The World Health Organization (WHO) recommends that breastfeeding be initiated within the first hour after birth and practiced exclusively for the first 6 months of life. After that, to meet their evolving nutritional requirements, infants should receive nutritionally adequate and safe complementary foods while continuing breastfeeding for up to 2 years of age or beyond (World Health Organization, 2002).
Solid and consistent evidence has documented a constellation of benefits of breastfeeding for the health and development of children and the health of their mothers. Breastfeeding reduces infant morbidity and mortality (Sankar et al., 2015; Victora et al., 2016) and the risk of chronic diseases during adolescence and adulthood (Horta et al., 2015b) and has significant health benefits for the mother who breastfeeds, reducing the risk of breast and ovarian carcinoma, and developing type 2 diabetes (Chowdhury et al., 2015; Mazariegos et al., 2019; Unar-Munguía et al., 2017).
A recent systematic review demonstrates that breastfeeding positively affects cognitive scores in later childhood and showed a modest dose-dependent increase in cognitive scores in breastfed children (McGowan & Bland, 2023). This positive effect was also found in the PROBIT randomized trial (Kramer et al., 2008) and longitudinal studies with follow-ups during childhood and adolescence (Horta et al., 2015a). Overall, these findings indicate that breastfeeding offers economic benefits to individuals, families and countries as a higher intelligence level, driven by breastfeeding, leads to higher schooling attainment, income and productivity (Renfrew et al., 2012; Straub et al., 2016; Victora et al., 2015).
Further research is needed to confirm or refute the relationship between breastfeeding and intelligence, considering internal and external validity considerations. First, the relationship between breastfeeding and intelligence could be potentially confounded due to selection bias since it has been well-documented that women who decide to breastfeed have different characteristics than those who do not. For example, mothers who breastfeed are more likely to be health conscious than those who do not (Bürger et al., 2022; Horta et al., 2007; Turcksin et al., 2014; Vieira et al., 2016). Because these maternal attributes are usually not measured in population surveys, it is challenging to include them in secondary data analyses as confounding factors. Second, these studies often do not measure other variables that may also be associated with breastfeeding and child development or mediate this relationship, such as cognitive stimulation and parenting practices and children's home environment (Hackman & Farah, 2009). Third, most studies that relate breastfeeding and intelligence come from developed countries, where women with a higher socioeconomic status (SES) breastfeed for longer compared with women of lower SES. Because the opposite is true in countries with lower levels of economic development (i.e., women of lower SES breastfeed for longer), the directionality of confounding is in the opposite direction than in higher-income countries.
To strengthen their internal validity, data analyses from observational studies require controlling for confounders and measurement errors (Hernán & Robins, 2020). Thus, we aimed to use appropriate statistical analyses to control for possible selection bias or confounding. These analytical models have not been used to accurately estimate the association between breastfeeding and intelligence in a middle-income country setting. Therefore, the primary goal of this study was to estimate the association between predominant breastfeeding (PBF) and intelligence in school-age children that participated on a panel of the Mexican population, addressing the potential selection bias. The secondary goal was to estimate the effect of increasing the duration of PBF to 6 months on the intelligence of school-age children of low SES and quantify how much of the poverty-related intelligence gap could be reduced through breastfeeding improvements.
METHODS Study design and sample sizeWe conducted secondary data analyses using the Mexican Family Life Survey (MxFLS), a longitudinal, multithematic survey representative of the Mexican population at the national, urban, rural and regional levels. The data set included a baseline sample of 8400 households with 35,000 individuals conducted in 150 communities of Mexico, representing national, regional and urban–rural strata. The MxFLS contains information from 10 years, collected in three rounds: 2002 (MxFLS-1), 2005–2006 (MxFLS-2), and 2009–2012 (MxFLS-3) (Rubalcava & Teruel, 2013).
We analysed data from children 0–3 years with baseline data (MxFLS-1), who had information on breastfeeding practices, and intelligence scores from Raven coloured progressive matrices (CPM) test at 6–12 years of age (MxFLS-3). In the absence of valid data, such as Raven's score of 0 or a missing Raven test in MxFLS-3 (10.85%), we used the intelligence scores from MxFLS-2 (165 children).
Outcome variable IntelligenceThe Raven CPM test and the Raven Standard Progressive Matrices (SPM) test are designed to measure fluid intelligence (Smirni, 2020). However, the CPM test has been proven to be more appropriate for younger children and people without literacy since there is no need for language comprehension and the requirement for instruction is lower (Raven et al., 1993). The basic assumption in explaining the formation and evolution of cognitive skills is that Raven's test is a reliable measurement of cognitive ability to perform tasks of abstract reasoning because they were designed to measure the skills to build relationships by analogy, regardless of language and education (Carpenter et al., 1990). This test has an advantage over other tests in that it does not imply any a priori assumption that intellectual ability development in childhood is necessarily uniform or symmetrically distributed (Raven et al., 1993). Other studies have validated the abbreviated Raven CPM in adults (Raven, 2000) and among children (Qiu et al., 2020; Ramírez-Benítez et al., 2013).
MxFLS administered the abbreviated Raven CPM test made up of 18 CPM, which does not require literacy and involves matching patterns. The interviewers administered with pen and paper the same abbreviated CPM test to children between 6 and 12 years old. A Raven CPM test compounded with 12 black and white progressive matrices was administered to adults between 13 and 65 years old.
The complete Raven test comprises 60 items grouped into five series. The abbreviated Raven CPM test was composed of three sets/scales (A, Ab, and B), with six items each for children and four sets for adults. Each item was scored as 1 if correct and 0 if incorrect. These scores were summed to generate the general score with a maximum equivalent value of 18 points for children and 12 for adults. The items were organized in ascending difficulty throughout the sets. The test gradually introduces new and more complex types of reasoning, so that previous items prepare the individual for constructing the subsequent logical-associative strategies needed to solve the increasingly tricky items (Muniz et al., 2016). Set A assesses the ability to complete a continuous pattern, set Ab the ability to identify discrete figures as an interconnected whole, and set B assesses reasoning by analogies (Raven et al., 1993).
For this analysis, the individual's test scores were normalized as a percentage deviation from the mean Raven CPM test score among all children or adults from the MxFLS-3 survey, as recommended by other studies (Lawlor et al., 2006; Rubalcava & Teruel, 2004). A continuous Raven z-score was obtained with mean 0 and variance 1 without truncation at either tail. Scores were not standardized by age group as the sample size would be reduced for each age group, but we did adjust the statistical models by age to account for the fact that older children perform better on the test and earn higher scores than younger children.
Exposure variable PBF durationWHO considers infants to be predominantly breastfed when breast milk (including milk expressed or from a wet nurse) is the primary source of nourishment. Predominantly breastfed infants may receive water and water-based drinks, fruit juice, ritual fluids, oral rehydration salts, drops or syrups, including vitamins, minerals and medicines (World Health Organization, 2010). We adapted this indicator and measured PBF duration as the number of months the infant received breastmilk, water and tea. Using the maternal self-reported responses to the following questions asked at baseline when their children were 0–36 months: (1) How long did you feed (name of child) by breastfeeding, water or tea? (2) How old was (name of child) when you fed him/her other liquids, such as juice or formula in addition to breast milk? (3) How old was (name of child) when you first fed him/her solid foods, such as porridge?
With the first question, we determined the duration of PBF considering how long the child was fed only by breastmilk, water or tea. If a child received juice or formula in addition to breastmilk or any solid food, it was not considered PBF. Therefore, we considered questions 2 and 3 to verify and correct inconsistencies with the duration of PBF. Some mothers reported more duration in question 1 but answered in question 2 or 3 consumption of other beverages or solid foods, therefore PBF duration was adjusted towards the lower range. No information about ritual fluids, oral rehydration salt drops or syrups, including vitamins, minerals and medicines, was available.
Of the children at baseline with breastfeeding information, 20.9% were still breastfeeding on the day of the interview, and their actual breastfeeding duration was unknown. Therefore, to predict PBF duration among children with censored data, we used information from children no longer breastfeeding in the baseline round and estimated a Poisson regression model with robust standard errors (SE) adjusted for SES, area of residence, mother's employment, education, indigenous, age, place and type of birth.
PBF duration was modelled as a continuous variable (number of months) and as a categorical variable (never, <1 month, 1–3 months, 4–6 months and >6 months) to estimate the association with intelligence.
Child, maternal and sociodemographic variables Child characteristicsWe included baseline age (months) and follow-up age (years), sex (0 = boy, 1 = girl), birthweight (low birthweight <2500 g, normal ≥2500 g and do not recall), the type of delivery (natural or caesarean), and place of delivery (private or public clinic/hospital, or at the house).
Maternal characteristicsWe considered variables associated with the child's intelligence and the maternal decision to breastfeed, including baseline age (years), intelligence (z-score of the abbreviated Raven CPM test), smoking during pregnancy (yes/no), supplementation with iron, calcium and vitamins during pregnancy (yes/no), ethnicity defined as recognizing herself as part of an indigenous or ethnic group (yes/no), marital status (no partner, free union, married), employment defined as having worked during the last week (yes/no) and education (none, primary, secondary, high school, university or superior).
SESWe estimated a wealth index using polychoric principal component analysis considering the following household variables; dwelling's material, including the roof (waste material, cardboard, metal or asbestos sheet, palm or straw, wood, tiles, or concrete slabs), the floor (dirt, cement, wood planks, ceramic tiles or other covering), number of rooms, complete kitchen (yes/no), access to drinking water and drainage inside the household (yes/no), type of fuel used for cooking (firewood, coal, gas) and possession of material goods such as a car and electrical appliances (washing machine, refrigerator, stove, television, telephone, etc.). This information was obtained from self-reports and direct observation of household reports in the survey. The index was categorized into terciles (low, medium and high).
Area of residenceLocalities were classified according to the number of inhabitants as rural (<2500) or urban (≥2500).
Statistical analysesBaseline and follow-up characteristics of children and their mothers are described as means ± SE and proportions. We used the Heckman selection model as a statistical approach to estimate the association of PBF duration on intelligence correcting for selection bias. We used these results to simulate an increase in intelligence due to breastfeeding improvements among low SES. As sensitivity analyses, we estimated multiple linear regression (MLR) models, which do not adjust for selection bias. Also, we estimated an Instrumental Variables (IV) regression model and the Heckman selection model without predicting PBF duration. Finally, we analysed the outcomes of children with intelligence measures at MxFLS-2 and MxFLS-3 with random effects generalized least square (REGLS) model. The objective and motivation of the sensitivity analysis are to explore and contrast the models adjusted for selection bias and no adjustment. Also, to compare the estimates with and without PBF prediction.
Heckman selection modelThis model was estimated in two stages using the correction proposed by Heckman (Heckman, 1979) to address the selection bias (mothers who decided to breastfeed have different characteristics than those who decided not to breastfeed, and this difference is correlated with intelligence). In the first stage, we used a probit model to model the mothers' decision to breastfeed categorized as a dummy variable with 0 (never breastfed) and 1 (breastfed) based on the following measured explanatory variables: child's sex, mother's intelligence, smoking during pregnancy, supplementation with iron, calcium, vitamins during pregnancy, mother's age at baseline, ethnicity, marital status, employment, education, child's birthweight, type of birth, SES, area of residence (urban/rural), the mean price of powdered milk at the locality level and breastfeeding counselling during prenatal care. From the first stage, we predicted the inverse Mills ratio (IMR), the ratio of the probability density function to the cumulative distribution function. In our study, the IMR reflects the probability of the mother's decision to breastfeed. The second stage was conducted using a linear regression model to assess the association between the duration of PBF and intelligence only among children who breastfed, adjusted by child, maternal and sociodemographic variables and the IMR as an additional explanatory variable to control for selection bias. The Heckman two-step selection method uses the IMR as a proxy variable that captures the omitted part; those unobservable variables determining whether a woman breastfeeds (Koné et al., 2019). Additionally, we stratified the Heckman selection model by SES (low, medium and high) as an effect modifier.
The parameter ρ measures the correlation between the error terms and selection models. A value ρ closer to 0 would suggest estimates by unrelated processes (i.e., less evidence of selection bias) (Clark & Houle, 2014; Morrissey et al., 2016). The Wald test of independent equations assesses the null hypothesis that mothers who decided to breastfeed are no different from those who decided not to breastfeed.
Simulating the effect of increasing PBF duration on low-SES children's intelligenceWe used the stratified Heckman selection model estimates among children of low SES. Considering the breastfeeding duration of each child, we estimated the months that must be increased to achieve 6 months of PBF. We then added the estimated increase in intelligence among children predominantly breastfed 4–6 months compared with other categories (never, <1 month, 1–3 months, >6 months) to each child's actual Raven's z-score. We analysed the simulated Raven z-score distribution. We compared the mean z-score among SES groups to determine how much the poverty intelligence gap between low and high SES children would be reduced.
Sensitivity analyses MLR models (without correcting for selection bias)Adjusted linear regression models using ordinary least squares were used to estimate the association between PBF duration and intelligence in all children and only among children who had been breastfed. Both models were adjusted for child, maternal and sociodemographic variables.
REGLS modelWe estimated mixed models on a subsample of children with two measures of intelligence on the MxFLS-2 and the MxFLS-3. When multiple outcomes must be analysed, a model would be required to extend beyond the correlation between repeated outcome measures and imply an association structure between repeated measures of the same participant (Fieuws et al., 2007; Yngman et al., 2022). For this reason, mixed models were estimated. Random effects models are well-established for analyzing continuous and discrete longitudinal data (Fieuws et al., 2007). Due to the nature of our data, the GLS model was extended with random effects. A Fixed effect model was impossible to estimate since breastfeeding did not vary with survey rounds.
Heckman selection model without estimated prediction of PBF durationFor these estimates, we used information from PBF duration without considering predictions of breastfeeding among children who were still breastfeeding at baseline.
IV regression modelWe estimated a two-stage least square linear model considering the mean price of powdered milk at the locality level and breastfeeding counselling at pregnancy as instruments associated with the decision to breastfeed but not associated with the child's intelligence (Ertefaie et al., 2017).
For all models, a p-value < 0.05 was defined as statistically significant. All analyses were performed using Stata 14.0 statistical program (Stata Corporation). The SVY module was used for complex samples, and weights were used in all models except for GLS regression random effects that did not accept sample weights.
RESULTSOf the 2816 children with information on the MxFLS-1, 504 (17.8%) were lost due to incomplete data on maternal characteristics or breastfeeding practices, 387 (13.7%) did not have follow-up information, and 329 (11.6%) did not have an intelligence test score (Figure 1). The analysed sample was 1761 Mexican children belonging to the MxFLS with complete information. Children excluded from the analysis were younger, had lower PBF duration, higher SES and maternal education than those included in the sample (results not shown). The characteristics of the studied population are shown in Table 1. The mean age of children at baseline was 2.1 years, at follow-up was 5.9 years (range: 5–8 years) in MxFLS-2 and 9.0 years (range: 7–12 years) in MxFLS-3. Regarding breastfeeding practices, 7.9% were never breastfed and 26% were predominantly breastfed for 4–6 months. From infants aged >6 months, 7.6% were predominantly breastfed which means that no complementary feeding had been introduced for these children. The mean age of children who were still being breastfed was 10 months (SE 0.66). Nearly 65% of children were born in a public sector hospital or clinic, and 37.5% had low SES.
Figure 1. Flow diagram of the study sample of children with breastfeeding and intelligence information in the Mexican Family Life Survey, 2002, 2005–2006, and 2009–2012.
Table 1 Household, maternal and individual characteristics of children with breastfeeding and intelligence information in the Mexican Family Life Survey, 2002, 2005–2006, and 2009–2012.
Sample size (expanded N) | 1761 (6,545,844) |
Child's characteristics | percentage (95% CI) or mean (±SE) |
Baseline age, years (mean) | 2.1 ± 0.04 |
Follow-up age, years (mean) | 9.0 ± 0.06 |
Sex (female) (%) | 49.5 (45.6–53.3) |
Predominant breastfeeding, months (mean) | 3.5 ± 0.11 |
Predominant breastfeeding (%) | |
Never | 7.8 (6.1–10.0) |
<1 month | 14.5 (12.0–17.3) |
1–3 months | 34.6 (31.2–38.3) |
4–6 months | 35.2 (31.6–39.0) |
>6 months | 7.6 (5.7–10.1) |
Raven's CPM test scores, (max. = 18) (mean) | 10.8 ± 0.13 |
Raven's z-score (mean) | 0.1 ± 0.9 |
Birth type, (%) vaginal delivery | 62.5 (58.7–66.2) |
Birthplace (%) | |
Private hospital | 22.8(19.6–26.3) |
Public hospital | 62.7 (58.9–66.4) |
Home | 14.5 (11.9–17.5) |
Birthweight (%) | |
Normal weight (≥2500 g) | 81.4 (77.9–84.5) |
Low weight (<2500 g) | 7.6 (5.5–10.5) |
Do not recall | 10.9 (8.6–13.7) |
Mother's characteristics at baseline | |
Age, years (mean) | 29.1 ± 0.25 |
Raven test scores (max. = 12) (mean) | 5.5 ± 0.11 |
Indigenous (%) | 15.7 (13.0–19.0) |
Breastfeeding counseling during pregnancy (%) | 57.4 (53.5–61.1) |
Supplementation during pregnancy (%) | |
Iron | 68.1 (64.4–71.7) |
Calcium | 64.1 (60.2–67.8) |
Vitamins | 74.2 (70.8–77.4) |
Marital status (%) | |
Marriage | 67.4 (63.7–70.8) |
Free union | 24.1 (20.9–27.6) |
No partner (single, separated, divorcee, widow) | 8.6 (6.8–10.7) |
Education (%) | |
No education or without information | 6.7 (5.0–8.8) |
Primary | 43.7 (39.9–47.5) |
Middle school | 31.4 (28.1–35.0) |
Senior high school | 12.4 (10.0–15.1) |
College or superior | 5.8 (4.4–7.7) |
Smoking during pregnancy (%) | |
Ex-smoker | 3.6 (2.5–5.3) |
Smoker | 2.6 (1.8–3.8) |
Employment, yes (%) | 26.2 (23.1–29.5) |
Household characteristics | |
Socioeconomic status (%) | |
Low | 37.5 (33.9–41.2) |
Middle | 37.5 (33.8–41.4) |
High | 25.0 (22.0–28.3) |
Area of residence (%) | |
Rural (<2500 inhabitants) | 24.6 (21.7–27.7) |
Urban (>2500 inhabitants) | 75.4 (72.3–78.2) |
Results from the Heckman selection model show that a 1-month increase in PBF duration was positively and significantly associated with a 0.02 SD increase in the Raven z-score (p < 0.05) (Table 2, column A). Children predominantly breastfed from 4 to 6 months had a 0.16 SD higher Raven z-score than children who breastfed <1 month (p < 0.05). No association with intelligence was found among children who were predominantly breastfed from 1 to 3 months and for more than 6 months compared with those <1 month (Table 2, column B). Estimations from the MLR model, which do not consider the potential selection bias, showed no association between PBF and intelligence (Table 2, columns C and D).
Table 2 Association between breastfeeding and intelligence (Raven's z-score): Results from the Mexican Family Life Survey, 2002, 2005–2006, 2009–2012.
Heckman selection Models | Multiple linear regression models (MLR) | |||
(A) | (B) | (C) | (D) | |
Predominant BF continuousa | Predominant BF categoricalb | Predominant BF continuous | Predominant BF categorical | |
Sample size (n) | 1761 | 1761 | 1761 | 1761 |
Predominant BF (months) | 0.02* | 0.011 | ||
Predominant BF (%) | ||||
Never | (Ref.) | |||
<1 | (Ref.) | −0.14 | ||
1–3 months | 0.11 | −0.04 | ||
4–6 months | 0.16* | −0.00 | ||
>6 months | 0.18 | −0.01 | ||
SES at baseline | ||||
Low | (Ref.) | (Ref.) | (Ref.) | (Ref.) |
Middle | −0.08 | −0.09 | −0.04 | −0.05 |
High | 0.17 | 0.17 | 0.18* | 0.17* |
Area of residence | ||||
Urban (>2500 residents) | (Ref.) | (Ref.) | (Ref.) | (Ref.) |
Rural (<2500 residents) | −0.26** | −0.26** | −0.24** | −0.24** |
Mother's characteristics | ||||
Age (years) | 0.00 | 0.00 | 0.00 | 0.00 |
Employment (yes = 1) | 0.16 | 0.16 | 0.12 | 0.12 |
Mother's education | ||||
No education | (Ref.) | (Ref.) | (Ref.) | (Ref.) |
Primary | −0.03 | −0.03 | −0.01 | −0.01 |
Middle school | −0.02 | −0.02 | −0.00 | −0.00 |
Senior high school | −0.08 | −0.06 | 0.01 | 0.02 |
College or superior | 0.13 | 0.13 | 0.21 | 0.21 |
Indigenous (yes = 1) | −0.28** | −0.27** | −0.19* | −0.19* |
Mother's Raven's z-score | 0.09* | 0.08* | 0.09* | 0.09* |
Marital status | ||||
No partner | (Ref.) | (Ref.) | (Ref.) | (Ref.) |
Partner | −0.00 | −0.00 | 0.05 | 0.05 |
Smoking status during pregnancy | ||||
Smoker | (Ref.) | (Ref.) | (Ref.) | (Ref.) |
Ex-smoker | 0.01 | 0.01 | 0.13 | 0.14 |
No smoker | −0.07 | −0.07 | −0.07 | −0.07 |
Supplementation during pregnancy | ||||
Iron | ||||
No | (Ref.) | (Ref.) | (Ref.) | (Ref.) |
Yes | 0.26*** | 0.26*** | 0.26** | 0.25** |
Calcium | ||||
No | (Ref.) | (Ref.) | (Ref.) | (Ref.) |
Yes | −0.12 | −0.12 | −0.18 | −0.17 |
Vitamins | ||||
No | (Ref.) | (Ref.) | (Ref.) | (Ref.) |
Yes | 0.08 | 0.08 | 0.08 | 0.08 |
Child's characteristics | ||||
Sex (female = 1) | −0.04 | −0.04 | −0.05 | −0.04 |
Birthplace | ||||
Private hospital | (Ref.) | (Ref.) | (Ref.) | (Ref.) |
Public hospital | −0.03 | −0.04 | −0.05 | −0.06 |
Home | −0.26* | −0.28* | −0.24*** | −0.25* |
Type of birth | ||||
Caesarean delivery | (Ref.) | (Ref.) | (Ref.) | (Ref.) |
Vaginal delivery | −0.07 | −0.06 | 0.00 | 0.01 |
Birthweight | ||||
Low weight (2.5 kg) | (Ref.) | (Ref.) | (Ref.) | (Ref.) |
Normal weight (>2.5 kg) | 0.03 | 0.02 | 0.01 | 0.00 |
Do not recall | −0.06 | −0.06 | −0.08 | −0.09 |
Follow-up age (years) | 0.17** | 0.17** | 0.20** | 0.20** |
IMR | ||||
Athrho (ρ) | −2.14** | −2.36** | ||
Insigma | −0.07*** | −0.07* | ||
Constant | −1.40** | −1.72** | −1.80** | −1.70** |
R2 adjusted | 0.28 | 0.28 |
Wald test for independent equations (rho = 0): χ2 (1) = 18.64 Prob > χ2 = 0.0000.
Wald test for independent equations (rho= 0): χ2 (1) = 11.03 Prob > χ2 = 0.0009.
The Heckman model confirmed the existence of a selection bias with a significant (p < 0.001) coefficient ρ (−2.14 and −2.36) (Table 2, columns A and B). Furthermore, the Wald test indicates that it is significant (p < 0.0001). Estimations from the first stage equation of the Heckman model are presented in Supporting Information: Table S1.
As a sensitivity analysis, we excluded 165 children with intelligence scores in MxFLS-2 (children with missing Raven test in MxFLS-3). The results from all models were similar; a 1-month increase in PBF duration was associated with a 0.02 SD increase in the Raven z-score (p < 0.05). The children who were predominantly breastfed for 4–6 months versus <1 month had 0.18 SD higher Raven z-score (p < 0.05). No associations were found using MLR models (results not shown).
Estimations from the Heckman selection model stratified by SES are shown in Figure 2. We observe that the association between PBF duration and intelligence (Raven z-score) was statistically significant for the low SES but not for the middle and high SES.
Figure 2. Association between predominant breastfeeding duration and intelligence in school-age children (Raven's z-score), stratified by socioeconomic status. Results from the Mexican Family Life Survey, 2002, 2005–2006, 2009–2012. Stratification analyses using Heckman selection models. Compared with [less than]1-month predominant breastfeeding duration. SD, standard deviation; SES, socioeconomic status. Statistically significant **p [less than] 0.01.
Simulated increase in breastfeeding duration on intelligence among low SES children. The mean Raven z-score in low SES children was −0.14 ± 0.05 SE, 0.15 ± 0.05 SE in medium SES, and 0.42 ± 0.06 SE in high SES. When simulating the effect of increasing the duration of PBF to 6 months in low SES, we found that the mean Raven z-score would increase to −0.07 ± 0.05 SE (Figure 3), reducing 24.1% the intelligence gap with medium SES children and 12.5% with high SES children.
Figure 3. Distribution of intelligence (Raven's z-score) among school-age children from the Mexican Family Life Survey by SES and simulated distribution by increasing breastfeeding in low SES. We simulated an increase in predominant breastfeeding duration to 6 months among children of low SES. We estimated its effect on intelligence using the stratified coefficient of the Heckman selection model in Figure 2. SES: Socioeconomic status Raven z-score: Normalized Raven Coloured Matrix test score.
No association between PBF and intelligence was found using the REGLS model (Supporting Information: Table S2, columns A and B), similar to the MLR model, which does not control for selection bias. The results from the Heckman selection model without considering censored data are slightly higher than those found in Heckman's model considering the predicted PBF duration (Supporting Information: Table S1, columns C and D).
Estimations from the IV regression indicated that instruments were weak. The R2 statistic of 0.07 was relatively low, and the F-statistic was below the frequently used threshold of 10 (Stock & Yogo, 2001) (Estimations not shown). Due to the weakness in the instruments, estimations from IV regression are not presented and main results are focused on Heckman selection model.
DISCUSSIONWe estimated the association between PBF in the 1st month of life and intelligence in school-age children participating in a longitudinal study from the MxFLS, controlling for selection bias. We found that increased PBF duration was positively and significantly associated with the Raven z-score at 6–12 years old. The most significant increase in Raven z-score was found among those who were predominantly breastfed for 4–6 months compared with children who were breastfed for less than 1 month. No statistically significant associations were found in models that did not control for selection bias.
Our findings confirm a positive association between breastfeeding and intelligence, also found by systematic reviews and meta-analyses, which included studies with a low risk of bias and less prone to residual confounding, publication bias and misclassification (Horta et al., 2015a; Hou et al., 2021). A recent network meta-analysis found that longer breastfeeding duration resulted in higher intelligence scores than children who had never been breastfed (breastfeeding <6 months: ratio of means: 1.04, 95% CI: 1.03–1.06, p < 0.05; breastfeeding >6 months: ratio of means: 1.06, 95% CI: 1.05–1.08, p < 0.05) (Hou et al., 2021). This positive effect has also been observed in a randomized trial, which found that intelligence quotient (IQ) at 6.5 years of age was, on average, 7.5 points higher among children allocated to breastfeeding promotion groups (Kramer et al., 2008). Research also suggests the long-term benefits of breastfeeding on intelligence in childhood and adolescence of 3.44 points higher IQ compared with never breastfed children (Horta et al., 2015a).
There are biological mechanisms that support the association between breastfeeding and intelligence. Breast milk ‘is a fluid that adapts to the nutritional and immunological requirements of the child as it grows and develops’ (Ministerio de Salud & Subsecretaría de Salud Pública, 2020). Breast milk has improved the bioavailability of lipids, proteins, trace elements, and the content of elements considered nonnutrients, such as immunoglobulins, lysozymes and nucleotides, among others (Guesnet & Alessandri, 2011). Unlike formula milk, or any milk whose nutritional content is standardized, human milk is a dynamic fluid that constantly changes (Grote et al., 2016; Koletzko et al., 2008; Wu et al., 2018). The content of polyunsaturated fatty acids in the colostrum has been associated with higher intelligence in children at 5–6 years old (Bernard et al., 2017). In addition, a study reported a dose–response relationship between early breast milk intake and later IQ and whole brain volume (Isaacs et al., 2010) and the thickness of the parietal cortex in adolescence (Kafouri et al., 2013). In addition to the chemical properties of breast milk, breastfeeding enhances the bonding between mother and child (Horta et al., 2007), which may contribute to the child's intellectual development.
Our estimates show that being a rural dweller or having low SES is negatively associated with school-age intelligence. In Mexico, studies indicate that where essential services such as potable water, drainage, electricity and medical coverage are lacking, there are severe deficiencies with significant and lasting impacts on the well-being of children and the future performance of adults, especially in cognitive skills (Mazariegos et al., 2019; Rubalcava & Teruel, 2004). An essential contribution of our study is that increasing PBF for up to 6 months can reduce the gap in intelligence between children with low versus higher SES and diminish poverty-driven inequities.
Our study has some limitations. First, due to the nature of the questions in the MxFLS, it was not possible to estimate exclusive breastfeeding and its duration, defined by WHO as giving only breast milk and no other liquid, formula or food to the baby (World Health Organization, 2002). Consequently, we focused on the duration of PBF, which means the infant's predominant source of nourishment has been breast milk. It allows the consumption of nonnutritive and water-based liquids such as tea, ritual fluids and medicines. Therefore, our results are conservative since exclusive breastfeeding supplies more nutrients and immunological components and provides higher health protection than predominant and partial breastfeeding (Kramer & Kakuma, 2012).
Second, we did not have information on validating the abbreviated Raven test for children and adults in the MxFLS, which could lead to a nondifferential misclassification error in intelligence. In this case, the measurement error is independent of the exposure (breastfeeding), leading to a dilution of the real effect, making it less likely to detect an association, even if it exists (Hernán & Robins, 2020). However, the abbreviated form of the Raven CPM test has been evaluated and validated in other studies (Bilker et al., 2012; Caffarra et al., 2003; Irrgang et al., 2019; Ramírez-Benítez et al., 2013; Raven, 2000). In China, the CPM and SPM test scores were highly correlated in a sample of children 8–16 years old (Qiu et al., 2020). In Cuba, an investigation that applied both the SPM and CPM tests in its abbreviated form to school children found high consistency between tests (Cronbach's ⍺ = 0.96), and 93% of the sample had similar results with the standard and the abbreviated test (Ramírez-Benítez et al., 2013). The CPM test has been validated among children in Aguascalientes, Mexico (Méndez & Palacios, 2008) and among indigenous children in the country with an overall acceptable reliability of the test (Cronbach ⍺ = 0.810 (Fernández & Mercado, 2014). It is considered a solution to the problem of applying an instrument to measure intelligence in large populations with shorter execution times in scholars (Ramírez-Benítez et al., 2013) and adults (Bilker et al., 2012).
Another consideration is that the Raven CPM test is one of many methods for assessing intellectual abilities, and this test does not evaluate verbal language capacity or acquired cultural knowledge. All Raven tests were developed, in theory, to determine the general factor (g factor) and a specific factor according to the bifactor theory or intelligence proposed by Spearman (Muniz et al., 2016; Sternberg, 2012). This consideration is essential when comparing our results with those found in other studies using alternative tests based on other theories of intelligence or other tests of intellectual abilities with different evaluation methods (verbal ability tests and nonverbal abstraction tests) (Stebbins, 2007). However, the Raven test measures nonverbal fluid intelligence that is culturally fair. It is preferred in settings where poor language skills are a barrier to performing another intelligence test, such as multicultural and rural areas (Mandlik et al., 2019).
We could not obtain information on preterm birth or intrauterine growth restriction from the survey data. Preterm children are at risk for neurologic injuries associated with immaturity, hypoxia, inflammation, painful procedures and stressful treatments. These injuries can have long-lasting effects, including cognitive, behavioural, motor language and neurosensory impairment (Peralta-Carcelen et al., 2018). These conditions have also been associated with poor breastfeeding practices (Demirci et al., 2013). Therefore, failure to consider these variables may bias the real association between breastfeeding and intelligence.
To the best of our knowledge, this is the first study to address the potential selection bias in the association between breastfeeding and intelligence using data from a panel survey (with a 10-year follow-up and national representativeness). Selection bias was accounted for using Heckman two-step selection method, which captures unobservable variables that determine whether a woman breastfeeds. This method also corrects for selection bias in samples with a high proportion of missing outcomes correlated with unobserved variables, and provides unbiased estimates of the analysed association (Koné et al., 2019). Additionally, we accounted for censored data for the duration of PBF, avoiding overestimation of the association with intelligence. Another strength of our study is that we included maternal intelligence as a control variable, separate from SES and maternal education. Maternal intelligence is a confounder in the relationship between breastfeeding and cognitive development (Cooper, 2015); studies that do not adjust for these variables may tend to overestimate the effect of breastfeeding (Horta et al., 2007). Studies that controlled for maternal IQ showed a smaller benefit from breastfeeding than those that did not control for this variable (Horta et al., 2015a).
For future research on this topic, we recommend including information on parenting in the 1st year of life and more aspects of the perinatal stage in the analyses to reduce residual confounding. Cognition and performance in intelligence tests are related to the stimulation received by the child, and breastfeeding mothers may be more prone to stimulating their children (Horta et al., 2007).
For example, a multicenter study of birth cohorts in six European countries suggests that many prenatal and childhood environmental risk factors indicate that unfavourable childhood nutrition, family crowding and indoor air pollution in children, and exposure to environmental tobacco smoke are adversely and cross-sectionally associated with cognitive function (Julvez et al., 2021). Therefore, it is essential to consider these variables for future analyses. Although the MxFLS survey does not have information on these variables, the statistical model used controlled for selection bias that could confuse the association between breastfeeding and intelligence.
Studies suggest that increasing breastfeeding rates can significantly affect the health burden of mothers and children, leading to economic benefits for society (Renfrew et al., 2012; Straub et al., 2016; Unar-Munguía et al., 2019; Victora et al., 2015). With the results of our study, economic estimates can be used to calculate the potential benefits of increased intelligence attributed to improved breastfeeding rates.
The results of this study strengthen the evidence of the benefits of breastfeeding that can be used to support public policies to increase breastfeeding in Mexico. Although progress has been made in legislation and policies for the promotion, protection, and support of breastfeeding in Mexico, other public policies should be implemented with a specific budget so that all Mexican women, especially the poorest, can breastfeed according to recommended practices (Unar-Munguía et al., 2021). Some evidence-based policies and interventions can effectively promote breastfeeding on a large scale; these include strengthening the Baby-Friendly Hospital Initiative, improving maternity benefits, promoting a breastfeeding-friendly workplace environment, and enforcing the WHO Code of Marketing of Breastmilk Substitutes (Hernández-Cordero & Pérez-Escamilla, 2022).
CONCLUSIONWe showed that breastfeeding has a significant effect on intelligence in childhood. Children with low SES would benefit more from an increased duration of breastfeeding. Policies to support breastfeeding should be strengthened to ensure that children, especially the more deprived, achieve their development potential and reduce the existing gap.
AUTHOR CONTRIBUTIONSLidia Sarahi Peña-Ruiz, Mishel Unar-Munguía, Mónica Arantxa Colchero, Fernando Alarid-Escudero, and Rafael Pérez-Escamilla conceived the project. Lidia Sarahi Peña-Ruiz and Mishel Unar-Munguía are responsible for developing the overall research plan and overseeing the study. Lidia Sarahi Peña-Ruiz analysed the data and wrote the first draft. Mónica Arantxa Colchero, Fernando Alarid-Escudero, and Rafael Pérez-Escamilla added important intellectual content. Lidia Sarahi Peña-Ruiz and Mishel Unar-Munguía are primarily responsible for the final content. All the authors have read and approved the final manuscript as submitted.
ACKNOWLEDGEMENTSThe authors would like to thank Teresa Shamah Levy, Sonia Hernández Cordero, Filipa De Castro, Mauricio Hernández Fernández for their comments and review of the manuscript. This study is a part of Lidia Sarahi Peña-Ruiz's doctoral thesis; she received a PhD scholarship from the Mexican National Council of Science and Technology (CV 486035). The funder had no role in the study design, analysis, or interpretation.
CONFLICT OF INTEREST STATEMENTAll authors declare no conflict of interest.
DATA AVAILABILITY STATEMENTThe data that support the findings of this study are available from the corresponding author upon reasonable request.
ETHICS STATEMENTThis study was based on a secondary analysis of MxFLS data. The Ethics and Research Committee of the National Institute of Public Health (INSP for its Spanish acronym) approved the original protocol in the MxFLS-1 and the National Institute of Perinatology (INPer for its Spanish acronym) in MxFLS-2 and MxFLS-3. All participants signed an informed consent approved by INPer and INSP.
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
© 2023. This work is published under http://creativecommons.org/licenses/by-nc/4.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Abstract
Breastfeeding has been consistently associated with higher intelligence since childhood. However, this relation could be confounded due to maternal selection bias. We estimated the association between predominant breastfeeding and intelligence in school-age children considering potential selection bias and we simulated the intelligence gap reduction between low versus higher socioeconomic status children by increasing breastfeeding. We analysed predominant breastfeeding practices (breastmilk and water-based liquids) of children 0–3 years included in the Mexican Family Life Survey (MxFLS-1). Intelligence was estimated as the z-score of the abbreviated Raven score, measured at 6–12 years in the MxFLS-2 or MxFLS-3. We predicted breastfeeding duration among children with censored data with a Poisson model. We used the Heckman selection model to assess the association between breastfeeding and intelligence, correcting for selection bias and stratified by socioeconomic status. Results show after controlling for selection bias, a 1-month increase in predominant breastfeeding duration was associated with a 0.02 SD increase in the Raven z-score (p < 0.05). The children who were predominantly breastfed for 4–6 months versus <1 month had 0.16 SD higher Raven z-score (p < 0.05). No associations were found using multiple linear regression models. Among low socioeconomic status children, increasing predominantly breastfeeding duration to 6 months would increase their mean Raven z-score from −0.14 to −0.07 SD and reduce by 12.5% the intelligence gap with high socioeconomic status children. In conclusion, predominant breastfeeding duration was significantly associated with childhood intelligence after controlling for maternal selection bias. Increased breastfeeding duration may reduce poverty-driven intelligence inequities.
You have requested "on-the-fly" machine translation of selected content from our databases. This functionality is provided solely for your convenience and is in no way intended to replace human translation. Show full disclaimer
Neither ProQuest nor its licensors make any representations or warranties with respect to the translations. The translations are automatically generated "AS IS" and "AS AVAILABLE" and are not retained in our systems. PROQUEST AND ITS LICENSORS SPECIFICALLY DISCLAIM ANY AND ALL EXPRESS OR IMPLIED WARRANTIES, INCLUDING WITHOUT LIMITATION, ANY WARRANTIES FOR AVAILABILITY, ACCURACY, TIMELINESS, COMPLETENESS, NON-INFRINGMENT, MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Your use of the translations is subject to all use restrictions contained in your Electronic Products License Agreement and by using the translation functionality you agree to forgo any and all claims against ProQuest or its licensors for your use of the translation functionality and any output derived there from. Hide full disclaimer
Details





1 Department of Maternal, Child and Adolescent Nutrition, Center for Research on Health and Nutrition, National Institute of Public Health, Cuernavaca, Morelos, Mexico
2 Department of Health Economics, Center for Research on Health Systems, National Institute of Public Health, Cuernavaca, Morelos, Mexico
3 Department of Health Policy, School of Medicine, Stanford University, CA, US; Center for Health Policy, Freeman Spogli Institute, Stanford University, CA, US
4 Department of Social and Behavioral Science, Yale School of Public Health, Yale University, New Haven, Connecticut, USA