The shortest way to reach nutritional goals is to adopt Mediterranean food choices: evidence from computer-generated personalized diets
The shortest way to reach nutritional goals is to adopt Mediterranean food choices: evidence from computer-generated personalized diets1-3
Background: Dietary guidelines can be derived from dietary patterns known to be healthy such as the traditional Mediterranean diet. They can also be deduced by translating a set of nutrient recommendations into food combinations. However, the latter may vary depending on the decisions made by different expert committees. Objective: The objective was to compare the effect of removing or adding selected nutrient recommendations on the dietary changes needed to fulfill a whole set of nutrient recommendations. Design: For each adult participating in the French INCA dietary survey (Enque?te Individuelle et Nationale sur les Consommations Alimentaires) (n = 1171), a nutritionally adequate diet was modeled that simultaneously met a whole set of nutrient goals (proteins, fiber, essential fatty acids, 10 vitamins, 9 minerals, sodium, saturated fatty acids, free sugars) while deviating the least from the observed diet in terms of food content. Eight sets of models were developed according to the inclusion or not of constraints on total fats, total carbohydrates, total MUFAs, and cholesterol. Results: Compared with the observed intakes, fulfilling the whole set of nutrient constraints systematically decreased total fats and increased total carbohydrates, even in the absence of specific constraints on those macronutrients. For whichever model used, a strong consistency was observed in the dietary changes needed to fulfill the constraints, and the greatest increases were seen for unsalted nuts, unrefined grains, legumes, fruit, fish and shellfish, and vegetables. Conclusion: Whether recommendations on total fats, MUFAs, or total carbohydrates are included or not in the definition of overall nutrient adequacy, foods typical of the Mediterranean diet are needed to reach overall nutrient adequacy. Am J Clin Nutr 2011;94:1127-37.
Nutrient recommendations try to synthesize, through nutrients,
all of the knowledge available on the relations between diet and
). However, people consume foods, not nutrients. It is
thus necessary to advise the general public about healthy food
patterns through dietary guidelines. Such advice can be based on an
existing dietary pattern known to be associated with good health,
such as the traditional Mediterranean diet (
). The advice can
also be based on dietary guidelines specifically developed to help
people from a given population meet their nutrient needs in a way
that is culturally and socioeconomically acceptable (
Identifying which food should be consumed and in what
quantity to fulfill the nutrient needs of a target population is
a complex task that requires sophisticated approaches (
Individual diet modeling, in particular, is able to translate nutrient
recommendations into combinations of real food choices while
taking into consideration the individual variability of eating
patterns in a given population (
One key problem in deriving dietary guidelines from nutrient
recommendations is that the latter may vary depending on the
positions adopted by different expert committees. This is
particularly true for macronutrient intakes, for which clear evidence
of a causal effect on health and disease is often lacking (
Thus, low-fat (15?30% of energy) high-carbohydrate (55?75% of
energy) diets with a low cholesterol content (,300 mg/d) have
long been considered as the standard for a healthy diet and were
widely recommended (
). However, the harmful consequences
of very high refined carbohydrate and sugar intakes, in terms of
obesity, diabetes, and cardiovascular disease (
), have led
international committees to increase the acceptable upper intake
level of total fats (currently 35%) (21) and to decrease the
minimum recommended intake of total carbohydrates (currently
). Regarding specific lipids, such as MUFAs and
cholesterol, the debate is both about their desired level and whether
or not to set a specific recommendation for them (
Surprisingly, the consequence of modifying a given nutrient
recommendation on its feasibility and the dietary changes needed
to meet the recommended intakes for a whole set of nutrients
has seldom been addressed (
The objective of the present study was therefore to analyze the
effect of introducing recommendations on total fats, total
carbohydrates, MUFAs, and cholesterol and combinations of these
on the smallest required changes in food choices to reach overall
nutrient adequacy for each individual in a representative sample
of French adults.
The dietary data used in the present study were derived from the
7-d food records of adults (n = 1171; age .18 y) of the INCA
crosssectional dietary survey (Enque?te Individuelle et Nationale sur les
Consommations Alimentaires) conducted in 1999 (
). A food
composition database was specifically developed for this survey
). Alcoholic beverages, tea, and coffee were excluded from all
analyses. All foods in the database were aggregated into 7 groups
and 20 categories. For instance, the ?unsalted nuts,? ?vegetables,?
and ?fruit? categories (including fruit juices) were collapsed into the
?fruit & vegetables? group. The ?mixed dishes? category included
food such as couscous-based dishes, paella, soups, and mixed meat
and vegetable casseroles, and the ?sandwich & pizza? category
included food such as pizzas, quiches, and sandwiches; both were
collapsed into the ?mixed & sandwiches? group.
Individual diet optimization
The aim of individual diet modeling is to design, for each
individual in a population, an optimized diet that is both
individual-specific and nutritionally adequate (
). As previously
), for each INCA participant, the optimization
process started from his or her observed diet (ie, 7-d food record
and nutrient intakes) to design a new weekly modeled diet that
simultaneously conformed with a whole set of nutrient
constraints and that was the closest to his or her usual eating pattern
to optimize acceptability.
As previously reported (
), the objective function was aimed
at 1) preferentially choosing foods already declared as
consumed by each individual in the 7-d food record, 2) minimizing
the reduction of foods already consumed, and 3) introducing (in
the lowest quantity possible) foods that were not declared as
consumed (only if necessary for the fulfillment of the
constraints) and preferentially selecting foods most frequently eaten
by the French population to favor acceptability.
Constraints on food
Constraints on food quantities were introduced to ensure realism
in the optimized diets. As previously described (
), an upper limit
was set for the quantity of each particular food, each food category,
and each food group. For a given food, the 95th percentile of
consumers? intake distribution (ie, the distribution of quantities
consumed by adults who consumed the food) was calculated and
considered as an upper bound for each food in the database. For
food groups and food categories, these limits were defined by the
95th percentile of the population intake distribution calculated for
each sex. The total weight of optimized diets could not exceed
115% of the corresponding observed diet weight.
Constraints on nutrients
A common set of nutritional constraints was introduced in each
case to ensure the nutritional adequacy of the optimized diets. As
done and discussed before (
), each optimized diet was
isocaloric to its corresponding observed diets and met a whole set of
nutritional constraints (Table 1). The latter constraints were
based on the WHO recommendations for proteins (
linoleic acid, a-linolenic acid, DHA and EPA, total omega-3 fatty
acids, total PUFAs (21), cholesterol, saturated fatty acids
(SFAs), and free sugars (
); on the French recommendations
) for fiber, 10 vitamins, and 9 minerals; and on the Nordic
Nutrient Recommendations (
) for sodium.
For most nutritional constraints, the constraint levels were
identical for all individuals, except for some vitamins and minerals.
As previously described (
), safe upper limits for niacin, folate,
ascorbic acid, vitamin A, vitamin B-6, vitamin E, zinc, and selenium
were also applied in the optimized diets.
In the present study, 2 modifications were introduced in the set
of nutritional constraints as compared with previous related
). First, the French Recommended Dietary
Allowance for fiber used in this study was 30 g/d for men and
women instead of 25 g/d. Second, the vitamin D constraint
defining its minimum required level in the optimized diets was
removed, given the difficulty to obtain sufficient amounts of
vitamin D without relying on fortified dietary sources (
Consequently, in this study, we assumed that the inadequate
vitamin D intake would be possibly corrected by either food
fortification or supplementation, as generally agreed on (
Definition of several models derived from a BASAL model
Different optimized models were developed to test the effect of
nutritional constraints, defined by fat and carbohydrate
contributions to total energy intake, on the dietary changes needed to
reach nutrient adequacy. Indeed, debated issues on those
macronutrients have led international committees to recently revise their
acceptable range for a healthy diet. We therefore used recent ranges
?ie, 20?35% of energy and 50?75% of energy from total fats (
and total carbohydrates (
), respectively. We also tested the effect
of introducing constraints on cholesterol and MUFA. Even though
most committees recommend diets with a low cholesterol content
3, 18, 36
), others do not specify any recommended cholesterol
intake for the general population (
1, 26, 37
). As for MUFA, most
committees do not recommend a specific intake of MUFAs (
whereas others do specify a desired range of MUFA intake (
For cholesterol, we used an upper bound of 300 mg/d, because,
when specified, this maximal value was generally set for
cholesterol intake (
). For MUFA, we used the recommendation
given by both the Spanish Heart Foundation and the French Food
Security Agency?ie, 15?20% of total energy intake (
Depending on the addition or subtraction of these constraints (or
a combination of them), 8 different models were obtained (Table 1).
The BASAL model was the first designed and included all of the
constraints except for MUFA (which is not usually included in
nutrient recommendations). Then, 4 other models were derived from
the BASAL model by subtracting separately or simultaneously the
constraints on total fats, total carbohydrates, and cholesterol (noted as
?- Lip?, ?- Carb,? ?-Chol,? ?-[Lip,Carb,Chol]?). Finally, 3 models
(noted ?+MUFA?) were also defined in reference to the traditional
Mediterranean diet pattern (
) and the new French
recommendations for fatty acids in which a range of 15?20% of energy
from MUFA is recommended (27). For each model, the modeling
process generated 1171 optimized diets from the 1171 observed diets.
Choice of a model: the SELECTED model
The 8 models were compared on the basis of their effect on the
macronutrients (ie, proteins, total fats, carbohydrates, MUFAs), on
[M ,b ile
+ 2 ra e e e p
e e e p
L C r r r
, F F F A
]- ] D
LA FA ra T ie ie
S U ,C C e e p p
A M ip EL re re p p
B [ F F A A
BA [M t(
L A ) d d d d
A F l ie ie ie ie
a l l l l
S U to p p p p
p p p p
- l o
L , h
, R en in e eh I ly
ts , s r t O s
e m a e t u
i A r e w n e o
d D o r h i
4A edv ;R fd ehw lrsa ccou ten repv
r e un , e a h
2 RD seb tak bo /gd inm tno w ree
i A w
a 03 9 k D
i ea ch
am t n
ti :a re
v s F
d w e
ic llo th
i s s
n a e
t e lu
o r a
n w .
p d 4
, se (1
c p A
ia im D
n s R
n le e
v ev th
a l d
R -6 eh A
A B t
E s R
C fl n
. o . m a
t b u
h ir m m R
ig , iu i
ew ien len inm eEA
tla iam sed eh th
to th n T en
, a .t e
A ,e n w
in in i e
d tr b
m o u s
a i n a
+ ls A A A A ls ls ls ls ls ls ls ls ls
e e e e e e e e e d
d d d d d d d d d e
o o o o o o o o o rv
m m m m m m m m m e
ll ll ll ll ll ll ll ll ll b
a a a a a a a a a o
f . it ,r h w
to to to to to to to to to o l) v e c t
C re p e u
d d d d d d d d d
SBAA i[pL ,rabC iledpp ee ee ee ee ilepp ilepp ilepp ilepp ilepp ilepp ilepp ilepp ilepp 511% agN senw ,ccop rfIo aom
C A rF rF rF rF A A A A A A A A A dn ,6 im izn eO teh
a i(e tia , th n
% v m f e
5 g u o h
d d 8 m 10 is t w
een 56 eh eng nuo IO
t 9 r o in l
tirsannoC 2?001 3?052 7?055 030 2?015 .9?25 .05 .025 .2?50 1?61 01 01 72?00551 IrooROA ilrfrcoom rseevdbo iaexhmm rsaebfiw .iehnTE ttitrsaoon ttlseaehR lirsaen9m
E ca ,I t r s
iso ;O sa fo tiam on ,ta teh
e id w A v .C R d
re ilp l) D d s A n
y r )
ttlf)reaeongy ttlf)raeeooygn tlt(frseeaonog% /)dg tlt)reaeonyg iltt(fceaedono% tltf(reeaoogny% /()gdA ttlf)raeeooygn ltt)raeeonyg itlt(fsacadyoo% tltf)reeaooygn ) 3,ilirssaen9nmm ,liitseezdoodpm irraeeeeegnuqRm ,if(e2o7597gm ,tttrrsceenuhyduF ,ii-rscca1n2boB lfsaeu7o7=RD% tlIrseaaehoOww ttitfredyuov10h
ittrsenu itr(senoo% lttf(saao% trreaabohyd ltlr(seeohom (foFAU% ili-ecanonL liiiecacond lspuPEAH -(eag3m% (foFAU% ttttrfaeaaud rr(sseaeug% i/(udogdmm ,iitrea10bv 1llIanm titsaedvAm2ttlaoA3tIenh ,tltiseaavom4vERA tehnhRwA itisssendh
N P T C C M a L D O P S F S F E f E u
C R n e E a
w , a lu
itsed ;tipL agN recnh ,icad avA tenh isanm
the cholesterol content of the optimized diets, and on the dietary
changes needed to achieve the whole set of constraints.
The distribution (5th percentile, median, 95th percentile) of
proteins, total fats, carbohydrates, MUFAs, and cholesterol were
graphically represented among observed diets and optimized
diets, separately for each model. Due to their skewed
distributions, the nonparametric Wilcoxon?s test for paired samples
was used to compare macronutrients and cholesterol between the
observed diets and optimized diets for each model. For each
model, food categories were ranked depending on the median of
their percentage weight variation, between the observed diets and
optimized diets, among consumers (ie, individuals who
consumed the food categories).
One model, called the SELECTED model, was chosen by
comparing the results obtained with the 8 models tested in terms
of macronutrient distribution and rank of food category
variations. The rationale for choosing the SELECTED model is
therefore presented in Results.
Sensitivity analysis of the nutritional constraints in the
BASAL and SELECTED models
The optimization process produced a coefficient called ?dual
value? for each constraint. A nonnull dual value implied that the
corresponding constraint was active. In other words, a nonnull dual
value, induced by the optimization process, meant that fulfilling the
constraint had a direct influence on food selection (ie, deviation
from observed individual food choices). In contrast, a null dual
value indicated that the constraint was inactive. In the present study,
the strength of each nutritional constraint in a given model was
estimated by the percentage of optimized diets in which a nonnull
dual value was recorded for that constraint. The strength of each
nutritional constraint was statistically compared between BASAL
and SELECTED models by using the McNemar test for paired data.
Identification of the dietary changes induced by the BASAL and SELECTED models
The variations recorded between observed and optimized diets
for each food category were graphically represented for the
BASAL and SELECTED models. For each food category, the
percentage variation (and confidence limit) was calculated only
for consumers (ie, individuals who consumed the food category).
In addition, for each food category, the number of nonconsumers
for whom the optimization process introduced a given food
category into their modeled diet was calculated.
The mean quantity of each food group and food category was
calculated for the observed diets and for the optimized diets
obtained with the BASAL and SELECTED models. For each
food group or food category, the difference between observed
diets and diets optimized with the BASAL and SELECTED
models was compared by using a pairwise Student?s t test.
The Operational Research and the STAT packages of SAS
version 9.2 (SAS Institute) were used to run linear programming
models and to perform statistical analyses, respectively. An a
level of 5% was used for all statistical tests.
Macronutrient and cholesterol distributions in the observed diets and in the diet optimized with the 8 models
A high variability of macronutrients and cholesterol intakes
was noted in the observed diets of the subjects involved (Figure
1). Regarding the modeled optimized diets, they always
contained a maximum of 20% energy from proteins because this
constraint was imposed on all models. In addition, whatever the
model used, the median values of total fats and cholesterol
contents were lower in the optimized diets than in the observed
diet, and the median carbohydrate content was higher.
More specifically, diets optimized with the BASAL model
(including all constraints except on MUFA) contained no more
than 35% energy from total fats, 50% of energy from
carbohydrates, and no more than 300 mg cholesterol/d, as imposed by
the 3 corresponding constraints included in this model (Figure 1).
Compared with the BASAL model, removing the total fats
constraint (BASAL-[Lip]) had no effect on total fats distribution.
Removing the carbohydrates constraint (BASAL-[Carb])
decreased the carbohydrate content (5th percentile = 45% compared
with 50% of energy). Removing the cholesterol constraint
(BASAL-[Chol]) increased the cholesterol content (95th
percentile = 530 compared with 300 mg/d), and removing the 3
constraints simultaneously (BASAL-[Lip,Carb,Chol]) increased the
95th percentiles of total fats (from 35% to 37% of energy) and
cholesterol (from 300 to 520 mg/d), and decreased the 5th
percentile of carbohydrates (from 50% to 44% of energy). Diets
optimized with the first 5 models contained less MUFA than did
the observed diets (the median decreased from 14% in the
observed diets to values lower than 12% in the first 5 models, and
there was a shift of the whole distribution toward lower values).
In the last 3 models (BASAL+[MUFA];
BASAL+[MUFA][Lip,Carb]; BASAL+[MUFA]-[Lip,Carb,Chol]), a constraint was
included to impose a minimum content of 15% of energy from
MUFA to the optimized diets (Figure 1). In the (BASAL+[MUFA])
model, a very narrow distribution of values (ie, almost no
variability) was observed for each macronutrient. In particular, all of
the 1171 optimized diets contained exactly 15% MUFA and 50%
total carbohydrates, showing that this model was unrealistic and by
far the most constrained. Removing the total fats and
carbohydrates constraints (BASAL+[MUFA]-[Lip,Carb]) allowed more
variability in the macronutrient composition of the optimized diets,
and then removing the cholesterol constraint
(BASAL+[MUFA][Lip,Carb,Chol]) increased the cholesterol content (95th percentile =
520 compared with 300 mg/d).
Effect of each model on dietary changes needed to reach nutritional adequacy
The ranks of food category variations between observed and
optimized diets were calculated (Table 2). To allow for comparison
between models, food categories were sorted according to their
rank in the BASAL model. An increase in unsalted nuts was
recorded in all models, and this increase was the greatest recorded as
compared with the other food categories. Fruit, fish, vegetables,
unrefined starches and grains, legumes, fresh dairy products, and
vegetable oils were also invariably increased and were always in
the top 10 categories (of 20). In contrast, refined grains and
potatoes either increased or were unchanged, depending on the model,
and were in an intermediate position (ranks 7?12). Except in the
(BASAL+[MUFA]) model, soft drinks, poultry, salted snacks, eggs,
sweets, and red meats either decreased or were unchanged and
were in an intermediate-low position (ranks 10?16). Mixed dishes,
cheese, deli meats, and animal fats were always decreased and in
the last 4 positions and in the same order (
). The last 3
models with MUFAs were characterized by a high increase in
vegetable oils (rank 2 or 3). The dietary variations induced by the
last 2 models were almost the same, except for eggs, which
recorded a less favorable ranking (13 instead of 11) in the model
including the cholesterol constraint (BASAL+[MUFA]-[Lip,Carb]).
Choice of the SELECTED model
The (BASAL+[MUFA]-[Lip,Carb]) model was selected for
further analysis and comparison with the BASAL model and then
called the SELECTED model. The rationale for choosing this
model among the 8 models was as follows: 1) a constraint on
MUFAs was necessary because, without it, MUFAs were always
reduced (Figure 1); 2) among the 3 models with MUFA, the
(BASAL+[MUFA]) model was not retained because it was too
unrealistic; 3) among the other 2 models with MUFAs, the
model that included the cholesterol constraint
(BASAL+[MUFA]-[Lip,Carb]) was preferred because the majority of official
guidelines include a recommendation on cholesterol (
Without an upper limit on cholesterol, the median content will
decrease in the optimized diets, but some of these diets will still
contain very high amounts of cholesterol (95th percentile = 520
mg/d in the BASAL+[MUFA]-[Lip,Carb,Chol] model).
Sensitivity analysis of nutritional constraints in the BASAL and SELECTED models
The strength of each nutritional constraint was estimated by the
percentage of optimized diets in which the constraint was active (ie,
had a nonnull dual value). These percentages were compared both
within and between the 2 models (Table 3). Keeping dietary energy
equal to the observed levels was the most difficult constraint to fulfill:
it was the only constraint that was active for 100% of individuals.
Detailed analysis of the dual values obtained for that constraint
indicated that the models would prefer to increase the energy content
for 48% and 58% of individuals (and thus would prefer to decrease it
for 52% and 42% of these individuals) in the BASAL and
SELECTED models, respectively. The next constraints that were
difficult to fulfill were those on the maximal content of SFA and
sodium, the minimal contents of a-linolenic acid and calcium, and
the maximal bound on total diet quantity (ie, ,115% the observed
weight). These constraints were active in 70% of the optimized
diets. All other constraints were active in less than two-thirds of the
optimized diets, and their strengths were comparable between the 2
models. In particular, in the BASAL model, the constraint on total
fats was quite inactive (active in only 4% of diets). In addition,
constraints on carbohydrates and cholesterol were active in 56% and
51% of diets, respectively. In the SELECTED model, the constraint
on MUFA was very strong (it was active in 92% of the diets
optimized with this model) and had an effect on the ease to fulfill the
constraints for other lipids. Thus, in the SELECTED model, as
] ] e A ,s n b
FA lo rv B id o m
U ,hC ()+ ()+ ()+ ()+ ()+ ()+ ()+ ()= ()= ()+ ()= ()= ()2 ()= ()2 ()= ()2 ()2 ()2 ()2 seb ,l]o lip itsn enu
[+M rab 1 7 3 6 4 5 8 12 9 2 13 10 16 11 15 14 17 19 18 20 on hC ltao trsa ,th
L C ee [- t n N
tebw SLA tson tcuo l;reo
se BA ian itho tse
a ; r l
re se ts w oh
ce ta n d c
d r o n l
t d c a ta
l l f
d L a ta o
a A o to r
S t e
n hy tu s to
fo + ,C ca ob toh FA d
rse LA i[pL ()+ ()+ ()+ ()+ ()+ ()+ ()+ ()= ()= ()+ ()= ()= ()2 ()= ()2 ()2 ()2 ()2 ()2 ()2 ignfi rac iw UM ,san
snum SBA ]-FA 1 7 3 6 4 5 8 11 9 2 12 10 16 13 15 14 17 19 18 20 issa ttlao leod son trae
co U tae on m tin yhd
g [M ic s L ra o
n d t A t b
o in in S sn ra
m ra A o c
a ) t B c l
lny ]A ;(2 son ,l] ith tto
do FU itse tcu hoC lew ,sd
laend [+LM )(1+ )(6+ )(8+ )(2+ )(5+ )(4+ )(7+ )(11+ )(9+ )(3+ )(21= )(412 )(512 )(31= )(10+ )(712 )(612 )(912 ()812 )(022 leedd tiohw ,,rabC dom ililap
o A d l p L to
m S o ed i A t
A m o [L S n
- A o
d m L
a B ts
e a L A , n
r d A S ] ia
o e S A rb r
f , rv A B a ts
tised i[-pL l]oh ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) seob ,]B ;isd ,ipC con
ed L ,C (1+ (2+ (3+ (5+ (4+ (6+ (8+ (0+ (7+ (9+ (2= (1= (5= (3= (62 (4= (72 (82 (92 (02 en rab ilp -L tou
z SA rab 1 1 1 1 1 1 1 1 1 1 2 e C l h
i [ a A it
tw - t F
e L to w
m A C
b A n d
e S o [M n
d a A s + s
an l] rce .oB itan LA FA
rseebovd -[hoLCA ()1+ ()3+ ()2+ ()4+ ()5+ ()6+ ()9+ ()8+ ()7+ ()01+ ()31= ()11= ()61= ()41= ()21= ()51= ()712 ()812 ()912 ()022 iticannfi rsazew ttrscoun ;sFSBAA tsnoUM
en SA isgn itse tho U ian
e B d i
tag SA ipL (1+ (2+ (3+ (4+ (5+ (6+ (7+ (8+ (9+ (0+ (1= (2= (3= (42 (52 (62 (72 (82 (92 (02 SA een -L o B
n A [- 1 1 1 1 1 1 1 1 1 1 2
1 Nutritional constraints represent the minimal bound for omega-3, linoleic acid, proteins, total fats, vitamin B-12, and
phosphorus. Upper limits for MUFAs, omega-3, total carbohydrates, niacin, vitamin B-6, folates, ascorbic acid, vitamin E,
selenium, and zinc are not presented because they were not active constraints. All nutritional constraints included in the
BASAL model are listed in Table 1. The SELECTED model comprises the BASAL model with constraints on MUFAs and
without constraints on total lipids and total carbohydrates. Constraints were considered as ?active? when they had a nonnull
dual value. NA, the constraint was not applied in this model.
2 The minimum or maximum levels imposed on each nutritional constraint are listed either in Table 1 or in reference 14. =
indicates that the energy constraint was equal to the observed energy intake; < indicates that the nutrient constraint imposed an
upper bound; > indicates that the nutrient constraint required a minimal content.
3 For both the BASAL and the SELECTED models, nutritional constraints were sorted by decreasing order of the
percentage of nonnull dual values recorded in the BASAL model. The McNemar test was used to compare the strength of
each constraint between the BASAL and SELECTED models. All P values were significant except for iodine, ascorbic acid,
iron, vitamin A, panthotenic acid, selenium, folates, niacin, and total amount.
4 Detailed analysis of the dual values obtained for the energy constraint showed that the models would prefer
increasing the energy content in 48% and 58% of individuals in the BASAL and SELECTED models, respectively, and thus
would prefer decreasing it in 52% and 42% of individuals in the BASAL and SELECTED models, respectively.
5 The total amount of each optimized diet was constrained between 85% and 115% of the weight of the corresponding
compared with the BASAL model, it was slightly easier to find
a sufficient amount of PUFAs and vitamin E and it was more difficult
to maintain the amount of linoleic acid below the desired level. This
is probably because MUFAs are positively associated with total
PUFAs, including linoleic acid, and with vitamin E.
Dietary changes induced by the BASAL and
As shown in Figure 2, when a food category was increased by
the BASAL (Figure 2A) or the SELECTED (Figure 2B) models,
the mean variation between consumers in that specific category
was always ,100% (ie, less than double the amount); the only
exception was observed for unsalted nuts, which recorded very
high variations, particularly in the SELECTED model (Figure
2B). The number of nonconsumers of each food category for
whom the optimization process added that category in the
corresponding optimized diet, for the BASAL (Figure 2C) and
SELECTED models (Figure 2D), respectively, is also shown in
Figure 2. Among the 13 food categories (out of 20) which had
.5% of nonconsumers (ie, .60 nonconsumers), unsalted nuts,
fruit, fresh dairy products, and fish were added for a large
majority (ie, .75%) of nonconsumers. In fact, fish was added to all
of them. Then, unrefined grains, legumes, poultry, and cheese
were added to 17?30% of nonconsumers, depending on the
model and on the category. Conversely, potatoes, soft drinks,
salted snacks, eggs, and deli meats were never (or almost never)
added, no matter which model was applied.
The average changes in daily food quantities induced by the
optimization process with the BASAL and SELECTED models
are shown in Table 4. For most food groups and food categories,
the changes induced by the 2 models were in the same direction
and were of similar magnitude (although they could be
significantly different). Thus, whatever the model used, the average
optimized daily diets contained ;540 g fruit and vegetables,
260 g fresh dairy products, 60 g fish, and 70 g red meat. Mixed
and sandwiches and sweets (including soft drinks) were reduced
to ;140 g/d each, and animal fats were almost totally removed.
Although unsalted nuts, unrefined starches, and legumes were
among the food categories that increased the most (see Figure
2), their mean absolute content in the optimized diets was quite
low (,25 g each) because of their low observed intakes. In
contrast, for fruit, vegetables, and fish, both the variations and
absolute quantities were large. The main differences between
BASAL and SELECTED diets were observed for starches and
grains (higher in the BASAL model) and for poultry, mixed
dishes, and vegetable oils (higher in the SELECTED model).
In this study, we used individual diet modeling to investigate
the smallest dietary changes needed to fulfill a whole set of
nutrient recommendations by each individual in a representative
1 All values are means 6 SDs. All nutritional constraints included in the BASAL model are listed in Table 1. The
SELECTED model comprises BASAL model with constraints on MUFAs and without constraints on total lipids and total
carbohydrates. a,bIndicates the significance of the differences computed by using Student?s t test (Bonferroni corrected).
Values sharing the same letter were not significantly different.
2 Fruit juices were included in the fruit category.
3 Soups were included in the mixed-dishes category.
sample of French adults. Eight different models were designed to
investigate the specific effects of recommendations on total
carbohydrates, total fats, MUFAs, and cholesterol and
combinations of them on the necessary minimal food changes needed to
reach overall nutrient adequacy in adults. As compared with
observed intakes, fulfilling the whole set of nutrient constraints
systematically decreased total fats and increased total
carbohydrates in the different models. This was obtained even in the
absence of specific constraints on those macronutrients, thus
underlining the importance of counteracting the nutrition
transition-driven tendency of reducing carbohydrate intakes (
addition, in whichever model that was applied, the food
categories most increased were those typical of the traditional
Mediterranean diet, suggesting that, in this Western population,
following Mediterranean steps was the shortest way to reach
It has been proposed that dietary guidelines should be deduced
from nutrient recommendations (
). However, the intensity of
the debate on some nutrient goals, especially those on total fats
17, 19, 41
), suggests that such recommendations could
have a huge effect on the definition of healthy eating. A surprising
finding of this study was that the various sets of nutrient
recommendations tested generated rather consistent food changes.
Thus, both ranking and direction of food category variations were
consistent across models. The greatest increases were always
seen for unsalted nuts, unrefined grains, fish/shellfish, fruit,
vegetables, legumes, and fresh dairy products, and the greatest
decreases were consistently observed for cheese, mixed dishes,
deli meats, and animal fats. This supports the basic principles of
dietary guidelines (
) and the validity of the traditional diet
pattern promoted by the Mediterranean diet pyramid (
only exceptions were the somewhat anomalous dietary changes
induced by the model (BASAL+[MUFA]) in which all nutritional
recommendations were imposed (it was the only model in which
poultry was decreased and sweets increased), suggesting that the
total fats limit (,35% of energy intake) and the total
carbohydrates goal (.50% of energy intake) need to be relaxed to allow
for an increase in MUFA intakes in the studied sample.
A strong consistency was also observed in terms of nutrient
content. In whichever model that was applied, the optimized diets
always contained less energy from total fats and more energy
from carbohydrates than did the observed diets, even in the
absence of constraints specifically related to those nutrients. This
means that respecting the full set of constraints on fiber,
micronutrients, and essential fatty acids and maximum limits on
saturated fatty acids, sodium, and free sugars automatically
induced an improvement (ie, decrease) of the lipid:carbohydrate
ratio in the optimized diets. In addition, sensitivity analysis
showed that constraints on total fats and total carbohydrates were
not the most severe ones, thus supporting the growing evidence
that, to prevent chronic diseases, the quality of fats and
carbohydrates is probably more important than their quantity (
). In contrast, imposing a maximum amount of cholesterol
(,300 mg/d) was needed to prevent high cholesterol intake,
which may present a risk for some individuals (
imposing a minimum energy percentage from MUFAs (ie, .15%)
was necessary to avoid a systematic decrease in MUFA intake
compared with the observed diets (as observed with all modeled
diets without this constraint). This strongly suggests that a
minimum intake of MUFAs should be explicitly listed in nutrient
recommendations. In that respect, it is rather surprising that the
WHO expert panel did not include a precise recommendation on
MUFAs but recognized and stated their importance for
maintaining insulin sensitivity and healthy concentrations of HDL
and LDL cholesterol (
Sensitivity analysis showed that constraints on nutrients such
as SFA, sodium, and free sugars as well as calcium, omega-3
fatty-acids, and iodine had the greatest effect on the food choices
needed to reach overall nutrient adequacy. The constraint on
MUFA was also very strong. In fact, increasing MUFAs by .15%
seemed unrealistic, even in the absence of a constraint on total
fats. When the MUFA constraint was applied, it was more
difficult to maintain the amount of linoleic acid below the desired
level, suggesting that MUFA and linoleic acid are associated in
the French diet. Also, contrary to what is assumed by some
), it was not easier to limit SFA intake when
a constraint on MUFA was included (SFA was the most difficult
constraint to achieve in both the SELECTED and the BASAL
models,ie, whether or not the constraint on MUFA was
included). Although very strong, the constraint on MUFAs had an
effect on only a limited number of food categories (especially
vegetable oils). This explains that the diets optimized with the
BASAL and the SELECTED models were in fact rather
comparable. The common point was the key importance of foods
typical of the Mediterranean diet pattern (especially unsalted
nuts, unrefined grains, fish/shellfish, legumes, fruit, vegetables,
and vegetable/olive oil) in the diets optimized with the 2 models.
This observation gives additional support to the healthy
Mediterranean diet pyramid, which was initially based on
observational data (
), and to the recently updated ?Mediterranean diet
pyramid, a lifestyle for today? (
The main limitations of this study lie in our definition of an
adequate diet, both in terms of nutrients and in terms of foods.
First, by defining nutritional adequacy by the fulfillment of
nutrient recommendations, we excluded nutritional
characteristics of a healthy diet for which a quantitative recommendation
should be set, such as having a low glycemic index or a high
content of phytomicronutrients. Yet, given the large amount of
fruit and vegetables and the increases in legumes and unrefined
cereals in the optimized diets, our results suggest that specific
constraints on the glycemic index and on phytomicronutrient
content might not be necessary when all other nutritional
constraints are fulfilled. Second, based on the observation that small
dietary changes are more feasible to achieve than are large ones
), our approach insists on minimizing the deviation from the
observed diets (
). However, this decision might induce some
bias. As an example, foods strongly recommended for healthy
eating, such as unrefined cereals (
), unsalted nuts (
), were greatly increased but had their mean
absolute content remain quite low in the optimized diets (because
of their initially very low observed intakes and the small
percentage of consumers). Note that the amount of legumes in our
optimized diets (24 g/d) was nevertheless greater than the
legumes intake considered as beneficial in the original
Mediterranean Diet Score (10 g/d) (39). In addition, due to our
decision to depart the least from observed food intakes to favor
acceptability, the amount of certain unfavorable food groups
remained quite high in the optimized diets when compared with
the traditional Mediterranean diet pattern (
). Thus, our results
suggest that the Mediterranean diet pyramid should be interpreted
as a goal showing the right direction rather than as a strict model
to apply. Our results also support the conclusion of recently
published studies showing that healthy diets can include more
foods with an unfavorable nutrient profile than is generally
recommended if foods with a favorable nutrient profile are also
present in sufficient amounts (
In conclusion, in the present study, we went further than the
observation that traditional Mediterranean foods (
), recipes (
and diet pattern (
) were associated with a high nutrient adequacy.
Indeed, we showed that improvements in food habits by following
food choices typical of the Mediterranean diet pattern were strictly
needed to fulfill nutrient recommendations in the French
population. Finally, we suggest that applying the present diet-modeling
approach to populations of non-Mediterranean origins could be
useful in identifying food choices that are both nutritionally and
culturally appropriate for them, considering that Mediterranean
food choices could be irrelevant to such populations (
We are grateful to Jean-Luc Volatier (Agence Franc?aise de Se?curite?
Sanitaire des Aliments) for providing us the access to the INCA database.
The authors? responsibilities were as follows?MM, CI, DL, and ND:
conceived the study; MM: conducted data modeling and statistical analyses; and
FV: participated in the conception of the study and in data modeling. All of the
authors participated in the interpretation of the results and in writing of the
research paper. None of the authors had a potential conflict of interest.
1. Martin A . Apports nutritionnels conseille?s pour la population franc?aise. [Nutritional recommendations for the French population .] Paris, France: Lavoisier, 2001 (in French).
2. Institute of Medicine. Dietary Reference Intakes: applications in dietary planning . Washington DC: Institute of Medicine, Food and Nutrition Board , National Academic Press, 2003 . Available from: www. nap. edu (cited 23 August 2011 ).
3. Institute of Medicine. Dietary Reference Intakes for energy, carbohydrate, fiber, fat, fatty acids, cholesterol, protein and amino acids . Washington, DC: Institute of Medicine, Food and Nutrition Board , National Academic Press, 2005 . Available from: www.nap. edu (cited 23 August 2011 ).
4. Willett WC , Sacks F , Trichopoulou A , Drescher G , Ferro-Luzzi A , Helsing E , Trichopoulos D. Mediterranean diet pyramid: a cultural model for healthy eating . Am J Clin Nutr 1995 ; 61 : 1402S - 6S .
5. Reguant-Aleix J , Arbore MR , Bach-Faig A , Serra-Majem L . Mediterranean heritage: an intangible cultural heritage . Public Health Nutr 2009 ; 12 : 1591 - 4 .
6. Sofi F , Cesari F , Abbate R , Gensini GF , Casini A . Adherence to Mediterranean diet and health status: meta-analysis . BMJ 2008 ; 337 : a1344 .
7. Trichopoulou A , Bamia C , Trichopoulos D . Anatomy of health effects of Mediterranean diet: Greek EPIC prospective cohort study . BMJ 2009 ; 338 : b2337 .
8. Bach-Faig A , Berry EM , Lairon D , Reguant J , Trichopoulou A , Dernini S , Burlingame B , Medina FX , Battino M , Miranda G , et al. Mediterranean diet Pyramid today . Science and cultural updates . Public Health Nutr (in press).
9. WHO/FAO. Preparation and use of food-based dietary guidelines. Report of a joint expert consultation . World Health Organ Tech Rep Ser 1998 ; 880 : 1 - 108 .
10. EFSA. Scientific opinion of establishing food-based dietary guidelines . The Panel on Dietetic Products , Nutrition, and Allergies (NDA). EFSA J 2010 ; 8 : 1 - 42 . Available from: http://www.efsa.europa.eu/fr/scdocs/ doc/1460.pdf (cited 23 August 2011 ).
11. Britten P , Marcoe K , Yamini S , Davis C . Development of food intake patterns for the MyPyramid Food Guidance System . J Nutr Educ Behav 2006 ; 38 : S78 - 92 .
12. Katamay SW , Esslinger KA , Vigneault M , Johnston JL , Junkins BA , Robbins LG , Sirois IV , Jones-Mclean EM , Kennedy AF , Bush MA , et al. Eating well with Canada's Food Guide ( 2007 ) : development of the food intake pattern . Nutr Rev 2007 ; 65 : 155 - 66 .
13. Biltoft-Jensen A , Trolle E , Christensen T , Ygil KH , Fagt S , Matthiessen J , Groth MV , Tetens I . Development of a recommended food intake pattern for healthy Danish adolescents consistent with the Danish dietary guidelines, nutrient recommendations and national food preferences . J Hum Nutr Diet 2008 ; 21 : 451 - 63 .
14. Maillot M , Vieux F , Amiot J , Darmon N. Individual diet modeling translates nutrient recommendations into realistic and individual-specific food choices . Am J Clin Nutr 2010 ; 91 : 421 - 30 .
15. Soden PM , Fletcher LR . Modifying diets to satisfy nutritional requirements using linear programming . Br J Nutr 1992 ; 68 : 565 - 72 .
16. Sklan D , Dariel I . Diet planning for humans using mixed-integer linear programming . Br J Nutr 1993 ; 70 : 27 - 35 .
17. Hite AH , Feinman RD , Guzman GE , Satin M , Schoenfeld PA , Wood RJ . In the face of contradictory evidence: report of the Dietary Guidelines for Americans Committee . Nutrition 2010 ; 26 : 915 - 24 .
18. WHO-FAO . Diet, nutrition and the prevention of chronic diseases. Joint WHO-FAO expert consultation . World Health Organ Tech Rep Ser 2003 ; 916 : 1 - 149 .
19. Marantz PR , Bird ED , Alderman MH . A call for higher standards of evidence for dietary guidelines . Am J Prev Med 2008 ; 34 : 234 - 40 .
20. Weinberg SL . The diet-heart hypothesis: a critique . J Am Coll Cardiol 2004 ; 43 : 731 - 3 .
21. WHO-FAO . Interim summary of conclusions and dietary recommendations on total fat & fatty acids . From the Joint FAO/WHO Expert Consultation on Fats and Fatty Acids in Human Nutrition, November 10-14 , 2008 . Geneva, Switzerland: WHO, 2008 .
22. Mann J , Cummings JH , Englyst HN , Key T , Liu S , Riccardi G , Summerbell C , Uauy R , van Dam RM , Venn B , et al. FAO/WHO scientific update on carbohydrates in human nutrition: conclusions . Eur J Clin Nutr 2007 ; 61 ( suppl 1 ): S132 - 7 .
23. Trichopoulos D. In defense of the Mediterranean diet . Eur J Clin Nutr 2002 ; 56 : 928 - 9 .
24. Sanderson P , Gill JM , Packard CJ , Sanders TA , Vessby B , Williams CM . UK Food Standards Agency cis-monounsaturated fatty acid workshop report . Br J Nutr 2002 ; 88 : 99 - 104 .
25. Fernandez ML , Calle M. Revisiting dietary cholesterol recommendations: does the evidence support a limit of 300 mg/d? Curr Atheroscler Rep 2010 ; 12 : 377 - 83 .
26. McDonald BE . The Canadian experience: why Canada decided against an upper limit for cholesterol . J Am Coll Nutr 2004 ; 23 : 616S - 20S .
27. Agence Franc?aise de Se? curite? Sanitaire des Aliments (French Food Safety Agency) . Avis de l' Agence franc?aise de se?curite? sanitaire des aliments relatif a` l'actualisation des apports nutritionnels conseille?s pour les acides gras. [Opinion of the French Food Safety Agency on the update of French population reference intakes (ANCs) for fatty acids . ] Saisine n 2006-SA-0359 2010 . 1 March 2010 (in French).
28. Gao X , Wilde PE , Lichtenstein AH , Tucker KL . The 2005 USDA Food Guide Pyramid is associated with more adequate nutrient intakes within energy constraints than the 1992 Pyramid . J Nutr 2006 ; 136 : 1341 - 6 .
29. Maillot M , Vieux F , Ferguson EF , Volatier JL , Amiot MJ , Darmon N. To meet nutrient recommendations, most French adults need to expand their habitual food repertoire . J Nutr 2009 ; 139 : 1721 - 7 .
30. Volatier J-L. Enque?te INCA ( Individuelle et Nationale sur les Consommations Alimentaires) . Agence Franc?aise de Se?curite? Sanitaire des Aliments, ed. Paris, France: Lavoisier, 2000 .
31. Maillot M , Darmon N , Darmon M , Lafay L , Drewnowski A . Nutrient-dense food groups have high energy costs: an econometric approach to nutrient profiling . J Nutr 2007 ; 137 : 1815 - 20 .
32. WHO/FAO. Protein and amino acid requirements in human nutrition. Joint WHO-FAO expert consultation . World Health Organ Tech Rep Ser 2002 ; 935 .
33. Martin A . Nutritional recommendations for the French population . Sci Aliments 2001 ; 21 : 317 - 458 . Available from: http://www.lavoisier.fr/livre/ notice.asp? id=3LKW2OAOA2AOWD (cited 23 August 2011 ).
34. Pietinen P , Valsta LM , Hirvonen T , Sinkko H. Labelling the salt content in foods: a useful tool in reducing sodium intake in Finland . Public Health Nutr 2008 ; 11 : 335 - 40 .
35. Holick MF . Vitamin D deficiency . N Engl J Med 2007 ; 357 : 266 - 81 .
36. European Heart Network. Heart matters. 6: Quarterly Bulletin of the European Heart Network [monograph on the Internet] . In: L Busk, ed. European Heart Health Initiative , October 2002 . Available from: http: //www.ehnheart.org/publications/heart-health-nutrition. html (cited 14 March 2011 ).
37. NHMRC. Nutrient reference values for Australia and New Zealand: including Recommended Dietary Intakes [monograph on the Internet] 2005 . Endorsed by the National Health and Medical Research Council, September 2005 . Available from: http://www.nhmrc.gov.au/_files_ nhmrc/file/publications/synopses/n35. pdf (cited 14 March 2011 ).
38. Dixon LB , Ernst ND . Choose a diet that is low in saturated fat and cholesterol and moderate in total fat: subtle changes to a familiar message . J Nutr 2001 ; 131 : 510S - 26S .
39. Trichopoulou A , Costacou T , Bamia C , Trichopoulos D . Adherence to a Mediterranean diet and survival in a Greek population . N Engl J Med 2003 ; 348 : 2599 - 608 .
40. Drewnowski A , Popkin BM . The nutrition transition: new trends in the global diet . Nutr Rev 1997 ; 55 : 31 - 43 .
41. Woolf SH , Nestle M. Do dietary guidelines explain the obesity epidemic? Am J Prev Med 2008 ; 34 : 263 - 5 .
42. Programme National Nutrition Sante?. Ministe`re de la Sante?. 2001 . Available from: http://www.sante.gouv. fr (cited 23 August 2011 ).
43. World Health Organization Europe. CINDI dietary guidelines . Copenhagen. 2000 . Available from: www.euro.who. int/Document/E70041. pdf (cited 5 January 2010 ).
44. USDA/DHHS. Dietary Guidelines for Americans . US DA , Center of Nutrition Policy and Promotion . [Monograph on the Internet.] 2010 . 7th ed. Washington, DC: US Government Printing Office, 2010 . Available from: http://www.cnpp.usda.gov/Publications/DietaryGuidelines/2010/ PolicyDoc/PolicyDoc.pdf (cited 14 March 2011 ).
45. Hu FB , Willett WC . Optimal diets for prevention of coronary heart disease . JAMA 2002 ; 288 : 2569 - 78 .
46. Siri-Tarino PW , Sun Q , Hu FB , Krauss RM . Meta-analysis of prospective cohort studies evaluating the association of saturated fat with cardiovascular disease . Am J Clin Nutr 2010 ; 91 : 535 - 46 .
47. Katan MB , Beynen AC , de Vries JH , Nobels A . Existence of consistent hypo- and hyperresponders to dietary cholesterol in man . Am J Epidemiol 1986 ; 123 : 221 - 34 .
48. Brug J , Campbell M , van Assema P. The application and impact of computer-generated personalized nutrition education: a review of the literature . Patient Educ Couns 1999 ; 36 : 145 - 56 .
49. Harris KA , Kris-Etherton PM . Effects of whole grains on coronary heart disease risk . Curr Atheroscler Rep 2010 ; 12 : 368 - 76 .
50. Sabate ? J, Ang Y. Nuts and health outcomes: new epidemiologic evidence . Am J Clin Nutr 2009 ; 89 : 1643S - 8S .
51. Flight I , Clifton P . Cereal grains and legumes in the prevention of coronary heart disease and stroke: a review of the literature . Eur J Clin Nutr 2006 ; 60 : 1145 - 59 .
52. Arambepola C , Scarborough P , Rayner M. Validating a nutrient profile model . Public Health Nutr 2008 ; 11 : 371 - 8 .
53. Maillot M , Drewnowski A , Vieux F , Darmon N. Quantifying the contribution of foods with unfavourable nutrient profiles to nutritionally adequate diets . Br J Nutr 2010 ; 105 : 1133 - 37 .
54. Maillot M , Drewnowski A . Energy allowances for solid fats and added sugars in nutritionally adequate u .s. Diets estimated at 17-33% by a linear programming model . J Nutr 2011 ; 141 : 333 - 40 .
55. Vasilopoulou E , Trichopoulou A . The micronutrient content of traditional Greek foods . Mediterr J Nutr Metab 2009 ; 2 : 97 - 102 .
56. Issa C , Salameh P , Batal M , Vieux F , Lairon D , Darmon N. The nutrient profile of traditional Lebanese composite dishes: comparison with composite dishes consumed in France . Int J Food Sci Nutr 2009 ; 60 ( suppl 4 ): 285 - 95 .
57. Serra-Majem L , Bes-Rastrollo M , Roman-Vinas B , Pfrimer K , SanchezVillegas A , Martinez-Gonzalez MA . Dietary patterns and nutritional adequacy in a Mediterranean country . Br J Nutr 2009 ; 101 ( suppl 2 ): S21 - 8 .
58. Bere E , Brug J . Is the term 'Mediterranean diet' a misnomer? Public Health Nutr 2010 ; 13 : 1 - 3 .