Insulin resistance is a common pathological state in which target cells (liver, muscle and adipose tissue) fail to respond to normal levels of circulating insulin(Reference Mlinar, Marc and Janez1). Concomitant conditions that are associated with insulin resistance include impaired glucose tolerance, impaired fasting glucose, elevated insulin concentrations, low HDL-cholesterol concentrations and elevated TAG concentrations(Reference Rader2). Diet is believed to play a key role in the development of insulin resistance. To date, various aspects of diet, such as fat, carbohydrate, fibre, whole grains, Mg, glycaemic index and load(Reference Hu, van Dam and Liu3–Reference Rumawas, McKeown and Rogers7), have been related to insulin resistance. Most observational studies relate individual nutrients or foods to various outcomes and although these studies contribute to our understanding on the role of diet in disease, there are several potential limitations. First, due to the high correlation between components of a diet, it is often difficult to examine the separate effect for a specific nutrient or food. Second, as interactions may exist between components of a diet, the findings of single-nutrient or food analysis may be misleading. Finally, the effect of a single nutrient or food on a disease outcome or risk factor may be too small to detect, particularly in smaller cohorts(Reference Jacques and Tucker8, Reference Hu9).
The use of dietary patterns to capture the overall dietary habits of populations has received much attention in recent years. Instead of focusing on the effect of individual nutrients or foods, a dietary pattern approach examines the effect of an overall diet on health outcomes, thereby representing a complex combination of foods and nutrients(Reference Hu, Rimm and Smith-Warner10). Thus, from a public health perspective, results from dietary pattern analysis can be easily translated and incorporated into dietary guidelines for the public(Reference Hu9). A number of observational studies have examined the association between dietary patterns and either the metabolic syndrome, or components of the metabolic syndrome(Reference Esmaillzadeh, Kimiagar and Mehrabi11–Reference Lutsey, Steffen and Stevens16). Although insulin resistance is a key underlying risk factor associated with this syndrome, few studies have used a dietary pattern approach to directly examine the relationship between diet and surrogate measures of insulin resistance in healthy adults(Reference Esmaillzadeh, Kimiagar and Mehrabi11, Reference Nettleton, Steffen and Mayer-Davis17). To our knowledge, no studies have evaluated dietary patterns derived by cluster analysis in relation to insulin resistance in a population without diabetic patients. The purpose of the present study was to determine if specific dietary patterns derived by cluster analysis are related to insulin-resistant phenotypes among adults without diabetes.
Experimental methods
Participants
The present investigation examined the cross-sectional relationship between dietary patterns and insulin-resistant phenotypes using data from the Framingham Offspring Study. The Framingham Study was initiated in 1948 as a longitudinal population-based study of CVD. In 1971, 5135 offspring of original participants of the Framingham Study and spouses of the offspring were recruited to participate in the Framingham Offspring Study. Members of the Framingham Offspring Study have returned on average every 4 years for a physical examination, questionnaires, laboratory tests, and assessment of cardiovascular and other risk factors(Reference Feinleib, Kannel and Garrison18). During the fifth examination cycle (between 1991 and 1995), a total of 3799 participants underwent a standardised medical history and physical examination. Valid FFQ were available for 3418 participants. Participants were excluded from this analysis if they had previously diagnosed (n 212) diabetes, or if their fasting glucose concentrations were ≥ 7·0 mmol/l, or 2 h post-challenge glucose concentrations were ≥ 11·1 mmol/l (n 134). We also excluded those who were taking cholesterol-lowing medications (n 197). After exclusions, the final sample size was 2875 participants (1278 men and 1597 women). The institutional review boards for human research at Boston University and Tufts Medical Center reviewed and approved all study protocols and procedures.
Dietary assessment
Usual dietary intake for the previous year was assessed at the fifth examination using a 126-item semi-quantitative FFQ(Reference Rimm, Giovannucci and Stampfer19). The questionnaires were mailed to participants before the examination, and the participants were asked to bring the completed questionnaire with them to their examination at the Framingham Heart Study. The FFQ consists of a list of foods with a standardised serving size and a selection of nine frequency categories ranging from never or less than one serving per month to more than six servings/d. Separate questions about the use of vitamin and mineral supplements and the type of breakfast cereal most commonly consumed were also included in the FFQ. FFQ with reported energy intakes less than 2·51 MJ/d (600 kcal/d) for men and women, or more than 16·74 MJ/d (4000 kcal/d) for women or more than 17·57 MJ/d (4200 kcal/d) for men, or with more than twelve food items left blank were considered invalid. Nutrient intakes were calculated by multiplying the frequency of consumption of each unit of food from the FFQ by the nutrient content of the specified portion. The relative validity of the FFQ has been reported for both nutrients and foods(Reference Rimm, Giovannucci and Stampfer19, Reference Willett, Sampson and Stampfer20). Individual foods from the 126-item FFQ were collapsed into forty food groups based on the similarity of food and nutrient composition (Table 1). Dietary data were used to calculate the average glycaemic index of a subject's diet, which is the mean glycaemic index of all the food items in the FFQ weighted by the content of carbohydrate from each food item. The dietary guidelines adherence index, a measure of adherence to the key dietary intake recommendations in the 2005 Dietary Guidelines for Americans, was calculated as a measure of diet quality. The score ranged from 2·5 to 17·5 with a mean of 9·1 in the present study population(Reference Fogli-Cawley, Dwyer and Saltzman21).
(Mean values with their standard errors)
a,b,c,d Values within a row with unlike superscript letters are significantly different (P < 0·05; Tukey–Kramer's adjustment for multiple comparisons).
* Adjusted for age and sex.
† Lowest mean in the row.
‡ Highest mean in the row.
Outcome measurements
Height, weight and waist circumference were measured in the standing position by a trained technician. BMI was calculated as weight in kilograms divided by the square of height in meters. Blood samples were drawn after the participants had fasted for at least 8 h for the measurement of fasting glucose and insulin, and lipids. A 75 g oral glucose tolerance test (OGTT) was administered according to WHO standards to measure the 2 h post-challenge glucose and insulin(22). Plasma glucose concentrations were measured in fresh specimens with a hexokinase reagent kit (A-gent glucose test; Abbott Laboratories, Inc., South Pasadena, CA, USA); the intra-assay CV was < 3 %. Plasma insulin concentrations were measured using the Coat-A-Count 125I-radioimmunoassy (Diagnostic Products, Los Angeles, CA, USA); this assay has cross-reactivity with proinsulin at the midcurve of 40 %, the intra- and inter-assay CV were 5–10 %. Insulin sensitivity index (ISI0,120) was calculated using the following formula(Reference Gutt, Davis and Spitzer23):
where
MPG is the mean of fasting and 2 h post-OGTT glucose concentrations (mg/dl) and MSI is the mean of fasting and 2 h post-OGTT insulin concentrations (mU/l).
Serum lipid measures included the enzymic measurement of TAG(Reference McNamara and Schaefer24) and the measurement of the HDL-cholesterol fraction after precipitation of LDL- and VLDL-cholesterol with dextran sulfan magnesium(Reference Warnick, Benderson and Albers25). Blood pressure was measured twice after the participants sat for 5 min or more.
Covariates
Physical activity was measured using a standardised questionnaire to determine estimates of activity based on a 24 h history. A physical activity score was calculated from the number of hours spent doing specific activities that were categorised (sleep, sedentary, slight activity, moderate activity and heavy activity) and weighted according to oxygen consumption required to perform them (metabolic equivalent task (MET)-h per d)(Reference Kannel and Sorlie26). Additional covariate information included age, sex, smoking dose (0, 1–15, 16–25, or >25 cigarettes/d) and years of education.
Statistical analysis
All statistical analyses were performed using the Statistical Analysis Systems statistical software package (version 9.1; SAS Institute, Cary, NC, USA). For cluster analysis, the PROC FASTCLUS procedure was used to derive dietary patterns. We used the K-means method to classify participants into a predetermined number of mutually exclusive, non-overlapping groups (clusters) by comparing Euclidean distances between each subject and each cluster centre. It aims to minimise the differences within clusters while maximising the differences between clusters(Reference Newby, Muller and Tucker27, Reference Wirfalt and Jeffery28). In order to minimise the effect of outliers on the cluster analysis, the FASTCLUS procedure was performed on a predefined number of clusters (n 20). Clusters with less than five individuals were removed and the remaining clusters were used as seeds for the subsequent analysis. The analysis was re-run with varying number of clusters, ranging from three to six using the seeds. In addition the analysis was re-run using sex-specific clusters. However, due to the observation of similar clusters in men and women and lack of any significant sex interaction with respect to our current clusters and outcomes, we chose to retain the clusters that included both men and women to maximise statistical power.
The four-cluster set (Table 1) was selected as the final because it provided the most distinct eating patterns. The food groups, represented by the percentage of energy contributed from each food group, were the input variables in the cluster analysis. We decided to treat the food group variables as a percentage of total energy as this helps to remove extraneous variation due to differences in sex, age, body size and physical activity(Reference Willett, Howe and Kushi29). The ‘low-calorie soda’ (i.e. diet soda) food group was excluded from cluster analysis because it does not provide energy, leaving thirty-nine food groups for inclusion in the cluster analysis. For descriptive purposes only, daily diet soda consumption was presented as servings per d.
To describe food intake across the four clusters (dietary patterns), the age- and sex-adjusted means of energy contribution from each food group were calculated using analysis of covariance (PROC GLM; SAS Institute, Inc.). Analysis of covariance was also used to test differences in nutrient intakes and insulin-resistant phenotypes, including waist circumference, BMI, fasting and 2 h post-challenge insulin, ISI0,120, HDL-cholesterol, TAG, and systolic and diastolic blood pressure, across the four dietary patterns. Tukey–Kramer's method was used to adjust for the multiple comparisons for all possible pairwise comparisons among the four patterns. Many of the markers of the insulin-resistant phenotypes were positively skewed. Consequently, we performed analysis with and without natural logarithmic data transformation for these markers. Since the finding and statistical inference were the same using both the transformed and untransformed data, we present only the results based on the untransformed data to simplified the data presentation. The covariates included age, sex, waist circumference, smoking, physical activity, treatment of hypertension and total energy intake. Additional adjustment for education (used to capture socio-economic status(Reference Cagney and Lauderdale30)) did not change the results.
In order to examine whether the relationships between dietary patterns and insulin-resistant phenotypes were modified by sex or obesity status (BMI ≥ 30 or < 30 kg/m2), we examined first-order interactions for each of these associations. Bonferroni adjustment was used to adjust for the multiple comparisons resulting from the examination of these interactions for multiple outcome variables. No statistically significant interactions were observed for any of these outcome variables.
Results
The mean age of the study participants was 54 years for both men and women. Men had significantly higher waist circumference (98·5 v. 85·6 cm; P < 0·01), BMI (27·9 v. 26·3 kg/m2; P < 0·01), physical activity score (36·1 v. 33·7 metabolic equivalent task (MET)-h per d; P < 0·01), energy intake (8407 v. 7306 kJ/d; P < 0·01), alcohol intake (15·5 v. 7·2 g/d; P < 0·01) and prevalence of hypertension (19·8 v. 16·1 %; P = 0·01) than women. There was no difference between men and women with respect to the proportion of smokers (19·4 % for men and 19·7 % for women).
Four clusters were characterised based on the food or foods that were distinct contributors to total energy intake in each cluster. We named the four clusters as follows: ‘Fruits, Reduced Fat Dairy and Whole Grains’ pattern (n 577), ‘Refined Grains and Sweets’ pattern (n 1846), ‘Beer’ pattern (n 242) and ‘Soda’ pattern (n 210). The ‘Fruits, Reduced Fat Dairy and Whole Grains’ pattern was characterised by a relatively higher energy contribution from reduced-fat dairy, whole grains, fruits, vegetables, fish and seafood, and a lower energy contribution from high-fat dairy, meat, chocolate, and sweet baked foods (Table 2). The ‘Refined Grains and Sweets’ pattern was characterised by a relatively high percentage of energy from sweet baked foods (9 %) and also contained relatively greater energy contributions from high-fat dairy, refined grains, and candy. Energy contributions presented from beer and soda were greatest in the ‘Beer’ and the ‘Soda’ patterns, respectively, and they were the only foods to contribute more than 15 % of energy in any of the four patterns. In addition, the ‘Beer’ and ‘Soda’ patterns both had a relatively lower energy contribution from many healthier food choices, such as reduced-fat dairy, fruits, vegetables, fish, whole grains; the ‘Soda’ pattern also had higher contributions from meat, chocolate and miscellaneous sweets. With respect to diet soda consumption, individuals in the ‘Soda’ pattern consumed significantly less diet soda compared with those in the other dietary patterns.
(Mean values with their standard errors)
DGAI, dietary guidelines adherence index.
a,b,c,d Values within a row with unlike superscript letters are significantly different (P < 0·05; Tukey–Kramer's adjustment for multiple comparisons).
* Adjusted for age, sex and total energy.
† Lowest mean in the row.
‡ Highest mean in the row.
The four dietary patterns differed significantly with respect to sex, nutrient intakes, dietary glycaemic index and diet quality (Table 2). Participants in the ‘Fruits, Reduced Fat Dairy and Whole Grains’ pattern were more likely to be female (67·2 %), and participants in the ‘Beer’ pattern were more likely to be male (87·2 %). Participants in the ‘Fruits, Reduced Fat Dairy and Whole Grains’ pattern had the lowest total energy intake, lowest energy contribution from total fat and saturated fat, lowest cholesterol intake, highest energy contribution from protein intake, highest intake of dietary fibre, highest intake of K, Mg and Ca, and highest overall diet quality. Individuals in the ‘Refined Grains and Sweets’ pattern had the highest energy contribution from total fat, including saturated fat, monounsaturated fat and polyunsaturated fat, and the highest Na and cholesterol intake. Individuals in the ‘Beer’ pattern had the highest alcohol intake, and the lowest energy contribution from carbohydrate and sucrose intakes and glycaemic index. Individuals in the ‘Soda’ pattern had the highest energy intake, highest energy contribution from carbohydrate and sucrose, and glycaemic index. In addition, this dietary pattern had the lowest energy contribution from protein, the lowest intakes of fibre, K, Mg and Ca, and the lowest overall diet quality.
Multivariate-adjusted means of insulin-resistant phenotype markers were compared across the four dietary patterns (Table 3). Compared with individuals in the ‘Fruits, Reduced Fat Dairy and Whole Grains’ pattern, those in the ‘Refined Grains and Sweets’ pattern had significant higher waist circumference and BMI after adjusting for multiple comparisons and potential confounders, including age, sex, smoking, physical activity, treatment of hypertension and total energy intake. In comparison with individuals in the ‘Fruits, Reduced Fat Dairy and Whole Grains’ pattern, fasting insulin concentrations were significantly higher in those individuals in the ‘Soda’ pattern. In comparison with individuals in the ‘Fruits, Reduced Fat Dairy and Whole Grains’ pattern, HDL-cholesterol concentrations were significantly higher among those in the ‘Beer’ pattern. No significant associations were found between dietary pattern and ISI0,120, TAG, or systolic and diastolic blood pressure.
(Mean values with their standard errors)
a,b,c,d Values within a row with unlike superscript letters are significantly different (P < 0·05; Tukey–Kramer's adjustment for multiple comparisons).
* Adjusted for age, sex, smoking, physical activity, treatment of hypertension and total energy.
† Lowest mean in the row.
‡ Highest mean in the row.
§ Adjusted for age, sex, waist circumference, smoking, physical activity, treatment of hypertension and total energy intake.
∥ Excluded those taking blood pressure medications.
Discussion
In this cross-sectional study of healthy adults in the Framingham Offspring Study, four dietary patterns were derived using cluster analysis (‘Fruits, Reduced Fat Dairy and Whole Grains’, ‘Refined Grains and Sweets’, ‘Beer’ and ‘Soda’). The ‘Refined Grains and Sweets’ dietary pattern comprised the largest number of individuals, while the ‘Soda’ pattern comprised the smallest number. The four patterns derived in the present study are similar to those identified using cluster analysis in other American populations(Reference Tucker, Dallal and Rush31–Reference Liese, Schulz and Moore34). For example, a ‘heart healthy’ pattern and an ‘empty calorie’ pattern, which resemble our ‘Fruits, Reduced Fat Dairy and Whole Grains’ and ‘Soda’ patterns, were identified as two of six patterns in women participating at the third examination cycle of the Framingham Offspring Study(Reference Millen, Quatromoni and Copenhafer33). Consistent with the present study, a ‘healthy’ and ‘alcohol’ pattern were identified among five patterns using cluster analysis in 459 men and women in the Baltimore Longitudinal Study of Aging(Reference Newby, Muller and Hallfrisch32). Often, an ‘alcohol’ pattern is captured in dietary pattern analysis, particularly when men are included in the analysis(Reference Newby, Muller and Hallfrisch32, Reference Tucker, Chen and Hannan35, Reference Millen, Quatromoni and Gagnon36). In the present study, beer was the predominant contributor to alcohol intake and thus the rationale for naming it the ‘Beer’ dietary pattern. Although dietary patterns derived in different studies are assigned different names, they often reflect similar food behaviours and nutrient profiles. For instance, our ‘Fruits, Reduced Fat Dairy and Whole Grains’ pattern is comparable in nutrient content with the ‘dark bread, rice and pasta, vegetables’ pattern identified in the Insulin Resistance Atherosclerosis Study(Reference Liese, Schulz and Moore34) and a ‘milk, cereals, and fruits’ pattern identified among elderly Boston area residents(Reference Tucker, Dallal and Rush31).
In the present study, compared with individuals in the ‘Fruits, Reduced Fat Dairy and Whole Grains’ dietary pattern, those in the ‘Refined Grains and Sweets’ dietary pattern had significantly higher waist circumference and BMI. The ‘Refined Grains and Sweets’ pattern also had a relatively higher energy contribution from high-fat dairy, meat, refined grains, sweet baked foods, candy, and had a lower energy contribution from reduced-fat dairy, fruits, vegetables and whole grains compared with the ‘Fruits, Reduced Fat Dairy and Whole Grains’ pattern. This result is consistent with previous findings that a diet high in refined grains, high-fat dairy, meat and sweets, and low in whole grains, fruits and vegetables, was associated with a high risk of obesity(Reference Esmaillzadeh and Azadbakht37, Reference Murtaugh, Herrick and Sweeney38). Prospective studies also found that consuming a diet high in fruits, vegetables, reduced-fat dairy and whole grains was associated with lower increases in BMI and waist circumference(Reference Newby, Muller and Hallfrisch32, Reference Quatromoni, Copenhafer and D'Agostino39–Reference Mendez, Popkin and Jakszyn42). In the present study, none of the dietary patterns was associated with improved blood pressure after the exclusion of those on hypertensive medication.
We found individuals in the ‘Soda’ pattern had significantly higher fasting insulin concentrations than those in the ‘Fruits, Reduced Fat Dairy and Whole Grains’ pattern, consistent with our previous finding that the consumption of sugar-sweetened drinks is positively associated with fasting insulin(Reference Yoshida, McKeown and Rogers43). Individuals in the ‘Soda’ pattern also had a lower dietary fibre intake and a higher dietary glycaemic index. The protective effect of dietary fibre on insulin resistance(Reference McKeown, Meigs and Liu4, Reference Liese, Schulz and Fang6, Reference Buyken and Liese44), the metabolic syndrome(Reference McKeown, Meigs and Liu5) and type 2 diabetes mellitus(Reference Liese, Schulz and Moore34, Reference Salmeron, Manson and Stampfer45) has been well documented. McKeown et al. (Reference McKeown, Meigs and Liu46) observed that a higher dietary glycaemic index was unfavourably associated with surrogate measures of insulin resistance. In contrast, no association was found between dietary glycaemic index and measures of insulin resistance in either the Insulin Resistance Atherosclerosis Study(Reference Liese, Schulz and Fang6) or the Zutphen Elderly Study(Reference van Dam, Visscher and Feskens47). Inconsistencies between studies may in part be attributed to the different methods used to determine insulin sensitivity, the different populations studied, and methods of dietary assessment.
Some, but not all, observational studies have found a significant association between sugar-sweetened soft drinks and obesity(Reference Ludwig, Peterson and Gortmaker48–Reference Dhingra, Sullivan and Jacques50). Using cluster analysis, Wirfalt & Jeffery(Reference Wirfalt and Jeffery28) observed that individuals in the ‘soft drinks’ pattern had a significantly higher mean BMI compared with individuals in the‘skim milk’ and‘meat–cheese’ patterns. The ‘Soda’ pattern had the highest energy contribution from carbohydrate and the highest glycaemic index. Previous studies have demonstrated that high-carbohydrate diets contribute to decreasing HDL-cholesterol and increasing TAG concentrations(Reference Reaven51, Reference Reaven52). Findings from several observational studies have consistently found that glycaemic index is inversely related to HDL-cholesterol concentration(Reference Frost, Leeds and Dore53, Reference Ford and Liu54).
Using data from an earlier examination in the Framingham Offspring Cohort, Sonnenberg et al. observed that obese and non-obese women in the ‘empty calories’ cluster had a higher prevalence of the metabolic syndrome(Reference Sonnenberg, Pencina and Kimokoti14). Despite differences in the type of FFQ administered and the defining of the clusters, this dietary pattern characterised by a higher intake of sugar-sweetened beverages and lower intakes of dietary fibre and vegetables is similar to the ‘Soda’ dietary pattern derived in the present study. Individuals in the ‘Soda’ pattern displayed several metabolic syndrome abnormalities, i.e. higher waist circumference and TAG and lower HDL-cholesterol. However, these differences were not statistically significant compared with the ‘Fruits, Reduced Fat Dairy and Whole Grains’ pattern, in part due to the small number of subjects in this pattern (n 210).
Two prospective studies have observed that diet soda consumption is positively associated with the metabolic syndrome(Reference Lutsey, Steffen and Stevens16, Reference Dhingra, Sullivan and Jacques50). It is possible that diet soda consumption may be associated with a particular dietary pattern, or alternatively a reflection of a dietary change due to some underlying metabolic disease. In the present study, only those individuals in the ‘Soda’ cluster differed with respect to diet soda consumption, i.e. they consumed significantly less diet soda compared with the other clusters. We previously found that diet soda consumption was not associated with any surrogate measures of insulin resistance after adjustment for potential confounders(Reference Yoshida, McKeown and Rogers43).
In the present study, individuals in the ‘Beer’ pattern had significant higher HDL-cholesterol concentration compared with those in the ‘Fruits, Reduced Fat Dairy and Whole Grains’ pattern. The inverse association between alcohol consumption and CHD has been observed in many observational studies around the world(Reference Lucas, Brown and Wassef55). It is estimated that approximately 50 % of the benefit of moderate alcohol consumption could be explained by the direct effect of alcohol on HDL-cholesterol(Reference Langer, Criqui and Reed56, Reference Rimm and Moats57). A meta-analysis of experimental studies found that a dose of 30 g alcohol per d increased the concentration of HDL-cholesterol by 39·9 (95 % CI 32·5, 47·3) mg/l(Reference Rimm, Williams and Fosher58). Although a higher beer intake was associated with a better HDL-cholesterol level, individuals in the ‘Beer’ pattern had a poorer overall diet, for instance, fewer fruits, vegetables, whole grains, and less reduced-fat dairy. In addition, individuals in the ‘Beer’ pattern had a significantly lower overall diet quality score compared with individuals in the ‘Fruits, Reduced Fat Dairy and Whole Grains’ pattern.
Strengths of the present study include its large sample size and inclusion of multiple insulin-resistant phenotypes measured in a clinical setting. The main limitation is the cross-sectional study design. Since the diet and insulin-resistant phenotypes data were collected at the same time, the causal relationship cannot be assessed due to uncertainty regarding the timing of exposure and outcome, so further prospective studies are needed to examine the effect of diet on insulin resistance using dietary patterns to characterise exposures. Although we adjusted for physical activity, residual confounding caused by lifestyle behaviours and social economic status factors may have arisen in the present study. The characterisation of the food groups used in the present study is limited by the dietary data obtained by self-reported FFQ. Although the FFQ has limitations in estimating absolute intake for individuals, it is a feasible and valid method to rank (or differentiate) individuals according to their usual diet. Differences in food and nutrient intake profile across clusters, as depicted in Tables 1 and 2, demonstrate the utility of the cluster analysis for discriminating dietary exposure within the cohort. In addition, dietary patterns derived by cluster analysis using FFQ data have been validated against 3 d food records specifically in the Framingham Offspring Study(Reference Millen, Quatromoni and Copenhafer33, Reference Quatromoni, Copenhafer and Demissie59). Because the participants of the Framingham Offspring Study are predominantly white Americans, the results from the present study may not be readily generalised to other populations who have different dietary behaviours.
In conclusion, our findings suggest that the consumption of a diet rich in fruits, vegetables, whole grains and reduced-fat dairy protects against insulin-resistant phenotypes; displacing these healthy choices with meat, refined grains, high-fat dairy, sweet baked foods, candy and sugar-sweetened soda promotes insulin-resistant phenotypes.
Acknowledgements
The present study was supported in part by the US Department of Agriculture, under agreement no. 58-1950-7-707, and with support from the Framingham Heart Study of the National Heart Lung and Blood Institute of the National Institutes of Health (contract no. N01-HC-25195). J. B. M. is supported by the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) K24 DK080140. R. S. V. is supported in part by 2K24 HL04334. N. M. is supported in part by a grant from the General Mills Bell Institute of Health and Nutrition.
Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the view of the US Department of Agriculture.
We are grateful to the Framingham Study participants and staff for data collection. We thank Gail Rogers for data management.
The authors' responsibilities were as follows. E. L. designed and performed the statistical analyses and drafted the manuscript. P. F. J., N. M. M. and P. K. N. contributed to the design of the analyses, the interpretation of the data, and the writing of the manuscript. J. B. M., P. A. Q. and R. S. V. contributed to the interpretation of the data and the writing of the manuscript. R. B. D. served as liaison with Framingham Study investigators, provided data, and contributed to the interpretation of the data and the writing of the manuscript. All authors participated in the revision and approval of the manuscript.
N. M. M. has served as an independent consultant for Cadbury. None of the other authors has any conflict of interest.
The supplementary material for this article can be found at http://www.journals.cambridge.org/bjn
