The rankings of foods using the algorithm are presented for each of the modifications as follows:
 baseline algorithm with unit basis modification (NDS1): Section 4.2.1
 modification removing unsaturated fat (NDS2): Section 4.2.2
 modification removing added sugars (NDS3): Section 4.2.3
 modification with vitamins (only vitamin C presented, NDS4C): Section 4.2.4
 modification with whole grains (NDS5): Section 4.2.5
 modification with vitamin C and whole grains (NDS6): Section 4.2.6

4.2.1 Ranking of Foods and Food Groups for the Baseline Algorithm, Comparing the Unit Basis (NDS1)

The mean algorithm scores and distributions of scores of major food groupings using the baseline algorithm are presented in Figures 42 and 43 on a per 100 kcal basis (NDS1KCAL) and a per RACC basis (NDS1ACC), respectively. Positive scores mean that positive nutrients outweigh negative nutrients included in the algorithm. Likewise, negative scores mean that negative nutrients outweigh positive nutrients. The range of scores is much larger for NDS1KCAL than for NDS1RACC. The mean score of all 570 foods on a per RACC basis was lower than the mean score per 100 kcal.
Figure 42. Box Plot of Nutrient Density Scores per 100 Kcal for Baseline Algorithm (NDS1KCAL)
Figure 43. Box Plot of Nutrient Density Scores per RACC for Baseline Algorithm (NDS1RACC)
Fruits had the highest mean score among the food groups on a per 100 kcal basis and legumes had the highest mean score on a per RACC basis. Vegetables and eggs had the third and fourth highest mean scores. The mean score of grains was fifth for NDS1KCAL and eighth for NDS1RACC. The grain group contained cookies, cakes, and pastries, which have RACC servings greater than 100 kcal, and the high fat and sugar content resulted in lower scores for these foods. Fats, oils, and dressings scored higher on a per RACC basis because their portion sizes are smaller than 100 kcal portions. The sweets and beverages group had the lowest mean score on a per RACC basis.
Some of the major food groupings contain a diverse variety of foods; therefore, it is useful to examine mean scores of subgroupings for these foods (Tables E1 and E2 in Appendix E) and scores of selected foods (Table E3). In the dairy group, rankings of individual foods were similar between both unit bases (Table E3). As expected, the milks had decreasing scores with increasing fat content. Chocolate milk, even the lowfat version, scored lower than regular whole milk. Among the yogurts, the nonfat versions scored the highest. The fruit variety of wholemilk yogurt had the lowest ranking for both unit bases but had a more extreme lower score per RACC because the large portion contributes more sugar and fat to the score.
Among legumes and nuts, there were higher scores for nuts on a per RACC basis because the larger RACC serving size, 1 ounce, is greater than 100 kcal, resulting in higher positive nutrient values, particularly for unsaturated fat.
For many of the grain subgroups, similar rankings of foods on a per RACC and per 100 kcal were seen for some subgroups, particularly when the RACC serving size was close to 100 kcal (e.g., breads, cereals, cookies, crackers, and salty snacks). On the other hand, highcalorie and highfat cakes and mixed foods such as pizza resulted in more extreme lower values on a per RACC basis. For example, a RACC serving of chocolate cake provides 468 kcal, 30% of the daily saturated fat value, and 116% of the daily added sugar value.
The vegetable group had an overall negative mean score. Some foods in this group were condiments that contained a large amount of sodium (e.g., salsa, catsup, pickles). Among the highest scoring vegetables were leafy greens. The NDS1KCAL scores were very high because of the large serving size of 100 kcal portions (e.g., 435 g raw spinach). Pickles had the lowest ranking per 100 kcal because of the large serving (833 g) per 100 kcal and high sodium content. Potato chips are included in the vegetable group and, interestingly, scored higher than baked potatoes regardless of the unit basis. Potato chips have higher unsaturated fat and vitamin E. However, in the nutrient database used for NHANES, baked potatoes have higher sodium than potato chips because the baked potatoes have salt added by default. If the sodium value for baked potatoes without salt were used (5 mg/100 kcal), then the NDS1KCAL score would be 3.11 rather than −1.16 (with salt), which is higher than the score for potato chips (1.93).
Among the fats, oils, and salad dressings group, lowcalorie salad dressings had very low scores on a per 100 kcal basis because of high sodium content. For example, the lowest scoring salad dressing was fatfree, reducedcalorie Italian dressing (−33.41), providing 100% of the daily sodium value on a per 100 kcal basis. On a RACC basis, the score was −4.71, which is 14% of the daily sodium value. In contrast, scores for regular Italian dressing were similar with the different unit bases (−8.98 and −7.84 on a per 100 kcal and RACC basis, respectively).
Among beverages, coffee and tea scored very high on a per 100 kcal basis because of their low calorie content, resulting in large portions on a per 100 kcal basis and high potassium values driving the high scores. Sugarsweetened beverages were similar between both unit bases because a RACC of 240 g is equivalent to ~100 kcal. Sugarfree, caloriefree beverages were assigned a 1 kcal/100 g value to obtain a score on a per 100 kcal basis, which resulted in relatively high nutrient values on a per 100 kcal basis.


4.2.2 Ranking of Foods and Food Groups with Modified Algorithm, Removing Unsaturated Fat (NDS2)

The modification removing unsaturated fat from the algorithm results in changes in scores regardless of whether the food contains unsaturated fat because the total number of positive nutrients in this algorithm is 7 and the sum of the positive nutrient percentages was divided by 7 instead of 8 in the baseline algorithm. In general, if a food did not contain unsaturated fat, then the score increased because the positive nutrients were now divided by 7 rather than 8. Conversely, in general, if a food contained unsaturated fat, then the score decreased when it was removed from the algorithm. The mean and distribution of algorithm scores by major food groupings using the modified algorithm scoring foods on a per 100 kcal basis (NDS2KCAL) and a per RACC basis (NDS2RACC) are presented in Figures F1 and F2.
The overall mean scores of the 570 foods with the modified algorithms were similar to the overall mean scores for foods using the baseline algorithm. The rankings of the mean scores from the major USDA food groupings changed slightly. Fruits ranked first for both algorithms, although the mean score for fruits increased with the modified algorithm. The scores for most fruits increased because they do not contain unsaturated fat; however, the score for avocado decreased with the modified algorithm because avocados are high in unsaturated fat (i.e., from 4.42 to 3.02 on a per 100 kcal basis and from 9.89 to 6.76 on a per RACC basis). Vegetables and eggs reversed within the second and third rankings on a per kcal basis with the modification, and scores for eggs decreased because they contain some unsaturated fat. As in NDS1, potato chips still had higher scores than baked potatoes, but the gap was smaller because the score for potato chips decreased and the score for baked potatoes increased with the removal of unsaturated fat from the algorithm. Dairy scores generally increased because they contain relatively small amounts of unsaturated fat and high amounts of other positive nutrients. A few foods in the dairy group that contained more unsaturated fats (e.g., cream substitutes) had lower scores with the modified algorithms (NDS2KCAL and NDS2RACC). The mean score for legumes/nuts decreased slightly. Although the scores for all of the nut foods decreased, some of the bean scores increased. The topscoring food in the legume/nut group was almonds; its score decreased from 7.33 to 6.11 on a per 100 kcal basis and from 12.65 to 10.54 on a per RACC basis. The largest decrease in mean food group scores with the modified algorithm was for the fats, oils, and dressings group because of the high unsaturated fat content of many of the foods in this group. For example, the topranking food, olive oil, decreased from 1.14 to −1.39, and margarine decreased from −1.78 to −4.06 on a per 100 kcal basis. The fats and oils remained the group with the lowest mean score on a per 100 kcal basis but decreased from the fifth to the eighth ranking on a per RACC basis.


4.2.3 Ranking of Foods and Food Groups with Modified Algorithm, Removing Added Sugars (NDS3)

The baseline algorithm was modified to remove the added sugars component from the negative nutrients for a total of two negative nutrients instead of three in the baseline algorithm. The average score of the 570 foods with the modified algorithm was slightly lower than the mean score of foods using the baseline algorithm (Figures F3 and F4). In general, foods that contained the other two negative nutrients, saturated fat and sodium, had lower scores when added sugar was removed from the algorithm.
The scores for most fruits and fruit juices did not change; however, canned fruits containing added sugar had higher scores when added sugar was removed from the algorithm. Scores of foods in the sweets/beverages group increased substantially. The mean score of grains increased slightly on a per 100 kcal basis and decreased slightly on a per RACC basis. Among the grain products, scores for cakes, cookies, and cereals increased, and scores for crackers and mixed foods decreased. The decrease in scores for mixed foods was greater on a per RACC basis than on a per 100 kcal basis. For example, macaroni and cheese decreased from −10.10 to −24.07 because the high amounts of saturated fat and sodium on a per RACC basis were divided by two instead of three in the modified algorithm; on a per 100 kcal basis, the score decreased from −2.06 to −4.90. Among dairy foods, scores for yogurts and ice cream generally increased, while scores for cheeses decreased. Lowfat chocolate milk scored higher than white whole or 2% milk, whereas in the original algorithm all chocolate milks scored lower than all of the white milk varieties. Scores for fats, eggs, and meats decreased because these foods do not contain sugar, and the other negative nutrients were more heavily weighted because of dividing by two rather than three negative nutrients.


4.2.4 Ranking of Foods and Food Groups with Modified Algorithm, Adding Vitamin C (NDS4)

The mean scores of the 570 foods with vitamin C added to the algorithm were higher than the respective baseline algorithms (Figures F5 and F6).
The scores for fruits increased substantially with the addition of vitamin C to the algorithm; the average score increased from 3.06 to 8.77 on a per 100 kcal basis and from 2.47 to 7.49 on a per RACC basis. Fruits with the highest vitamin C content were capped at 100% of daily value to avoid extreme scores. One of these foods was raw oranges; its score increased from 5.86 to 16.31 on a per 100 kcal basis and from 3.85 to 14.53 on a per RACC basis. The highest score using both the original and modified algorithm was orange juice with calcium added; its score increased from 10.15 to 20.12 on a per 100 kcal basis when vitamin C was added to the algorithm. Scores for vegetables increased moderately. The mean scores of the other food groups were similar to the original scores.


4.2.5 Ranking of Foods and Food Groups with Modified Algorithm, Adding Whole Grains (NDS5)

In this algorithm, whole grains are expressed as the percentage of total grains compared with the recommendation to consume 50% of grains as whole grains. A second method of expressing whole grains was tested based on the recommendation of three servings per day of whole grains. In this section we present a summary of the scores for the former method of expressing whole grains as the percentage of total grains (Figures F7 and F8).
Except for increases in scores for the grain products, the mean scores and rankings of the food groupings remained similar to the original algorithm. Grain products increased from the fourth to the second highest ranking group on a per 100 kcal basis and from the eighth to the fifth ranking group on a per RACC basis. Scores for some breads increased substantially; for example, multigrain bread increased from −0.59 to 10.02 and wheat bread increased from −0.29 to 6.81 on a per 100 kcal basis. White bread decreased slightly, from −1.79 to −2.15. Granola bars increased from −4.49 to 6.32 on a per 100 kcal basis. In the original algorithm, fruitbased cereal bars had slightly higher scores than granola bars, but this was reversed with the modification because of a higher whole grain content. The score for Cheerios increased from 8.23 to 18.03 on a per 100 kcal basis. It should be noted that whole grains contribute other positive nutrients that are included in the algorithm, such as fiber and vitamin E.


4.2.6 Ranking of Foods and Food Groups with Modified Algorithm, Adding Vitamin C and Whole Grains (NDS6)

The mean score of the 570 foods with vitamin C and whole grains added on a per 100 kcal basis was the highest mean score of all modifications (−2.13). The mean and distribution of algorithm scores by major food groupings using the modified algorithm scoring foods on a per 100 kcal basis (NDS6KCAL) and a per RACC basis (NDS6RACC) are presented in Figures F9 and F10.
Fruits, vegetables, and legumes were the food groups with the highest ranking mean scores with both unit bases, and the rankings of mean scores of all food groups were the same for both unit bases.


4.2.7 Synopsis of Algorithm Performance in Ranking the Nutritional Quality of Foods

Based on examination of food scores using the various algorithms, it is difficult to assess the differences of the various algorithms to determine which nutrients should be included in an algorithm; however, the process does provide insights into the effects of various modifications. Examination of food scores using the various algorithms demonstrates that the algorithm discriminates between more and less healthy versions of foods; for example, lowfat dairy products score higher than highfat dairy products, and fruits score higher than sweetened beverages. Food scores are sensitive to the nutrients in the algorithm; for example, scores for nuts, which are high in unsaturated fat, decreased when unsaturated fat as a positive nutrient was removed from the algorithm. Some anomalies became evident when comparing the scores based on per 100 kcal or RACC basis; for example, green leafy vegetables score very high on a per 100 kcal basis because of their low energy content but 100 kcal is a very large portion size for these foods.
As noted previously, we did not conduct formal validity tests focused on the food scores themselves. We did conduct formal validity tests of the ability of the algorithms to predict overall diet quality by scoring foods in diets and comparing with the HEI, a measure of diet quality. Section 4.3 describes the results of regression models predicting HEI scores using the various algorithms to score foods in diets.
