 |
|
 |
|
 |
 |
|
 |
 |
 |
 |
 |
 |
 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ARTICLE |
a Terry Sanford Institute of Public Policy, Duke University, Durham, North Carolina
b National Center for Children and Families, Teacher’s College, and College of Physicians and Surgeons, Columbia University, New York, New York
 |
ABSTRACT |
|---|
 TOP ABSTRACT METHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
PATIENTS AND METHODS. With data on 1645 American-born mothers participating in a longitudinal birth cohort study, we used linear regression to determine the influence of breastfeeding for at least 1 month on the Peabody Picture Vocabulary Test-Third Edition scores of 3-year-old children. Models were adjusted for an extensive set of demographic characteristics, including mother’s Peabody Picture Vocabulary Test and the Home Observation for Measurement of the Environment score. Mothers were categorized into 1 of 3 educational-status groups: no high school diploma, high school diploma, and some post–secondary education.
RESULTS. In unadjusted mean comparisons, breastfed children had Peabody Picture Vocabulary Test scores that were 6.6 points higher than children who were not breastfed. After adjusting for demographic characteristics and maternal verbal ability, the coefficient dropped to 1.72. Among mothers with education beyond high school, the children’s Peabody Picture Vocabulary Test scores in adjusted models were 2.2 points higher for breastfed children. Among mothers with a high school diploma or less, there were no significant differences in the children’s Peabody Picture Vocabulary Test scores by breastfeeding status. These results were consistent in white, black, and Hispanic children.
CONCLUSIONS. Maternal Peabody Picture Vocabulary Test scores mediate much of the association between breastfeeding and child verbal abilities. The beneficial effects of breastfeeding on children’s cognition may emerge only when breastfeeding is done in conjunction with other positive parenting behaviors. The advantageous effects of breastfeeding do not seem to be solely attributable to the superior nutrient content of breast milk.
Key Words: breastfeeding • cognition • maternal educational status
Abbreviations: PPVT—Peabody Picture Vocabulary Test • FF—Fragile Families and Child Wellbeing Survey • HOME—Home Observation for Measurement of the Environment • LCPUFA—long-chain polyunsaturated fatty acids
Although breastfeeding is widely regarded as beneficial for child and maternal health,1 its effect on children’s cognitive development remains an open question. Mothers who breastfeed have more education, higher incomes, and better mental and physical health than those who do not,2–5 and failure to adequately adjust for these differences can overestimate the effects of breastfeeding. Maternal verbal ability seems particularly important, because it can account for the entire association between breastfeeding and child cognition.6,7 Nevertheless, many studies have found a positive effect of breastfeeding on mental functioning,8–14 and a 20-study meta-analysis found that breastfeeding increased cognition scores by 5 points for low birth weight infants and 3 points for normal birth weight infants.15 Two recent literature reviews16,17 also concluded that breastfeeding was beneficial, particularly for low birth weight children.
The majority of previous studies have assumed that breastfeeding has a direct effect on child cognition; most commonly, this effect is ascribed to the superior nutrients found in breast milk.2,18,19 Yet, it is possible that breastfeeding is beneficial because it occurs in the context of other positive parenting practices. Breastfeeding mothers may respond more to the needs of their child, spend more time in mother-child interaction, or engage in more stimulating activities. These advantageous parenting practices, together with breastfeeding, may produce cognitive benefits.
Advantageous parenting practices, particularly cognitively stimulating parenting, are strongly correlated with maternal education,20,21 in part because better-educated mothers have more opportunities to provide stimulation for their children. In this study we stratified our sample according to education level to analyze the association between breastfeeding and children’s verbal abilities. We would expect that among better-educated mothers, those who breastfeed, relative to those who do not, will have children with higher cognition scores. This association may not exist among less-educated mothers, because they may not be able to draw on the same resources as mothers with more education.
We examined the effect of breastfeeding on a measure of children’s verbal ability, the Peabody Picture Vocabulary Test-Third Edition (PPVT-III). Data come from the Fragile Families and Child Wellbeing Survey (FF), a longitudinal birth cohort study drawn from 20 cities in 15 states. We first present estimates of breastfeeding that have been adjusted for maternal PPVT-III scores, hypothesizing that maternal verbal ability mediates much of the effect of breastfeeding. Next, we test the hypothesis that breastfeeding is moderated by maternal education by estimating separate models for 3 education groups: no high school diploma, high school diploma, and some post–secondary education.
|
METHODS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
3700 unmarried and 1200 married new parents. Couples were sampled from 75 hospitals in 20 large cities in 15 different states. Mothers were interviewed in the hospital shortly after giving birth, and fathers were interviewed either in the hospital or wherever they could be located. When the child was 1 and 3 years old, parents were surveyed by telephone. The FF surveyed parents on topics related to child and family well-being, and additional details of the study, including city and hospital selection, are published elsewhere.22 At 3 years, an in-home assessment was conducted, which included measures of both mother and child cognitive ability and the quality of the child’s home environment. Data used in this article come from the baseline survey, first- and third-year surveys, and the in-home assessment.
At baseline, information was collected from 4898 mothers, of which 4231 (86%) participated in the year-3 telephone survey. Of the 4231 families, 3336 (79%) participated in some portion of the in-home assessment, but only 2182 (52%) participated in the child observation. Nonresponse on the child observation was a result of family relocation (85% of nonrespondents) or refusal (15% of nonrespondents). Of these 2182 observations, an additional 221 were missing data on either the cognitive or the mother-child interaction outcomes, resulting in 1961 completed in-home assessments. Cases were also deleted if they were missing data on breastfeeding (n = 62), the child had a physical disability (n = 58), or the mother was not born in America (n = 196), which resulted in a final sample size of 1645.*
Informed consent was obtained from all participants. The survey was approved by the internal review boards of Princeton and Columbia Universities.
Subjects who completed the year-3 telephone survey versus those who completed the in-home assessment had few differences in demographic characteristics, with the exception that they had lower household incomes and younger children. Mothers who participated in both the in-home assessment and the child observation, as compared with those who participated in the in-home assessment without the child observation, were more likely to be black and have older children.
Measures and Procedure
Maternal and child cognitive ability was measured at 3 years with the PPVT-III, a measure of receptive vocabulary.23 The PPVT-III correlates well with standardized measures of intelligence such as the Wechsler Intelligence Scale-Third Edition.23 The PPVT-III was administered in both English and Spanish by interviewers who had received appropriate training. As has been done elsewhere,24 extreme PPVT-III scores were imputed; scores below 55 were imputed to 55 for 104 children and 12 mothers. Because the average age at assessment was 36 months, 6% of children had PPVT-III scores of <55. Because the PPVT-III can only be administered to children 
3 years, this left-censoring reflects a basal effect. Results do not change if scores below 55 are not imputed. The correlation between maternal PPVT-III and child PPVT-III, education, and income was 0.42, 0.47, and 0.46, respectively.
Breastfeeding behaviors, collected at year 1, were based on mother’s report of how long she breastfed the child. On the basis of the work of Jain et al,25 and because mothers may change feeding methods within the first few weeks of life, mothers were classified as having breastfed if they did so for at least the first month.
Maternal educational status was classified into 3 groups: those who did not complete high school, those who graduated from high school or obtained their general equivalency degree, and those who had some education beyond high school.
Educational status was originally measured at baseline, but cases were coded to reflect any additional education received.
Home environment was measured through 2 interviewer-rated scales of the Home Observation for Measurement of the Environment (HOME) scale.26 The nonpunitive/hostility scale consisted of 5 items on which the interviewer rated the mother on the absence of antagonistic or aggressive behaviors (eg, if the mother did not shout or spank the child). The emotional-responsiveness scale consisted of 6 items that measured positive verbal maternal interactions (eg, if the mother spontaneously praised the child). For both scales, higher scores mean better home environments.
Demographic variables included maternal and paternal race/ethnicity and age, paternal education (measured similarly to that of mothers), parity, and if the mother had a child subsequent to the child participating in the FF. Marital status was measured through 4 dichotomous variables: married to the birth father (the reference category), cohabiting with the birth father, no romantic relationship, and married or cohabiting with a new partner. Household income was the log of the mother’s household income averaged over the 3 rounds of the survey. The child characteristics included if the child was female, the child’s age (in months), and if the child had a low birth weight. Maternal health variables were pregnancy behaviors (smoked during pregnancy, saw a doctor during first trimester), a self-report of health status (if mother was is in good or excellent health), and depression. Depression was measured by using the Composite International Diagnostic Interview Short Form,27,28 following the developer’s guidelines,29 to classify mothers who had a score of 3 as having a probable major depressive episode (according to Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition criteria).
The correlations between the 2 HOME scales and maternal PPVT-III were low (r < 0.31 in all cases). Correlations between the HOME measures and income, and between HOME and maternal education, were all <0.20. The correlation between income and education was 0.43.
Most control variables were taken from the year-3 survey, because maternal PPVT-III and the HOME scores were only measured at that time. Race and ethnicity, parity, health behaviors during the pregnancy, and child’s birth weight status and gender were gathered during the baseline interview. In other models not shown, we substituted year-3 variables for baseline variables; using baseline measures of education, relationship status, age, and income did not substantively change our results.
The association between breastfeeding and PPVT-III scores was modeled by using hierarchical multivariate regression. The first model estimated a bivariate association between breastfeeding and PPVT-III scores, and in the second and subsequent models, additional blocks of covariates were added. Changes in the breastfeeding coefficient across models provided an indication of how breastfeeding was mediated by the other covariates. In all models, to account for clustering within city, we used Huber-White standard errors that correct for potential nonindependence across observations. To address the problem of missing data in the control variables, we used multiple imputation.30 In multiple imputation, every missing case is replaced by n simulated values, resulting in n complete data sets. These data sets can then be analyzed by using standard regression methods, and the results are combined and adjusted for the multiple imputation procedure. Rates of missing data were low, because only a few variables were missing in >4% of the cases.
|
RESULTS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
In bivariate associations (Table 2, model 1), the average breastfed child had PPVT-III scores that were 6.2 points higher than a child who was not breastfed. This difference remained significant at the 5% level even after adjusting for demographics, health behaviors, and the HOME measures (models 2 and 3). However, the coefficient decreased to 1.72 once the mothers’ PPVT-III scores were added (model 4), and its P value increased to .060.
|
The first 2 rows of the table indicate no significant bivariate association between breastfeeding and PPVT-III scores for mothers with a high school diploma or less (model 1). The coefficients were small (1.9 for mothers who did not complete high school, 2.1 for mothers who did), and the coefficients only decreased in subsequent models.
|
In additional analyses not shown, mothers were divided into education according to race and ethnicity groups. Although the sample sizes were small, the pattern of effects was similar. For non-Hispanic white, non-Hispanic black, and Hispanic children, it is only among mothers with more education that breastfeeding was associated with an increase in PPVT-III scores.
Because of the importance of considering the dose-response effects of breastfeeding, in other models not shown we classified breastfeeding according to duration: never breastfed (the reference category), breastfed for 1 month, breastfed for 2 to 5 months, and breastfed for 6 months. In these models, there were no significant breastfeeding effects among mothers without some post–secondary education. For better-educated mothers, the coefficients for all 3 dichotomous breastfeeding measures were positive, but only the coefficient associated with breastfeeding for 6 months was significant at conventional levels (b = 3.63; P < .01). We also modeled breastfeeding as a continuous variable. Breastfeeding had a significant and positive effect, but only for the upper-education group; for every month this group breastfed, their child’s PPVT-III score increased by 0.27 points (P < .05).
|
DISCUSSION |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
One of those advantages is that mothers who breastfeed have higher verbal abilities.6,7 Given the high heritability in cognition between parent and child,33 models that do not control for maternal abilities risk finding a spurious correlation between breastfeeding and cognitive functioning. Our findings underscore this point, because adjusting for maternal PPVT-III scores decreased the size of the breastfeeding coefficient in the full sample by 48% and resulted in a coefficient that was no longer statistically significant at the 5% level. The large effect of maternal PPVT-III is consistent with past research6,7 and leads us to conclude that measures of maternal ability must be included in studies on breastfeeding and child cognitive development.
Another potential advantage associated with breastfeeding is that it may be correlated with other positive parenting practices. In this study, on the basis of past research that has found a strong positive correlation between education and cognitively enriching parenting,21,34 we had hypothesized that these parenting practices might be occurring among the better-educated mothers and had stratified our sample on the basis of education group. We found that among mothers with some post–secondary education, children who were breastfed for at least 1 month had higher PPVT-III scores than those who were not. The gain was not large; once all relevant maternal and child factors were taken into account, the difference was 2 points. It is notable, however, that there were no differences in PPVT-III scores, even in unadjusted mean comparisons, for mothers with a high school diploma or less.
A 2-point difference is 0.14 of an SD, a small effect according to Cohen’s rules35 for effect-size magnitude. However, the difference is similar to the gap in child PPVT-III between married and unmarried mothers (0.18 of an SD). In addition, as explained by Rock and Stenner,36 small differences in a vocabulary-test score may not mean much for any one individual but can be consequential for groups of children (such as those entering into a kindergarten class).
These results are consistent with the one study that has looked at breastfeeding by maternal educational status. Oddy et al11 found higher PPVT-R (revised edition) scores for breastfed infants, but only for mothers with a college degree. The authors do not explain their education-breastfeeding interaction but instead attribute their significant findings to the beneficial properties of breast milk.11
Some limitations to our study should be noted. First, the FF cannot be generalized to the larger US population. The FF oversampled births from unmarried couples in cities across the United States, which resulted in a sample that is predominantly urban and low income. Therefore, these results pertain to relatively disadvantaged mothers who live in select cities, and the findings cannot be applied to the effect of maternal education for mothers with a bachelor’s degree or higher. Omitted variable bias may also be a problem. Although models controlled for a broad set of characteristics, there may be other relevant parenting behaviors and/or environmental factors that were not included.
Another disadvantage of the FF is that it does not contain measures of exclusive breastfeeding. It is unclear if mothers used formula, when they began to do so, and how much they used. This limitation has important implications, because well-educated mothers are more likely to exclusively breastfeed.4 Therefore, it is possible that the lack of observed effect among lower-educated women is because some breastfeeding mothers are actually supplementing with formula, and both groups are being conflated in the comparison between mothers who do and do not breastfeed. The effects among better-educated mothers could also be explained, because that might represent a more accurate comparison between infants who are breastfed and those who are not breastfed.
Research on breastfeeding exclusivity has found that the majority of mothers of all educational levels who initiated breastfeeding do so exclusively for at least the first month of life.4 On the basis of this evidence, we infer that most mothers who breastfeed for the first month are doing so exclusively. We are less sanguine about the results that use different durations of breastfeeding, because it is likely that some mothers classified as breastfeeding are supplementing with formula. Nevertheless, these results are consistent with our earlier findings.
There are other sources of potential bias. Because the breastfeeding data are retrospective, the results may be biased if better-educated mothers can report their breastfeeding behaviors more accurately. We are unaware of any literature that documented a connection between breastfeeding recall and education, and breastfeeding information was collected within 1 year of the child’s birth. Although the FF shows lower rates of breastfeeding among high school dropouts than in other surveys,4,37 this may reflect the minority nature of the sample. White mothers, when stratified according to educational level, breastfed at rates commensurate with those from other large-scale surveys (results not shown).38 Another source of bias could arise if interviewers assigned higher PPVT-III scores to better-educated mothers. However, information on education status was collected by telephone, whereas the PPVT-III scores were gathered in person by a different interviewer. We also note that neither of these biases could account for differences found by breastfeeding status within the group of better-educated mothers.
In the sample, mothers who completed the in-home assessment were also more likely to be black. Because results were replicated in models estimated separately for white, black, and Hispanic mothers, we do not believe this selection unduly bias our findings.
This work occurs in the context of other studies that have attributed the beneficial effects of breastfeeding to the presence of long-chain polyunsaturated fatty acids (LCPUFAs) in breast milk.2,11,12,14,39,40 LCPUFAs are necessary for retinal and neural development,3,18,41,42 and animal studies indicate that severe LCPUFA deprivation negatively affects cognitive and behavioral performance.43 Evidence showing a causal role between LCPUFAs and enhanced human neural functioning, however, is inconclusive.19,41,43,44 Although these findings do not relate directly to possible associations between LCPUFAs and cognition, they may indicate that LCPUFAs by themselves are insufficient to create cognitive gains.
|
CONCLUSIONS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Address correspondence to Christina M. Gibson-Davis, PhD, Terry Sanford Institute of Public Policy, Duke University, PO Box 90245, Durham, NC 27708. E-mail: cgibson{at}duke.edu
The authors have indicated they have no financial relationships relevant to this article to disclose.
* Immigrant mothers were omitted because it was unclear if their educational status would be adequately captured by the American-based measures. Including immigrant mothers does not substantially change our results, nor does including children with a physical disability. 
There were too few mothers who had a bachelor’s degree to constitute their own category.
The Hispanic sample includes 28 mothers who classified themselves as another race or ethnicity (eg, Asian or Pacific Islander).
In additional models, we included an interaction term between breastfeeding and education. The term was not significant at conventional levels, but its positive direction was consistent with what we would expect. Furthermore, in these models, the associated breastfeeding coefficient decreased by 50%, indicating that some of its effect was moderated by the interaction term.
|
REFERENCES |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONCONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
 |
|
 |
|
  N. Ribas-Fito, J. Julvez, M. Torrent, J. O. Grimalt, and J. Sunyer Beneficial Effects of Breastfeeding on Cognition Regardless of DDT Concentrations at Birth Am. J. Epidemiol., November 15, 2007; 166(10): 1198 - 1202. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
||
|
||
Home | My Pediatrics | Journal Information | Current Issue | Past Issues | Subscriptions & Services | Contact Us Click here for faster international access © 2006 American Academy of Pediatrics. All rights reserved. |
||
|
||
