Professional Documents
Culture Documents
Defining Nutrient Density Development and Validation of The Nutrient Rich Foods Index
Defining Nutrient Density Development and Validation of The Nutrient Rich Foods Index
Adam Drewnowski
To cite this article: Adam Drewnowski (2009) Defining Nutrient Density: Development and
Validation of the Nutrient Rich Foods Index, Journal of the American College of Nutrition, 28:4,
421S-426S, DOI: 10.1080/07315724.2009.10718106
The 2005 Dietary Guidelines for Americans and U.S. Department of Agriculture’s MyPyramid made a clear
distinction between foods that were energy dense and those that were nutrient rich. Consumers were advised to
seek out nutrient-rich foods (NRF) in preference to discretionary calories. However, the concept of nutrient
density was not formally defined. The Nutrient Rich Foods (NRF) index is a formal metric of nutrient density
that has been extensively compared with other methods and validated with respect to a healthy diet. The NRF
index is based on 9 nutrients to encourage (protein; fiber; vitamins A, C, and E; calcium; iron; potassium; and
magnesium) and 3 nutrients to limit (saturated fat, added sugar, and sodium). The NRF9.3 algorithm is the
unweighted sum of percentage daily values (DVs) for 9 nutrients to encourage, minus the sum of percentage
maximum recommended values (MRVs) for 3 nutrients to limit, calculated per reference amount and capped at
100% DV. The NRF index successfully ranks foods based on their nutritional value and can be applied to
individual foods, meals, menus, and even the daily diet. Development of the NRF index followed scientific
guidelines for nutrient profiling and was accompanied by extensive testing and validation.
Address correspondence to: Adam Drewnowski, 305 Raitt Hall 353410, University of Washington, Seattle, WA 98195-3410. E-mail: adamdrew@u.washington.edu
Presented in part at the American College of Nutrition Annual Meeting, October 4, 2008, Arlington, VA.
Financial interests: The Symposium was supported by the National Dairy Council.
Journal of the American College of Nutrition, Vol. 28, No. 4, 421S–426S (2009)
Published by the American College of Nutrition
421S
Defining Nutrient Density
Table 1. Nutrients to Encourage and Nutrients to Limit in Selected NRn and LIM Subscores
The basis of calculation typically has been 100 g or 100 kcal of disqualified from nutrition and health claims by the FDA if
food, or sometimes both [15]. In some cases, food rankings they contain above-specified amounts of fat, saturated fat, trans
generated by such models have been compared with expert fat, cholesterol, and sodium. European Union regulations
opinion [7,16,17]; in fewer cases, model scores were validated include sugar on the list of nutrients to limit, although without
with respect to measures of a healthy diet [18]. specifying if the sugar was added or total. Models developed in
The Nutrient Rich Food approach is unique because it is the United Kingdom and in France have used energy, saturated
conceptualized as a comprehensive food guidance system fat, total sugar, and sodium [5,6] or saturated fats, added
rather than a rating, numbering, or labeling system. Consumers sugars, and sodium [12], respectively.
make critical decisions about food choices at many points Since fat and sugar are highly correlated with energy density,
during the day: the NRF approach provides a useful tool for all nutrient profiles award high scores to foods that provide little
education and guidance. Extensive consumer research is being dietary energy per unit volume. In other words, energy density
conducted to ensure that consumers will find the NRF and nutrient density of foods are inversely linked.
approach appealing and useful in selecting NRF and building
nutritious meals and more healthful diets. Not limited to Development of the NRF Index
individual foods only, the NRF algorithm can be applied to
Table 1 shows the initial development and testing of the
food groups, meals, menus, and total diets. Describing the
NRF family of scores. First to be tested were the NRn
science that helps consumers make informed decisions about
subscores, based on nutrients to encourage, where the number
their food choices—both inside and outside the store—is the
n of nutrients could vary from 5 to as many as 15 [1,23,24].
purpose of this review.
The NR6 model was based on the 6 nutrients that the FDA uses
to define healthy foods. The NR9 model added a further 3
What Is Nutrient Density? nutrients of concern as identified by the Dietary Guidelines for
The concept of nutrient density is built around the amounts Americans (vitamin E, magnesium, and potassium). The NR11
of key nutrients contained in 100 kcal of a given food [19,20]. model added another additional 5 nutrients of concern for at
Whereas some foods contain ‘‘empty’’ calories, NRF are those least a subset of the population (vitamin E, magnesium,
that contain relatively fewer calories than nutrients. The potassium, vitamin B12, and zinc). The NR15 model was the
selection of nutrients for inclusion in the NRF index was based original Naturally Nutrient Rich score, based on protein; fiber;
on current regulatory frameworks and dietary guidance vitamins A, C, D, E, B1, B2, and B12; folate; calcium; iron;
[21,22]. The 2005 Dietary Guidelines for Americans [2] potassium; and zinc [1]. Two subscores based on nutrients to
identified fiber; vitamins A, C, and E; calcium; potassium; and limit were the LIM score, based on saturated fat, added sugar
magnesium as underrepresented in the U.S. diet. For children and sodium, and the LIMt score, based on total fat, total sugar,
and adolescents, nutrients of concern were fiber, vitamin E, and sodium [12].
calcium, potassium, and magnesium. Vitamin B12, iron, folic All nutrient amounts were converted to percentage daily
acid, and vitamin D were nutrients of concern for some values (%DV) per specified reference amount of food.
population subgroups. More formally, the U.S. Food and Drug Percentage DVs were capped at 100% so that foods containing
Administration (FDA) bases its definition of healthy foods on very large amounts of a single nutrient would not obtain a
their content of protein, fiber, vitamins A and C, calcium, and disproportionately high index score [1]. For nutrients to limit,
iron. The credit side of the nutrient density equation typically maximum recommended values were 20 g for saturated fat,
includes protein and fiber and a range of vitamins and 125 g for total sugar, 50 g for added sugar, and 2400 mg for
minerals. sodium in a 2000-kcal/d diet, as based on a variety of sources
Fat, sugar, and sodium are key nutrients to limit based on [12]. Reference daily values for use on nutrition labels are
multiple authoritative sources. For example, foods are summarized in Table 2 [21,22].
Table 2. Reference Daily Values for Nutrients, Based on 2000- regulatory structure of the FDA. Serving sizes, officially
kcal Diet described as the reference amounts customarily consumed
(RACC), are the basis of calculating nutrient %DVs on the
Nutrient Reference Daily Value
FDA Nutrition Facts panel. RACC values are related to the
Protein 50 g energy density of foods and are lower for sugar (4 g) and fats
Fiber 25 g
and oils (15 g) than for cheese (30 g), meats (85 g), vegetables
Monounsaturated fatty 20 g
acids and fruit (120 g), or milk, juices, and other beverages (240 g).
Vitamin A 5000 IU Models based on U.S. serving sizes may not find universal
Vitamin C 60 mg applications. In the European Union, for example, govern-
Vitamin D 400 IU (10 mg)
ment-mandated serving sizes do not as yet exist, and dietary
Vitamin E 30 IU (20 mg)
Thiamin 1.5 mg information is provided per 100 g [15,23]. The nutrient-
Riboflavin 1.7 mg profiling scheme developed by the British Food Standards
Vitamin B12 6 mg Agency (FSA) is based on 100 g, whereas the SAIN,LIM
Folate 400 mg scheme developed for the equivalent French agency AFSSA is
Calcium 1000 mg
based on a combination of 100 g and 100 kcal.
Iron 18 mg
Zinc 15 mg Although a system based on 100 g would seem to have
Potassium 3500 mg universal appeal, model profiles based on 100 g of foods make
Magnesium 400 mg no allowances for the fact that different foods and beverages
Fat 65 g
are consumed in very different amounts. Models in which
Saturated fat 20 g (10% energy of 2000-kcal diet)
Monounsaturated fat 20 g (10% energy of 2000-kcal diet) saturated fat, added sugar, and sodium are calculated per 100 g
Cholesterol 300 mg of food or beverage tend to penalize energy-dense foods
Sugars, total 125 g/50 g* consumed in small quantities (nuts, dried fruit, cheese) while
Sugars, added 50 g (10% energy of 2000-kcal diet) giving overly favorable scores to sugared beverages. A system
Sodium 2400 mg
based on 100 g can be very lenient toward caloric beverages
* 125 g from Institute of Medicine report; 50 g from World Health Organization unless special provisions are made. To account for these
report.
discrepancies between the 100 g standard and the larger
beverage portions normally consumed, the latest version of the
Table 3 summarizes the algorithms for selected families of British FSA score multiplies the negative subscore for
models. Raw NRF9.3 scores were unweighted sums of beverages by a factor of 2 [16]. This approximates calculations
reference %DVs for 9 nutrients to encourage minus the sum based on RACC in which added sugar is calculated per 245 g
of maximum %MRVs for 2 nutrients to limit. reference amount.
Calculations based on 100 kcal best reflected the original
concept of nutrient density of foods, defined as the ratio of
Basis of Calculation: 100 g, 100 kcal, or Serving Size?
nutrients to calories. One caution is that the 100-kcal base of
Three different bases of calculating nutrient density were calculations leads to 100 kcal and tends to give overly high
used in developing the NRF model: 100 g, 100 kcal, and scores to very low-energy-density salad greens, such as
serving size. Serving sizes were defined based on the current spinach, lettuce, endive, watercress, and cabbage. This is
Table 3. Algorithms for NRn and LIM Subscores and the NRF Index Score, Calculated per 100 kcal
Reference
Model Algorithm Amount Comment
Subscores NRn
P
NRn_100kcal 1{n (Nutrienti =DVi )=ED|100 100 kcal Nutrienti 5 content of nutrient i in 100-g edible portion;
DVi 5 daily values for nutrient i (see Table 1); n 5 the
number of nutrients; ED 5 energy density
Subscores LIM
P
LIM_100kcal 1{3 (Li =MRVi )=ED|100 100 kcal Li 5 content of limiting nutrient i in 100 g of edible portion
of food, I 5 1 to 3; ED 5 energy density
Composite models, NRF
NRFn.3_100kcal NRn_100kcal 2 LIM_100kcal 100 kcal Difference between sums
MRV 5 maximum recommended value; NRF 5 nutrient rich food; LIM 5 limited nutrient score.
Fig. 1. A plot of median NRF9.3 scores (x-axis) shown in relation to energy density of foods (ED 5 kcal/100 g) by major USDA food group.
because 100 kcal of lettuce, with an energy density of 0.1 kcal/ scores whenever the negative component tended toward zero.
g, translates into 1 kg of greens by weight. In excess of any Based on model performance, the optimal NRF9.3 algorithm
portion size, the 1 kg amount is then used to calculate the was the unweighted sum of %DVs for 9 nutrients to encourage,
amounts of vitamins and minerals. However, calculations minus the sum of %MRVs for nutrients to limit.
based on the 100-kcal reference amount led to the highest Fig. 1 summarizes how the NRF algorithm ranks foods
percentage of variance accounted for in validation studies. across and within food groups. The range of scores extends
from sugar to spinach. Most important, the NRF approach does
Alternative NRF9.3 Algorithms not make distinctions between good and bad foods; all foods
fall along a continuum. Their place on the continuum depends
Alternative NRF models were then validated against
on their nutrient content in relation to calories.
independent measures of a healthy diet. The present diet quality
measure was based on the revised Healthy Eating Index scores
[25] that were calculated for participants in the 1999–2002
Consumer Applications
NHANES dietary survey. The best validation results were Helping the consumer to identify the most NRF at the point
obtained for the NRF9.3 model composed of a positive subscore of sale is one application of nutrient profiling. Depending on
based on protein; fiber; vitamins A, C, and E; calcium; iron; the system chosen, individual foods have been awarded from 0
potassium; and magnesium and a negative subscore based on to 3 stars (effectively a 4-point scale) or were ranked from 0 to
saturated fat, added sugar, and sodium [26]. The composite 100 points to denote their nutritional value. Another proposal
algorithm tracked diet quality better than metrics based is to label foods with nutrition symbols and logos based on
exclusively on nutrients to avoid, a finding that may have their content of sugar and fat. In an alternative system,
implications for nutrition education and dietary guidance. The launched by the National Heart Lung and Blood Institute,
composite NRF9.3 model based on nutrients to encourage and foods are assigned into Go, Slow, and Whoa categories based
nutrients to limit performed better than the negative LIM on their perceived sugar and fat content [27].
subscore based on saturated fat, added sugars, and sodium only. The NRF approach to food guidance stands apart from the
The preferred algorithms were those that were based on conventional nutrient-profiling systems. First, the NRF index
subtracting the negative from the positive subscore (NRn 2 focuses on nutrients to encourage, while taking into account
LIM), rather than a ratio between the two (NRn/LIM). The three key nutrients to limit. Second, the NRF algorithm can be
ratio model, in particular, yielded a nonnormal distribution and applied to individual foods, meals, menus, and the total diet.
had least discriminating power. Ratio-based algorithms The NRF approach can therefore assist food choice decisions
exaggerated small differences and gave unrealistically high at different places and times during the day, not only in the
19. Hansen RG: An index of food quality. Nutr Rev 31:1, 1973. 24. Drewnowski A: What’s next for nutrition labelling and health
20. Hansen RG, Wyse BW, Sorenson AW: ‘‘Nutritional Quality Index claims: an update of nutrient profiling in the European Union and
of Foods.’’ Westport, CT: AVI Publishing Co., 1979. the US. Nutr Today 42:206–214, 2007.
21. U.S. Food and Drug Administration: Code of Federal Regulations, 25. Guenther P, Reedy J, Krebs-Smith SM, Reeve P, Basiotis P: The
Title 21-Food and Drugs (Vol. 2), Chapter I-Food and Drug revised Healthy Eating Index: a tool for measuring change in diet
Administration, Department of Health and Human Services, Part quality. J Am Diet Assoc 106:A68, 2006.
101-Food Labeling-Table of Contents, Subpart A-General Provi- 26. Fulgoni VL III, Keast DR, Drewnowski A: Development and
sions, Sec. 101.14 Health claims: General Requirements and validation of the Nutrient-Rich Foods index: a tool to measure
Subpart E-Specific Requirements for Health Claims. Washington, nutritional quality of foods. J Nutr 139:1549–1554, 2009.
DC: U.S. Government Printing Office. Revised as of April 1, 27. Department of Health and Human Services, National Institutes of
2002. Health: Go, Slow and Whoa foods. Accessed August 8, 2009, at:
22. U.S. Food and Drug Administration: Code of Federal Regulations. http://www.nhlbi.nih.gov/health/public/heart/obesity/wecan/live-it/
Sec 101.9 Nutrition Labeling of Food. Accessed at: http://www. go-slow-whoa.htm
accessdata.fda.goc/scripts/cdrh/cfdocs/cfcfr/cfrsearch.cfm?fr5109.9 28. Zelman K, Kennedy E: Naturally nutrient rich: putting more power
23. Drewnowski A, Maillot M, Darmon N: Testing nutrient profile on Americans’ plates. Nutr Today 40:60–68, 2005.
models in relation to energy density and energy cost. Eur J Clin
Nutr 63:674–683, 2009. Received December 8, 2009.