Objectives The association between breastfeeding and child cognitive development is conflicted by studies reporting positive and null effects. This relationship may be confounded by factors associated with breastfeeding, specifically maternal socioeconomic class and IQ.
Design Systematic review of the literature.
Setting and participants Any prospective or retrospective study, in any language, evaluating the association between breastfeeding and cognitive development using a validated method in healthy term infants, children or adults, was included.
Primary and secondary outcome measures Extracted data included the study design, target population and sample size, breastfeeding exposure, cognitive development assessment tool used and participants’ age, summary of the results prior to, and following, adjustment for confounders, and all confounders adjusted for. Study quality was assessed as well.
Results 84 studies met our inclusion criteria (34 rated as high quality, 26 moderate and 24 low quality). Critical assessment of accepted studies revealed the following associations: 21 null, 28 positive, 18 null after adjusting for confounders and 17 positive—diminished after adjusting for confounders. Directionality of effect did not correlate with study quality; however, studies showing a decreased effect after multivariate analysis were of superior quality compared with other study groupings (14/17 high quality, 82%). Further, studies that showed null or diminished effect after multivariate analysis corrected for significantly more confounders (7.7±3.4) as compared with those that found no change following adjustment (5.6±4.5, p=0.04). The majority of included studies were carried out during childhood (75%) and set in high-income countries (85.5%).
Conclusions Much of the reported effect of breastfeeding on child neurodevelopment is due to confounding. It is unlikely that additional work will change the current synthesis. Future studies should attempt to rigorously control for all important confounders. Alternatively, study designs using sibling cohorts discordant for breastfeeding may yield more robust conclusions.
- Public Health
- Reproductive Medicine
This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 3.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/3.0/
Statistics from Altmetric.com
Although most published data support the association between breastfeeding and IQ of the offspring, debate remains whether this is a causal relationship or an association with favourable parental characteristics.
We conducted a systematic review of the literature investigating the association between breastfeeding and cognitive outcomes of healthy term infants.
Over 80 studies addressing this issue were published with conflicting results.
Studies where the initial positive effect of breastfeeding on IQ disappeared or diminished after multivariate analysis controlled for significantly more confounders than studies showing no such change.
Much of the reported effect of breastfeeding on child cognitive abilities is due to the maternal cognitive and socioeconomic effects.
Strengths and limitations of this study
The significant heterogeneity in study design and rigour precluded the conduct of a formal meta-analysis.
The potential of breast milk to enhance cognitive development has been the focus of numerous studies since Hoefer and Hardy's9 initial observation in 1929. It is generally agreed that children who breastfeed are more intelligent; however, debate remains whether this is a causal relationship or merely an association with favourable parental socioeconomic class and IQ. The beneficial effects of breastfeeding on the child's neurodevelopment are hypothesised by some to be mediated by long-chain polyunsaturated fatty acids (PUFA)10 which are present in human milk, but not in cow's milk or most infant formulas.11 However, a recent systematic review of all randomised trials where mother's diet was supplemented with PUFAs during pregnancy has failed to confirm such an effect.12
The pendulum of opinion has swung back and forth with different investigators showing inconsistent results depending upon study design and rigour. The Achilles heel of most of these studies, and the probable explanation for the conflicting results, is the difficulty in controlling for confounders that may affect child development. Furthermore, the ability to clarify this relationship is hindered by ethical considerations, which preclude randomised controlled trials (RCT), given that breastfeeding has other protective effects and the highly personal nature of the decision to breastfeed.
Well-established confounders in breastfeeding research include demographic and IQ differences between mothers who breastfeed and those who choose not to.13 Parents who score high on a range of cognitive abilities have children with above average IQ scores.13 In parallel, advantage in mother's IQ more than doubles the odds of breastfeeding.13 Thus, some of the published data demonstrates the disappearance of the breastfeeding effect on child's cognition after correction for maternal IQ.
In an attempt to partially overcome these sources of bias, a few randomised trials have been published, with randomisation to breastfeeding promotion14 ,15 or in preterm infants.16 In the breastfeeding promotion intervention trial (PROBIT Trial) by Kramer et al15, IQ scores and academic performance tests were more favourable in the intervention group; however, statistically significant differences were only shown for some of the subscores.
Systematic reviews examining the impact of breastfeeding on cognitive abilities have reached conflicting results.13 ,17–19 The meta-analysis by Jain et al18 suggests that less than 25% of studies in this area have adjusted for sociodemographic confounders.
There is a paucity of literature critically assessing the current published evidence within this field. In trying to address these challenges, the objective of the present work was to conduct a systematic review of published studies investigating the association between breastfeeding and neurodevelopmental outcome of healthy infants born at term.
The study was conducted based on a prospectively prepared protocol, using the Preferred Reporting Items for Systematic reviews and Meta-Analyses guidelines.20
Searches were conducted in the following databases (all from inception to July 2011): MEDLINE(R) with Daily Update, Cochrane Central Register of Controlled Trials, Cochrane Database of Systematic Reviews, Database of Abstracts of Reviews of Effects, Health Technology Assessment, NHS Economic Evaluation Database, EMBASE and PsycINFO using the OvidSP interface and on Science Citation Index Expanded (SCI-EXPANDED), Social Sciences Citation Index (SSCI), Conference Proceedings Citation Index-Science (CPCI-S), and Conference Proceedings Citation Index—Social Science & Humanities (CPCI-SSH) using the Web of Knowledge interface.
A search strategy was developed based on the MEDLINE, EMBASE and PsycINFO database subject headings and the ‘Used for’ synonyms listed in the scope notes for the terms ‘breast feeding’, ‘human milk’, ‘breast milk’, ‘infant formula’, ‘artificial milk’, ‘cognition’, ‘intelligence’ and ‘intelligent quotient’. Reference lists of identified studies, textbooks, previously published systematic reviews and review articles were also searched. No language restrictions were applied and studies in languages other than English were translated for incorporation into the study.
A second complimentary literature search was carried out on April 2013 for all studies published during the period August 2011–December 2012.
Prospective and retrospective studies (RCTs, non-randomised controlled clinical trials, cohort studies, longitudinal studies and case-control studies) were included if
One of the study aims was to address the question of breastfeeding and cognitive development;
The authors used reliable validated methods to evaluate cognitive development (eg, Bayley scales of infant development, Wechsler Intelligence Scale for Children Revised (WISC-R));
Prospective or retrospective documentation of use and duration of breastfeeding was used;
The authors focused on healthy term infants and not those at increased biological risk for developmental delays (eg, prematurity, intrauterine growth restriction).
Studies were excluded if
The study group included preterm or small for gestational age babies.
Evaluation of cognitive development was carried out using only a non-reliable or subjective tool (such as school grades, or maternal report).
Dietary patterns and breastfeeding were not evaluated since birth.
Titles and abstracts were reviewed for possible exclusion by two reviewers (AW and CS). If both reviewers excluded a citation, it was eliminated from further review. If at least one reviewer included the citation or if there was insufficient information to make a determination from the title and abstract, the full article was obtained for review. Full text articles were reviewed by three authors (AW, CS and AC) for suitability for inclusion. Disagreements regarding study eligibility were resolved by consensus.
Study quality grading
Quality assessment of individual studies was performed by two authors (AW and CS) using the three category summary grading system (A, B, C) suggested by Ip et al.19 Their system defines a generic grading system that is applicable to each type of study design including RCTs and cohort and case-control studies as follows:
A (good): A study that adheres mostly to the commonly held concepts of high quality including the following: clear description of the population, setting, interventions and comparison groups; clear description of the comparison groups; appropriate measurement of outcomes; appropriate statistical and analytic methods and reporting; no reporting errors; less than 20% dropout; clear reporting of dropouts and appropriate consideration and adjustment for potential confounders.
B (fair/moderate): Category B studies do not meet all the criteria in category A because they have some deficiencies, but none of them are likely to cause major biases. The study may have suboptimal adjustment for potential confounders. The study may also be missing information, making it difficult to assess limitations and potential problems.
C (poor): Category C studies either did not consider potential confounders or did not adjust for them appropriately. These studies may have serious shortcomings in design, analysis or reporting; have large amounts of missing information, or discrepancies in reporting.
Extracted data were compiled in an evidence table. The table includes a description of the studies that addressed the key question according to the inclusion/exclusion criteria. The table provides information about study design, target population and sample size, description of breastfeeding exposure and method of categorisation, nature of the comparison group, cognitive development assessment tool used and participants’ age, summary of the results prior to adjustment for confounders, a list of all confounders adjusted for, differences in IQ between the groups after adjustment for possible confounders (if available), and study quality grading according to the scale described above.19
Comparison of studies based on their results or quality was performed by χ² or analysis of variance as appropriate.
The flow of the literature search is displayed in figure 1. Of the 1696 potentially relevant citations identified, 84 studies met the a priori inclusion criteria for this systematic review (table 1).9 ,13 ,15 ,21–101 The overall agreement between reviewers on the inclusion of studies was 100%.
Out of these 84 publications, 34 were rated as high quality (grade A), 26 as moderate (grade B) and 24 as low quality (grade C). Overall, based on the primary endpoint of cognitive function, there were 21 studies showing no association between IQ and breastfeeding,24 ,25 ,30–32 ,35 ,36 ,39 ,54 ,56 ,62 ,63 ,73 ,74 ,86 ,87 ,89 ,91 ,97 ,100 ,101 28 positive studies,9 ,15 ,21 ,27–29 ,37 ,38 ,42 ,48 ,51 ,57 ,59 ,61 ,64 ,68 ,71 ,72 ,75 ,79–82 ,85 ,90 ,94 ,95 ,98 18 initially positive studies that became negative after accounting for confounders,13 ,23 ,26 ,33 ,43–46 ,49 ,50 ,52 ,55 ,65 ,66 ,69 ,78 ,84 ,96 and 17 studies where the initial positive effect was diminished but remained statistically significant after accounting for confounders (table 2).22 ,34 ,40 ,41 ,47 ,53 ,58 ,60 ,67 ,70 ,76 ,77 ,83 ,88 ,92 ,93 ,99
In general, the directionality of the results did not correlate with the quality of the studies. However, the studies showing a decrease in the effect after multivariate analysis were of superior quality compared with the rest of the studies (ie, 14 of 17 had a quality score of A—table 2).
Different studies corrected in their analyses for different potential confounders, ranging from 0 to 16 total confounders (table 1). Confounders commonly considered in these studies were socioeconomic status, maternal education, birth weight, gestational age, birth order and gender. Some considered the quality and quantity of stimulation of the child to be crucial confounders but did not consider maternal or paternal intelligence and other important factors. Studies that showed null or diminished effect in their multivariate analysis controlled for significantly more confounders (7.7±3.4) as compared with those that found no change following adjustment for confounders (5.6±4.5, p=0.04). Furthermore, many of the studies did not have a clear definition of breastfeeding or cumulative breast milk exposure.
Table 3 divides the included studies according to their settings: developed versus developing world. The majority of included studies were set in the developed world (71/84, 85.5% vs 13/84, 15.5% in the developing world). The quality of the studies set in developing countries were generally poorer, given our criteria: 46% graded A+B (6/13) in studies set in developing countries, compared with 76% (54/71) in studies set in developed countries. Developing country studies were also more likely to reach a null association or null association after adjustment for confounding compared with developed country studies (8/13, 61% vs 31/84, 37%, respectively).
A large variety of cognitive assessment tools were used and study outcomes were measured anywhere from 8 days of age into adulthood. Table 4 divides the included studies according to age groups of participants: infancy, childhood and adulthood, with the corresponding direction of results and study quality. The majority of included studies measured intelligence during the childhood period (age 1–18 years: 70/93 studies, 75%). Studies performed during childhood and reaching an initial positive association, weakened after adjustment, were generally of higher quality than other studies (12/14 quality grade A, 86%, table 4). Studies performed during infancy or adulthood were more likely to find a null association (before or after adjustment) compared with studies performed during childhood (Infancy group—61%, adulthood—60%, childhood 43%, table 4).
The significant heterogeneity in study design and rigour precluded the conduct of a formal meta-analysis.
The continuing debate of whether breastfeeding imparts direct advantage on child cognition, or whether this is merely an association with favourable familial socioeconomic status and cognition, is not purely theoretical. From a public health perspective, if breastfeeding has biological effects on a child's IQ, this will be one of the very few cost-effective means to significantly improve a child's neurodevelopment. If, on the other hand, there is no such effect, in the case where breastfeeding is either impossible or not sought by the mother, this will allow these women to rest assured that their choice will not have long-term developmental consequences.
In the case of other comparable therapeutic dilemmas, conflicts are typically resolved through RCTs, which are not ethically feasible in this case, given that breastfeeding has other protective effects and the highly personal nature of the decision to breastfeed.
The closest comparison to a formal RCT in reducing selection bias would be sibling-pair analysis, when cognition of breastfed infants is compared with that of their siblings who were formula fed. This design ensures similar socioeconomic and maternal characteristics. Unfortunately, the few studies that have followed this design reached conflicting results.13 ,40
The second closest design to RCT was employed in the PROBIT study by Kramer et al,14 ,15 who cluster randomised women in Eastern Europe to receive or not receive formal education about the advantages of breastfeeding. This study did show favourable effects, but it has been argued that the mothers randomised for the breastfeeding promotion arm might have been influenced not only in providing higher rates of breastfeeding, but also by improving other positive health behaviours.
Our analysis reveals that there are over 80 studies addressing this issue and that their results divide almost evenly between positive and negative associations. The quality of ‘positive’ or ‘negative’ studies did not differ, except for higher quality on average in studies that showed an apparent decrease in effect after multivariate analysis.
We have shown that studies where the initial positive effect of breastfeeding on IQ disappeared or substantially diminished after multivariate analysis controlled for significantly more confounders than studies showing no such change. When compared with a meta-analysis conducted 14 years ago,17 it appears that many more new studies did attempt to control for confounding measures of socioeconomic status and parental education, among others.
Given that more tight control of confounders resulted in greater likelihood of disappearance of breastfeeding effect, it can be argued that the remaining positive effect reflects residual uncontrolled bias, as shown by Der et al13 in their large study. In that study, before adjustment, breastfeeding was associated with an increase of around 4 points in mental ability. Post hoc analysis revealed that adjustment for maternal intelligence accounted for most of this effect—where full adjustment for a range of relevant confounders yielded a small (0.52) and non-significant effect size (95% CI −0.19 to 1.23).
In our systematic review, a similar effect was recorded by a total of 18 studies, and in addition 17 studies showed substantially diminished effect after adjustment.
When we examined studies based on setting (table 3), we found that the majority of the 84 included studies were set in the developed world (85%). Studies completed in middle-income and low-income countries were nearly twice as likely to find a null association (before or after adjustment) compared with studies set in developed countries (61% vs 43.5%, respectively).
This may be due to the fact that in many low-income and middle-income countries high rates of some degree of breastfeeding exist102 and comparisons between breastfed and non-breastfed populations may examine more homogeneous study groups (ie, parental socioeconomic status, income and parental IQ).34 In contrast, studies originating in the developed world exhibit a greater heterogeneity between breastfed and non-breastfed populations13 as the choice to breastfeed is associated with a family's socioeconomic status, maternal education, maternal intelligence and social advantage.13 ,17–19 If a biological effect truly exists between breastfeeding and infant IQ, one would expect this relationship to exist in multiple settings, including the developing world. The fact that this relationship is less apparent in developing countries suggests that much of the observed relationship may be due to parental social advantage, confounding the choice to breastfeed.
This systematic review includes studies using a large variety of cognitive assessment tools and age span. The majority of included studies measured intelligence during the childhood period (age 1–18 years, 75%). Studies performed during infancy or adulthood were more likely to find a null association (before or after adjustment), although the number of included studies is small. Possible explanations for this finding include reduced accuracy of IQ evaluation in infancy (<1 year) on the one hand, and a variety of additional factors influencing IQ at an older age (>18 years), on the other.
Another factor that needs to be seriously considered in our review is the existence of bias against the null hypothesis. The likelihood of studies not detecting a significant effect in pregnancy to be submitted and published in the peer review literature is substantially lower than that of positive studies.103 ,104 This can create a distorted balance that may seriously affect the conclusions on effects of interventions.
In conclusion, this systematic review suggests that much of the reported effect of breastfeeding on child cognitive abilities is due to the maternal cognitive and socioeconomic effects. When considered together with the fact that a recent systematic review failed to corroborate a biological effect of milk PUFA on brain development, it is quite likely that breastfeeding does not, by itself, directly affect child IQ.
Although it is unlikely that additional studies will change substantially the current synthesis, future studies in this field should attempt to rigorously control for all important confounders even if they are difficult to obtain (eg, parental IQ). Alternatively, study designs using sibling cohorts discordant for breastfeeding may yield more robust conclusions to further clarify this dilemma.
Contributors AW has participated in all phases of this study including literature search, data collection and review, quality grading, interpretation of the results, and has written a part of the manuscript and the revised manuscript. CS has participated in all phases of this study including literature search, data collection and review, quality grading, and revision of the manuscript. AC has participated in the literature search, data collection and review, translation of foreign language manuscripts, and has taken a significant part in the manuscript preparation and revision. GK participated in all phases of this study. He initiated the study and supervised actively throughout its conduct. Specifically, he was involved in data interpretation and statistical analysis, and has written and revised a substantial part of the manuscript.
Funding This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement The study protocol is available with the authors upon request.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.