Testing nutrient profile models in relation to energy density and cost

Article (PDF Available)inEuropean journal of clinical nutrition 63(5):674-83 · March 2008with33 Reads
DOI: 10.1038/ejcn.2008.16 · Source: PubMed
Abstract
Nutrient profiling of foods is defined as the science of classifying foods based on their nutrient content. Food rankings generated by nutrient profile models need to be tested against objective reality as opposed to public opinion. To test the performance of selected nutrient profile models in relation to the foods' energy density (kcal g(-1)) and energy cost (Dollar per 1000 kcal). Analyses were based on 378 component foods of a food frequency instrument. The models tested were the French nutrient adequacy models NAS23 and NAS16 and nutrient density models NDS23 and NDS16; and a family of nutrient-rich models (NR(n), where n=5-7; 10-12, and 15). Also tested were LIM scores and a modified British Food Standards Agency model WXYfm. Profiles were calculated based on 100 g, 100 kcal and on Reference Amounts Customarily Consumed. Food rankings generated by different models were correlated with each other and with the foods' energy density and energy cost. Nutrient profile models based on protein, fiber, vitamins and minerals showed an inverse correlation with energy density that diminished as more micronutrients were introduced into the model. Models based on fat, sugar and sodium were highly correlated with energy density. Foods classified as healthier were generally associated with higher energy costs. Not all models accurately reflected the foods' content of nutrients known to be beneficial to health. High correlations with energy density meant that some models classified foods based on their energy density as opposed to nutrient content.
ORIGINAL ARTICLE
Testing nutrient profile models in relation to energy
density and energy cost
A Drewnowski
1
, M Maillot
2,3,4
and N Darmon
2,3,4
1
Center for Public Health Nutrition and the Nutritional Sciences Program, School of Public Health and Community Medicine,
University of Washington, Seattle, WA, USA;
2
INRA, UMR1260, Nutriments Lipidiques et Pre
´
vention des Maladies Me
´
taboliques,
Marseille, France;
3
INSERM, U476, Marseille, France and
4
Univ Aix-Marseille 1, Univ Aix-Marseille 2, Faculte
´
de Me
´
decine, IPHM-IFR
125, Marseille, France
Background: Nutrient profiling of foods is defined as the science of classifying foods based on their nutrient content. Food
rankings generated by nutrient profile models need to be tested against objective reality as opposed to public opinion.
Objective: To test the performance of selected nutrient profile models in relation to the foods’ energy density (kcal g
1
) and
energy cost (Dollar per 1000 kcal).
Subjects/Methods: Analyses were based on 378 component foods of a food frequency instrument. The models tested were the
French nutrient adequacy models NAS23 and NAS16 and nutrient density models NDS23 and NDS16; and a family of nutrient-
rich models (NR
n
, where n ¼ 5–7; 10–12, and 15). Also tested were LIM scores and a modified British Food Standards Agency
model WXYfm. Profiles were calculated based on 100 g, 100 kcal and on Reference Amounts Customarily Consumed. Food
rankings generated by different models were correlated with each other and with the foods’ energy density and energy cost.
Results: Nutrient profile models based on protein, fiber, vitamins and minerals showed an inverse correlation with energy
density that diminished as more micronutrients were introduced into the model. Models based on fat, sugar and sodium were
highly correlated with energy density. Foods classified as healthier were generally associated with higher energy costs.
Conclusions: Not all models accurately reflected the foods’ content of nutrients known to be beneficial to health. High
correlations with energy density meant that some models classified foods based on their energy density as opposed to nutrient
content.
European Journal of Clinical Nutrition (2009) 63, 674–683; doi:10.1038/ejcn.2008.16; published online 20 February 2008
Keywords: Food analysis; nutrition value; nutrition requirements; energy intake; food supply; food legislation
Introduction
Nutrient profiling, defined as the science of categorizing
foods according to their nutritional composition (Rayner
et al., 2004), will shortly become the basis for regulating
nutrition and health claims in the European Union. As
proposed, only foods with favorable nutrient profiles will be
allowed such claims, whereas foods with unfavorable profiles
will be disqualified (The European Parliament and the
Council of the European Union, 2006). Given the high
stakes, the development of nutrient profile models has been
the focus of much research effort (Rayner et al., 2005b;
Scarborough et al., 2007b). The UK Food Standards Agency
(FSA) has made its nutrient profiles available in peer-review
publications (Scarborough et al., 2007b) and online reports
(Rayner et al., 2004, 2005a, c; Stockley, 2007). The corre-
sponding French agency (Agence Franc¸aise de Se
´
curite
´
Sanitaire des Aliments (AFSSA)) will soon release its own
model profile (AFSSA, in press). Nutrient profiles from
Australia (Gazibarich and Ricci, 1998), the US (Drewnowski,
2005; Zelman and Kennedy, 2005), France (Darmon et al.,
2005; Labouze et al., 2007; Maillot et al., 2007), the UK
(Rayner et al., 2004, 2005a, c; Stockley, 2007; Scarborough
et al., 2007b) and the Netherlands (Netherlands nutrition
center, 2007; Nijman et al., 2007) have been the topic of
international symposia (IFN, 2006), workshops (FSA, 2005b;
ILSI, 2006; EFSA, 2007) and reviews (Drewnowski, 2007).
Received 20 September 2007; revised 25 November 2007; accepted 7 January
2008; published online 20 February 2008
Correspondence: Professor A Drewnowski, Center for Public Health Nutrition
and the Nutritional Sciences Program, University of Washington, 305 Raitt Hall
#353410, Seattle, WA 98195-3410, USA.
E-mail: adamdrew@u.washington.edu
European Journal of Clinical Nutrition (2009) 63, 674683
&
2009 Macmillan Publishers Limited All rights reserved 0954-3007/09 $
32.00
www.nature.com/ejcn
How well such nutrient profiles perform is not always
clear, since the criteria for selecting one model over another
have not been established (Drewnowski, 2007). The chief
criterion of success was whether or not the set of food
rankings generated by the model looked ‘right’. In some
studies, health professionals were asked to rate some 100
arbitrarily chosen foods for their perceived health value
(Azais-Braesco et al., 2006; Scarborough et al., 2007a, c). The
rankings generated by each model were then compared to
those subjective survey results. Another way to establish
relative validity was to compare rankings generated by
different models with each other (Azais-Braesco et al., 2006;
Scarborough et al., 2007a, c).
Nutrient profile models need to be tested against objective
reality as opposed to subjective opinion. One important
question is whether nutrient profiles should provide more
information about the nutrient content of foods that is
provided by the measure of calories per unit weight. On one
hand, the concept of energy density (kcal g
1
) alone may
be sufficient to promote healthier food choices and spur
industry innovation. On the other hand, the very notion of
nutrient profiling does imply going beyond calories to
consider the total nutrient package. Article 4 of the European
Union (EU) proposal specified the inclusion of nutrients
known to be beneficial to health.
Our research had previously identified strong inter-
relations among nutrient density of foods, energy density
and energy cost (h per 100 kcal or $ per 1000 kcal) (Darmon
et al., 2004; Andrieu et al., 2006; Drewnowski et al., 2007;
Maillot et al., 2007). We therefore tested the performance of
selected nutrient profile models against energy density and
energy cost of 378 frequently eaten foods that were broadly
representative of the US diet.
Methods
The food list
This study was based on 378 component foods of the food
frequency questionnaire (FFQ) developed by the Fred
Hutchinson Cancer Research Center (Patterson et al.,
1999). This food list excluded diet beverages with energy
density o0.1 kcal g
1
, noncaloric tea, coffee, drinking water,
medical foods and vitamin supplements but included ready
to eat cereals, fortified beverages and fortified products such
as liquid formula diets. The foods were also aggregated into
nine major food groups, following US Department of
Agriculture (USDA) codes (USDA, 2006).
Nutrient composition databases and food prices
Each FFQ component food was first translated to a specific
food item in purchasable form, using a software nutrient
composition database of over 27 000 food items (Monsivais
and Drewnowski, 2007; USDA, 2007). Food prices (in US
dollars) were obtained in May–July 2006 from three
supermarkets in the Seattle metropolitan area using in-store
visits and supermarket websites. Prices of fast foods were
obtained at local branches of national fast-food restaurants
(Monsivais and Drewnowski, 2007). For each food, price per
100 g was calculated taking into account the edible portion
or yield, based on the US Department of Agriculture Hand-
book 102 (USDA, 1975).
Index nutrients and reference amounts
All calculations were based on the USDA nutrient composi-
tion database and on Food and Drug Administration (FDA)
reference amounts summarized in Table 1 (FDA, 2002,
2007a). Scores based on nutrient profile models developed
in France (Darmon et al., 2005; AFSSA, in press; Maillot et al.,
2007) were recalculated using FDA-specific criteria. The FSA
WXYfm point score was based on the criteria specified by its
authors (Rayner et al., 2005b). The USDA database does not,
for the most part, flag fortified foods.
Table 1 Daily values and maximum recommended values used in
calculation of nutrient profiles, based on 2000 kcal per day
Desirable nutrients Daily values
Protein 50 g
Fiber 25 g
MUFA 20 g
Linoleic acid 9 g
Linolenic acid 1.8 g
DHA 0.11 g
Vitamin A 5000 IU
Vitamin C 60 mg
Vitamin D 400 IU (10 mcg)
Vitamin E 30 IU (20 mg)
Vitamin K 80 mg
Thiamin 1.5 mg
Riboflavin 1.7 mg
Niacin 20 mg
Vitamin B6 2.0 mg
Vitamin B12 6 mcg
Folate 400 mcg
Pantothenic acid 10 mg
Calcium 1000 mg
Iron 18 mg
Magnesium 400 mg
Zinc 15 mg
Phosphorus 1000 mg
Selenium 70 mcg
Copper 2.0 mg
Potassium 3500 mg
Nutrients to limit Maximum recommended values
Fat 65 g
Saturated fat 20 g
Sugars, total 125 g
Sugars, added 50 g
Sodium 2400 mg
Abbreviations: DHA, docosahexanoic acid; MUFA, monounsaturated fatty
acid.
Testing nutrient profile models
A Drewnowski et al
675
European Journal of Clinical Nutrition
Calculation basis
Calculations can be based on 100 g, 100 kcal or per serving of
food. Those based on 100 kcal have the effect of assigning the
highest scores to foods with the highest water content and
lowest energy density. Those based on 100 g ignore the often
substantial differences in portion size and may penalize foods
that are consumed in small amounts. Partly because the EU
lacks a harmonized standard definition of portion size, many
models were based on 100 kcal, 100 g or both. In the US, the
FDA operates using Reference Amounts Customarily Consumed
(RACC), based on dietary survey data. RACC values reflect
real-life eating patterns and have important and immediate
applications to food labeling. Values based on portion sizes may
provide a better way to communicate the concept of nutrient
density to the consumer. For that reason, we also calculated
model scores based on FDA gram values for RACC.
Families of nutrient profile models
Table 2 summarizes the index nutrients included in the
models tested. The models included protein, fiber and a wide
range of micronutrients, mostly vitamins and minerals.
Some models also included monounsaturated fats and
essential fatty acids. Table 3 summarizes the algorithms for
selected families of models calculated per 100 g, per 100 kcal
and per RACC.
The nutrient adequacy scores family of models (NAS23 and
NAS16). These profiles, previously published under the
name of nutrient adequacy scores (NAS) (Darmon et al.,
2005), were based on unweighted arithmetic means of
percentage recommended daily values for a number of index
nutrients. Index nutrients ranged from 5 to 23 and calcula-
tions were based on 100 g of edible food.
The nutrient density score family of models (NDS23, NDS16,
NDS5). Dividing NAS scores by the energy density of food
yielded nutrient density scores (Darmon et al., 2005). The
nutrient density score (NDS) models in published research
(Darmon et al., 2005; AFSSA, in press; Maillot et al., 2007)
were based on unweighted arithmetic means for positive
Table 2 Nutrient basis of selected nutrient profile models
Nutrient profile
model
References Macronutrients Vitamins Minerals Nutrients to limit
NAS23, NDS23 (Maillot et al.,
2007)
Protein, fiber, linoleic acid,
linolenic acid, DHA
Vitamin A, C, D, E, thiamin,
riboflavin, B6, B12, folate, niacin
Ca, Fe, Zn, Mg,
Cu, Se, K, Ph (I)
a
NAS16a, NDS16a (AFSSA, in press) Protein, fiber, linolenic
acid, DHA
Vitamin C, D, E, thiamin, riboflavin,
B6, folate
Ca, Fe, Mg, Zn, K
NAS16b,
NDS16b
(Darmon et al.,
2005)
Protein, fiber Vitamin A, C, D, E, thiamin,
riboflavin, B6, B12, folate, niacin,
panthotenic acid
Ca, Fe, Mg
NNR15 (Drewnowski,
2005)
Protein, fiber, MUFA Vitamin A, C, D, E, thiamin,
riboflavin, B12, folate
Ca, Fe, Zn, K
NR12 Protein, fiber Vitamin A, C, E, thiamin,
riboflavin, B12
Ca, Fe, Zn, K
NR11 Protein, fiber Vitamin A, C, E, B12 Ca, Fe, Mg, Zn, K
NR10 Protein, fiber Vitamin A, C, E, B12 Ca, Fe, Zn, K
NR7 Protein, fiber Vitamin A, C, E Ca, Fe
NR6 Protein, fiber Vitamin A, C Ca, Fe
NR5, NDS5 (AFSSA, in press) Protein, fiber Vitamin C Ca, Fe
LIM (Maillot et al.,
2007)
Saturated fat, added sugar, Na
LIMtot Total fat, total sugar, Na
FSA WXYfm (Rayner et al.,
2005a)
Protein, fiber
F þ V þ nuts (g)
Energy, saturated fat, total
sugar, Na
Abbreviations: AFSSA, Agence Franc¸aise de Se
´
curite
´
Sanitaire des Aliments; DHA, docosahexanoic acid; F, fruits; FSA, Food Standards Agency; LIM, limited nutrient
score; MUFA, monounsaturated fatty acid; Na, sodium; NAS, nutrient adequacy score; NDS, nutrient density score; NR, nutrient rich. V, vegetables;
a
Iodine, which was originally included in the published NAS23 and NDS23 scores, has been replaced by phosphorus in the present calculation.
Testing nutrient profile models
A Drewnowski et al
676
European Journal of Clinical Nutrition
nutrients, calculated per 100 kcal. Data presented in this
study are for NDS profiles based on 5, 16 and 23 nutrients.
NDS models were also calculated per RACC.
The nutrient-rich food family of models. The nutrient-rich
food models (NR
n
, where n ¼ 5–7; 10–12, 15) were based on a
variable number of index nutrients, calculated per 100 kcal
and per RACC. They were derived from the naturally
nutrient-rich score, which was based on protein, fiber,
monounsaturated fat and 12 vitamins and minerals
(Drewnowski, 2005). The NR models were equivalent to the
NDS family, since both were based on a nutrient to energy
ratio. The NDS5 and NR5 models were, in fact, the same.
The limited nutrient score. The published limited nutrient
score (LIM) (Maillot et al., 2007) was based on three
nutrients to limit, calculated per 100 g of food. Index
nutrients were saturated fat, added sugar and sodium.
Reference amounts were based on maximum recommended
values (MRVs) for the French population: 10% of
energy intake for each added sugars and saturated fats
and 6 g per day for sodium. Translated to the FDA
benchmarks, these values became 20 g of saturated fat,
50 g of added sugar and 2400 mg of sodium. A variant
LIMtot score was based on total fat, total sugar and
sodium. LIM scores were also calculated per 100 kcal and
per RACC.
The Food Standards Agency model WXYfm. The FSA model
WXYfm was based on four negative and three positive
nutrients, all calculated per 100 g. The sum of desirable or
positive components (subscore A) was subtracted from the
sum of the negative components (subscore C) to yield the
final score, unless the sum of negative components exceeded
11, in which case it remained the final score (Rayner et al.,
2005b). The FSA WXYfm subscore C was based on energy,
saturated fat, added sugar and sodium, calculated per 100 g.
The FSA subscore A, in addition to protein and fiber, awarded
points based on the foods’ content of fruits, vegetables and
nuts, all calculated per 100 g. The final score had reverse
polarity, with numbers below zero denoting the more
nutritious foods. The present modification used a simpler
method for calculating the amounts (in g) of fruit, vegetables
or nuts present in soups and some infrequent mixed foods
that were part of subscore A.
Statistical analyses
All analyses were performed using SAS software version 9.1.
(SAS institute, Cary, NC, USA) and the Statistical Package for
the Social Sciences version 11.0. Relation between total score
and score components and analyses of colinearity among
nutrient profiles scores was conducted using Pearson’s
correlations. All variables were log transformed for correla-
tion analyses with the exception of the FSA WXYfm subscore
Table 3 A comparison of algorithms and calculation basei for selected nutrient profile models
Nutrient profile model Algorithm Amount Comment
NAS23, NAS16, NAS5 NAS
n
¼ (
P
1n
(Nutrient
i
/DV
i
)/n) 100 100 g A nutrient adequacy measure based on 100 g of food
Nutrient
i
, content of nutrient i in 100 g edible portion
DV
i
, daily values for nutrient i (see Table 1)
n, the number of nutrients
NDS23, NDS16, NDS5 NDS
n
¼ (NAS
n
/ED) 100 100 kcal A nutrient density score based on 100 kcal of food
NDS
n
, NAS
n
divided by ED
NDS23, NDS16, NDS5 NDS
n
¼ (
P
1n
(Nutrient
i
/DV
i
)/n) RACC RACC
a
A nutrient density score based on portion size of food
Nutrient
i
, content of nutrient i in 100 g edible portion
LIM, LIMtot LIM ¼ (
P
13
(L
i
/MRV
i
)/3) 100 100 g Based on MRV
i
(see Table 1) and on food weight (100 g)
L
i
, content of limiting nutrient i in 100 g of edible portion
LIM, LIMtot LIM ¼ (
P
13
(L
i
/MRV
i
)/3) RACC RACC
a
Based on MRV
i
(see Table 1) and on portion sizes
NR
n
NR
n
¼ (
P
1n
(Nutrient
i
/DV
i
)/n) 100 100 kcal Nutrient
i
, content of nutrient i in 100 kcal edible portion
NR
n
per 100 kcal are equivalent to NDS
n
per 100 kcal
NR
n
NR
n
¼ (
P
1n
(Nutrient
i
/DV
i
)/n) RACC RACC Nutrient
i
, content of nutrient i in 100 g edible portion
NR
n
per RACC are equivalent to NDS
n
per RACC
FSA WXYfm See ref (Rayner et al., 2005a) Total score ¼ sum of negative nutrients (C)sum of
positive nutrients (A), unless C411. Complex scoring
system for % nuts, vegetables and fruits
Abbreviations: DV, daily values; ED, energy density; FSA, Food Standards Agency; LIM, limited nutrient score; MRV, maximum recommended values; NAS, nutrient
adequacy scores; NDS, nutrient density score; NR, nutrient rich; RACC, reference amounts customarily consumed.
a
RACC in gram.
Testing nutrient profile models
A Drewnowski et al
677
European Journal of Clinical Nutrition
A and FSA WXYfm total score, which already displayed
approximatively normal distributions. An a-level of 0.05 was
used to determine statistical significance.
Results
Nutrient profile models, energy density and energy cost
As shown in Table 4, the models’ relation to energy density
depended on the type and the number of index nutrients. All
models that were exclusively based on total or saturated fat,
total or added sugar and sodium were highly correlated with
energy density. Generally, the relation to energy density
weakened as fat and sugar was removed and more micro-
nutrients were introduced into the model. Models based on
protein, fiber and three vitamins and minerals were more
strongly linked to energy density than those based on 412
micronutrients.
The calculation basis changed the models’ relation to
energy density in a predictable manner. NAS, calculated per
100 g, were positively linked to energy density and fat
content of foods. NDS and nutrient-rich (NR) score,
calculated per 100 kcal, were inversely related to energy
density and to the fat and added sugar content of foods. So
were NDS and NR score calculated based on RACC. The
relation to energy density weakened as more micronutrients
was introduced into the model.
The relation between model scores and energy cost also
depended on the type and the number of nutrients included.
A strong relation between a model and energy density
usually meant a strong relation between model scores and
energy costs. In other words, such models gave unfavorable
ratings to inexpensive and favorable ratings to more
expensive foods. Thus, both the FSA subscore C and the
LIM score were strongly associated with high energy density
and low energy cost. There were also significant correlations
Table 4 Pearson’s correlations between scores generated by selected nutrient profile models and energy density, percent energy from fat, total
saturated fats (g), added sugars (g), cost per 100 g and cost per 1000 kcal (all values log transformed)
Nutrient profile models Reference
amount
Energy density, Energy from fat Saturated fats Added sugars Cost Energy cost
kcal per 100 g % g per 100 g g per 100 g $ per 100 g $ per 1000 kcal
NAS23 100 g 0.43 0.45 0.39 0.12* 0.48 0.05*
NAS16a 100 g 0.37 0.37 0.30 0.11* 0.47 0.09*
NAS16b 100 g 0.24 0.26 0.16 0.11* 0.43 0.16
NAS5 100 g 0.01* 0.03* 0.08* 0.12* 0.37 0.30
NDS23 100 kcal 0.42 0.01* 0.41 0.36 0.24 0.55
NDS16a 100 kcal 0.48 0.09* 0.49 0.34 0.21 0.57
NDS16b 100 kcal 0.53 0.17 0.53 0.31 0.17 0.58
NR15 100 kcal 0.49 0.06* 0.47 0.32 0.19 0.56
NR12 100 kcal 0.56 0.18 0.54 0.31 0.18 0.61
NR11 100 kcal 0.57 0.20 0.55 0.33 0.17 0.62
NR10 100 kcal 0.57 0.19 0.54 0.34 0.18 0.62
NR7 100 kcal 0.61 0.00* 0.36 0.35 0.05* 0.55
NR6 100 kcal 0.63 0.02* 0.37 0.37 0.03* 0.54
NR5, NDS5 100 kcal 0.62 0.07* 0.39 0.29 0.12 0.61
NDS23 RACC 0.01* 0.28 0.09* 0.42 0.31 0.26
NDS16a RACC 0.06* 0.19 0.01* 0.40 0.28 0.29
NDS16b RACC 0.16 0.09* 0.12 0.38 0.24 0.33
NR15 RACC 0.08* 0.22 0.01* 0.39 0.27 0.29
NR12 RACC 0.19 0.07* 0.14 0.38 0.25 0.37
NR11 RACC 0.22 0.05* 0.17 0.39 0.25 0.40
NR10 RACC 0.23 0.05* 0.17 0.40 0.25 0.40
NR7 RACC 0.28 0.20 0.03* 0.27 0.13 0.34
NR6 RACC 0.30 0.17 0.04* 0.29 0.11 0.34
NR5, NDS5 RACC 0.30 0.11 0.08* 0.20 0.19 0.40
LIM 100 g 0.72 0.52 0.79 0.58 0.13 0.47
LIMtot 100 g 0.77 0.52 0.75 0.25 0.19 0.47
LIM 100 kcal 0.33 0.38 0.57 0.36 0.00* 0.27
LIMtot 100 kcal 0.05* 0.15 0.10 0.15* 0.06* 0.01*
LIM RACC 0.52 0.47 0.70 0.06* 0.06* 0.38
LIMtot RACC 0.40 0.41 0.54 0.44 0.05* 0.29
FSA subscore A
a
100 g 0.34 0.11 0.13 0.15* 0.26 0.06*
FSA subscore C 100 g 0.77 0.36 0.64 0.24 0.17 0.49
FSA total
a
100 g 0.71 0.43 0.66 0.38 0.20 0.41
Abbreviations: FSA, Food Standards Agency; LIM, limited nutrient score; NAS, nutrient adequacy score; NDS, nutrient density score; NR, nutrient rich; RACC,
reference amounts customarily consumed.
*All P-values o0.05, unless indicated.
a
Not log transformed.
Testing nutrient profile models
A Drewnowski et al
678
European Journal of Clinical Nutrition
with the fat and added sugar content of foods. The
composite FSA model performed much the same as the
subscore C. There were high and significant correlations
between FSA subscore C, FSA total score, the LIM scores and
energy density.
Correlations among nutrient profile models
Table 5 shows Pearson’s correlations among nutrient profile
models. The NDS and NR nutrient density profiles were
highly correlated with each other. The correlations at 0.93
level even as the number of nutrients in the model was
reduced from 23 to 10. Correlations were reduced further
(r ¼ 0.78) as the number of positive nutrients was stripped
down to basic five. In other words, nutrient profile models
in the NDS family yielded similar results whether based on
10 or 23 nutrients. The addition of further vitamins and
minerals beyond some optimum of 10 or 11 may not alter
nutrient profiles very much.
The NDS and NR nutrient profiles, based on 100 kcal, were
negatively correlated with LIM scores. It was to be expected
given that LIM scores were so closely linked to the energy
density of foods. However, the surprise was that the FSA
WXYfm model subscore C and the FSA WXYfm total score
were highly correlated with LIM and even more so with
LIMtot.
Nutrient profiles for major food groups
Figure 1 (top panel) shows the relation between median
nutrient density scores and median energy density for the
nine food groups. First, it can be seen that there was an
inverse relation between energy and nutrient densities
ratings. Sweets and snacks, added fats and cereals were
energy dense and had lower nutrient density scores.
Vegetables and fruit were energy poor but nutrient rich.
Milk and meat products had energy density in the 150–
200 kcal per 100 g range and mid-range nutrient density
scores. Caloric beverages were an exception, since they were
low-energy density and had low nutrient density scores.
It can be seen that the nutrient density models, based on
23, 11 and 5 nutrients behaved similarly. Each awarded
highest ratings to vegetables and beans followed by fruits,
meat and dairy. Because the NDS23 model included poly-
unsaturated fats among desirable nutrients, median scores in
the meat group were actually higher than those in the fruit
group. Breads and cereals, including fortified ones, received
medium scores. Sweets and snacks, fats and beverages tended
to receive the lowest scores. The inverse relation to energy
density was strengthened as the number of nutrients in the
model was reduced from 23 to 5.
Figure 1 (bottom panel) shows the relation between
median nutrient density scores and median LIM scores for
the same nine food groups. It can be seen that the LIM scores
were not substantially different from energy density scores.
Figure 2 (top panel) shows a very strong relation between
LIM scores and energy density of foods. The highest ratings
were given to fats, sweets and snacks, and to dry breads and
cereals. Meats, milk and milk products received medium
scores, whereas low-energy-density fruits and vegetables
received the most favorable scores. The emphasis on energy
density meant that low-energy-density caloric beverages
scored better as compared to yogurts or cheese products
though not as well as vegetables and fruit.
Figure 2 (bottom panel) shows the inverse relation
between LIM scores and energy cost. As expected, foods
with high energy density were associated with lower energy
costs.
Table 5 Pearson’s correlation between scores generated by NDS/NR models and the FSA WXYfm model, calculations based on 100 kcal (all values log
transformed)
NDS23 NDS16 a NDS16 b NR15 NR12 NR11 NR10 NR7 NR6 NR5 FSA good
subscore A
FSA bad
subscore C
FSA
total
LIM
100 kcal 100 kcal 100 kcal 100 kcal 100 kcal 100 kcal 100 kcal 100 kcal 100 kcal 100 kcal 100 g 100 g 100 g 100 g
NDS23 1.00
NDS16a 0.97 1.00
NDS16b 0.94 0.94 1.00
NR15 0.96 0.93 0.96 1.00
NR12 0.93 0.93 0.99 0.97 1.00
NR11 0.93 0.93 0.98 0.96 0.99 1.00
NR10 0.93 0.92 0.98 0.96 0.99 1.00 1.00
NR7 0.89 0.93 0.89 0.95 0.96 0.97 0.97 1.00
NR6 0.87 0.92 0.87 0.95 0.95 0.96 0.96 0.99 1.00
NR5, NDS5 0.78 0.85 0.91 0.83 0.90 0.90 0.89 0.90 0.94 1.00
LIM 0.34 0.41 0.42 0.37 0.43 0.45 0.44 0.48 0.49 0.49 0.24 0.72 0.67 1.00
LIMtot 0.30 0.35 0.40 0.32 0.40 0.41 0.40 0.42 0.44 0.46 0.40 0.81 0.74 0.83
FSA subscore A
a
0.18 0.18 0.19 0.17 0.17 0.15 0.15 0.15 0.11 0.19 1.00
FSA subscore C 0.47 0.50 0.53 0.51 0.54 0.55 0.54 0.58 0.59 0.55 0.26 1.00
FSA total
a
0.43 0.48 0.51 0.43 0.50 0.52 0.51 0.54 0.55 0.58 0.18 0.63 1.00
Abbreviations: FSA, Food Standards Agency; LIM, limited nutrient score; NDS, nutrient density score; NR, nutrient rich.
a
Not log transformed.
Testing nutrient profile models
A Drewnowski et al
679
European Journal of Clinical Nutrition
Discussion
Classifying foods based on their nutrient profiles have a
number of regulatory and marketing applications, but not
in the EU (COI, 2004; FSA, 2005a). The US FDA may also
consider the use of nutrient profile-based symbols to
communicate nutrition information to the consumer (FDA,
2004, 2007b). Both the 2005 Dietary Guidelines for Americans
and the 2005 USDA MYPyramid featured nutrient density as a
useful tool for consumer education (USDA, 2005; USDA/
DHHS, 2005; ADA Reports, 2007). Similar principles were used
to develop nutrition standards for schools (IOM, 2007).
As more nutrient profiles are developed, it is good to show
how various classes of models behave. These are the first tests
0
50
100
150
200
250
300
350
400
0 2 4 6 8 10 12 14 16 18 20
Nutrient scores, %/100kcal
Energy density, kcal/100g
NDS23 NR11 NDS5
BEVERAGE
FRUITS
VEGETABLES & BEANS
MIXED FOODS
MILK & MILK PRODUCTS
MEATS/POULTRY/FISH/EGGS/NUTS
CEREALS
ADDED FATS
SWEETS & SNACKS
0
5
10
15
20
25
0 2 4 6 8 10 12 14 16 18 20
Nutrient Scores, %/100kcal
LIM score, %/100g
NDS23
NR11
NDS5
ADDED FATS
SWEETS & SNACKS
CEREALS
MEATS/POULTRY/FISH/EGGS/NUT
MILK & MILK PRODUCTS
MIXED FOODS
BEVERAGE
FRUITS
VEGETABLES & BEANS
Figure 1 Relation between energy density and nutrient density scores (top panel) and between limited nutrient score (LIM) and nutrient
density scores. Nutrient density scores calculated for nine major food groups using model profiles NDS23, NR11 and NDS5.
Testing nutrient profile models
A Drewnowski et al
680
European Journal of Clinical Nutrition
of nutrient profile models against the foods’ energy density
and energy cost. Nutrient profile models based on protein,
fiber, vitamins and minerals were, in essence, nutrient to
calorie ratios. Their algorithms operationalized the FDA
definition that nutritious foods should provide ‘substantial’
amounts of desirable nutrients in relation to ‘few’ calories
(FDA, 2007b). Not surprisingly, these models were negatively
related to energy density and positively related to nutrient
cost. The relation to energy density was stronger when the
model was based on few micronutrients and was attenuated
as more vitamins and minerals were introduced into the
model. The one surprise was that adding more vitamins and
0
5
10
15
20
25
0 50 100 150 200 250 300 350 400
Energy Density, kcal/100g
LIM score, %/100g
ADDED FATS
SWEETS & SNACKS
CEREALS
MEATS/POULTRY/FISH/EGGS/NUT
MILK & MILK PRODUCTS
MIXED FOODS
BEVERAGE
FRUITS
VEGETABLES & BEANS
0
5
10
15
20
25
Ener
g
y cost, $/1000kcal
LIM score, %/100g
BEVERAGE
FRUITS
VEGETABLES
& BEANS
MIXED FOODS
MILK & MILK PRODUCTS
MEATS/POULTRY/FISH/EGGS/NUT
CEREALS
ADDED FATS
SWEETS & SNACKS
02461357
8
Figure 2 Relation between limited nutrient score (LIM) and energy density (top panel) and LIM and between energy cost (bottom panel) for
nine main food groups (based on 378 foods).
Testing nutrient profile models
A Drewnowski et al
681
European Journal of Clinical Nutrition
minerals to the model beyond a certain limit had little
additional impact on food-group rankings. In other words, a
model based on 23 positive nutrients provided rankings
similar to those generated by a model based on 9 or 11
positive nutrients, with correlation levels exceeding 0.90.
This is an important consideration, since regulatory agencies
would most likely prefer a minimal number of nutrients
for the ease of enforcement, whereas models based on an
‘optimal’ number might show higher correlations with a
healthy diet.
Those models that were based exclusively on nutrients to
limit were highly correlated with energy density but less well
with the nutrient content of foods, as tracked by alternative
models. This was true of the LIM and LIMtot scores and of
the FSA WXYfm model subscore C. The close links to energy
density meant that the models provided little additional
information beyond calories. The FSA WXYfm total score, in
particular, was little more than a function of energy density,
with beneficial micronutrients contributing relatively little
to the total score.
For the most part, energy-dense foods were associated with
lower energy costs, whereas nutrient-dense foods were
associated with higher energy costs, consistent with our past
research (Darmon et al., 2005; Maillot et al., 2007). The cost
issue bears watching because the foods deemed to be
undesirable or unhealthy are typically cheaper (per calorie)
as compared to the more nutrient-dense options. For
example, foods deemed ‘healthy’ by the FSA model were
far more costly than those deemed less healthy and less
desirable (Drewnowski, 2007).
The present analyses were intended to demystify the
process of nutrient profiling. Comparing model scores to
energy density and cost should be considered tests of model
performance, rather than efforts at validation, since they did
not make use of dietary intake data. By contrast, a validation
technique (ILSI, 2006) compared rankings generated by three
models with a set of indicator foods that were positively or
negatively associated with ‘healthy’ diets, as defined using
Eurodiet criteria and national dietary surveys from five EU
countries. Development of validation methods ought to be a
high research priority.
Additional studies need to link nutrient profiles of foods
to other determinants of food choice—such as food
preferences and food costs (Drewnowski, 2007). We must
remember that foods are consumed for more than just
sustenance and nutrition. Food choices and eating habits
provide opportunities for social interaction and food is a
source of considerable pleasure to consumers. Finally,
the usefulness of the chosen model needs to be tested
among nutrition professionals and among consumer
groups of different socioeconomic status (COI, 2004).
Regulatory agencies should act only when they are satisfied
that the scientific process has been followed; that the
algorithms are transparent, and the profile model has
been validated with respect to objective measures of a
healthy diet.
Acknowledgements
A Drewnowski was supported in part by funds from the
National Cattlemen’s Beef Association and the National
Dairy Council. This work was carried out with financial
support of the ANR-Agence Nationale de la Recherche—The
French National Research Agency under the Programme
National de Recherche en Alimentation et nutrition
humaine, project ANR-07-PNRA-018, ALIMINFO.
References
ADA Reports (2007). Practice paper of the American Dietetic
Association: nutrient density: meeting nutrient goals within
calorie needs. J Am Diet Assoc 107, 860–869.
AFSSA. Les Profils nutritionnels. Report (in press).
Andrieu E, Darmon N, Drewnowski A (2006). Low-cost diets: more
energy, fewer nutrients. Eur J Clin Nutr 60, 434–436.
Azais-Braesco V, Goffi C, Labouze E (2006). Nutrient profiling:
comparison and critical analysis of existing systems. Public Health
Nutr 9, 613–622.
COI (2004). Food promotion and marketing to children: views of low
income consumers. Communications on Behalf of Food Standard
Agency. [Monograph on the internet] 2004 Research Works Ltd.:
London, Available at: http://www.food.gov.uk/multimedia/pdfs/
foodpromoreport.pdf.
Darmon N, Briend A, Drewnowski A (2004). Energy-dense diets are
associated with lower diet costs: a community study of French
adults. Publ Health Nutr 7, 21–27.
Darmon N, Darmon M, Maillot M, Drewnowski A (2005). A nutrient
density standard for vegetables and fruits: nutrients per calorie
and nutrients per unit cost. J Am Diet Assoc 105, 1881–1887.
Drewnowski A (2005). Concept of a nutritious food: toward a
nutrient density score. Am J Clin Nutr 82, 721–732.
Drewnowski A (2007). What’s next for nutrition labeling and health
claims: an update of nutrient profiling in the European Union and
the US. Nutr Today 42, 206–214.
Drewnowski A, Monsivais P, Maillot M, Darmon N (2007). Low-
energy-density diets are associated with higher diet quality and
higher diet costs in French adults. J Am Diet Assoc 107, 1028–1032.
European Food Safety Authority (EFSA) (2007). EFSA 9th scientific
colloquium—nutrient profiling for foods bearing nutrition and
health claims. Parma, Accessed at:www.efsa.europa.eu.
FDA (2002). Code of federal regulations, title 21-Food and drugs
(volume 2), Chapter I-Food and Drug administration, Department
of Health and Human Services, Part 101-Food Labeling-Table of
Contents, Subpart A-General Provisions, Sec 101.14 Health claims:
General requirements and Subpart E-Specific requirements for
Health Claims. US Government Printing Office, Revised as of 1 April
2002.
FDA (2004). Calories Count. Report of the Working Group on Obesity.
[Monograph on the Internet]. Available at: www.cfsan.fda.gov/
Bdms/nutrcal.html.
FDA (2007a). Code of Federal Regulations, Sec 101.9 Nutrition labeling
of food. Accessed at: http://www accessdata.fda goc/scripts/cdrh/
cfdocs/cfcfr/cfrsearch. cfm?fr ¼ 109. 9.
FDA (2007b). Food labeling: use of symbols to communicate
nutrition information, consideration of consumer studies and
nutritional criteria; public hearing, request for comments. FDA
Docket No. 2007N-0277.
FSA (2005a). Nutrient profiling, Agenda item 7. 13 October 2005
[Monograph on the Internet] Available at: http://www.food.
gov.uk/multimedia/pdfs/fsa051004.pdf.
FSA (2005b). Scientific workshop to assess the Food Standards
Agency’s proposed approach to nutrient profiling. 25th February
Testing nutrient profile models
A Drewnowski et al
682
European Journal of Clinical Nutrition
2005. Available at: http://www.food.gov.uk/multimedia/pdfs/
nutprofworkshop250205.pdf.
Gazibarich B, Ricci PE (1998). Towards better food choice: the
nutritious food index. Aust J Nutr Diet 55, 10–20.
IFN (2006). Confe
´
rence de l’Institut Franc¸ais pour la Nutrition sur
les profils nutritionnels des aliments: actualites, enjeux et
perspectives. Symposium 12 June 2006. Available at: http://www.
alimentation-sante.com/public/actes/juin06r.htm.
ILSI (2006). International Life Sciences Institute Europe Workshop
on nutritional characteristics of foods: science-based approach to
nutrient profiling. Mallorca, 25–27 April 2006. Available at: http://
www.ilsi.org/ Accessed March 10, 2007.
IOM (2007). Nutrition standards for foods in schools: leading the
way toward healthier youth. Food and Nutrition Board, Institute of
Medicine. National Academies Press: Washington DC.
Labouze E, Goffi C, Moulay L, Azais-Braesco V (2007). A multi-
purpose tool to evaluate the nutritional quality of individual
foods: Nutrimap. Public Health Nutr 10, 690–700.
Maillot M, Darmon N, Darmon M, Lafay L, Drewnowski A (2007).
Nutrient-dense food groups have high energy costs: an econo-
metric approach to nutrient profiling. J Nutr 137, 1815–1820.
Monsivais P, Drewnowski A (2007). The rising cost of low energy
density foods. J Am Diet Assoc 107, 2071–2076.
Netherlands nutrition center (2007). Criteria for the nutritional
evaluation of food. The Netherlands tripartite classification model
for foods Available at:www. voedingscentrum. nl/NR/rdonlyres/
0AF85A19-79B1-4DB5-A0E8-C8BFFD44B089/0/Criteriaengelssite.
pdfAccessed 8 June 2007.
Nijman CA, Zijp IM, Sierksma A, Roodenburg AJ, Leenen R,
van den KC et al. (2007). A method to improve the nutritional
quality of foods and beverages based on dietary recommendations.
Eur J Clin Nutr 61, 461–471.
Patterson RE, Kristal AR, Tinker LF, Carter RA, Bolton MP,
Agurs-Collins T (1999). Measurement characteristics of the
Women’s Health Initiative food frequency questionnaire. Ann
Epidemiol 9, 178–187.
Rayner M, Scarborough P, Stockley L (2004). Nutrient
profiles: options for definitions of use in relation to food
promotion and children’s diets. Final report, 2004 [Monograph
online]. Available at: http://www.food.gov.uk/multimedia/pdfs/
nutrientprofilingfullreport.pdf.
Rayner M, Scarborough P, Stockley L (2005a). Nutrient profiles:
applicability of currently proposed model for uses in relation
to promotion of food to children aged 5–10 and adults.
2005. [Monograph online]. Available at: http://www.food.gov.uk/
multimedia/pdfs/nutprofmodelforadults.pdf.
Rayner M, Scarborough P, Boxer B, Stockley L (2005b). Nutrient
profiles: development of final model. [Monograph online] Avail-
able at: http://www.food.gov.uk/multimedia/pdfs/nutprofr.pdf.
Rayner M, Scarborough P, Stockley L, Boxer A (2005c). Nutrient
profiles; further refinement and testing of model SSCg3d. Final
Report 2005. [Monograph online]. Available at: http://www.food.
gov.uk/multimedia/pdfs/npreportsept05.pdf.
Scarborough P, Boxer A, Rayner M, Stockley L (2007a). Testing
nutrient profile models using data from a survey of nutrition
professionals. Public Health Nutr 10, 337–345.
Scarborough P, Rayner M, Stockley L (2007b). Developing nutrient
profile models: a systematic approach. Public Health Nutr 10,
330–336.
Scarborough P, Rayner M, Stockley L, Black A (2007c). Nutrition
professionals’ perception of the ‘healthiness’ of individual foods.
Public Health Nutr 10, 346–353.
Stockley L (2007). Nutrition profiles for foods to which nutrients
could be added, or on which health claims could be
made. Experiences from other countries and testing
possible models. Final Report Prepared for the UK Food Standards
Agency. 2003 Available at: http://www.food.gov.uk/multimedia/
pdfs/nutritionclaims.pdfAccessed 30 March 2007.
The European Parliament and the Council of the European Union
(2006). Regulation (EC) No 1924/2006 of the European Parliament
and of the Council of 20 December 2006 on nutrition and health
claims made on foods. OJEU 404, 9–25.
USDA (1975). Food Yields Summarized by Different Stages of Preparation.
USDA Agricultural Research Service Handbook # 102: Washington,
DC 1975.
USDA (2005). The Food Guide Pyramid. Washington; 2005–7.
Available at: http://www.mypyramid.gov/.
USDA (2006). USDA food and nutrient database for dietary studies,
2.0. (2006). [Monograph on the Internet]. Agricultural Research
Service, Food Surveys Research Group: Beltsville, MD, Accessed 8
June 2007\.
USDA (2007). Composition of foods raw, processed, prepared.
USDA National nutrient database for standard reference
SR-19. Available at: www.nal.usda.gov/fnic/foodcomp/Data/SR19/
sr19_doc.pdfAccessed 12 August 2007.
USDA/DHHS (2005). Dietary guidelines for Americans by US
department of health and human services and US department of
agriculture [Monograph on the Internet] 6th edn., US Govern-
ment Printing Office: Washington, DC January 2005, http://
www.health.gov/dietaryguidelines/dga2005/document/Accessed 7
June 2007.
Zelman K, Kennedy E (2005). Naturally nutrient richyputting more
power on Americans’ plates. Nutr Today 40, 60–68.
Testing nutrient profile models
A Drewnowski et al
683
European Journal of Clinical Nutrition
    • "Those three nutrients were also used in the creation of the limiting nutrients (LIM subscore) in both the French SAIN, LIM system [37] and the US-based Nutrient Rich Foods Index [39]. Comparative studies have also shown that the LIM subscore was highly correlated with the FSA-Ofcom nutrient profiling model [40], used in the UK to regulate advertising and marketing to children [10]. In the present analyses, we observed reductions in the LIM components as well as small reduction in the total LIM subscore (data not shown). "
    [Show abstract] [Hide abstract] ABSTRACT: Purpose To describe the Nestlé Nutritional Profiling System (NNPS) developed to guide the reformulation of Nestlé products, and the results of its application in the USA and France. Design The NNPS is a category-specific system that calculates nutrient targets per serving as consumed, based on age-adjusted dietary guidelines. Products are aggregated into 32 food categories. The NNPS ensures that excessive amounts of nutrients to limit cannot be compensated for by adding nutrients to encourage. A study was conducted to measure changes in nutrient profiles of the most widely purchased Nestlé products from eight food categories (n = 99) in the USA and France. A comparison was made between the 2009–2010 and 2014–2015 products. Results The application of the NNPS between 2009–2010 and 2014–2015 was associated with an overall downwards trend for all nutrients to limit. Sodium and total sugars contents were reduced by up to 22 and 31 %, respectively. Saturated Fatty Acids and total fat reductions were less homogeneous across categories, with children products having larger reductions. Energy per serving was reduced by <10 % in most categories, while serving sizes remained unchanged. Conclusions The NNPS sets feasible and yet challenging targets for public health-oriented reformulation of a varied product portfolio; its application was associated with improved nutrient density in eight major food categories in the USA and France. Confirmatory analyses are needed in other countries and food categories; the impact of such a large-scale reformulation on dietary intake and health remains to be investigated.
    Full-text · Article · Feb 2016
    • "Vegetables and raw vegetables are nutrient-rich foods (NRFs) as confirmed by their 45 th and 70 th rank in the logarithmic scale of nutritional density that ranges between 1 and 100 [Drewnowski & Fulgoni, 2008; Drewnowski, 2009] . Moreover, the high nutritional value of the dishes prepared in our study was validated by the analysis of nutrient profile models (NPMs), indicating that the models based on protein, fibre, vitamins and minerals correlate positively with nutrient density and inversely with energy density of food [Drewnowski, 2007; 2009; Drewnowski et al., 2009] . The content of protein and fibre per 100 g of chicken salad amounted to 10.81 g and 3.47 g, respectively; and in spaghetti with tomatoes to 4.72 g and 2.86 g, respectively. "
    [Show abstract] [Hide abstract] ABSTRACT: The research was stimulated by the lack of data on the nutritional value of several dishes preferred by athletes. The aim of this study was to determine the nutritional value of selected designed dishes, grilled chicken salad and spaghetti with tomatoes and parmesan. The examined material was analysed for contents of protein, fat, fibre, and dry matter according to respective standards. Energetic value was calculated using Atwater factors. The energy value of spaghetti with tomatoes and parmesan cheese amounted to 81.1 kcal/100 g and was statistically significantly higher (P<0.001) than that of the grilled chicken salad (67.0 kcal/100 g of product), which was associated with a significantly higher content of total carbohydrates (15.54 vs. 2.77 g/100 g), and significantly lower contents of protein (3.83 vs. 7.25 g/100 g) and fat (1.33 vs. 4.04 g/100g). The content of dietary fibre in examined dishes was similar, and amounted to 2.32 g/100 g and 2.33 g/100 g in the spaghetti and salad, respectively (P<0.001). The ratio of saturated to monounsaturated and polyunsaturated acids per 100 g of the chicken salad and spaghetti with tomatoes and parmesan cheese was 0.68:2.98:0.38 g and 0.54:0.61:0.18 g, respectively. High contents of protein in grilled chicken salad and digestible carbohydrates in spaghetti with tomatoes and parmesan, as well as favourable fatty acid profile substantiates their use as part of the balanced diet for sportspersons.
    Article · Dec 2013
    • "Although there is already a considerable body of research examining consumer responses to food nutrition labels in general, and specifically to FOP labelling schemes (Kraus et al. 2010; Kelly et al., 2007; Jones and Richardson, 2007; Kozup et al., 2003), evidence of the impact on consumer behaviour is marginal at best (Nayga 2008). Additionally, few articles have considered ways different (voluntary) nutritional criteria or nutrient profile models may influence the adoption of FOP labelling or food product formulation (notable examples are Moser et al. 2010; Azaïs-Braesco et al. 2009; Drewnowski et al. 2009; Nijman et al. 2007; Young and Swinburn 2002). Studies measuring actual nutritional quality of products using voluntary FOP labels have been limited by the rapidity, scope and proprietary nature of product (re)formulations. "
    [Show abstract] [Hide abstract] ABSTRACT: The adoption of voluntary front-of-pack (FOP) nutrition labels by UK food retailers and manufacturers is explored. These labels highlight key nutrients, facilitating product comparisons. Information for 2,201 products launched between 2007 and 2009 was analysed. Binary and multinomial logistic regression models explore drivers of FOP label use. Products introduced more recently by retailers and certain food categories were more likely to carry FOP labels. Increasing the content of sodium and sugar decreased odds of FOP use in some categories, but with limited significance. Discussion includes policy options to optimise firm response and implications for evolving mandatory FOP labelling proposals.
    Article · Oct 2011
Show more