“Trying to measure dietary exposures is very difficult,” sayd McKeown. The nature of nutrition is extremely complex, and she quotes Walter Willet on the complexity (which makes me like her already). “A single nutrient may be confounded by an overall dietary pattern.”
Single-nutrient approaches ignore complexity of diets, biological interactions, and there is difficulty to detect small effects and observe health effects of single components.
How do we research with this complexity? We may be interested in a dietary pattern, a food group, an individual food, single nutrients, and bioactives (a top-down approach).
Why study dietary patterns? They represent the interactions and cumulative effects of dietary components on disease risk. They capture potential foods and nutrient synergy. They help generate hypotheses about biological mechanisms, and they can translate to dietary pattern recommendations.
In epidemiology, there are two ways to derive dietary pattern approaches are theoretical (hypothesis-oriented) and empirical (exploratory-oriented). “The data drives the patterns,” she says. Most of the time it’s based on food frequency questionnaires. The input variables may be frequency, weight, daily percent of energy contribution.
With a factor analysis, the goal is to identify common factors that explain variance in the dietary data. it’s based on correlation/covariance matrix of the food groups. It aggregates specific food groups on the basis of degree of use.
She then shows examples where various food groups are weighted based on factor loading of top contributing foods. A Western dietary pattern, for example, is higher in meat, processed meat, and butter in comparison to a prudent dietary pattern.
Cluster analysis (which you may “hear a lot about in the literature”) is based on aggregates of individuals into distinct food groups. It’s sensitive to extreme outliers so treatment of input variables is important.
She gives a few examples of cluster analysis, such as:
Cluster 1: reduced-fat dairy, fruits, and whole-grains.
Cluster 2: refined grains and sweets.
Cluster 3: Beer (lots of men were in this category).
Cluster 4: Soda (people in this category were found to have higher fasting insulin)
From a research standpoint, there have been a few diets with great interest including the Mediterranean diet and low-carb diet. There have been indexes developed for these two diets as well as the Healthy Eating Index.
The indexes can be used to create a “diet score,” each of which are based on cut-points of various food groups with criteria that determine weight of each food that contribute to the score. [The diet scores look like a handy tool for helping people stay on track.]
The 2005 Dietary Guidelines for Americans Adherence Index (DGAI) are based on food intake recommendations and it penalizes for overconsumption of discretionary energy and energy-dense foods (chips and French fries). They also had a “variety score” when people had various foods in the diet.
When individuals adhere to the DGAI, they had a lower prevalence of metabolic syndrome. Based on the scoring approach, McKeown says, it’s possible for people to get the same score and have different dietary patterns.
Regarding the Mediterranean dietary pattern, McKeown reminds us that “there is no single Mediterranean diet,” but it’s based on patterns, so a diet score can be useful for adhering to the dietary pattern.
When discussing low-carb dietary patterns, they have similar macronutrient composition, but may have different dietary quality. There should be scores indicated depending on the choices of fat — animal or vegetable. She is careful to note that it’s important to consider that when you talk about macronutrients to consider substitutions for foods eliminated.
She closes by saying that there is subjectivity in defining cut points in indices. Indices may be a good index of diet quality but not of disease risk. High scores are rare, but average scores can be achieved a number of different ways. Eating patterns associate with other health behaviors.