Infant Feeding and Childhood Cognition at Ages 3 and 7 Years: Effects of Breastfeeding Duration and Exclusivity

Importance  Breastfeeding may benefit child cognitive development, but few studies have quantified breastfeeding duration or exclusivity, nor has any study to date examined the role of maternal diet during lactation on child cognition.

Objectives  To examine relationships of breastfeeding duration and exclusivity with child cognition at ages 3 and 7 years and to evaluate the extent to which maternal fish intake during lactation modifies associations of infant feeding with later cognition.

Design, Setting, and Participants  Prospective cohort study (Project Viva), a US prebirth cohort that enrolled mothers from April 22, 1999, to July 31, 2002, and followed up children to age 7 years, including 1312 Project Viva mothers and children.

Main Exposure  Duration of any breastfeeding to age 12 months.

Main Outcomes and Measures  Child receptive language assessed with the Peabody Picture Vocabulary Test at age 3 years, Wide Range Assessment of Visual Motor Abilities at ages 3 and 7 years, and Kaufman Brief Intelligence Test and Wide Range Assessment of Memory and Learning at age 7 years.

Results  Adjusting for sociodemographics, maternal intelligence, and home environment in linear regression, longer breastfeeding duration was associated with higher Peabody Picture Vocabulary Test score at age 3 years (0.21; 95% CI, 0.03-0.38 points per month breastfed) and with higher intelligence on the Kaufman Brief Intelligence Test at age 7 years (0.35; 0.16-0.53 verbal points per month breastfed; and 0.29; 0.05-0.54 nonverbal points per month breastfed). Breastfeeding duration was not associated with Wide Range Assessment of Memory and Learning scores. Beneficial effects of breastfeeding on the Wide Range Assessment of Visual Motor Abilities at age 3 years seemed greater for women who consumed 2 or more servings of fish per week (0.24; 0.00-0.47 points per month breastfed) compared with less than 2 servings of fish per week (−0.01; −0.22 to 0.20 points per month breastfed) (P = .16 for interaction).

Conclusions and Relevance  Our results support a causal relationship of breastfeeding duration with receptive language and verbal and nonverbal intelligence later in life.

Strong evidence supports the relationship between breastfeeding and health benefits in infancy, including prevention of gastrointestinal tract infections and otitis media.¹ The extent to which breastfeeding leads to better cognitive development is less certain. While observational studies^1-4 have reported positive associations of breastfeeding with later intelligence, breastfeeding is strongly related to determinants of child intelligence, such as maternal intelligence, and developmental stimulation received by the child; residual confounding by such shared determinants may have led observational studies^1,2,5 to overestimate the effect of breastfeeding on child intelligence. Another limitation of prior investigations is the classification of infant feeding as ever vs never breastfed.⁴ Failure to account for partial vs exclusive breastfeeding or breastfeeding duration could lead to underestimation of the true effect of breastfeeding on child intelligence. Detailed data regarding breastfeeding exposure and adequate control for confounding factors are necessary for valid estimates of the relationship between breastfeeding and later intelligence, but no study to date has fulfilled these requirements.

Nutrients in breast milk, such as n-3 fatty acid docosahexaenoic acid (DHA), may benefit the developing brain. A major determinant of breast milk DHA content is the mother’s diet,⁶ and fish is a rich source of DHA. In pregnancy, greater maternal fish intake (particularly fish low in mercury contamination) is associated with better childhood cognitive outcomes,⁷ but the extent to which maternal fish intake during lactation accounts for the relationship between breastfeeding and cognition has not been reported. The aims of our study were 2-fold: (1) to examine relationships of breastfeeding duration and exclusivity with child cognition at ages 3 and 7 years and (2) to evaluate the extent to which maternal fish intake during lactation modifies associations of infant feeding with later cognition.

We studied participants in Project Viva, a prospective, longitudinal cohort study designed to examine prenatal factors in relation to pregnancy and child health. From April 22, 1999, to July 31, 2002, Project Viva enrolled pregnant women attending prenatal care at 8 obstetrical offices of a multispecialty group practice in eastern Massachusetts. Exclusion criteria included multiple gestation, inability to answer questions in English, gestational age of at least 22 weeks at the initial prenatal care appointment, and plans to move away from the area before delivery. Recruitment and follow-up details at birth,⁸ at 6 months,⁹ and at 3 years¹⁰ have been reported. Follow-up data collection at age 7 years was completed in December 2010. Human investigation committees of the Harvard Pilgrim Health Care Institute, Brigham and Women’s Hospital, and Beth Israel Deaconess Medical Center approved the study, and mothers of all participating children gave written informed consent.

Of 2128 women who delivered a live infant, we excluded 45 children born at a gestational age of less than 34 weeks, 325 children who were missing breastfeeding status at age 6 months and breastfeeding duration at age 12 months, and 446 children who were missing cognitive measures at ages 3 and 7 years. Therefore, our sample for this analysis comprised 1312 Project Viva mothers and children (1224 at age 3 years and 1037 at age 7 years).

When the participating child was ages 6 and 12 months, we asked the mother the questions listed in Table 1. To determine breastfeeding exclusivity at ages 6 and 12 months, we asked detailed questions about the age at which solid foods and non–breast milk liquids were introduced.

When children were age 3 years, trained research staff administered the Peabody Picture Vocabulary Test–Third Edition (PPVT-III),¹¹ a test of receptive language correlated (Pearson R = 0.90) with intelligence tests, such as the Wechsler Intelligence Scale for Children III. We also administered the Wide Range Assessment of Visual Motor Abilities (WRAVMA)¹² pegboard (fine motor), matching (visual spatial), and drawing (visual motor) subtests. Subtest scores are reported individually and combined as a visual motor composite score.

At age 7 years, we administered the WRAVMA drawing subtest and the Kaufman Brief Intelligence Test–Second Edition (KBIT-II), which measures verbal and nonverbal intelligence and is correlated (Pearson R = 0.89) with the Wechsler Intelligence Scale for Children III.¹³ In addition, we assessed memory and learning with the Wide Range Assessment of Memory and Learning (WRAML)¹⁴ design memory and picture memory tests. Scores were summed to yield a visual memory combined score.

Study staff administering cognitive tests were unaware of the children’s breastfeeding status. The PPVT-III, WRAVMA, and KBIT-II are scaled to a mean (SD) score of 100 (15).

We collected data from mothers regarding parental and child demographic, social, economic, and health information through self-administered questionnaires and interviews in pregnancy and shortly after delivery.¹⁵ At 6 months’ post partum, we administered a brief, validated food frequency questionnaire,¹⁶ including questions about the mother’s mean weekly fish intake (canned tuna fish, shellfish, and dark meat fish [eg, mackerel, salmon, sardines, bluefish, or swordfish], as well as other fish [eg, cod, haddock, or halibut]) since the infant’s birth. To measure maternal intelligence, we administered to mothers the PPVT-III when the child was age 3 years and the KBIT-II when the child was age 7 years. We also administered the Home Observation Measurement of the Environment short form (HOME-SF),¹⁷ which measures cognitive stimulation and emotional support in the child’s environment. Higher scores (range, 0-22) indicate more favorable environments.

Our main exposures were the following: (1) duration of any breastfeeding in months; (2) duration of exclusive breastfeeding in months, defined as feeding breast milk but no solid foods or non–breast milk liquids (except water) to age 6 months; and (3) breastfeeding status at age 6 months, categorized as “formula only, never breast fed,” “formula only, weaned,” “mixed formula and breast milk,” and “breast milk only, no formula.” Our outcome measures were the PPVT-III and WRAVMA scores at age 3 years and the KBIT-II, WRAVMA, and WRAML scores at age 7 years. To examine the effect of potential confounders on estimated relationships of breastfeeding measures with cognitive outcome measures, we adjusted for 4 models in linear regression. Model 0 adjusted for child age and sex. Model 1 adjusted for covariates in model 0 plus gestational age and birth weight z score.¹⁸ Model 2 adjusted for covariates in model 1 plus child race/ethnicity and maternal age, parity, smoking status, depression at 6 months’ post partum, and employment and child care at age 6 months, as well as primary language, annual household income, and parental educational level and marital status. Model 3 adjusted for covariates in model 2 plus HOME-SF score. Model 4 adjusted for covariates in model 3 plus maternal PPVT-III or KBIT-II score.

To compare our results with those of other studies, we estimated the difference in cognitive test scores between children ever vs never breastfed. To examine the extent to which maternal fish intake modified relationships of breastfeeding with outcome measures, we stratified by fish intake (<2 vs ≥2 servings per week) and calculated the P value for an interaction term (breastfeeding duration × fish intake) in linear regression.

All covariates were not observed in all participants. Using only individuals with all data observed would have resulted in a smaller sample size, with most excluded participants missing only 1 or 2 values, leading to lost information and possibly a selected subset. Therefore, we used multiple imputation to generate several plausible values for each missing value.¹⁹ To generate imputation data sets, we used a set of variables chosen from the thousands available in Project Viva to reflect demographic and other factors that we deemed plausibly related to potential missingness mechanisms and to the exposures and outcome measures. A “completed” data set includes the observed data and an imputed value for each missing value. The analysis was replicated across completed data sets and then combined in a structured fashion that accurately reflects the true amount of information in the observed data. This method assumes that the exposures and outcome measures are missing completely at random given the observed variables and the imputed covariates. This is a reduced assumption relative to that made in studies that use only complete cases. Using a statistical program (Proc MI ANALYZE in SAS, version 9.3; SAS Institute, Inc), we generated 50 complete data sets and combined multivariable modeling results for all 2128 participants in the Project Viva cohort. For this analysis, we excluded participants born at a gestational age of less than 34 weeks and those missing observed exposure or outcome data.

Table 2 summarizes characteristics of participants included at ages 3 and 7 years and characteristics of the excluded participants. Compared with those included in the analysis, mothers of excluded participants were less educated, had lower annual household income, were more likely to be of nonwhite race/ethnicity, and breastfed for a shorter duration. For 1224 participants included at age 3 years, the mean duration of any breastfeeding was 6.4 months and of exclusive breastfeeding was 2.4 months; values were similar for participants included at age 7 years. The PPVT-III mean score at age 3 years was 103.7, and the KBIT-II verbal mean score at age 7 years was 112.5.

Table 3 summarizes the effect of covariate adjustment on estimated relationships between breastfeeding duration and child cognitive outcomes. Adjusting for child age and sex (model 0), longer breastfeeding duration was associated with higher PPVT-III score at age 3 years (0.58; 95% CI, 0.40-0.76 points per month breastfed). This relationship was similar with additional adjustment for fetal growth and gestational age (model 1) and attenuated with adjustment for demographic variables (model 2) and HOME-SF score (model 3). With further adjustment for maternal IQ (model 4), the association diminished to 0.21 (95% CI, 0.03-0.38) points per month breastfed. We observed a similar pattern of attenuation for the KBIT-II verbal and nonverbal scores at age 7 years.

In Table 4, we give fully adjusted associations of any and exclusive breastfeeding with all cognitive test scores at ages 3 and 7 years. Associations of breastfeeding duration (any and exclusive) with the PPVT-III score and the KBIT-II verbal and nonverbal scores were positive, and 95% CIs excluded 0. The Figure shows the adjusted KBIT-II scores at age 7 years by category of any breastfeeding duration (<1, 1-3, 4-6, 7-9, 10-11, and ≥12 months). Associations of breastfeeding duration with WRAVMA scores were null (with narrow 95% CIs).

Estimated cognitive test mean score differences according to breastfeeding status at age 6 months are given in Table 5. Compared with children fed breast milk only, the PPVT-III score at age 3 years was approximately 3 points lower for children never breastfed and approximately 2 points lower for weaned children and those receiving mixed feedings (P = .01 for trend). We found a similar trend for the KBIT-II verbal and nonverbal scores at age 7 years but observed no appreciable trend for the WRAVMA or WRAML scores.

Compared with children who were never breastfed, the fully adjusted PPVT-III score at age 3 years was 1.45 (95% CI, −0.98 to 3.87) points higher for children who were ever breastfed, and the KBIT-II verbal score at age 7 years was 3.75 (1.17-6.33) points higher. The WRAVMA and WRAML scores were not statistically different (data not shown).

Stratifying by maternal postpartum fish intake (<2 vs ≥2 servings per week), the relationship between breastfeeding duration and the WRAVMA score at age 3 years seemed stronger in children of women with higher vs lower fish intake (Table 6), but the interaction was not statistically significant (P = .16 for interaction). For other cognitive outcomes, associations with breastfeeding duration were not appreciably stronger among children of women who consumed more fish.

We found that longer duration of breastfeeding and greater exclusivity of breastfeeding were associated with better receptive language at age 3 years and with higher verbal and nonverbal IQ at age 7 years. At age 7 years, the effect size of 0.35 verbal IQ points per month of any breastfeeding translates to 4.2 points, or almost one-third of an SD during 12 months, whereas the effect size of 0.80 verbal IQ points per month of exclusive breastfeeding translates to almost 5 points over 6 months. Effects were similar in direction but somewhat weaker in magnitude for nonverbal IQ and receptive language at age 3 years. We found no important main association of breastfeeding with visual motor skills or visual memory.

While numerous investigations have demonstrated associations of breastfeeding with later cognition, some studies^1,4,5 have had methodological flaws. In particular, adequate control for confounding factors is critical because breastfeeding and child cognition share many determinants, including maternal characteristics and environmental factors. A 2007 meta-analysis¹ identified maternal intelligence and the home environment as key confounders that are frequently overlooked and found only 1 prior study⁵ with appropriate adjustment, an analysis of data from the US National Longitudinal Survey of Youth (NSLY) in which the association of breastfeeding (ever vs never) with achievement scores at ages 5 to 14 years was attenuated from 4.7 to 1.3 points after adjustment for maternal intelligence and diminished to only 0.5 points after adjustment for sociodemographic and other variables, including the HOME-SF score. We also adjusted for maternal intelligence and the HOME-SF score, as well as numerous other potential confounders, and nevertheless found a substantially stronger association (3.75 points) of ever vs never breastfed with verbal IQ at age 7 years.

It is possible that differences in the degree of breastfeeding exclusivity explain why we observed a stronger association of breastfeeding with cognition than was seen in the NSLY.⁵ By classifying breastfeeding as ever vs never, the NSLY may have included in their breastfed group a substantial number of infants who received formula and breast milk, biasing results toward the null, but the authors did not report the degree of mixed feedings. Differences in breastfeeding duration may also explain our discrepant results. In a secondary analysis, the NSLY found that the achievement scores of children breastfed for at least 29 weeks were 1.5 points higher than those of children never breastfed (P = .01), but the authors considered their data about breastfeeding duration “less reliable” than data about whether a child was ever breastfed. Finally, variable outcome measures (achievement test score in the NSLY vs IQ in our study) may explain our different results.

We identified 4 additional observational studies^20-23 that adjusted for maternal intelligence and the HOME-SF score. While we found a modest association of breastfeeding with verbal intelligence at age 3 years, neither of the other 2 preschool studies found an important association with cognitive outcomes (McCarthy General Cognitive Index^21,22 and PPVT-Revised²² at age 4 years). Of the studies reporting school-age outcomes, one study²¹ found a 1.3-point (95% CI, −2.3 to 4.9) advantage of ever vs never breastfeeding on the Wechsler Full-Scale IQ at age 7 years; another study²³ found a 0.7-point (0.2-1.3) advantage on the same outcome at age 11 years; and the other study²² found no association with verbal or performance IQ at age 11 years (effect estimate not reported). All those effect estimates are smaller than ours, but none of the studies accounted for breastfeeding duration or exclusivity.

Studies of cohorts with different confounding patterns are also informative. Brion et al²⁴ analyzed associations of breastfeeding duration with IQ at age 8 years in 2 cohorts. In one cohort (the British Avon Longitudinal Study of Parents and Children), breastfeeding duration and child IQ were strongly predicted by measures of socioeconomic position, whereas in the other cohort (Pelotas, Brazil) child IQ was predicted by socioeconomic factors, but breastfeeding duration was not. In both cohorts, child IQ was strongly associated with breastfeeding duration, suggesting that confounding alone did not explain the relationship.

The results of our study are also consistent with a large cluster randomized trial²⁵ of breastfeeding promotion in which verbal IQ at age 6.5 years was 7.5 points (one-half of an SD) higher in the breastfeeding promotion group. By design, that study minimized confounding by measured and unmeasured factors; however, nonblinding of clinicians assessing the cognitive outcomes to participant breastfeeding status suggests the potential for bias. Together, the results of the well-controlled observational studies^20-23 (including ours), the analysis of cohorts without social patterning of breastfeeding (eg, in the Pelotas cohort),²⁴ and the large randomized trial²⁵ suggest that confounding does not account fully for the observed association of breastfeeding with later cognition.

In analyses stratified by fish intake, the beneficial effects of breastfeeding on visual motor ability at age 3 years seemed greater for women who consumed 2 or more servings compared with less than 2 servings per week, although the interaction was not statistically significant. This observation is consistent with the hypothesis that 1 or more nutrients in fish transfer to breast milk and account for some of the observed beneficial effect and is relevant to optimizing the maternal diet during lactation. Docosahexaenoic acid is incorporated in large amounts into cell membranes of the developing retina and brain. Its content in breast milk is variable²⁶ and depends on DHA sources in the maternal diet,^6,27 including fish; infant DHA status in turn depends on the DHA content of ingested breast milk.²⁷ Randomized trials of DHA supplementation during lactation have found beneficial effects of DHA on early motor skills²⁸ and sustained attention²⁹ but not visual motor function or general cognition.^28,30 Our observation may be explained by DHA or nutrients in fish other than DHA. It may also be a chance finding.

Strengths of our study include a prospective design, detailed contemporaneous measurement of duration and exclusivity of breastfeeding, and measurement of numerous potential confounding variables, including the home environment and maternal IQ. As in all observational studies, confounding by unmeasured factors is possible and may have led us to overestimate the true effect of breastfeeding, although our results are consistent with data from a randomized trial²⁵ of breastfeeding promotion that eliminates confounding by design. We measured cognition at school age, which tends to be stable through adulthood³¹ compared with measurement in preschool or earlier. The elevated socioeconomic status and high breastfeeding rate of our cohort may limit generalizability of the study findings. In addition, we followed up only a subset of the original Project Viva cohort to ages 3 and 7 years. The children we observed tended to be of higher socioeconomic status and were less likely to be of minority race/ethnicity than the children we did not follow up, which could have led to overestimates if the effect of breastfeeding on cognition was much weaker or in the opposite direction in those who dropped out, situations we find unlikely. Finally, for the statistically significant associations of breastfeeding with later cognition, 95% CIs were narrow and exclude a null result, but the lower confidence limits include values with little clinical importance.

In summary, our results support a causal relationship of breastfeeding in infancy with receptive language at age 3 and with verbal and nonverbal IQ at school age. These findings support national and international recommendations to promote exclusive breastfeeding through age 6 months and continuation of breastfeeding through at least age 1 year.

Accepted for Publication: February 6, 2013.

Corresponding Author: Mandy B. Belfort, MD, MPH, Division of Newborn Medicine, Boston Children’s Hospital, Harvard Medical School, Hunnewell Room 438, 300 Longwood Ave, Boston, MA 02115 (mandy.belfort@childrens.harvard.edu)

Published Online: July 29, 2013. doi:10.1001/jamapediatrics.2013.455.

Author Contributions:Study concept and design: Belfort, Kleinman, Gillman, Oken.

Acquisition of data: Bellinger, Gillman, Oken.

Analysis and interpretation of data: Belfort, Rifas-Shiman, Kleinman, Guthrie, Bellinger, Taveras, Oken.

Drafting of the manuscript: Belfort.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Belfort, Rifas-Shiman, Kleinman, Guthrie.

Obtained funding: Belfort, Gillman, Oken.

Administrative, technical, and material support: Gillman.

Study supervision: Bellinger, Taveras, Gillman, Oken.

Conflict of Interest Disclosures: None.

Funding/Support: This work was supported by grants K23 DK083817 (Dr Belfort), K24 HL68041 (Dr Gillman), K24 HD069408 (Dr Oken), R01 ES016314, R01 HD34568, and R01 HL64925 from the National Institutes of Health.