For more information, read an Overview of the Methods & Calculations.
The steps of the bivariate method method are:
Deriving Sums. For each HEI component, dietary constituents are summed together (e.g., Greens and Beans is created from the sum of dark green vegetables and legumes (beans and peas)).
For the HEI-2015, the components and associated dietary constituents are noted below:HEI-2015 Components and Associated Dietary Constituents for Bivariate Method
Components Typea Units Dietary Constituents From FPED (or other food-based database) Whole Fruits Episodic cup eq. Citrus, Melons, Berries + Other Intact Fruits Total Fruits Varies cup eq. Total Fruit Greens and Beans Episodic cup eq. Dark Green Vegetables + Legumes (Beans and Peas) in cup equivalents Total Vegetables Daily cup eq. Total Vegetables + Legumes (Beans and Peas) in cup equivalents Whole Grains Episodic oz. eq. Whole Grains Refined Grains Daily oz. eq. Refined Grains Dairy Daily cup eq. Total Dairy Seafood and Plant Proteins Episodic oz. eq. Seafood (high in n-3) + Seafood (low in n-3) + Soy Products + Nuts and Seeds + Legumes (Beans and Peas) in oz equivalents Total Protein Foods Daily oz. eq. Total Meat, Poultry, and Seafood (including organ meats and cured meats) + Eggs + Nuts and Seeds + Soy + Legumes (Beans and Peas) in oz equivalents Added Sugars Daily tsp. eq. Added Sugars From FNDDS (or other nutrient database) Fatty Acids Daily g (Total Monounsaturated Fatty Acids + Total Polyunsaturated Fatty Acids)/Total Saturated Fatty Acids Saturated Fats Daily g* Total Saturated Fatty Acids Sodium Daily mg Sodium Energy Daily kcal Total Energy
cup eq.=cup equivalents; oz. eq.=ounce equivalents; tsp. eq.=teaspoon equivalents; g=grams; g*= fatty acids are calculated in grams but converted to energy in the scoring process; mg=milligrams
a General guidance on daily vs. episodic. Should be examined for each individual dataset.
Modeling. Using a model that: a) transforms data to approximate normality, b) separates within-person from between-person variability, c) models the probability of consumption separately from the consumption day amount for episodically consumed foods, d) allows for correlation among the probability of consumption, the consumption-day amount, and energy for one dietary category; usual intakes for each dietary constituent are predicted.
Constructing Ratios. The appropriate ratios are constructed. Usually these are the ratios of the dietary constituents to 1000 kcal of energy, with the exception of fatty acids, which use the ratio of the sum of monounsaturated and polyunsaturated fatty acids to saturated fatty acids. (Also, note two components are expressed on a percent of calories basis. Therefore, grams of saturated fat should be multiplied by 9 to convert g to kcal, and added sugars should be multiplied by 16 to convert teaspoons to kcal, prior to dividing by total energy.) Step 4 and beyond are different depending on purpose; see two options below.
For Describing Dietary Intakes:
Estimating. Generate “pseudo-individuals” from the parameters estimated from the model fit in Step 1.
Scoring. For each of the pseudo-individuals, score the ratios according to the scoring standards for each component. The component scores are summed to calculate the total score. (Note that the bivariate method takes into account the correlation for each component, but not among all components so the total score should be used with caution.)
Calculating Means. Compute means and percentiles for the population of pseudo-individuals.
For Examining Association between Diet and Another Variable:
Scoring. For each individual, use adaptive Gaussian quadrature to predict usual intake for each individual and then score the ratios according to the scoring standards for each component. The component scores are summed to calculate the total score. (Note that the bivariate method takes into account the correlation for each component, but not among all components so the total score should be used with caution.)
Performing Regression Analysis. Use an appropriate regression model (with bootstrapping or balanced repeated replication for standard errors) to estimate the relationship between the HEI score and the outcome of interest.