Dietary patterns and cognitive achievement among school children in socio-cultural context, a case of Montevideo, Uruguay

Abstract

Purpose:

Benefits of a healthy diet on childreńs cognition have been described in several studies. However, many previous studies have analyzed the effect of diets on general cognitive domains (e.g., intelligence), used measures based almost exclusively on local examinations, and rarely consider social context.

Objective:

The objective of the present study was to examine the relationship between two diet patterns and contextualized cognitive performance measures of children aged 6–8 years from low-average income neighborhoods in Montevideo, Uruguay.

Methods:

270 first-grade children with complete data participated in the study. Consumption of foods was determined via two averaged 24-hours dietary recalls with the mother. Two dietary patterns were identified via principal component analysis: “processed (high calorie) foods” and “nutrient dense” foods. Childreńs cognitive performance, including general cognitive abilities, achievement in mathematics and reading, and discrepancy scores between predicted and actual achievement was assessed with the Woodcock-Muñoz Cognitive and Achievement scales. The association of dietary patterns and cognitive endpoints was analyzed in multilevel models, clustered by childreńs school. Sociodemographic and biological variables were used as covariates.

Results:

The nutrient dense foods pattern, characterized by higher consumption of dark leafy and red-orange vegetables, eggs, beans & peas, potatoes, was associated with better performance in reading 3.29 (95% CI: 0.03, 6.56). There was also an association between the nutrient dense foods factor and the discrepancy in reading, 2.52 (0.17, 4.88). Processed (high calorie) foods pattern, characterized by higher consumption of breads, processed meats, fats and oils, sweetened beverages, and sweetened yogurt/dairy products; reduced intake of milk, pastries and pizza dinners was not associated with cognitive performance.

Conclusions:

Nutrient dense food pattern was positively associated with children’s reading achievement. A nutrient-rich diet may benefit language acquisition at the beginning of schooling.

1. Introduction

Early school years are an intense learning period, with high demands placed on attention, memory, and executive functions (EF). The first year of primary school may be particularly challenging for children given the need to adapt to a formal educational setting, with more academic requirements and norms, and to regulate their behavior more independently and consistently than before. Prior to school entry, children start to build internal representational models of the external world, through the development of symbolic and abstract thought. A critical component of this process is the acquisition of verbal abilities that offer the child the conceptual tools to understand the environment and socialize with others. Those cognitive changes are underpinned by a spurt of maturation that occurs in the frontal lobe, mainly in the prefrontal cortex (PFC). During the school years, the PFC increases in volume (10% approximately), with a notable growth of 20% in white matter caused by the intensification of neuronal connections [1]. Increased myelination during the early school years strengthens the connections among neural networks, promoting high levels of brain plasticity [2]. Therefore, this developmental stage requires a fine synchronization between biological maturation, cognitive development, and environmental stimulation.

Optimal nutrition is one of the most critical environmental factors to support child development. Brain growth requires large amounts of energy and nutrients to support the high requirement of neural metabolic pathways and structural components (protein, iron, manganese, zinc, selenium, iodine, folate, vitamin A, choline, and long-chain polyunsaturated fatty acids among others) [3]. Considering that diet is the main source to meet those needs, the study of the relationship between nutritional patterns and cognitive achievement in the first years of school is highly relevant. There is good understanding that protein-energy malnutrition [4, 5], as well as specific nutrients [6, 7], affect neurobehavioral development and cognitive performance in childhood and adolescence. There is also growing research on the link between whole diets and these endpoints. Dietary patterns take into account that nutrients act synergistically in the body and not in isolation [8]. Tandon et al [9] reviewed 8 studies on the links between dietary patterns and general cognition in children aged 8–15 years. All the studies were secondary analyses from longitudinal cohorts and compared a “healthy” pattern (where fruits, vegetables and whole grains were important) against an “unhealthy” pattern (with high sugar and fat content). All studies found a positive effect of a healthy dietary pattern consumed in the first years (from 6 month to 3 years) on children’s general intelligence and in verbal abilities assessed in childhood and adolescence (from 5 to 15 years old). Some studies also found this association when dietary patterns were evaluated at the same time as cognitive measures at 5 [10], 8 [11] and 8.5 years old [12]. Other studies found an inverse association between an unhealthy diet pattern and intelligence [13, 14]. The Baltic Sea Diet Score (BSDS) and the Dietary Approaches to Stop Hypertension (DASH) score (both characterized by higher intake of fruits, vegetables and fish) were also positively associated with general intelligence in 6–8 year old children [15]. A “snacky” pattern (potatoes chips, salty snacks, sugar products) had a negative effect on cognitive abilities measured two years after assessing dietary consumption [16]. A recent narrative review [17] about health behaviors and brain health in children and adolescents includes some studies that evaluate the association between dietary patterns and achievement. A randomized controlled school meal trial showed that a healthy Nordic diet pattern, based on fruits, vegetables, fish and berries, was related to improved reading skills in 8–11 year old Danish children [18]. There was no effect on math performance or the ability to concentrate. Another study based on the BSDS and the DASH dietary pattern found a positive association with better academic performance in reading, but not in arithmetic, in 6–8 year old children [19]. Three studies found a positive association between a better diet quality [20] or the adherence to a Mediterranean Diet [21, 22] with general school performance in 10–12 year-olds.

While the benefits of a healthy diet on cognition and learning of school children has been described in several studies, several major issues deserve further consideration. First, previous studies have focused on global cognitive measures. For example, some studies assessed only general intelligence [9]. Although intelligence is a suitable index of general performance, it is limited to discriminate deficits in specific cognitive systems (for instance, language, reasoning or working memory). In relation to achievement, another difficulty is that most studies employ measures based on local examinations, limiting comparison among samples since different countries emphasize distinct contents in their academic curricula. Second, eating behavior and cognition occur in a broader context. Socioeconomic status (SES) is a robust predictor of children’s cognitive functioning including memory, executive functioning, IQ and verbal skills [23–26]. Geographically concentrated poverty, or neighborhood disadvantage, is also related to cognition [27]. For instance, language acquisition is strongly shaped by the socialization context [28]. Thus, it is necessary to include both individual and neighborhood level SES when evaluating childhood cognitive functioning. Particularly, when considering the influence of diet on school achievement in children from poor urban settings, the assessment of social context is highly relevant [10]. Finally, whereas most of the studies concerning nutrition and neurocognitive development have been conducted in the United States or Europe, studies in Asia, Africa and Latin America are lacking. The investigation into the impact of nutrition on achievement in Uruguay holds particular relevance as the country faces the challenge of a dual burden of undernutrition and obesity [29]. Given the strong correlation between this issue and socioeconomic variables, it is important to measure both factors - nutrition and sociocultural context - to discriminate the influence of each on cognitive development. In this manner, studying the Uruguayan case can offer valuable insights to other countries that share similar characteristics.

The objective of the present study was to examine the relationship between two diet patterns and the cognitive performance of children aged 6–8 years from low-average income neighborhoods in Montevideo, Uruguay. We hypothesized that a dietary pattern consisting of nutrient-dense foods will be positively associated with better cognitive abilities as well as learning. On the other hand, a pattern consisting of processed (high calorie) food would be negatively associated with these endpoints. With the purpose of discriminating between the effects of diet and SES, we controlled for individual and contextual factors that could affect the hypothesized associations.

2. Methods

2.1. Study Sample

This cross-sectional study was conducted in the context of an investigation into the link between environmental contaminants and neurocognitive development in children, carried out in Montevideo between July 2009 and August 2013 in Montevideo, the capital of Uruguay. To this end, 357 children and their families from city neighborhoods where lead exposure was reported or suspected, participated in the study. Our research later also found arsenic, manganese, cadmium and mercury exposure in children living in these areas of the city. In Uruguay, schools are classified based on the socioeconomic status of the populations they serve, from 1 (highest) to 5 (lowest). This study focused on levels 3–5 because of the established links between socioeconomic status and lead exposure (but a similar link exists with nutritional status) Recruitment was conducted through schools located in the geographical areas of interest. Parents met with study staff at the schools and received information about the study, including risks and benefits. They were given informed consent forms to either sign at the meeting or take home for further consideration. All first-grade children in the participating schools were eligible to take part in the assessments (n= 673). As there were and continue to be no diagnostic tests or care recommendations for children exposed to other metals, the sole exclusion criterion was a previous diagnosis of blood lead level > 45 μg/dL, a level which would have necessitated medical treatment involving chelation therapy. No children were excluded based on this criterion. The protocol of the study was described in detail elsewhere [30]. To complete all necessary assessments, study participants were required to be present on multiple occasions including two cognitive assessments of the child, a “clinic” and nutritional evaluation, and an interview with the mother (see Figure 1). While these activities took place in the child’s school, due to difficulties related to this population’s vulnerable socio-economic context, it was common for children to not attend school. As a result of missing data, we chose to only include cases with complete data. Of the 357 enrolled children, 270 (76%) had complete data on all the nutritional, cognitive and sociodemographic variables relevant to the planned analysis, comprising our final sample. The study was approved by the Ethics Committee for Research Involving Human Participants at the Pennsylvania State University (IRB# 28597), the Catholic University of Uruguay (IRB# B041108), the University of the Republic of Uruguay (IRB approval date: March 12, 2009), and the University at Buffalo (IRB#1066). The study was conducted in accordance with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. All parents signed an informed consent to participate in the study.

Figure 1.

Flow chart of complete-case participants

2.2. Measures

2.2.1. Nutrition measures

Childreńs dietary intake was determined through two 24-hour dietary recalls conducted by trained nutritionists with the mother or another caregiver familiar with the childś diet. The child participated in the interview to complement the information about the consumption of foods or snacks away from home. The first recall took place at the school and the second recall took place over the phone without prior appointment, at least 2 weeks later. Three contact attempts were made to conduct the second recall. Median [5%, 95%] time interval between recalls was 19 [13,47] days and 90% were conducted within 1 month of one another. Photographs and models of foods, plates and serving/eating utensils were presented to aid in the estimation of serving sizes and composition. Information about the name of the meals, time and place of consumption, preparation methods, recipe ingredients, brand names of commercial products and use of vitamin and mineral supplements were collected. The foods reported and their amounts were entered into a database containing the nutrient composition of typical Uruguayan foods (342 unique foods or preparations) to calculate nutrient intake. This database was created based on information provided by Uruguayan food manufacturers, recipes from the dietary guidelines published by the National Diet Institute and other published regional sources, including mineral fortification laws in Uruguay.

In addition to analyzing nutrient intakes, the consumption of foods and food preparations was extracted from each of the two recalls. The intake of foods was recorded in grams and of drinks in milliliters. These intakes were averaged over the two recall days and their distributions were examined. Because many of the individual foods did not represent normal distributions and were consumed by small numbers of children, they were combined based on similarities and frequency of consumption. Altogether, 25 food groups were created: 1) fruit, 2) dark leafy vegetables, 3) red and orange vegetables, 4) beans and peas, 5) other vegetables, 6) potatoes, 7) breads, 8) grains, 9) pasta, 10) red meats, 11) white meats, 12) processed meats, 13) eggs, 14) milks, 15) cheeses, 16) yogurt, 17) soy products, 18) fats and oils, 19) milk-based desserts, 20) sweets, 21) pastries, 22) potato chips and fried potatoes, 23) sweetened beverages, 24) pizzas and dinner pies, 25) sauces and condiments. These food groups were subjected to a principal component analysis (PCA). To determine the number of components to retain, first the scree plot was examined. As it was unclear from the plot whether two or three components should be retained, the number was chosen based on the explained variance (i.e., eigenvalues). When the 3-factor solution was compared to the 2-factor solution, the additional variance that was explained by having the third factor was minimal. Interpretation of the 2-factor solution was further improved by applying the Varimax rotation, an orthogonal rotation that enhances the difference between loadings. The two food patterns thus identified had specific foods (listed below) with absolute component loadings of at least 0.2: 1) “processed (high calorie) foods”—characterized by higher consumption of breads, processed/fried meats, fats and oils, chips and fried potatoes, sweetened beverages and milk-based desserts, yogurt; but reduced intake of milk, pastries and pizza dinners; 2) “nutrient dense”—characterized by higher consumption of dark leafy and red-orange vegetables, higher consumption of eggs, beans & peas, potatoes; reduced consumption of pasta and sauces/condiments (See Supplementary Table 1). The two food pattern scores obtained in PCA analysis were normally distributed and used as continuous variables in statistical models.

2.2.2. Cognitive measures

Childreńs cognition was assessed with the Woodcock-Muñoz III battery (WM) [31], the Spanish adapted version of the Woodcock-Johnson III battery [32], in two interviews. The WM is a widely used tool for cognitive assessment. It is based in the Cattell-Horn-Carroll theory of intelligence, and therefore, it is designed to measure three hierarchically ordered strata of cognitive abilities: specific or narrow abilities (stratum I), grouped into broad categories (stratum II) and a general factor (g) of cognitive ability similar to the IQ score (stratum III). This system allows the construction of a comprehensive cognitive profile, with a high level of reliability and validity, very suitable for comparing populations with different characteristics [33].

The WM consists of two major assessment instruments: WM Cognitive abilities (WM-COG) and WM Tests of Achievement (WM-ACH). Due to the size of the WM battery, selected testing is recommended in the research context [34] to promote efficient management of participants’ time and to avoid redundancy in assessment. Seven subtests of the WM-COG were chosen for the present study: verbal comprehension, visual-auditory learning, spatial relations, sound integration, concept formation, visual matching, and numbers reversed. These subtests allow the calculation of four cluster derived scores: 1) general intellectual ability (GIA), a measure an IQ-like score; 2) thinking ability, comprised of several thinking processes (long-term retrieval, visual-spatial thinking, auditory processing, and fluid reasoning; 3) cognitive efficiency, a sampling of two automatic cognitive processes (processing speed and short-term memory); and 4) verbal ability, language-based acquired knowledge, and verbal communication. The WM-ACH selected subtests were calculation, math facts fluency, applied problems, sentence reading fluency, letter word identification, and passage comprehension, which allow the calculation of a broad math domain cluster score (problem solving), a calculation skills score, and a broad reading domain cluster score.

2.2.3. Covariates

A set of covariables was selected considering empirical evidence of their influence in cognitive development observed in previous epidemiological studies [12, 19, 35, 36]. The covariables included in the models were age, sex, height for age z score, weight for age z score, hemoglobin level, blood lead level, family environment, maternal education, and year of recruitment into the study.

Sociodemographic variables were obtained through a questionnaire answered by the childś parents. Maternal education, one of the factors associated with academic achievement of children in Uruguay, was measured in years of formal education [37]. Family environment was measured through the Home Observation for Measurement of the Environment (HOME) inventory [38] that assesses the quality of support and developmental stimulation in the house. Higher scores indicate greater environmental enrichment. The HOME Inventory was administered by a social worker who visited the child’s home at a previously scheduled time. Since iron status has been associated with neurodevelopment, hemoglobin was measured in fasting whole blood using a portable hemoglobinometer (HemoCue Inc, Lake Forest, CA). Approximately 3 ml of fasting venous blood was also collected from each child and taken to the Toxicology Laboratory at the Faculty of Chemistry at the University of the Republic (CEQUIMTOX) in Montevideo, Uruguay for analysis. Blood lead concentrations were measured by Atomic Absorption Spectrometry (AAS, VARIAN SpectrAA-55B) using flame or graphite furnace ionization techniques, depending on the volume of whole blood available. The LOD were as follows for the years 2009–10 (FAAS 2.5 μg/dL, GFAAS 2.0 μg/dL and later 0.8 μg/dL), 2011–13 (FAAS 1.8 μg/dL, GFAAS 1.0–0.8 μg/dL). The laboratory participates in CDC’s Lead and Multi-Element Proficiency Program (LAMP) and the Interlaboratory Program of Quality Control for Lead in Blood, Spain (PICC Pb-S).

Height was considered as a covariate because taller children are treated differently in school and are thought to be smarter or more mature, and so tend to have better outcomes. Children’s height was measured in triplicate to the nearest of 0.1 cm, using a portable stadiometer (Seca 214, Shorr Productions, Colombia, MD) by a trained pediatric nurse or nutritionist. The three measurements were averaged and standardized for age and sex. For the same reason, weight was also considered as a covariate in the models. Children were weighed in triplicate to the nearest 0.1 kg using a digital scale (Seca 872, Shorr Productions, Colombia, MD). Both measurements were carried out by a trained pediatric nurse or nutritionist. Children were asked to remove their sweaters or jackets and shoes but did retain light clothing like school uniforms. The three measurements were averaged, and the final weight was calculated by subtracting standard weights of children’s clothing, to account for the clothes each child wore at the time of measurement.

Neighborhood disadvantage is an important contextual variable for scholastic achievement. To account for this effect, a neighborhood disadvantage factor (DF) was calculated [39]. The DF was extracted from the last census data (2011) for each census section (mean 1241 persons) including education, employment, housing quality and demographic characteristics variables. Factor values were then assigned to participants based on the census section wherein their home address was located at the time of the study. Nevertheless, when tested as a grouping variable in mixed models, the intraclass correlation coefficients (ICC) ranged between 0.0001–0.06, suggesting limited between-neighborhood differences in cognition. This might be explained by the low number of observations by cluster (as low as 2) as well as the socioeconomic homogeneity of participating families. Consequently, we decided to use children’s school (n=12) as a grouping variable of level 2 to control both for the quality of education received and to capture more subtle neighborhood context differences. Different schools participated in the study in different years, therefore the year of recruitment was also included in the statistical models.

2.3. Statistical analyses

All statistical analyses were performed using STATA 12.0 (Stata Corp, College Station, TX, USA). Of the 357 children enrolled into the study, 270 had complete data on the independent variables, outcomes, and covariates of interest. Descriptive analyses of sociodemographic characteristics, biomarker concentrations and cognitive tests were conducted among the complete case sample in comparison to children excluded from the analysis based on missing data (Table 1).

Table 1.

Sample characteristics and comparison between children included and excluded from analysis due to missing data.

Variable	Complete-case analysis (n=270)	N	Excluded from analysis
	M ± SD or %
Age (months)	81 ± 6.4	81	81 ± 6.2
Sex		83
Male	54.4%		57.8%
HOME score1	44.8 ± 8.2	62	43.8 ± 7.1
Maternal education (years completed)	9.1 ± 2.7	64	8.8 ± 2.8
Year of recruitment		81
2009	17.9%		11.1%
2010	32.2%		37.1%
2011	6.3%		18.5%
2012	22.9%		13.6%
2013	20.7%		19.7%
Height for age z score (HAZ)	0.5 ± 1.1	54	0.3 ± 1.1
Weight for age z score (WAZ)	0.7 ± 0.1	54	0.8 ± 0.1
Blood lead level (ug/dL)	4.1 ± 2 .1	51	4.3 ± 2.1
Hemoglobin level (g/dL)	13.2 ± 1.1	52	13.1 ± 1.1
Nutrient dense foods factor score	−0.03 ± 1.4	51	0.16 ± 1.4
Processed food factor score	−0.02 ± 1.5	51	0.21 ± 1.4
Woodcock- Muñoz battery, percentile score2		69
General intelligence ability (GIA)	31.9 ± 27.2		25.1 ± 28.5
Verbal ability	40.9 ± 26.6		37.6 ± 27.6
Thinking ability	43.5 ± 30.3		32.6 ± 31
Cognitive efficiency	25.6 ± 24.6		19.1 ± 21.9
Math calculation skills	34.6 ± 31.4	55	34.1 ± 33.6
Broad reading	52.7 ± 38.2	62	40.1 ± 39.4
Broad mathematics	36.3 ± 30.1	66	28.6 ± 32.4
Woodcock-Muñoz discrepancy score3
Calculation- discrepancy	−10.2 ± 24.9	54	−8.9 ± 20.3
Broad reading- discrepancy	2.1 ± 27.1	59	−9.2 ±29.7
Broad mathematics- discrepancy	−4.7 ± 18.1	65	−12.7 ± 22.4

¹Home Observation for the Measurement of the Environment Inventory (HOME) contains 59 items. Total obtainable score: 59.

²The percentile score indicates the percentage of individuals in the age-based standardization group that obtained a score as low or lower than the individual being assessed.

³Discrepancies between predicted achievement based on cognitive ability and actual achievement (raw scores); negative values mean worse performance than predicted, whereas positive values mean better performance than predicted.

The Nutrient Dense and Processed scores were split into tertiles to observe the differences in the sample characteristics between the extremes of their distribution. Normality tests were conducted to determine the distribution of each variable. Non-normally distributed variables were analyzed using the Kruskal-Wallis test, while parametrically distributed variables were analyzed using ANOVA. Categorical variables were analyzed using chi-square tests.

Raw scores on the W-M tests were converted into age and sex standardized scales based on a norming sample of more than 1400 Spanish-speaking children from the United States and Latin America [31], called the W score. The W score is based on the Rasch logit scale, an equal interval measurement that represents both a person’s ability and task difficulty. The W scale for each test is centered on a value of 500, which approximates the average performance at age 10 years, 0 months in the norming sample. For each age group a median W score is derived, that indicates the level of difficulty at which 50% of the sample responded correctly. The performance of each participant is compared against this median value. Finally, to obtain values comparable with peer performance, W scores were converted to percentile scores.

A unique feature of the W-M is that a discrepancy score between cognitive abilities and achievement (the two batteries) can be calculated for each person by comparing the actual score of each achievement cluster with the predicted achievement score, based on performance of others in the norm sample who were of the same age and had the same level of cognitive ability. Therefore, the discrepancy score provides an objective measure that considers how school context may influence a child’s academic achievement relative to their cognitive abilities [40]. In this study, discrepancy scores were calculated for broad math, broad reading, and skills calculation clusters. All calculations of derived scores were made with the WM batterýs software Compuscore^®. Positive discrepancy scores indicate better performance on the achievement test than predicted by cognitive abilities, while negative discrepancy scores indicate poorer than predicted performance. Scores close to zero indicate performance that is on par with cognitive abilities.

The two dietary patterns, “nutrient dense” and “processed (high calorie) foods”, were modeled as predictors of ten outcome variables: the four clusters of the WM-COG (GIA, thinking ability, verbal ability, and cognitive efficiency), the three clusters of WM-ACH (broad math, broad reading, and math calculation skills) and the three discrepancy scores between expected and actual achievement. Both dietary patterns were included as predictors in the same models. Dietary patterns co-exist. All study children received scores on each of the two patterns. Some children may be low on nutrient dense foods and high on processed (high calorie) foods or vice-versa. Some children may have middle-of-the-range scores on both, meaning that they do not eat predominantly one way or the other. The patterns in this study were computed via orthogonal rotation, thus they are uncorrelated; to understand how one pattern related to cognitive ability and learning independently of the other, we included both in the same models. All models were adjusted for the following first-level covariates: for child age, sex, hemoglobin level, and blood lead level, as well as the HOME Inventory score, maternal education, and year of recruitment into the study. For the sake of parsimony, final models are presented without lead because its inclusion/exclusion from the models did not change inferences.

Our data was clustered in nature, with children nested within schools. Therefore, we employed linear mixed-effect regression models. All multilevel models were estimated using restricted maximum likelihood. Schools were included as the level-2 random effect. Intercept-only models indicated that the variation in cognitive achievement between schools was notable, with an intraclass correlation coefficient (ICC) ranging from 0.06 to 0.19, supporting the use of a multilevel model. The variance corresponding to the random intercept (schools) was reported using the variance partition coefficient (VPC), and their significance was tested through likelihood-ratio (LR) chi-square tests comparing the specified multilevel model with the equivalent linear model. Logistic multilevel models were performed to estimate the odds ratios (OR) of cognitive performance in relation to the consumption of the dietary patterns. The cut-off point for the cognitive variables was set at the upper tertile (tertile 1- higher scores) of performance. We used an alpha level of 0.05. The percent of explained variation in WM scores from the covariable adjusted models according to dietary patterns was also calculated.

Due to the reduction of participants with complete information, we performed multiple imputation for the covariates to be able to include all 312 children with cognitive outcomes and dietary patterns assessments in the analysis. Imputed variables were hemoglobin level, HOME score, height for age z score, weight for age z score and mother’s education. The other covariates (age, sex, and year of recruitment) were complete in the sub-sample, and thus did not require imputation. We utilized the chained imputation command in STATA, specifying 50 imputations and predictive mean matching estimation (pmm STATA command combining 5 neighbor matches) to perform the multiple imputation process. Subsequently, we re-ran the analytical models using STATA’s mi estimate, which applies Rubin’s rules to combine the beta coefficients and 95% confidence intervals into a single estimated value.

3. Results

Several variables had missing observations and children with missing data (n=83) were excluded from the analysis. The complete-case sample consisted of 270 children. In general, those children who were excluded had similar characteristics to the complete-case sample.

Sociodemographic, nutritional, and cognitive characteristics of the sample are shown in Table 1. The study sample had a mean age of 81 months (standard deviation/SD= 6.4) and 54.4% of boys. The level of education completed by mothers was low (9.1 ± 2.7 years) considering that the lowest compulsory level of education in Uruguay is 9 years. Study children had adequate iron status, with mean hemoglobin level of 13.2 (± 1.1) g/dL, and there was no stunting.

Table 2 presents cognitive performance and sociodemographic predictor variables stratified according to tertiles of the nutrient dense diet pattern. All WM scores were higher in tertile 3 of the nutrient dense diet pattern than in tertile 1, particularly in the reading cluster where the mean score for Broad reading in tertile 3 was 60.9 compared with 47.2 for the tertile 1. A Kruskal-Wallis H test showed that the difference in Broad reading was statistically significant (Chi square = 7.23, p = 0.02, df = 2). Post hoc contrasts showed greater difference between tertiles 1 and 3 (p = 0.007). A smaller difference among tertiles was observed regarding Math calculation skills (Chi square = 5.87, p = 0.05, df = 2) and Verbal ability (Chi square = 5.19, p = 0.06, df = 2). Considering the discrepancy between the predicted achievement based on cognitive ability and actual achievement in the Broad reading domain, a similar pattern was observed. In the highest tertile of the nutrient dense diet pattern, the discrepancy score was 7.2 indicating better than predicted performance, whereas it was −0.8 in tertile 1. Nevertheless, none of the comparisons among tertiles with respect to the discrepancies variables were statistically significant.

Table 2:

Sample characteristics according to the level of consumption of Nutrient dense factor (factor tertiles)

	Nutrient dense foods factor
	Tertile 1 [−4.3, −0.6] 1 (n= 90)	Tertile 2 [−0.5, 0.3] (n= 90)	Tertile 3 [0.4, 5.4] (n= 90)
Variables	Mean ± SD or %	Mean ± SD or %	Mean ± SD or %	P – value4
Age (months)	79.9 ± 5.3	81.2 ± 7.1	82.1 ± 6.7	0.08
Sex (%)
Male	58.9	46.7	57.8	0.19
HOME score1	44.1 ± 8.9	46.5 ± 7.1	43.9 ± 8.4	0.12
Maternal education (years completed)	8.8 ± 2.5	9.3 ± 2.6	9.3 ± 2.9	0.47
Year of recruitment (%)				0.001
2009	11.1	15.6	26.7
2010	26.7	28.9	41.1
2011	6.7	6.7	5.6
2012	35.6	20	13.3
2013	20	28.9	13.3
Height for age z score (HAZ)	0.4 ± 1.1	0.5 ± 1	0.6 ± 1.1	0.42
Weight for age z score (WAZ)	0.6 ± 1.4	0.8 ± 1.2	0.9 ± 1.3	0.18
Hemoglobin level (g/dL)	13.1 ± 1.1	13.2 ± 1.1	13.2 ± 1.1	0.19
Blood lead level (ug/dL)	3.9 ± 2.1	4.1 ± 1.9	4.3 ± 2.2	0.55
Woodcock- Muñoz battery (percentile score)2
General intelligence ability (GIA)	28.1 ± 24.2	31.2 ± 26.9	36.4 ± 29.9	0.21
Verbal ability	35.6 ± 24.2	43.1 ± 27.1	44.2 ± 27.8	0.06
Thinking ability	40.5 ± 31.3	42 ± 29.8	48.1 ± 29.7	0.17
Cognitive efficiency	22.6 ± 19.4	24.2 ± 23.7	29.4 ± 29.1	0.88
Math calculation skills	28.1 ± 29.3	35.5 ± 32.6	40.1 ± 31.2	0.05
Broad reading	47.2 ± 37.6	49.7 ± 39.4	60.9 ± 36.6	0.02
Broad mathematics	31.4 ± 28.7	35.5 ± 30.1	42.1 ± 30.6	0.08
Woodcock-Muñoz discrepancies3
Calculation- discrepancy	−13.1 ± 25.3	−9.9 ± 25.1	−7.5 ± 24.3	0.27
Broad reading- discrepancy	−0.8 ± 28.2	−0.2 ± 28.2	7.2 ± 24.1	0.11
Broad mathematics- discrepancy	−5.7 ± 17.9	−4.9 ± 17.5	−3.5 ± 18.9	0.53

¹Tertile cutoff points of factor loadings for the nutrient dense foods factor

²The percentile score indicates the percentage of individuals in the age-based standardization group that obtained a score as low or lower than the individual being assessed;

³Discrepancies between predicted achievement based on cognitive ability and actual achievement (raw scores); negative values mean worse performance than predicted, whereas positive values mean better performance than predicted.

⁴Statistical significance for the comparisons between tertiles. Kruskal-Wallis tests were used for comparison for variables that did not present normal distribution, ANOVA for those with a parametric distribution and Chi square for categorical variables.

Dietary tertiles did not differ on sociodemographic or biological variables, exception for the Year or recruitment (Pearson Chi square = 25.47, p = 0.001, df =4). A higher proportion of tertile 1 participants than expected were recruited in 2012 and a higher proportion of tertile 3 participants in 2009. Sample characteristics stratified according to tertiles of the processed diet pattern are presented as Supplementary Material-Table 2, showing no significant differences in the reported variables.

The nutrient dense and the processed foods diet patterns were tested for associations with WM cognitive scores in multilevel models (Table 3). In covariate-adjusted models, the nutrient dense diet pattern foods factor was statistically associated with better performance in Broad reading, where one-unit difference in the nutrient dense diet pattern score was associated with 3.29 (95% CI: 0.03, 6.56, p = 0.04) points higher in the percentile score. This nutrient dense diet pattern explained 13% of the variance in Broad reading in the adjusted model. There was also an association between the nutrient dense diet pattern and the Discrepancy in broad reading with a coefficient of 2.52 (0.17, 4.88, p = 0.03). The nutrient dense diet pattern explained 5% of the variance in the Discrepancy in Broad reading in the adjusted model. A positive relationship, although not statistically significant, was observed between the nutrient dense diet pattern and the remaining cognitive outcomes (except for math and calculation discrepancies scores). Regarding the processed foods diet pattern, several coefficients were negative but none of them reached statistical significance (Table 3).

Table 3.

Association in multi-level models¹ between dietary patterns and cognitive endpoints among 5–8-year-old boys and girls from Montevideo, Uruguay (n=270)

		Nutrient dense food factor1,2		Processed foods factor1,2
Cognitive outcomes3	VPC, (LR P-value)6	β [95% CI], P-value	OR5 [95% CI], P-value	β [95% CI], P-value	OR5 [95% CI], P-value
General intelligence ability	0.31, <0.0001*	0.59 [−1.61, 2.78], 0.60	1.16 [0.93, 1.45], 0.16	−0.59 [−2.58, 1.39], 0.55	0.93 [0.77, 1.13], 0.48
Verbal ability	0.20, 0.008*	1.73 [−0.48, 3.95], 0.12	1.22 [0.98, 1.52], 0.06	−1.69 [−3.70, 0.31], 0.09	0.87 [0.71, 1.06], 0.16
Thinking ability	0.17, 0.04*	0.72 [−1.83, 3.28], 0.57	1.05 [0.85, 1.28], 0.64	−0.16 [−2.47, 2.15], 0.89	0.92 [0.76, 1.10], 0.38
Cognitive efficiency	0.33, <0.0001*	1.08 [−0.97, 3.14], 0.30	1.08 [0.87, 1.35], 0.43	0.01 [−1.85, 1.85], 0.99	1.01 [0.83, 1.21], 0.96
Math calculation skills	0.21, 0.01*	1.42 [−1.32, 4.18], 0.31	1.15 [0.94, 1.41], 0.16	0.09 [−2.40, 2.58], 0.94	0.96 [0.80, 1.15], 0.69
Broad reading	0.23, 0.007*	3.29 [0.03, 6.56], 0.04*	1.31 [1.06, 1.62], 0.01*	0.89 [−2.05, 3.84], 0.55	1.15 [0.95, 1.40], 0.13
Broad mathematics	0.21, 0.01*	1.26 [−1.34, 3.86], 0.34	1.29 [1.04, 1.59], 0.01*	0.23 [−2.12, 2.58], 0.85	0.91 [0.75, 1.11], 0.36
Calculation-discrepancy4	0.06, 0.38	−0.49 [−2.74, 1.75], 0.66	0.93 [0.76, 1.13], 0.48	0.70 [−1.34, 2.73], 0.50	1.01 [0.84, 1.21], 0.88
Broad reading-discrepancy	0.25, <0.0001*	2.52 [0.17, 4.88], 0.03*	1.23 [1.00, 1.52], 0.05	0.60 [−1.52, 2.72], 0.58	1.14 [0.94, 1.37], 0.16
Broad mathematics-discrepancy	0.08, 0.26	−0.97 [−2.62, 0.66], 0.24	0.97 [0.79, 1.18], 0.77	0.84 [−0.64, 2.33], 0.26	1.01 [0.85, 1.21], 0.84

¹School was used as a clustering variable in two-level models.

²Models adjusted by age, sex, hemoglobin level (g/dL), height for age z score (HAZ), weight for age z score (WAZ), HOME inventory score, maternal education, and year of recruitment to the study.

³Woodcock-Johnson cluster scores. Each cluster score is calculated based on the following tests: VERBAL ABILITY- Verbal Comprehension, THINKING ABILITY- Visual-auditory learning, Spatial relations, Sound blending and Concept formation, COGNITIVE EFFICIENCY- Visual matching and Numbers reversed, GIA- all the previous, MATH CALCULATION SKILLS- Calculation and Math fluency, BROAD MATHEMATICS- Calculation, Math fluency and Applied problems, BROAD READING- Letter word identification, Reading fluency and Passage comprehension.

⁴Discrepancy between performance predicted based on general cognitive abilities and actual performance. Positive scores indicate better performance than predicted.

⁵Association between adherence to a particular dietary pattern and the likelihood of improved cognitive performance, as evidenced by being in the upper tertile of scores.

⁶Variance partition coefficient (VPC) corresponding to the variance attributed to the random intercept (school) in models with continuous outcomes, with the significance value of the likelihood-ratio (LR) chi-square test comparing the multilevel model with the equivalent linear model.

^*P-value <0.05.

Logistic multilevel models were run to estimate the association between adherence to a dietary pattern and the likelihood of improved cognitive performance, as evidenced by being in the upper tertile of cognitive scores (tertile 1- higher scores). Each unit of the nutrient-dense pattern was associated with 31% higher likelihood of being in the upper tertile of performance on Broad reading (95% CI: 1.06, 1.62, p = 0.01) and with 23% higher likelihood of being in the highest tertial of Discrepancy in this domain (1.00, 1.52, p = 0.05), meaning that children with higher adherence to the nutrient-dense pattern demonstrated the greatest disparity between their predicted performance based on general intellectual ability and their actual, superior performance in reading. A statistically significant association was also observed between the nutrient-dense pattern and likelihood of higher level performance in Broad mathematics, OR 1.29 (1.04, 1.59, p = 0.01).

The same models, with continuous and categorical outcomes, were repeated in the imputed sample (n = 312) (Supplementary Table 3). The results remained consistent, with similar coefficients. Additionally, an association was found between the consumption of the nutrient-dense pattern and the likelihood of being in the upper tertile of Verbal ability, OR 1.26 (1.03, 1.54, p = 0.02). There were no other difference between imputed and complete-case models.

To quantify the specific contribution of schools (random intercept) to the variance in the models the variance partition coefficient (VPC) was calculated and its effect was tested through likelihood-ratio chi square (Table 3). The contribution of the schools was significant in all models except for the Discrepancy scores in calculation and mathematics. Greater VPCs were observed in models for the following outcomes: Cognitive efficiency, GIA, and Broad reading discrepancy (0.33, 0.31 and 0.25, respectively) explaining a considerable portion of the total variance. Broad reading (0.23) and mathematics (0.21) VPCs were also relevant.

4. Discussion and conclusions

The influence of nutrients and dietary patterns on general cognitive abilities in children is well demonstrated. Nevertheless, how this influence is translated in better achievement and a more efficient acquisition of scholastic competences is still unknown. Among children from Montevideo, Uruguay, a nutrient rich diet was positively associated with children’s verbal skills at the beginning of schooling. This finding is of particular importance for countries facing dual challenges of obesity and undernutrition. This is also the first study in Latin America to examine the relationship between dietary patterns and academic achievement using an internationally standardized battery (and therefore not based in results of local school examinations) accounting for school-related contextual variables.

We estimated the association of two dietary patterns obtained through PCA, nutrient dense and processed foods, with children’s performance on several tests of general cognition, achievement, and the discrepancies between them, measured with the Woodcock-Muñoz battery. As hypothesized, children who had higher scores on the nutrient dense diet pattern had better verbal ability and reading achievement, although the magnitude of these associations was modest (~10% of the SD in scores and 5–13% of variability in scores explained by the nutrient dense pattern). These results are nevertheless in line with previous studies. A dietary pattern of fresh food consumption at five years-old was associated with better vocabulary, but not with non-verbal reasoning ability measured at the same age in a Scottish birth cohort (Growing Up in Scotland) [10]. Northstone and colleagues [12] found in a sample of more than 7000 children of the ALSPAC cohort that a nutrient dense food consumption pattern (characterized by higher consumption of salads, fruit, vegetables, fish, pasta and rice) was positively associated with Weschler’s IQ at 8.5 years-old, but this relationship was stronger for verbal IQ than for performance IQ. Similarly, a cohort of children of Southampton (UK) who had consumed a dietary pattern that included high consumption of fruit, vegetables and home-prepared foods in the first year of life, had a better verbal IQ at 4 years-old but the associations with other cognitive abilities were not significant [41].

Learning to read is a complex cognitive process that implies the recognition of visual patterns (letters), their association with a sound and the blending of the sounds to extract meaning. To accomplish these tasks, several brain structures are modified. For example, the left ventral occipito-temporal cortex that reacts before literacy acquisition to faces is “recycled” to respond to letters in children who have learned to read [42]. The development of tract-specific white matter pathways is also necessary to connect brain areas responsible for visual analysis with others responsible for phonological conversion, as well as circuits of semantic storage. These changes have been observed in the increment of fiber alignment, activation, and organization of the left arcuate fasciculus and increases in grey-matter density in bilateral fusiform gyri [43]. To complete those anatomical and functional changes in the brain, there is a high demand of nutrients (proteins, iron, long-chain polyunsaturated fatty acids, etc.). [3]. Neuroimaging tools have been used to investigate the impact of nutrients on brain development. The findings describe that several nutrients play a crucial role in myelinization and neural proliferation in the brain. These include iron, manganese, zinc, selenium, iodine, folate, vitamin A, choline, copper, protein, amino acids, long-chain polyunsaturated fatty acids, vitamin A, and retinol. These nutrients are fundamental for the structural development of the prefrontal and parietal cortex, cerebellum, and hippocampus from birth to 5 years of age [44]. Although we did not find neuroimaging studies that focus on dietary patterns in children, research suggests that better diet quality, characterized by high intake of vegetables, fruits, grains, and fish, is positively associated with larger brain volume, gray matter volume, white matter volume, and hippocampal volume in adults, which highlights the biological significance of these essential nutrients throughout the life cycle [45, 46] A nutrient dense dietary pattern is more likely to satisfy that nutrient requirement.

Interestingly, the nutrient dense diet pattern was also associated with the discrepancy score for reading. Higher scores on the nutrient dense diet pattern were associated with a reading performance that was better than expected based on reading ability alone. The discrepancy score reflects if the person is performing as well as one would expect, given his or her measured levels of associated cognitive abilities, in this case the verbal ability cluster. This discrepancy could be attributed to extrinsic factors, other than the cognitive domain itself, such as motivation, instruction, family or socio-economic level [47] that may include the provision of a more nutritious diet, which in our study was operationalized as higher consumption of nutrient dense foods. This dietary pattern may not only provide children with nutrients at a critical point of brain transformation that underpins learning the written language, but it may also represent opportunities for verbal stimulation around food/diet related topics [48, 49], as well as the availability of nutritious foods. Studies have indicated that families who consume a diet rich in healthy foods also engage in increased shared mealtime activities, which encompass shopping, cooking, and eating together. These activities provide a valuable opportunity for talking about food and daily life and have been shown to be positively associated with language development. Specifically, mealtime talk has been linked to the acquisition of new vocabulary and the gradual utilization of abstract concepts [50, 51]. This relationship has been extensively analyzed within the Latino cultural context [52]. Therefore, it is possible that the provision of nutrients via a nutrient rich diet at six years of age directly supports brain and language development. It is also possible that, despite adjustment for maternal education, the HOME score and the school, our nutrient-rich variable still represents better socioeconomic status and opportunities for enrichment and their effect on learning.

Given that food has several meanings, as discussed above – nutritional and socio-cultural -- it is relevant to discriminate the influence of dietary patterns from other contextual elements with respect to achievement, which is also a complex construct. Our results suggest that the effect of diet is independent of the familial (maternal education, HOME score) and school influences. At the level of fixed factors, models were adjusted for variables associated with the socioeconomic background of the participants, but the association of the nutrient dense diet pattern with reading ability remained. Nevertheless, school-related variables displayed a significant relationship with academic achievement, as indicated by the substantial amount of variance accounted for this factor in the models. As previously discussed, sociocultural contexts exert a crucial influence on learning and cognitive development. Comparative research has underscored the strength of this relationship in Latin America [53], where family socioeconomic status represents a robust predictor of children’s academic performance upon completion of primary school. In Uruguay [54, 55], the sociocultural context exerts a notable influence on achievement in specific domains. For instance, the proportion of students who exhibit proficient reading skills is four times higher among those from favorable contexts than among their disadvantaged counterparts. In our study, neighborhoods were relatively homogeneous in terms of socioeconomic status, but schools differed on infrastructure and support for students. While some schools operated on a full-day schedule, offering extra time for addressing students’ learning difficulties and engaging in recreational activities, others operated on a single morning or afternoon shift. Given these disparities, differences in sociocultural resources among families may be associated with access to better schools. Therefore, the possibility of residual confounding remains, and future investigations should explore a wider range of explanatory socioeconomic and sociocultural factors, as they play an important role in cognitive development.

Contrary to our expectations, we found no association between diet patterns and general cognitive abilities. Most of the studies that found an association between diet patterns and general cognition have a longitudinal design. That allows them to identify the early nutritional influences on cognitive development and predict achievement in childhood or adolescence. Basic cognitive processes such as intelligence and executive functions are strongly influenced, besides sociocultural context, by genetic [56] and early nutritional status (from pregnancy to the second year) [41, 57]. Our study was only able to investigate the concurrent relationship between dietary patterns and achievement at ~7 years of age. At that age, a richer nutrition can improve achievement (domain-specific learning process mediated by instruction), but it is difficult that it would modify domain-general abilities (that influence all learning processes such as intelligence). Regarding math scores, the effect was only significant when considering the best performance tertile. This may be explained by the fact that in Uruguay the main educational goal in the first grade is for children to learn how to read. The emphasis on reading over the acquisition of other skills and competencies in the first year of school is typical in countries that have transparent written systems, where there is a direct relation between the orthography and segmental phonology, like Spanish. The focus on reading is also reflected in the public curricula [58]. Considering that, it is reasonable to observe the effect of richer nutrition in the domain more stimulated by school and that requires more cognitive effort on the part of the child. Additionally, in the initial years of school mathematical achievement is associated with more basic cognitive abilities (reasoning and self-regulation) while reading requires mainly domain-specific competences (vocabulary, phonological decoding, speed) [61].

Another unexpected result was the absence of a negative association between the processed food pattern and achievement, which had been observed in other studies (for a review, see 15). Nevertheless, studies that reported such association have some methodological differences with our study that can explain the divergent results. For example, some of the studies found associations between an early (0 to 3 years-old) processed food pattern and a decrement of some measure of cognition at 8 to 10 years of age [15, 17], but they do not find this association when diet and cognition were measured concurrently in older children.

The observations found in this study need to be interpreted considering some limitations. Firstly, this is a cross-sectional study, which means that a causal relationship between dietary patterns and scholastic achievement cannot be established. Second, previous developmental stages where nutrition is relevant for cognitive development were not measured. Third, the diet pattern scores were based on two 24-hour recalls administered at least 2 weeks apart. It is possible that these scores do not reflect children’s diets long term. Although changes in diet across time are probable, there is evidence that those changes involve individual foods, but patterns tend to remain more stable [62]. Another concern is that most sweet/fat/snack foods are eaten outside of the home and adult supervision; where mothers or caregivers responded to the 24hr recalls, misclassification of consumption of those foods is possible. To reduce this problem, however, we administered the questionnaire to the mother in the presence of the child to correct or add information. Finally, considering the large number of the variables analyzed, the loss of participant data at various stages of the assessment protocol we had limited statistical power to detect other effects than reported. In addition, the limited sample size likely contributed to the wide confidence intervals around estimated associations, indicating a level of imprecision. Despite this, when we re-run the models using the imputed, somewhat larger sample, the findings were consistent with complete-case analyses.

Despite these limitations, the study measured the association of dietary patterns on cognitive achievement through a standardized battery at an important time during child development, when the brain is drastically changed by school learning through socialization and the acquisition of written language. In a similar vein, the two 24hr recalls were conducted by trained nutritionists and included neutral probing questions to capture the consumption of snacks, beverages and sweets that may otherwise be overlooked. We modeled both nutrient-dense and processed food patterns in a single regression model to more fully account for their additive influence on achievement. Finally, we accounted for the influence of the familial environment by including maternal education and a measure of the developmental stimulation in the home, as well as for differences in scholastic achievement that could be due to differences in the school/teaching environment.

In conclusion, a pattern characterized by higher consumption of dark leafy and red-orange vegetables, eggs, beans & peas, potatoes and reduced consumption of pasta and sauces/condiments was modestly but positively associated with children’s reading achievement. Our findings suggest that a nutrient rich diet may be one way to benefit children’s learning processes, such as reading, at the beginning of schooling. Further large, longitudinal studies on the relation among dietary patterns and child cognition and achievement that measure school variables together with individual/family ones are necessary to provide more elements to inform interventions in this area.