 15277177.
- Newby PK, Muller D, Hallfrisch J, Qiao N, Andres R, Tucker KL. Dietary patterns and changes in body mass index and waist circumference in adults. Am J Clin Nutr. 2003 Jun;77(6):1417-25. PMID: 12791618.
- Quatromoni PA, Copenhafer DL, D'Agostino RB, Millen BE. Dietary patterns predict the development of overweight in women: The Framingham Nutrition Studies. J Am Diet Assoc. 2002 Sep;102(9):1239-46. PMID: 12792620.
- Ritchie LD, Spector P, Stevens MJ, Schmidt MM, Schreiber GB, Striegel-Moore RH, Wang MC, Crawford PB. Dietary patterns in adolescence are related to adiposity in young adulthood in black and white females. J Nutr. 2007 Feb;137(2):399-406. PMID: 17237318.
- Schulze MB, Fung TT, Manson JE, Willett WC, Hu FB. Dietary patterns and changes in body weight in women. Obesity (Silver Spring). 2006 Aug;14(8):1444-53. PMID: 16988088.
- Togo P, Osler M, Sørensen TI, Heitmann BL. A longitudinal study of food intake patterns and obesity in adult Danish men and women. Int J Obes Relat Metab Disord. 2004 Apr;28(4):583-93. PMID: 14770197.
Research Design and Implementation
For a summary of the Research Design and Implementation results, click here.
Boggs DA, Palmer JR, Spiegelman D, Stampfer MJ, Adams-Cambell LL, Rosenberg L. Dietary patterns and 14-year weight gain in African American women. Am J Clin Nutr. 2011; 94: 86-94.
Duffey KJ, Steffen LM, Van Horn L, Jacobs DR, Popkin BM. Dietary patterns matter: diet beverages and cardiometabolic risks in the longitudinal Coronary Artery Risk Development in Young Adults (CARdia) Study. Amer J Clin Nutr. 2012; 95: 909-915.
Hosseini-Esfahani F, Djazaieri SA, Mirmiran P, Mehrabi Y, Azizi F. Which food patterns are predictors of obesity in Tehranian adults? J Nutr Educ Behav. 2011. [Epub ahead of print] PMID: 21652267.
McNaughton SA, Mishra GD, Stephen AM, Wadsworth ME. Dietary patterns throughout adult life are associated with body mass index, waist circumference, blood pressure, and red cell folate. J Nutr. 2007 Jan; 137(1): 99-105.