The ASPE/FDA team selected Option 1, the nutrient density-based scoring system to test for this project. The modifications they were most interested in testing were
- unit basis,
- nutrients such as fat quality and added sugar, and
- categorization of scores.
RTI devised a detailed plan describing the algorithm, modifications, and methods of testing. The baseline algorithm for a nutrient density-based score (NDS) is depicted in Figure 3-1.
Figure 3-1. Algorithm for a Nutrient Density-Based Score (NDS) on a per 100 Kcal Basis
(Protein g per kcal/50 g + Fiber g per kcal/25 g + Vitamin E IU per kcal/30 IU + Vitamin D IU per kcal/400 IU + Calcium mg per kcal/1,000 mg + Iron mg per kcal/18 mg + Potassium mg per kcal/3,500 mg + Unsaturated fat g per kcal/44 g) * 100/8
− (Saturated fat g per kcal/20 g + Added sugars g per kcal/50 g + Sodium mg per kcal/2,400 mg) * 100/3
|a NDS1KCAL is an abbreviation for the baseline algorithm per 100 kcal. Positive nutrients were capped at 100% of the recommended intake.|
We selected the nutrients to be included in the baseline algorithm for these reasons:
- positive nutrients of importance that were identified as consumed in less than optimal amounts and are of concern for the entire U.S. population as specified in the 2010 DGA (vitamin D, calcium, potassium, and fiber);
- some positive nutrients that were previously identified as low in the American diet in the 2005 DGA report (iron and vitamin E);
- some negative nutrients that are overconsumed as defined by both 2005 and 2010 DGA (saturated fat and sodium);
- added sugars because of their contribution to excess energy intakes;
- protein because of its dietary importance; and
- unsaturated fat because of the recognition of its importance and recommendation in the 2010 DGA to replace saturated fatty acids with polyunsaturated and monounsaturated fatty acids.
We identified additional positive nutrients that were identified as nutrients of concern in the American diet in the 2005 DGA to be tested in modified versions of the algorithm (as described further in Step 5 below).
The unit basis for the baseline algorithm is per 100 kcal. The basis for the denominators for most of the nutrients is the daily value (DV) or daily reference value (DRV) (Code of Federal Regulations, 2010a). For unsaturated fat, the denominator was derived from the total fat DRV of ~30% of total kcal minus the saturated fat DRV of ~10% total kcal; therefore, 20% of total energy from unsaturated fat or 44 g was used as the basis for unsaturated fat. There is no daily value for added sugars; we used as a basis the recommendation 10% of total energy (WHO, 2003). Positive nutrients were capped at 100% of DV to avoid undue influence by one nutrient and to discourage fortification to yield high scores. The total contribution from the positive and negative nutrients is equalized by dividing the total positive portion of the score by the number of positive nutrients (eight) and the total negative portion of the score by the number of negative nutrients (three).
We followed a step-wise approach to testing modifications to the base algorithm. Details of the methodology used to test the algorithm are presented in Section 4.
Step 1: Test the baseline algorithm. Baseline algorithm scores were calculated for a set of foods based on nutrient values per 100 kcal (500 foods most commonly consumed from the NHANES survey). The rankings of foods were examined across and within food groupings. Major food groupings will be selected based on the USDA food coding scheme (for example, dairy, meat/poultry/fish, legumes, grains, fruits, vegetables, fats/oils, and sweets). We looked for discriminations between foods to determine if expected rankings are evident. Finally, we determined the amount of variation explained by the algorithm in accounting for HEI scores. Details of the methods are presented in Section 4.
Step 2: Test the unit basis. The unit basis for the amount of nutrient components was calculated "per RACC" serving, and Step 1 was repeated.
Step 3: Test a "fat quality" component. We removed the unsaturated fat component of the algorithm and repeated Step 1 to determine if inclusion of a "good" fat improved discrimination of food rankings and variation explained in HEI scores.
Step 4: Test an added sugar component. We removed the added sugar component of the algorithm and repeated Step 1 to determine if it improved discrimination of food rankings and variation explained in HEI scores.
Step 5: Test other positive nutrients. We added other positive nutrients individually into the algorithm. We proposed nutrients that have been previously identified as being important nutrients to encourage in the diet of Americans: magnesium, vitamin A, vitamin C, folate, and vitamin B12.
Step 6: Test the concept of whole grains. We added the percentage of grains that are whole grains to the positive nutrients in the base algorithm. The data for this component were derived from the MyPyramid database by dividing the ounce-equivalents of whole grains by the ounce-equivalents of total grains. The recommendation that at least 50% of grains should be consumed as whole grains was used as the "daily value."
Step 7: Decide on final algorithm based on modification results. We determined which nutrient components to keep in a final algorithm. We planned to base this on the results of testing in the previous steps; however, we developed a new statistical approach to determine the final algorithm. The details of the new statistical approach and the resulting final algorithm are described in detail in Section 4.4. We tested the unit basis again to verify the final algorithm.
Step 8: Convert continuous scores to categorical rankings. Using the final algorithm, we examined tertile and quintile rankings and categorization of foods across and within food categories. Such a categorization could be considered, for example, for an FOP system with an overall "low, medium, high" rating scheme.
Step 9: Conduct additional validations with subpopulations. We conducted additional validations of the algorithm and HEI scores among different subgroups of the population (e.g., age, ethnicity, or health conditions).
Results of testing are described in Section 4.