Increase Font Size Decrease Font Size View as PDF Print

Are prevailing patterns of diet behavior in a population, assessed using factor or cluster analysis, related to risk of type 2 diabetes?

Conclusion

Limited and inconsistent evidence from epidemiological studies indicates that in adults, dietary patterns derived using factor or cluster analysis, characterized by vegetables, fruits, and low-fat dairy products tend to have an association with decreased risk of type 2 diabetes and those patterns characterized by red meat, sugar-sweetened foods and drinks, French fries, refined grains, and high-fat dairy products tended to show an increased association for risk of type 2 diabetes. Among studies, there was substantial variation in food group components and not all studies with similar patterns showed significant association.
 

Grade

III– Limited

 

Key Findings

  • Cluster and factor analyses are data-driven approaches that describe the dietary patterns consumed by the study population. High variability in the studies included in this review, including populations, case number, sample size, dietary assessment techniques, methods used to define and retain factors and clusters, confounders considered and the statistical analysis employed, made comparisons among studies challenging.
  • Studies focused on intermediate outcomes were too few and too diverse in methodology to draw a conclusion.

Evidence Summary Overview

Description of the Evidence
Factor and cluster analyses are data-driven approaches that empirically derive food intake patterns (appendix A). Fifteen prospective cohort studies conducted between 2004 and 2012 were included. Seven studies received a positive quality and eight a neutral quality rating. Sample sizes ranged from 690 to 75,512 participants (2 studies <2,000; 6 studies 3,000 to 6,500; 4 with 20,000 to 45,000; and 3 studies with >65,000 subjects). Study duration ranged from 4 to 23 years (6 <10 years; 7 between 10 to 15 years; and 2 >15 years). Eight studies were conducted in the United States and two in Japan, while the remaining were conducted in Australia, Finland, Hong Kong, the Netherlands, and the United Kingdom.
 
Population: Eleven studies were conducted in both men and women (Bauer, 2012; Brunner, 2008; Duffey, 2012; Erber, 2009; Hodge, 2007; Lau, 2009; Montonen, 2005; Morimoto, 2012; Nanri, 2013; Nettleton, 2008; Yu, 2011), and two of these studies analyzed health outcomes separately by gender (Erber, 2009; Nanri, 2013).Three studies included U.S. women only (Fung, 2004; Kimokoti, 2012; Malik, 2012), and one study included U.S. men only (Van Dam, 2002). Age range at baseline spanned from 18 to 84 years. Fourteen studies analyzed middle-aged and older populations (above 35 years), and one study analyzed young adults 18 to 30 years at baseline (Duffey, 2012). Three studies identified race/ethnic subgroups within their cohort (Erber, 2009; Hodge, 2007; Nettleton, 2008).
Dietary Pattern Methodology: Thirteen of fifteen studies assessed dietary intake using a baseline food frequency questionnaire (FFQ), and two studies from the Nurses’ Health Study aggregated data from FFQs completed at four separate time points (Fung, 2004; Malik, 2012). Two studies used a diet history approach (Duffey, 2012; Montonen, 2005). In general, individual food and beverage items were consolidated into food groups based on established criteria, and dietary patterns were then generated using factor analysis in 12 studies (Bauer, 2012; Erber, 2009; Fung, 2004; Hodge, 2007; Lau, 2009; Malik, 2012; Montonen, 2005; Morimoto, 2012; Nanri, 2013; Nettleton, 2008; Van Dam, 2002; Yu, 2011), and cluster analysis in 3 studies (Brunner, 2008; Duffey, 2012; Kimokoti, 2012). Once dietary patterns were defined using factor analysis, pattern scores were calculated for each participant and after multivariate adjustment, the association between dietary pattern scores and type 2 diabetes risks by quintile or quartile were assessed. Generally, studies adjusted for baseline BMI, total energy intake, physical activity, sex, age, and smoking, but additional factors were considered by individual studies. Only Nettleton et al. (2008) controlled for change in body weight or waist circumference over the course of the study. When cluster analysis was used, a reference group was defined and analysis was conducted to assess the relationship between cluster group and risk of T2D.
 
Outcomes: The studies in this body of evidence evaluated associations between dietary patterns and endpoint outcomes and intermediate outcomes.
 
Endpoint Outcome: Twelve studies evaluated the association between dietary patterns and incidence of T2D. Only one of these studies used cluster analysis to define dietary patterns (Brunner, 2008). Factor and cluster name reflect those assigned by the author, followed by food components characteristic of the pattern.
 
Intermediate Outcome: One study measured fasting blood glucose with a cutoff of ≥6.1 mmol/L (Duffey, 2012); another study measured plasma glucose with a cutoff of ≥5.1 mmol/L (Kimokoti, 2012), while a third study measured plasma glucose after an overnight fast and after a standard 75 g oral glucose tolerance test (Lau, 2009).  Table 2 provides a general overview of the study characteristics, dietary assessment methods, dietary patterns identified using factor and cluster analysis, and their association with plasma glucose levels. Factor and cluster name reflect those assigned by the author, followed by food components characteristic of the pattern.
 
Themes
T2D incidence: Twelve prospective cohort studies examined dietary patterns and their association with T2D incidence (table 4-C-II-1). Eleven studies used factor analysis and one used cluster analyses to identify one to four dietary patterns per study and a total of 35 diverse dietary patterns within the body of evidence. Studies ranged in size from 690 to 75,512 subjects, were conducted in the United States (five), Japan (two), the Netherlands, Australia, Finland, China, and the United Kingdom, and ranged in duration from 4 to 23 years. Results were mixed. There were many null findings, particularly among studies with duration of less than 7 years (Malik, 2012; Hodge, 2007; Nanri, 2012; Nettleton, 2008). Patterns associated with lower risk of T2D were characterized by vegetables, fruits, low-fat dairy products, and whole grains, and those associated with increased risk of T2D were characterized by red meat, sugar-sweetened foods and drinks, French fries, refined grains, and high-fat dairy products. However, there was substantial variation in the food groups identified, even among patterns with the same name.
 
Intermediate outcomes: Three prospective cohort studies assessed the association between dietary patterns and plasma glucose levels (table 4-C-II-2). Two U.S. studies derived patterns using cluster analysis (Duffy, 2012; Kimokoti, 2012) and one study conducted in Denmark used factor analysis (Lau, 2009). Studies ranged in size from 1,146 to 5,824 adults. Duffey et al. (2012) identified two diet clusters: “Prudent Diet” and “Western Diet”; Kimokoti et al. (2012) identified five clusters: “Heart Healthier,” “Lighter Eating,” “Wine and Moderate Eating,” “Higher Fat,” and “Empty Calories”; and Lau et al. (2009) derived two factors: “Modern” and “Traditional.” Variations in population, dietary assessment methodologies, and methods used to derive patterns resulted in a highly variable set of dietary patterns, making it difficult to draw conclusions. No association with fasting plasma glucose was found with any of the nine dietary patterns identified. Lau (2009) assessed 2-hour plasma glucose concentration and found a dietary pattern characterized by high intake of vegetables, fruit, mixed vegetable dishes, vegetable oil and vinegar dressing, poultry, pasta, rice, and cereals associated with decreased T2D risk.
 
Table 4-C-II-1 Summary of Findings 
Table 4-C-II-2 Summary of Findings
Dietary patterns identified using factor (white rows) or cluster (colored rows) analysis and association with incidence of type 2 diabetes (T2D) in adults
 

Qualitative Assessment of the Collected Evidence

Quality and Quantity
Twelve prospective cohort studies evaluated the association between dietary patterns and incidence of T2D. Six studies were found to be of positive quality, and six received a neutral quality rating. Studies ranged in size from 690 to 75,512 subjects. Three prospective cohort studies with 1,146 to 5,824 subjects evaluated the association between dietary patterns and intermediate outcomes, specifically fasting plasma glucose. Fasting plasma glucose criteria varied between studies and only one study (Lau, 2009) measured post-challenge 2-hour plasma glucose, in addition to a fasting measure.
 
Consistency
Among studies that showed significant associations, there were substantial variations in food group components and not all studies with similar patterns showed significant associations. There were many null findings, particularly among shorter studies (less than 7 years). In general, dietary patterns derived using factor or cluster analysis, characterized by vegetables, fruits, and low-fat dairy products tended to have an association with decreased risk of type 2 diabetes, and those patterns characterized by red meat, sugar-sweetened foods and drinks, French fries, refined grains, and high-fat dairy products tended to show an increased association for risk of type 2 diabetes in adults.
 
Impact
It is challenging to infer public health implications from these studies, since the results from cluster analysis and factor analysis are based on a specific population and hard to translate into detailed dietary prescriptions, aside from broad generalizations. Results were clinically meaningful for associations made between factors and clusters and incidence of T2D. The methodologies used in the included studies varied substantially. Patterns using the same naming convention may contain very different foods or groups of foods (e.g., a pattern labeled “prudent” may or may not contain fish, nuts, legumes, whole grains, poultry, or low-fat dairy products). Variations in the number of study subjects, cases, and subjective decisions involved in deriving and retaining factors and clusters for analysis likely influence the ability to detect associations.
 
Generalizability/External Validity  
Studies recruited adult populations, and both men and women were well represented. Eight of twelve studies assessing T2D incidence were conducted outside the United States and one of three assessing intermediate outcomes. The majority of studies included White populations. Ethnicity and socioeconomic status were often not reported in analyses.
 

Limitations of the Evidence

  • Variation in methodology used to derive and analyze dietary patterns (e.g., factor versus cluster analysis, subjective decisions regarding groupings of foods, number of patterns retained and naming conventions, population characteristics, sample size and case numbers) make the analysis challenging. Even factors with the same naming convention (e.g., “vegetable” or “prudent”) included somewhat different foods or groups of foods.
  • Patterns derived from either factor or cluster analysis may not be reproducible because of variations in populations, sample sizes, dietary assessment methods, and decisions made to define food variables used in factor and cluster analysis, and factors and clusters differ across studies.
  • Differences in the statistical analysis approaches used to derive and retain factors and clusters influences power and the ability to detect an association.
  • Patterns derived from factor analysis and cluster analyses were analyzed differently. In factor analysis, “high” scores were generally compared with “low” scores of the same pattern, though it was not clear what characteristic differences there were in a “high” versus “low” score factor. In cluster analysis, one cluster was compared with another one, making it difficult to interpret results together.
  • Dietary patterns with significant association should not be construed as the best or worst possible diet associated with diabetes risk.
  • Most longitudinal studies included only baseline measure of dietary intake and did not account for changes in subject’s diets, availability, and variations in the food supply, which may have influenced the food components of patterns. Food frequency questionnaires may not accurately capture important elements of the diet.

Research Recommendations

  • Evaluate and standardize methods used to assess, organize, aggregate, and adjust food variables to facilitate interpretation of findings across studies.
  • Additional research is needed to examine if and how gender, age, SES, and ethnicity might influence the relationship between dietary patterns and risk for T2D.
  • Consider important confounders that may modify or explain the association between dietary intake and T2D, for example weight change.
 

References

  1. Bauer F, Beulens JW, van der A DL, Wijmenga C, Grobbee DE, Spijkerman AM, van der Schouw YT, Onland-Moret NC. Dietary patterns and the risk of type 2 diabetes in overweight and obese individuals. Eur J Nutr. 2012 Jul 28. [Epub ahead of print] PubMed PMID: 22972436.
  2. Brunner EJ, Mosdøl A, Witte DR, Martikainen P, Stafford M, Shipley MJ, Marmot MG. Dietary patterns and 15-y risks of major coronary events, diabetes, and mortality. Am J Clin Nutr. 2008 May;87(5):1414-21. PubMed PMID: 18469266.
  3. 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. PubMed PMID: 22378729; PubMed Central PMCID: PMC3302365.
  4. Erber E, Hopping BN, Grandinetti A, Park SY, Kolonel LN, Maskarinec G. Dietary patterns and risk for diabetes: the multiethnic cohort. Diabetes Care. 2010 Mar;33(3):532-8. Epub 2009 Dec 10. PubMed PMID: 20007939; PubMed Central PMCID: PMC2827503.
  5. Fung TT, Schulze M, Manson JE, Willett WC, Hu FB. Dietary patterns, meat intake, and the risk of type 2 diabetes in women. Arch Intern Med. 2004 Nov 8;164(20):2235-40. PubMed PMID: 15534160.
  6. Hodge AM, English DR, O'Dea K, Giles GG. Dietary patterns and diabetes incidence in the Melbourne Collaborative Cohort Study. Am J Epidemiol. 2007 Mar 15;165(6):603-10. Epub 2007 Jan 12. PubMed PMID: 17220476.
  7. Kimokoti RW, Gona P, Zhu L, Newby PK, Millen BE, Brown LS, D'Agostino RB, Fung TT. Dietary patterns of women are associated with incident abdominal obesity but not metabolic syndrome. J Nutr. 2012 Sep;142(9):1720-7. doi: 10.3945/jn.112.162479. Epub 2012 Jul 25. PubMed PMID: 22833658; PubMed Central PMCID: PMC3417833.
  8. Lau C, Toft U, Tetens I, Carstensen B, Jørgensen T, Pedersen O, Borch-Johnsen K. Dietary patterns predict changes in two-hour post-oral glucose tolerance test plasma glucose concentrations in middle-aged adults. J Nutr. 2009 Mar;139(3):588-93. Epub 2009 Jan 21. PubMed PMID: 19158222. (Glucose tolerance)
  9. Malik VS, Fung TT, van Dam RM, Rimm EB, Rosner B, Hu FB. Dietary patterns during adolescence and risk of type 2 diabetes in middle-aged women. Diabetes Care. 2012 Jan;35(1):12-8. Epub 2011 Nov 10. PubMed PMID: 22074723; PubMed Central PMCID: PMC3241320.
  10. Montonen J, Knekt P, Härkänen T, Järvinen R, Heliövaara M, Aromaa A, Reunanen A. Dietary patterns and the incidence of type 2 diabetes. Am J Epidemiol. 2005 Feb 1;161(3):219-27. PubMed PMID: 15671254.
  11. Nettleton JA, Steffen LM, Ni H, Liu K, Jacobs DR Jr. Dietary patterns and risk of incident type 2 diabetes in the Multi-Ethnic Study of Atherosclerosis (MESA). Diabetes Care. 2008 Sep;31(9):1777-82. Epub 2008 Jun 10. PubMed PMID: 18544792; PubMed Central PMCID: PMC2518344.
  12. Van Dam RM, Rimm EB, Willett WC, Stampfer MJ, Hu FB. Dietary patterns and risk for type 2 diabetes mellitus in U.S. men. Ann Intern Med. 2002 Feb 5;136(3):201-9. PubMed PMID: 11827496.
  13. Yu R, Woo J, Chan R, Sham A, Ho S, Tso A, Cheung B, Lam TH, Lam K. Relationship between dietary intake and the development of type 2 diabetes in a Chinese population: the Hong Kong Dietary Survey. Public Health Nutr. 2011 Jul;14(7):1133-41. Epub 2011 Apr 5. PubMed PMID: 21466742.
  14. Morimoto A, Ohno Y, Tatsumi Y, Mizuno S, Watanabe S. Effects of healthy dietary pattern and other lifestyle factors on incidence of diabetes in a rural Japanese population. Asia Pac J Clin Nutr. 2012;21(4):601-8. PubMed PMID: 23017319.
  15. Nanri A, Shimazu T, Takachi R, Ishihara J, Mizoue T, Noda M, Inoue M, Tsugane S. Dietary patterns and type 2 diabetes in Japanese men and women: the Japan Public Health Center-based Prospective Study. Eur J Clin Nutr. 2013 Jan;67(1):18-24. doi: 10.1038/ejcn.2012.171. Epub 2012 Oct 24. PubMed PMID: 23093343.



Research Design and Implementation
For a summary of the Research Design and Implementation results, click here.