Increase Font Size Decrease Font Size View as PDF Print

What is the relationship between adherence to dietary guidelines/recommendations or specific dietary patterns, assessed using reduced rank regression analysis, and cardiovascular disease?


Insufficient evidence due to a small number of studies was available to examine the relationship between dietary patterns derived using reduced rank regression and risk of cardiovascular disease. The disparate nature of the methods used made it difficult to compare results; therefore, no conclusions were drawn.


IV – Not Assignable


Key Findings

  • Four positive quality prospective cohort studies that used reduced rank regression to examine the relationship between dietary patterns and cardiovascular disease (CVD) status were included in this review. Comparison across studies is limited by the small number of studies, differences in methodologies used and in the populations studied. Therefore no conclusions were drawn.
  • More U.S. population-based research is needed to examine dietary patterns and risk of CVD using reduced rank regression, preferably with more consistent methods and response variables.

Evidence Summary Overview

Description of the Evidence

Four prospective cohort studies included in this systematic review (Drogan, 2007; Heroux, 2009; McNaughton, 2009; Meyer, 2011) used reduced rank regression (RRR) analysis (see appendix A) to examine the relationship between dietary patterns and CVD. All of the studies were rated positive quality. One study each was conducted in the United States, United Kingdom and Germany, and one included subjects from Europe. The sample sizes for these studies ranged from 981 participants to 26,238 participants (one study had less than 1,000 participants, one study had less than 8,000, one study had less than 14,000 and one study had more than 26,000). All four studies were conducted in adults. Three of the studies included both females and males, while one study included only males (Meyer, 2011). 
The studies in this review used different dietary assessment methods, including three-day diet records (Heroux 2009), a self-administered food-frequency questionnaire (FFQ) (Drogan 2007), a 127-item validated FFQ (McNaughton, 2009) and a seven-day dietary record (Meyer, 2011). The studies also examined a variety of CVD-related outcomes:
  • Drogan 2007 examined CVD morbidity and mortality
  • Heroux 2009 measured CVD disease and all-cause mortality
  • McNaughton 2009 examined death due to CVD and non-fatal incident of CHD
  • Meyer 2011 examined incidence of fatal or non-fatal MI and sudden cardiac death or mortality from CHD.
The independent variables in all four studies were dietary pattern scores derived using RRR analysis. Three studies used biomarkers as response variables (Heroux, 2009; McNaughton, 2009; Meyer 2011), while the fourth study used nutrients as response variables (Drogan, 2007). The response variables used and the respective dietary patterns extracted for each study are described in more detail below:

Evidence Summary Paragraphs

Drogan 2007 selected total fat, total carbohydrates and fiber as response variables. Three dietary patterns were extracted. Pattern 1 explained the greatest variation in all three response variables (53%). Patterns 2 and 3 explained only 21% and 10% of the variation. Only Pattern 1 was used in subsequent analyses to calculate a pattern score for each subject and included foods such as whole-grain bread, fresh fruit, fruit juices, grains, cereals and raw vegetables.
Heroux 2007 selected BMI, blood pressure, total blood cholesterol, HDL-cholesterol, triglycerides (mg per dL), fasting glucose, uric acid and white blood cell count as response variables. Five dietary patterns were extracted, which together explained 5.66% of the variation within the total biomarker index. Pattern 1 explained 4.33% of the overall variation, with the other four patterns only explaining an additional 1.33% between them. Thus, Pattern 1 accounted for 76.5% of the total variation and the pattern was labeled “Unhealthy Eating Index.” Pattern 1 was characterized by elevated consumption of processed and red meat, white potato products, non-whole grains, added fat and reduced consumption of non-citrus fruits.
McNaughton 2009 selected total cholesterol, HDL-cholesterol, and triglycerides as response variables. Three dietary patterns were extracted, and Patterns 1 and 2 explained the most variation in the response variables (Dietary Pattern 1 explained 7.14% of variation in HDL-cholesterol and 5.3% of variation in triglycerides, while dietary Pattern 2 explained 3.5% of variation in total cholesterol) and were used in subsequent analyses. Pattern 1 was characterized by high intakes of white bread, fried potatoes, sugar in tea and coffee, burgers and soft drinks and lower consumption of salad dressings and vegetables. Pattern 2 was characterized by higher consumption of red meat, cabbage, Brussels sprouts and cauliflower, and lower consumption of whole meal bread, jam, marmalade, tofu, buns, cakes, pastries, fruit pies and margarine.
Meyer 2011 selected C-reactive protein, Interleukin (IL)-6, and Interleukin (IL)-18 as response variables. Pattern 1 showed a high score of the RRR-derived pattern characterized by high intakes of meat, soft drinks and beer and low intakes of vegetables, fresh fruit, chocolates, cake, pastries, whole meal bread, cereals, muesli, curd, condensed milk, cream, butter, nuts, sweet bread spread and tea.   

Table 4-C-III-1 Summary of Findings

Studies examining the combinations of food intake (assessed using RRR) explain the most variation in risk of CVD.

Qualitative Assessment of the Body of Evidence

This review included four positive-quality prospective cohort studies. However, because there were so few studies available, variability in the methodology used in the studies that were reviewed and populations considered, there was insufficient information from which to assess consistency or draw conclusions about the relationship between dietary patterns derived using RRR and risk of CVD.

Limitations of the Evidence

Methodological Differences
  • Three out of the four studies used biomarkers and the fourth study used nutrients as response variables in the RRR analyses. Among the three studies that used biomarkers as response variables, there were differences in the type of biomarkers chosen, leading to the identification of dietary patterns that differed from study to study. As response variables, Heroux (2009) used change in BMI, mean arterial pressure, total blood cholesterol, HDL-cholesterol, triglycerides (mg per dL), fasting glucose and uric acid; Meyer (2011) used C-reactive protein, Interleukin (IL)-6 and Interleukin (IL)-18; and McNaughton (2009) used total cholesterol, HDL-cholesterol and triglycerides. The fourth study, Drogan (2007), used nutrients including total fat, total carbohydrate and fiber. Because the dietary patterns described in each study are directly linked to response variables chosen, the variation in the response variables used means that the resulting dietary patterns may not be comparable.
  • Dietary assessment methods were different across the studies. One study used three-day diet records (Heroux 2009); another used a self-administered FFQ (Drogan 2007); a third used a 127-item validated FFQ (McNaughton 2009); and the fourth study used a seven-day dietary record (Meyer 2011). It is unclear what impacts different dietary assessment methods have on the derivation of dietary patterns using reduced rank regression.
  • The studies were not consistent in their use of confounders in analyses. In particular, physical activity was not included as a confounder in the analyses by Meyer (2011), and Drogan (2007) did not include smoking as a confounder.
Population Differences

The studies were conducted in different countries (United States and several countries in Europe) and represented populations in different regions of the world, which limited the ability to compare and interpret the results due to potential differences in dietary patterns between these regions. From that perspective, the results may not be generalizable to some U.S. populations.

Research Recommendations

More research using reduced rank regression should be conducted. Additionally, standardization in methodology, such as food groupings and response variables used, are also needed.


  1. Drogan D, Hoffmann K, Schulz M, Bergmann MM, Boeing H, Weikert C. A food pattern predicting prospective weight change is associated with risk of fatal but not with nonfatal cardiovascular disease.  J Nutr. 2007 Aug;137(8):1961-7. PubMed PMID: 17634271.
  2. Héroux M, Janssen I, Lam M, Lee DC, Hebert JR, Sui X, Blair SN. Dietary patterns and the risk of mortality: impact of cardiorespiratory fitness. Int J Epidemiol. 2010 Feb;39(1):197-209. Epub 2009 Apr 20. PubMed PMID: 19380370; PubMed Central PMCID: PMC2912488
  3. McNaughton SA, Mishra GD, Brunner EJ. Food patterns associated with blood lipids are predictive of coronary heart disease: the Whitehall II study. Br J Nutr. 2009 Aug;102(4):619-24. Epub 2009 Mar 30. PubMed PMID: 19327192; PubMed Central PMCID: PMC2788758.
  4. Meyer J, Döring A, Herder C, Roden M, Koenig W, Thorand B. Dietary patterns, subclinical inflammation, incident coronary heart disease and mortality in middle-aged men from the MONICA/KORA Augsburg cohort study. Eur J Clin Nutr. 2011 Jul;65(7):800-7. doi: 10.1038/ejcn.2011.37. Epub 2011 Apr 6. PMID: 21468094.

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