Literature DB >> 35613097

To what extent does confounding explain the association between breastfeeding duration and cognitive development up to age 14? Findings from the UK Millennium Cohort Study.

Reneé Pereyra-Elías1, Maria A Quigley1, Claire Carson1.   

Abstract

BACKGROUND: Breastfeeding duration is associated with improved cognitive development in children, but it is unclear whether this is a causal relationship or due to confounding. This study evaluates whether the observed association is explained by socioeconomic position (SEP) and maternal cognitive ability.
METHODS: Data from 7,855 singletons born in 2000-2002 and followed up to age 14 years within the UK Millennium Cohort Study were analysed. Mothers reported breastfeeding duration, and children's cognitive abilities were assessed at 5, 7, 11, and 14 years using validated measures. Standardised verbal (age 5 to 14) and spatial (age 5 to 11) cognitive scores were compared across breastfeeding duration groups using multivariable linear mixed-effects models (repeated outcome measures).
RESULTS: At all ages, longer breastfeeding durations were associated with higher cognitive scores after accounting for the child's own characteristics. Adjustment for SEP approximately halved the effect sizes. Further adjustment for maternal cognitive scores removed the remaining associations at age 5, but not at ages 7, 11 and 14 (e.g.: verbal scores, age 14; breastfed ≥12 months vs never breastfed: 0.26 SD; 95%CI: 0.18, 0.34).
CONCLUSION: The associations between breastfeeding duration and cognitive scores persist after adjusting for SEP and maternal cognitive ability, however the effect was modest.

Entities:  

Mesh：

Year:  2022        PMID： 35613097      PMCID： PMC9132301          DOI： 10.1371/journal.pone.0267326

Source DB:  PubMed          Journal:  PLoS One        ISSN： 1932-6203            Impact factor:   3.752


Introduction

The association between breastfeeding and cognitive development has been extensively investigated. A systematic review found that on average breastfed infants scored 3.44 points higher in standardised intelligence tests than their non-breastfed peers [1], however, a causal relationship is still debated. It is argued that improved cognitive outcomes could be explained by other characteristics of the women who breastfeed their babies, principally socioeconomic position (SEP) and maternal intelligence [1-6]. In developed economies, including the UK, women from more socioeconomically advantaged backgrounds are more likely to breastfeed their infants [7], and to breastfeed for longer [8]. SEP refers to a range of factors that influence the position of individuals within society, and is reflected by education, occupation, wealth, and income, among others, which usually shape the distribution of health, social and wellbeing outcomes [9]. In this case, a higher SEP may be associated with a more favourable nurturing environment, less adversity, and more opportunities for intellectual stimulation, all of which may influence child cognitive outcomes [10]. Moreover, maternal intelligence also tends to be associated with longer breastfeeding durations [1-3] and is an important predictor of intelligence in the offspring [11]. Despite the potential influence of these variables on the association of interest, systematic reviews have shown that many studies do not conduct sufficient adjustment for potential confounders, with maternal intelligence being one of the most frequently overlooked variables [1-2]. For example, in the systematic review by Horta et al [1], only nine of the 17 studies included had adjusted for maternal cognitive ability. Additionally, the majority of studies that evaluate cognitive outcomes do so at young ages only (first years of life) and have relatively limited sample sizes [1-3]. Most of the studies only controlling for SEP find that longer breastfeeding durations are associated with higher cognitive scores [12-19]. However, others report no association after adjusting for SEP [20,21]. Given that the association between SEP and breastfeeding can vary by setting (no or inverse association in low- and middle-income countries) [22], one of the proposed ways to build the case for causal inference is to replicate this research question in populations with different confounding structures [4]. Results from a Brazilian sample, where breastfeeding patterns did not vary considerably with SEP, showed that the association persisted after adjustment for socioeconomic circumstances [5]. Additional adjustment for maternal intelligence removes the observed associations in some studies [23-25], but not in others [26-31]. The only experimental study conducted to date randomised a breastfeeding promotion intervention (rather than breastfeeding duration per se)—which by design would be controlling for SEP, maternal intelligence, and other confounders—and found a positive association between breastfeeding duration and child cognitive abilities at age 6 [32]. The aim of the present study was to evaluate how much of the association between breastfeeding duration and cognitive development is due to confounding by SEP and maternal cognitive scores among children from the UK Millennium Cohort Study (MCS). While this association has been explored previously in the MCS, there are gaps yet to be filled. Previous studies have only evaluated cognitive outcomes up to age 7 [16,33-36], while data on cognitive development is currently available at ages 11 and 14. Additionally, given that the maternal cognitive measures have been collected only recently, none of the previous studies have controlled for their effect. Therefore, the associations found on these studies [16,33-36] may be attributable to an important source of residual confounding.

Methods

Study population

The MCS is a nationally representative cohort study that recruited 18818 children aged around 9 months living in the UK in 2000–2002 and followed them up at ages 3, 5, 7, 11, 14 and 17 years [37] (). Our study sample included singleton births with a gestational age at birth ≥37 weeks. Multiple birth and length of gestation can influence both breastfeeding patterns [38], and cognitive outcomes [39-41] and thus could introduce substantial confounding, which was controlled by restriction. Additionally, it is possible that the effect of BF duration on cognitive outcomes might be different among children who were twins/multiples and those born premature [16,42]. Those participants for whom the mother was not the main survey respondent and those whose mothers did not speak English were also excluded. Given that this is a longitudinal study in which the outcome was measured on multiple occasions, children were only excluded if they were missing all data for the outcome measures. Further exclusions were made based on missing data for the exposure, and covariates, and were specific for the analysis of each cognitive outcome.

Ethical approval

The initial MCS protocol was approved by National Health Services (NHS) Research Ethics Committee (REC) of the South West. Further sweeps of data collection have also been approved by the NHS REC system. Data were pseudonymised to prevent participant identification and were subsequently made available at the UK Data Service platform, from where we accessed them (the data were anonymous to us). No formal ethical approval was required for this secondary analysis.

Data collection

All data collection was carried out during home interviews conducted by trained study personnel using computer-assisted personal interviewing [37,43]. Except for the outcomes—which were cognitive tests led by the interviewer or self-completed by the child (as will be described in the following section)—, data for all variables was provided by the mother as the main interviewee (and the partner in the case of their education and social class) [43].

Outcomes

The outcomes of interest were variables evaluating two domains of cognitive development: verbal abilities and spatial awareness.

Verbal cognitive development

Verbal ability was evaluated at ages 5, 7, 11 and 14 years. It was measured using the British Ability Scales Second Edition (BAS II), at ages 5 (BAS Naming Vocabulary, evaluating expressive verbal ability and vocabulary knowledge), 7 (BAS Word Reading, evaluating knowledge of reading), and 11 (BAS Verbal Similarities, evaluating verbal reasoning and knowledge) [43]. These tasks were led by the interviewer. At age 14, a “Word activity” self-completion instrument (evaluating the understanding of the meaning of words) [44] was used (. These instruments have appropriate construct validity, high test-retest reliability, and concurrent validity with other vocabulary tests [45,46].

Spatial cognitive development

Spatial ability was assessed at ages 5, 7 and 11 years [43]. The BAS Pattern Construction sub-test was administered at ages 5 and 7 and the Cambridge Neuropsychological Test Automated Battery (CANTAB] was administered at age 11 (). The BAS test (interviewer-led) evaluates spatial problem solving skills [45] and the CANTAB is a self-completion tool that evaluates the ability to memorise spatial information [46], considering two aspects: i) ‘Strategy’ (how systematic was the child while executing the test) and ii) the number of ‘Errors’ incurred during the test.

Standardisation

To allow meaningful comparison of effect sizes across different time points, continuous scores were standardised (mean = 0; standard deviation = 1) by sex and age at measurement [three-month intervals], considering the distribution in all infants that were included in the analysis.

Exposure

Breastfeeding duration was evaluated by maternal responses to the questions: “Did you ever try to breastfeed your baby?” and “How old was your baby when s/he last had breast milk?” at ages 9 months and 5 years of age (for those breastfeeding for longer than 9 months, approximately 10% of included participants). The latter captured the duration of any BF which was grouped as: Never BF; <2 months; ≥2 and <4 months; ≥4 and <6 months; ≥6 and <12 months; ≥12 months. Duration of exclusive breastfeeding (EBF) was defined as the time in which the child was fed with breast milk only, and was generated using a combination of the duration of BF and the age in which formula, cow’s milk, other types of milk and solid foods were first introduced. The final variable was classified as: Never BF; <2 months; ≥2 and <4 months, ≥4 months. Maternal report has shown to be a valid and reliable measure of breastfeeding duration [47,48], even up to age 6 years [48].

Potential confounders

After a literature review, a Directed Acyclic Graph (DAG) [49] was created in order to systematically represent the potential confounding factors that would need to be considered in the analysis [1,2] (). i) SEP: The markers of SEP used were social class (highest occupation between both parents: Managerial/Professional, Intermediate and Semi-routine/Routine; according to the National Statistics Socioeconomic Classification NS-SEC [50]) and maternal education [according to the National Vocational Qualification NVQ standards: Higher—NVQ 4 and 5 (University degree), Medium—NVQ3 (GCE A-level, national qualifications typically attained at age 18), Lower—NVQ 1 and 2 (national qualifications typically attained at the end of compulsory Secondary Education, age 16), Other/overseas qualifications, and “No formal education”. While social class and education are related, these two indicators capture different aspects of socioeconomic circumstances and adjusting for both reduces the risk of residual confounding [9]. ii) Other potential confounders (sociodemographic characteristics and variables related to pregnancy, childcare and health): Firstly, a basic set of confounders that included gestational age at birth (in weeks), maternal ethnicity (White vs Black, Asian and Minority Ethnic) and languages spoken in household (English only vs English and other language) was selected. Other potential confounders included maternal partnership status (married, cohabiting, single mother), mother working outside the home (yes, no), birth by Caesarean section (yes, no), having older siblings in the household (yes, no), maternal age (in years), and alcohol and tobacco use during pregnancy (yes, no). All these variables were measured at 9 months of age. iii) Maternal cognitive ability: MCS evaluated maternal verbal ability as a proxy for maternal intelligence, when the children were 14 years old. Mothers completed a “Word activity” questionnaire (similar to the Vocabulary test used for participants at age 14) [44]. This questionnaire was adapted from a standardised vocabulary test developed by the Applied Psychology Unit of the University of Edinburgh, and has been previously used in the 1970 British Birth Cohort [51,52]. These scores were standardised (mean = 0; standard deviation = 1).

Statistical analysis

The association between BF duration and categorical variables was evaluated using the chi-square test and its association with continuous variables was evaluated using the F-test. Given that cognitive development was assessed across two different domains (spatial, verbal), these were analysed separately. Crude and adjusted coefficients and 95% confidence intervals (95% CI) were calculated using unbalanced mixed-effects generalised linear models of the Gaussian family and identity link function, assuming an unstructured covariance matrix. The repeated measures of the outcome were clustered within each child. As the outcomes have been standardised, the coefficients are expressed in standard deviations (SD). The different durations of breastfeeding were compared to ‘never breastfed’ as the reference category [15]. Interactions between breastfeeding duration and the age at the time of cognitive assessment were fitted in all models and retained if they were statistically significant at the 5% level. Adjustment for different sets of confounders was completed sequentially. Model 1 adjusted for a basic set of confounders selected a priori (maternal age, ethnicity and language spoken at home). In Model 2, SEP markers were added (regardless of their association with the outcome). Model 3 incorporated a set of potential confounders into the model (from the previously described pool of variables). The process to select which potential confounders were included into the final models was guided by statistical criteria. Each potential confounder was added separately to Model 2 and remained if it showed an association with the outcome (p<0.10), after adjusting for other variables in the model. Adjustment for maternal cognitive scores was conducted as a final step (Model 4). All analyses were conducted in Stata 15.0 for Windows [53] and accounted for the sampling design and attrition at each sweep using the Stata ‘svy’ modules [54] and study weights [55]. E-values were calculated to evaluate the potential for residual confounding to explain the observed associations in the final, fully adjusted models [56].

Sensitivity analysis

All analyses were repeated using exclusive breastfeeding (EBF) duration (instead of any BF) as the exposure. Additionally, these associations might behave differently among children of White English-speaking mothers (a more homogeneous group that constitutes the majority of the UK population), so analyses restricted to these participants were conducted. Previous research has also explored similar research questions on the White British subpopulation of the MCS [16].

Results

Characteristics of the participants

A total of 7,855 and 7,582 participants were included in the analyses of verbal and spatial cognitive outcomes, respectively (). Half (49.8%) of the children were female; 25.9% of children had mothers in the highest education group. The majority of the mothers were White (90.2%) and lived in households in which only English was spoken (93.3%) ().

Flowchart of study participants included in the analysis, UK Millennium Cohort Study.

*Those not present at age 14 were excluded because data for maternal cognitive scores were assessed during that evaluation. **Data missing for the analysis of verbal scores. §mean (SD) for numerical variables. *chi2 / F test. These estimates consider the complex sampling design. The frequencies are unweighted counts, and the percentages are weighted using design and non-response weights.

Breastfeeding duration

Approximately, 33.9% of participants were never breastfed and 23.0% were breastfed for six months or longer. Almost all covariates explored were strongly associated with breastfeeding duration (). In particular, the longer the BF duration, the higher the probability of having a more educated mother or a higher parental social class. Children who were breastfed for longer were more likely to have non-smoking, married, and older mothers. They were also less likely to be born to White and English-speaking mothers. Mean maternal cognitive scores were higher among women who breastfed their babies for longer.

Verbal scores

Both maternal education and parental social class showed marked graded associations with verbal scores across the four sweeps. There were positive correlations between all verbal scores and the maternal cognitive score ().

Association between breastfeeding duration (any BF) and verbal scores

In the multivariable linear models, there was an interaction between breastfeeding duration and the age at which the verbal cognitive scores were evaluated (p<0.001). Therefore, the models retained the interaction term and coefficients for the outcome at each age are presented. The crude analysis at age 5 indicated that longer BF durations were associated with higher average BAS Vocabulary scores (). Children who were breastfed for ≥12 months had an average score 0.39 SD (95%CI: 0.30 to 0.47) higher than those never breastfed. After controlling for maternal age, ethnicity and the language spoken in the household (Model 1), the regression coefficients showed a slight increase. Adjustment for SEP markers (Model 2) reduced the effect sizes by approximately half (0.15 SD; 95%CI: 0.08 to 0.23, ≥12 months vs never breastfed). Further adjustment for the remaining potential confounders (Model 3) did not markedly change the coefficients. Further adjustment for maternal cognitive scores (Model 4) attenuated almost all coefficients to values that were not significantly different from zero (0.03 SD; 95%CI: -0.04 to 0.10, ≥12 months vs never breastfed).

Association between breastfeeding duration (any breastfeeding) and standardised cognitive verbal scores (mean: 0; SD: 1)between ages 5 and 14, UK Millennium Cohort Study (n = 7,855).

All categories of BF duration are compared to “Never breastfed” as the reference category. Model 1: Adjusted for gestational age at birth, maternal ethnicity and languages spoken in household. Model 2: Adjusted for Model 1 + Socioeconomic position (maternal education and highest social class in household). Model 3: Adjusted for Model 2 + other confounding factors (older siblings in household, maternal age, mother working outside the home, partnership status, and maternal smoking during pregnancy). Model 4: Adjusted for Model 4 + Maternal cognitive score. At age 7, adjustment for SEP and the rest of the confounders had a similar effect as at age 5. However, controlling for maternal cognitive scores did not fully explain the association of interest. In the fully-adjusted model, longer breastfeeding durations were associated with higher cognitive verbal scores in children (0.19 SD; 95%CI: 0.11 to 0.27, ≥12 months vs never breastfed). At ages 11 and 14, children breastfed for ≥6 months scored higher in verbal cognitive scores in comparison to children never breastfed, even after controlling for all confounders, including maternal cognitive scores. However, at age 11, the coefficients were smaller (0.08 SD; 95%CI: 0.00 to 0.16, ≥12 months vs never breastfed) than at age 14 (0.26 SD; 95%CI: 0.18 to 0.34, ≥12 months vs never breastfed) ().

Spatial scores

Both maternal education and parental social class showed graded positive associations with spatial scores in the three sweeps. All the spatial tests administered were positively correlated with maternal cognitive scores ().

Association between breastfeeding duration (any BF) and spatial scores

The interaction between the exposure and the age at which spatial scores were assessed (p<0.001) was retained and age-specific coefficients are presented. The crude analysis shows that longer BF durations were associated with higher average BAS Pattern Construction scores at age 5, although there was less of a gradient compared with verbal scores (). Children who were breastfed for 12 months or more had an average score 0.21 SD (95%CI: 0.13 to 0.29) higher than those never breastfed. After controlling for maternal age, ethnicity and the language spoken in the household, the regression coefficients again showed a slight increase. Adjustment for SEP markers considerably reduced the effect sizes and further adjustment for the remaining set of potential confounders only explained a small fraction of the associations. Adjustment for maternal cognitive scores considerably attenuated the estimates towards the null value. After full adjustment, there was not a clear gradient between BF durations and spatial scores. Those children who BF for 4 to <6 months had the highest average score, 0.11 SD (95%CI: 0.03 to 0.19) higher than those who were never breastfed. Meanwhile, the average score of those who were breastfed for ≥12 months was not different from the score of those never breastfed (-0.01 SD; 95%CI: -0.09 to 0.07).

Association between breastfeeding duration (any breastfeeding) and standardised cognitive spatial scores (mean: 0; SD: 1) between ages 5 and 11, UK Millennium Cohort Study (n = 7,582).

All categories of BF duration are compared to “Never breastfed” as the reference category. Model 1: Adjusted for gestational age at birth, maternal ethnicity and languages spoken in household. Model 2: Adjusted for Model 1 + Socioeconomic position (maternal education and highest social class in household). Model 3: Adjusted for Model 2 + other confounding factors (older siblings in household, mother working outside the home, partnership status, maternal alcohol use during pregnancy and smoking during pregnancy). Model 4: Adjusted for Model 3 + Maternal cognitive score. Results for ages 7 and 11 showed that longer BF durations were associated with higher average spatial cognitive scores, even after full adjustment (). Those breastfeeding for 4 to <6 also had the highest mean scores in comparison to those never breastfed. At age 11, the association was only present in the “Errors” Dimension of the CANTAB (as opposed to the “Strategy” Dimension, in which the coefficients were not statistically different from the null value).

Relative effect of key confounding factors and potential for residual confounding

To put the coefficients for the association between breastfeeding duration and the child’s cognitive scores into context, the coefficients for other variables that are associated with cognitive scores in children, such as maternal education (high vs low) or parental social class (high vs low) were approximately 0.20 in the fully adjusted models of our analysis. Meanwhile, a one-SD increase in the standardised maternal cognitive score was associated with a 0.21 SD increase in the child’s cognitive verbal score ().

Comparison of the coefficients for breastfeeding duration, markers of socioeconomic position and maternal cognitive scores on verbal and spatial cognitive outcomes* at ages 14 and 11, respectively.

The models adjusted for all potential confounders of the association between breastfeeding duration and cognitive scores, and include an interaction between breastfeeding duration and age of outcome measurement. *Standardised maternal cognitive scores (Mean: 0; SD: 1). To account for the observed associations found in this study (for example, a coefficient of 0.26 for breastfeeding ≥12 months for verbal scores at age 14, or 0.10 for 4 to <6 months for spatial scores at age 11), any given unmeasured confounder should have a coefficient of at least 0.68 or 0.39 (which corresponds to E-values of 1.85 or 1.43 on the risk ratio scale), respectively, with both exposure and outcome (). The analyses using EBF duration showed a similar pattern to those of any BF duration, as shown in (verbal scores) and (spatial scores). The analyses carried out on the restricted sample of children of white, English-speaking mothers yielded similar conclusions ( to ).

Discussion

Summary of key findings

This study assessed how much of the association between breastfeeding duration and cognitive development is due to confounding by SEP and maternal cognitive scores, based on the data of a nationally representative UK cohort study. The unadjusted associations showed that longer BF durations were associated with higher verbal and spatial cognitive scores up to ages 14 and 11, respectively. Adjustment for SEP explained approximately half of the initially observed associations. Further adjustment for maternal cognitive measures failed to completely remove the remaining associations at ages 7, 11 and 14. The fully-adjusted coefficients where there is evidence of an effect of breastfeeding on verbal cognitive scores varied between 0.08 (age 7; <2 months vs never breastfed) to 0.26 SD (age 14; ≥12 months vs never breastfed). For spatial scores, the coefficients varied between 0.08 (age 7; <2 months vs never breastfed) to 0.19 SD (ages 7 and 11; 4 to <6 months vs never breastfed). This suggests that while the association in this population is not completely due to confounding, the effect of breastfeeding on cognitive development is modest in this population.

The confounding effect of SEP and maternal cognitive scores

There are biologically plausible mechanisms through which breastfeeding could improve cognitive outcomes, such as the provision of myelination-inducing and neurodevelopment-enhancing long chain polyunsaturated fatty acids (PUFAs) and micronutrients (such as iron, folate, zinc, choline, among others). Human milk also contains microRNAs (miRNAs) that may be involved in epigenetic processes that promote the development of the brain and its functions [14,19,29,57]. Other potential mechanisms include the reduction of the risk of school absenteeism through protection from infectious diseases and the maternal attachment secondary to contact during breastfeeding [2,14,29]. However, it has also been argued that the observed associations are due to confounding [1-3]. SEP has been extensively described as one of the main confounders of this association [1-4]. Women from a higher SEP tend to have higher “health literacy”, be more receptive to health education, have better maternity benefits/working conditions, and stronger social networks [8,9], all of which could influence their decisions/ability to breastfeed. In this study, adjustment for SEP explained approximately half of the observed effect. This is consistent with systematic reviews on the topic which have found that adjustment for SEP, on average, reduces the effect sizes by a similar magnitude [1-3]. Despite this reduction, adjustment for SEP did not completely remove the observed associations in our study. While most studies that only control for SEP (and not maternal intelligence) also show positive associations with cognitive development in childhood [13-19], some others, such as two large cohort studies among Irish [20] and British [21] children found no association after adjusting for SEP and other sociodemographic confounders. This study found that adjustment for maternal cognitive measures explained a considerable proportion of the remaining associations, but did not remove them at ages 7, 11 and 14. As with SEP, higher maternal intelligence could favour better uptake of health information and consequently, increase the probability/duration of BF [1-3,23]. Prior evidence on its confounding effect is more heterogeneous than SEP. Several studies have found that adjustment for maternal measures of intelligence (in addition to SEP) completely removes the initially observed associations [20,21,24]. On the other hand, other studies found persistently significant positive associations after adjustment for this variable [26-31], albeit with very small effect sizes [28,29]. A 2015 meta-analysis pooled the estimates of the studies that controlled for maternal intelligence and found a positive association [1], however it has been criticised for overestimating the effect size due to sub-optimally addressed publication bias [58]. It is important to note that most of the previous reports have traditionally dichotomised breastfeeding duration as yes/no or with a temporal cut-off, which may hide important information. Conversely, our study uses several categories of duration, which helps to explore this relationship in a more nuanced way.

The association between breastfeeding duration and cognitive scores

Longer breastfeeding durations were associated with mean cognitive scores that were 0.08 and 0.26 SD higher than the mean cognitive score of those never breastfed. The meta-analysis by Horta et al. reported that, among those studies that controlled for maternal IQ, the pooled effect size was 2.62 IQ points (95% CI: 1.25, 3.98) [1]. Given that the IQ scale is expressed as a mean of 100 with an SD of 15, the coefficients in our study are similar to these combined estimates. While coefficients of 0.08 to 0.26 correspond to modest increases in the cognitive scores, the strength of this association should not be overlooked, as it may be comparable to the strength of the association for other recognised predictors of cognitive development, such as SEP and maternal cognitive ability in these models. We also considered the possibility that the remaining associations were explained by residual confounding produced by unmeasured confounders, such as paternal measures of cognitive ability or broader measures of maternal cognitive ability. This was assessed through the calculation of the E-values [56]. In order to explain the aforementioned associations, any unmeasured confounder should be associated with both BF duration and cognitive scores with coefficients of at least 0.39 (to fully explain a coefficient of 0.10) or 0.68 (to fully explain a coefficient of 0.26). Therefore, while there is room for the associations to be further explained, it is unlikely that all the observed associations could be explained in full by additional adjustment. Some findings deserve further investigation. The association at age 14 seems to be stronger than at other ages. The outcome was measured with a different instrument at age 14, which may contribute to the observed differences. However, these results seem to be in line with those of Kanazawa, who showed that the effect of BF on intelligence increased as children got older in the 1958 British Birth cohort [31]. On the other hand, the follow-up evaluation at age 16 among the sample from the PROBIT experimental study found that the cognitive benefits initially seen at younger ages largely disappeared, except for a modest effect in verbal scores [59]. Additionally, the association with spatial outcomes did not follow a gradient in our study. Sajjad et al. found a similar pattern in a Dutch cohort that evaluated the association between breastfeeding duration and non-verbal intelligence [60]. However, these findings could be due to chance or (less likely) to residual confounding and should be revisited in future studies. It is important to mention that, like several previous studies [1-3], we compare those who breastfed for a given period of time, versus those who were not breastfed (or were breastfed for a shorter duration in some studies). This, however, may not be comparable across studies and could partly explain the heterogeneity in the findings. Not breastfeeding/breastfeeding for shorter periods of time would entail using either formula, cow’s milk or other types of liquids and foods; and these would largely depend on the setting. Moreover, within formula-users, the composition of the formula will probably vary according to the geographical and temporal context [61]. Therefore, studies in different countries and time frames will likely differ in terms of what it means to “not be breastfed”, which should be considered when interpreting the results.

Limitations and strengths

There are some limitations that should be considered when interpreting the findings of the present study. Maternal cognitive tests in the MCS evaluate their understanding of the meaning of several words, which could be affected by further education and might not necessarily reflect broader intelligence. However, similar measures seem to be correlated with verbal ability and have been used as a proxy for intelligence in previous studies [1,2]. Previous studies have produced conflicting results irrespective of whether they have adjusted for verbal or global measures of maternal intelligence, [23-31]. Also, our results indicate that this variable explains the association more than maternal education alone, which suggests that this variable is capturing more than just educational attainment. Ideally, future studies should assess maternal intelligence at baseline using validated multi-dimensional intelligence tests. The proportion of participants who were excluded from the analysis due to missing data for covariates (item non-response) was approximately 15%. The variable that accounted for the highest proportion of missing data was the maternal cognitive measure (approximately 8%). However, non-participation (unit non-response) resulted in the exclusion of approximately half of the original cohort. This attrition was corrected by using non-response survey weights (adjusting for socio-demographic and clinical factors associated with non-response at baseline and subsequent surveys), thus minimising the effect of selection bias due to loss to follow up [55], which is a recommended approach when dealing with this frequent scenario in longitudinal studies [62]. This study also has important strengths. Using in a nationally-representative longitudinal study with a large sample size, we evaluated cognitive outcomes up to age 14 among the same participants thus allowing a comparison of the effects through childhood and early adolescence. Additionally, our study also evaluated two different aspects of cognition, and the confounding effect was apparent in both. Moreover, the findings using the duration of ‘any BF‘ as the exposure of interest were confirmed in the analysis of exclusive breastfeeding (EBF). Lastly, adjustment for SEP included both social class and education, leaving little room for residual confounding by social circumstances [9].

Conclusions

In conclusion, the positive associations between breastfeeding duration and cognitive development up to age 14 among children from the MCS were not explained in full after adjusting for SEP and maternal cognitive scores. While the size of the fully adjusted coefficients for breastfeeding duration was modest, they were comparable to the coefficients for SEP markers and maternal cognitive scores. This suggests that the role of breastfeeding on the child’s cognitive scores should not be underestimated. While a small increase in cognitive outcomes may not be clinically meaningful at the individual level, it has the potential to be influential at the population level. All future studies should ensure proper control of both socioeconomic factors and maternal intelligence. Breastfeeding should continue to be encouraged, as any improvements in child’s cognitive are only one aspect of the benefits it provides.

Association between exclusive breastfeeding duration* and standardised cognitive verbal scores between ages 5 and 14, UK Millennium Cohort Study (n = 7,754).

(PDF) Click here for additional data file.

Association between exclusive breastfeeding duration* and standardised cognitive spatial scores between ages 5 and 11, UK Millennium Cohort Study (n = 7,068).

(PDF) Click here for additional data file.

Verbal cognitive scores according to the characteristics of the study subjects, UK Millennium Cohort Study (n = 7,855).

(DOCX) Click here for additional data file.

Spatial cognitive scores according to the characteristics of the study subjects, UK Millennium Cohort Study (n = 7,582).

(DOCX) Click here for additional data file.

Calculation of E-values: Association between breastfeeding duration (any breastfeeding) and cognitive development, UK Millennium Cohort Study.

(DOCX) Click here for additional data file.

Association between breastfeeding duration (any breastfeeding) and standardised cognitive verbal scores (mean: 0; SD: 1) between ages 5 and 14 among children of white English-speaking mothers, UK Millennium Cohort Study (n = 6,834).

(DOCX) Click here for additional data file.

Association between breastfeeding duration (any breastfeeding) and standardised cognitive spatial scores (mean: 0; SD: 1) between ages 5 and 11 among children of white English-speaking mothers, UK Millennium Cohort Study (n = 6,608).

(DOCX) Click here for additional data file.

Association between exclusive breastfeeding duration* and standardised cognitive verbal scores (mean: 0; SD: 1) between ages 5 and 14 among children of white English-speaking mothers, UK Millennium Cohort Study (n = 6,752).

(DOCX) Click here for additional data file.

Association between exclusive breastfeeding duration* and standardised cognitive spatial scores (mean: 0; SD: 1) between ages 5 and 11 among children of white English-speaking mothers, UK Millennium Cohort Study (n = 6,530).

(DOCX) Click here for additional data file. 15 Mar 2022
PONE-D-22-03469
To what extent does confounding explain the association between breastfeeding duration and cognitive development up to age 14?  Findings from the UK Millennium Cohort Study
PLOS ONE Dear Dr. Pereyra-Elías, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. ============================== The authors present a longitudinal analysis of a large cohort of children, examining the influence of confounding on the relationship between breast-feeding variables and cognitive outcomes. The manuscript is clear and well-written and the questions addressed represent an important contribution to the field. As noted in the reviews, please could you provide further clarity on the methodologies of data collection as well as a discussion of the potential bias inherent in the tests of maternal intelligence used; and the difficulties of representing the breast-feeding variable. Other points of clarity have been raised in the reviewers’ comments. ============================== Please submit your revised manuscript by Apr 25 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:
If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols. We look forward to receiving your revised manuscript. Kind regards, Emma K. Kalk Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf. 2. In your ethics statement in the manuscript and in the online submission form, please provide additional information about the data used in your retrospective study. Specifically, please ensure that you have discussed whether all data were fully anonymized before you accessed them. 3. Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. [Note: HTML markup is below. Please do not edit.] Reviewers' comments: Reviewer's Responses to Questions Comments to the Author 1. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer #1: Yes Reviewer #2: Yes Reviewer #3: Yes ********** 2. Has the statistical analysis been performed appropriately and rigorously? Reviewer #1: Yes Reviewer #2: Yes Reviewer #3: Yes ********** 3. Have the authors made all data underlying the findings in their manuscript fully available? The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer #1: Yes Reviewer #2: Yes Reviewer #3: Yes ********** 4. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer #1: Yes Reviewer #2: Yes Reviewer #3: Yes ********** 5. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer #1: This study examined the relationship between breastfeeding duration and cognitive development until children were 14 years old using a longitudinal data of a UK cohort study. Verbal and spatial abilities of the children were repeatedly assessed at 5, 7, 11 and 14 years old. The results indicate that the associations were still positive at 14 years old even after controlling for full confounding factors measured. Although the manuscript includes a number of valuable data on the beneficial effects of breastfeeding, there were several points that should be clarified. [Introduction] (1) L59: (spell error?) ‘ould’ may be ‘could’ or ‘would’. [Methods] -Exposure (2) L111: How the study group collected the data of breastmilk, formula, solid foods, etc. The authors stated ‘Breastfeeding duration was evaluated by maternal responses to the questions: “Did you …”.’. Did they asked the questions when they visited a research centre for child’s assessment or received questionnaires afterwards by mail. The methodologies of the investigation could affect mother’s responses. -Potential confounders (3) In DAG (Fig.2), there were a space between ‘Maternal cognitive score’ and ‘Paternal cognitive score’. I first missed the above. The authors may want to remove the space. –Statistical analysis (4) L156: They used mixed-effects generalised linear models for explaining the repeated measures of cognitive assessments within each child. However, there seem to be no results of overall verbal or spatial ability. Please describe it more clearly. (5) L163: “Models 1 and 2 adjusted for a basic set of confounders (maternal age, ethnicity and language spoken at home) and SEP markers, respectively.” This could mean that model 1 adjusted for a basic set of confounders but not for SEP while model 2 adjusted for SEP but not for the basic set. However, actually, in the legends of Fig.4 and 5, “Model 2 adjusted for Model 1 + SEP”. Please describe them consistently. [Results] (6) In Table 1, ‘§’ may be missed at ‘Gestational age at birth’; ‘Maternal age (years)’; the four items of ‘Spatial cognitive scores’, in the ‘Characteristics’ column. (7) L231-234: The authers may want to write like this, “… and the language spoken in the household (model 1), … Further adjustment for maternal cognitive scores (model 4) attenuated …”. The descriptions are a bit hart to follow as it was. (8) L322: The authors may write like this ‘… yielded similar conclusions (data not shown).’ I cannot find the data. Are there the data elsewhere? [Discussion] (9) L311: ‘varied between 0.10 to 0.25’. I cannot understand what the values indicates. I guess that the ‘0.10’ indicates the association between Breastfeeding of ‘6 to <12 months’ and verbal ability at age 11 in Fig.4?? But I cannot find the ‘0.25’ (Maybe ‘0.26’ in the association between breastfeeding of ‘>=12 months’ and verbal at 14?). Most readers cannot identify the places of the values among lots of values in Fig.4 or 5. The same can be said for L366. (10) L402: Why the authors specifically mention the proportion of missing data only for covariates (n = 662)? Maternal cognitive scores were missed more (n=899) and could yield a risk of selection bias, too. (11) L374: The authors stated “We also considered the possibility that the remaining associations were explained by residual confounding produced by unmeasured confounders, such as paternal measures of cognitive ability or broader measures of maternal cognitive ability. In addition to them, child’s factors should be considered. If an infant at potential risk of developmental delay has less preference to breastfeeding, a superficial association can be produced between breastfeeding duration and better cognitive abilities that were assessed subsequently. Reviewer #2: The study finds that SEP approximately halved the effects of breastfeeding on cognitive development and that adjustment for maternal cognitive scores further diluted the effects, but did remove all associations. Although similar results have been observed in other studies, I do find that the quality of the maternal cognitive assessment is an important issue. The specific test applied required the mothers to explain the meaning of several words, and as the authors point out this specific test may no reflect broader intelligence. However, there is not only a validity problem, but also a reliability problem. Thus, it is possible that if a more comprehensive and more reliability test of maternal intelligence had been used, the effects of breastfeeding might have been non-significant after adjustment for maternal cognitive ability. The authors are aware of this problem, but I think an expanded discussion of the issue would improve the paper. Reviewer #3: Thank you for the opportunity to review this excellent manuscript by Pereyra-Elias and colleagues. The authors examined the associations between breastfeeding and children’s cognitive development in a large longitudinal cohort study which is nationally representative of the UK. This research question, i.e. does breastfeeding improve intelligence?, is old of near a century, but the field keeps improving with stronger and stronger research design to address it. To fill current gaps, the authors examined more closely i) the extent of confounding in the relationship, in particular by maternal intelligence, which is relatively rarely measured in large studies; and ii) potential long lasting effect up to the adolescence period using repeated cognitive outcomes. The authors found that even after adjusting for socioeconomic factors, and then maternal intelligence, remains an association between longer breastfeeding duration and children’s cognitive development at almost all ages and outcome measures from 5 to 14 y. The paper is also clearly structured and well written. My main comments are as follows: 1) Methods/Study population: lines 73-77 break the flow of the Study population section. It may read better in introduction or discussion or even in the statistical analysis section (when explaining the adjustment procedure). 2) Methods/Exposure: Although the authors cite relevant literature showing that mothers recall well the duration of breastfeeding of their child even at 5 or 6 years postpartum, this retrospective assessment of breastfeeding duration remains one limitation of the present study. On the same matter, the authors argue in the Discussion section that “most of the previous reports have traditionally dichotomised breastfeeding duration as yes/no or with a temporal cut-off [….]. Conversely, our study uses several categories of duration, which helps to explore this relationship in a more nuanced way”. While this is true overall, there are a couple of studies which used breastfeeding duration in an even more nuanced way, by using it as a continuous variable. See for example Tozzi et al. Dev Med Child Neurol 2012, Bernard et al. J Pediatr 2013, Bernard et al. J Pediatr 2017, and Belfort et al. JAMA Pediatr 20213. On top of the retrospective assessment, categorizing the breastfeeding variable is another limitation since it removes information. 3) Methods/Statistical analysis: The selection process of the confounders described on lines 167-170 is pretty contradictory with an a priori approach using DAGs, especially that the authors use a large dataset that does not suffer from a lack of statistical power. I would support forcing variables which are described as confounders in the literature, even when not associated with the outcome in the employed dataset. 4) Methods/Ethical approval: Unless it is among PLoS One requirements, this section would read better right after the Study population section. 5) Results/Breastfeeding duration: using Table 1, I found that 27.7% of participants were never breastfed and 25.7% were breastfed for six months or longer. Not 33.9 and 23.0, respectively. Please explain or correct discrepancies. 6) Results: I am not clear with how the authors consider the interactions between breastfeeding duration and the age at which the cognitive scores were evaluated. In my view, the interaction term is necessary, by default, in all repeated-outcome models. If it is significant, it means that the effect of breastfeeding on the outcome varies by age. If not significant, it means that the effect is quite the same across ages. I do not understand why one would remove the interaction term if not significant. 7) Discussion: The authors seem to sit on the fence regarding the relevance of the effect size which is observed (0.10 to 0.26 SD). They sometimes write it is modest, they sometimes write it should not be underestimated. In my view, it is modest at the clinical level, but it is huge at the population level. I made quick calculations using the presented data (based on the BF prevalence shown in Table 1 and the effect sizes from Model 4 at age 14 y in Fig 4). I transformed SD units into IQ points, because I find it more striking for public health messages. Were all children breastfed 12 months or longer, the average gain (provided the effect is causal) for the overall population would be around 3.2 IQ points. Let’s imagine you and I gain 3.2 IQ points starting tomorrow, it would change very little to our cognitive performance and everyday life. Let’s imagine the whole UK population gains 3.2 IQ points tomorrow, the benefit for the country is hard to believe. 8) Discussion: the authors are quite short on the potential underlying mechanisms of a causal effect of breastfeeding on cognitive development. They are quite a few studies on PUFAs. More development would be welcome. They could also discuss the fact that breastfeeding is compared to non-breastfeeding, i.e., infant formula, whose nutritional composition has largely changed over the last decades. In some way, we usually compare something to another thing that we wrongly consider stable and comparable across study settings. This could explain why findings vary across studies. No further comments ********** 6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy. Reviewer #1: No Reviewer #2: No Reviewer #3: Yes: Jonathan Y. Bernard [NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.
5 Apr 2022 Dear editor, We would like to thank you and the reviewers for the valuable comments provided. We believe these suggestions have helped us to improve the quality of our manuscript. Please, find below our responses to each of the comments (in Italics): 1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf. - We have adapted the manuscript formatting. 2. In your ethics statement in the manuscript and in the online submission form, please provide additional information about the data used in your retrospective study. Specifically, please ensure that you have discussed whether all data were fully anonymized before you accessed them. - Thank you for this comment. We have clarified that data were pseudonymised (before we accessed it) to prevent participant identification and subsequently uploaded to the UK Data Service. The data we accessed were anonymous to us. This information is in the “Ethical Approval” section of the Methods. 3. Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. - We have reviewed our reference list and we have not found any retracted articles. Reference 8, by Simpson DA, et al. has a correction, and we have mentioned the correction in that reference. We have also included additional references in the revised version of the manuscript. Reviewers' comments: Reviewer #1: This study examined the relationship between breastfeeding duration and cognitive development until children were 14 years old using a longitudinal data of a UK cohort study. Verbal and spatial abilities of the children were repeatedly assessed at 5, 7, 11 and 14 years old. The results indicate that the associations were still positive at 14 years old even after controlling for full confounding factors measured. Although the manuscript includes a number of valuable data on the beneficial effects of breastfeeding, there were several points that should be clarified. [Introduction] (1) L59: (spell error?) ‘ould’ may be ‘could’ or ‘would’. - We thank Reviewer 1 for spotting this mistake. The word is ‘would’. We have corrected it. [Methods] -Exposure (2) L111: How the study group collected the data of breastmilk, formula, solid foods, etc. The authors stated ‘Breastfeeding duration was evaluated by maternal responses to the questions: “Did you …”.’. Did they asked the questions when they visited a research centre for child’s assessment or received questionnaires afterwards by mail. The methodologies of the investigation could affect mother’s responses. - The questions (for all the data) were asked during home interviews conducted by trained personnel using computer-assisted personal interviewing. We have added this information in a new subsection (“Data Collection”) in the Methods section. -Potential confounders (3) In DAG (Fig.2), there were a space between ‘Maternal cognitive score’ and ‘Paternal cognitive score’. I first missed the above. The authors may want to remove the space. - We agree with the reviewer. We have removed the space. –Statistical analysis (4) L156: They used mixed-effects generalised linear models for explaining the repeated measures of cognitive assessments within each child. However, there seem to be no results of overall verbal or spatial ability. Please describe it more clearly. - In Comment 4, Reviewer 1 requests the inclusion of overall results; while Reviewer 3 in Comment 5, argues that interactions should always be included (which would require us to report separately for each group). Given that the positions of the reviewers seem to differ, we have written a common response for these two comments, explaining our decisions. We are not completely sure about what Reviewer 1 means by “overall results”. If they are requesting the mean cognitive scores by breastfeeding duration group and in the overall sample, these can be found in Table 1. If they are requesting the results for all ages combined, we decided not to show the overall results for verbal or spatial abilities because there was evidence of a statistical interaction between breastfeeding duration and age at which the outcome was assessed. Reviewer 3 also asks about the inclusion of interaction terms in the models, whereby the effect of breastfeeding on cognitive scores would differ by the age at which the outcome was assessed. We have included interaction terms between breastfeeding and age of outcome assessment in all of our models and all these interaction terms were statistically significant. This means that there is evidence that the effect of breastfeeding duration on cognitive outcomes differed according to the age at which the outcome was measured. We have kept all the interaction terms in the models, and we have therefore presented parameters for each age separately (5, 7, 11 and 14) and no overall estimates. (5) L163: “Models 1 and 2 adjusted for a basic set of confounders (maternal age, ethnicity and language spoken at home) and SEP markers, respectively.” This could mean that model 1 adjusted for a basic set of confounders but not for SEP while model 2 adjusted for SEP but not for the basic set. However, actually, in the legends of Fig.4 and 5, “Model 2 adjusted for Model 1 + SEP”. Please describe them consistently. - We agree with Reviewer 1. We have rewritten this description. [Results] (6) In Table 1, ‘§’ may be missed at ‘Gestational age at birth’; ‘Maternal age (years)’; the four items of ‘Spatial cognitive scores’, in the ‘Characteristics’ column. - We added the symbol to those variables. Thanks. (7) L231-234: The authers may want to write like this, “… and the language spoken in the household (model 1), … Further adjustment for maternal cognitive scores (model 4) attenuated …”. The descriptions are a bit hart to follow as it was. - We have specified the model in the description of these results. (8) L322: The authors may write like this ‘… yielded similar conclusions (data not shown).’ I cannot find the data. Are there the data elsewhere? - We have included these results as supplementary material (S6 to S9). [Discussion] (9) L311: ‘varied between 0.10 to 0.25’. I cannot understand what the values indicates. I guess that the ‘0.10’ indicates the association between Breastfeeding of ‘6 to <12 months’ and verbal ability at age 11 in Fig.4?? But I cannot find the ‘0.25’ (Maybe ‘0.26’ in the association between breastfeeding of ‘>=12 months’ and verbal at 14?). Most readers cannot identify the places of the values among lots of values in Fig.4 or 5. The same can be said for L366. - We thank the reviewer for the comment. We have clarified both instances in the revised text now reads: “The fully-adjusted coefficients where there is evidence of an effect of breastfeeding on verbal cognitive scores varied between 0.08 (age 7; <2 months vs never breastfed) to 0.26 SD (age 14; ≥12 months vs never breastfed). For spatial scores, the coefficients varied between 0.08 (age 7; <2 months vs never breastfed) to 0.19 SD (ages 7 and 11; 4 to <6 months vs never breastfed).”. (10) L402: Why the authors specifically mention the proportion of missing data only for covariates (n = 662)? Maternal cognitive scores were missed more (n=899) and could yield a risk of selection bias, too. - We corrected the statement (second paragraph of Limitations subsection). We have included the proportion of data missing for all covariates and for maternal cognitive scores. We have also adapted the Flowchart (Figure 3) to show the number of participants lost for missing data in a sequential manner. (11) L374: The authors stated “We also considered the possibility that the remaining associations were explained by residual confounding produced by unmeasured confounders, such as paternal measures of cognitive ability or broader measures of maternal cognitive ability.” In addition to them, child’s factors should be considered. If an infant at potential risk of developmental delay has less preference to breastfeeding, a superficial association can be produced between breastfeeding duration and better cognitive abilities that were assessed subsequently. - We agree with Reviewer 1. To avoid introducing confounding by child’s developmental impairment, we have excluded those children who are most likely to have those conditions, i.e. participants born premature. Reviewer #2: The study finds that SEP approximately halved the effects of breastfeeding on cognitive development and that adjustment for maternal cognitive scores further diluted the effects, but did remove all associations. Although similar results have been observed in other studies, I do find that the quality of the maternal cognitive assessment is an important issue. The specific test applied required the mothers to explain the meaning of several words, and as the authors point out this specific test may no reflect broader intelligence. However, there is not only a validity problem, but also a reliability problem. Thus, it is possible that if a more comprehensive and more reliability test of maternal intelligence had been used, the effects of breastfeeding might have been non-significant after adjustment for maternal cognitive ability. The authors are aware of this problem, but I think an expanded discussion of the issue would improve the paper. - We would like to thank Reviewer 2 for the comments. We have included more detail on the test used to measure maternal cognitive ability (Methods, “Potential confounders” subsection). We have also expanded the discussion on how the different maternal cognitive measures could affect the final estimates (first paragraph of the Limitations section). Reviewer #3: Thank you for the opportunity to review this excellent manuscript by Pereyra-Elias and colleagues. The authors examined the associations between breastfeeding and children’s cognitive development in a large longitudinal cohort study which is nationally representative of the UK. This research question, i.e. does breastfeeding improve intelligence?, is old of near a century, but the field keeps improving with stronger and stronger research design to address it. To fill current gaps, the authors examined more closely i) the extent of confounding in the relationship, in particular by maternal intelligence, which is relatively rarely measured in large studies; and ii) potential long lasting effect up to the adolescence period using repeated cognitive outcomes. The authors found that even after adjusting for socioeconomic factors, and then maternal intelligence, remains an association between longer breastfeeding duration and children’s cognitive development at almost all ages and outcome measures from 5 to 14 y. The paper is also clearly structured and well written. My main comments are as follows: 1) Methods/Study population: lines 73-77 break the flow of the Study population section. It may read better in introduction or discussion or even in the statistical analysis section (when explaining the adjustment procedure). - We thank the reviewer for the suggestion. These lines include the rationale behind some selection criteria of our sample. Therefore, we have decided to leave the paragraph as it currently is. The other two reviewers have not pointed this out, but if the Editor feels strongly about this, we could change it. 2) Methods/Exposure: Although the authors cite relevant literature showing that mothers recall well the duration of breastfeeding of their child even at 5 or 6 years postpartum, this retrospective assessment of breastfeeding duration remains one limitation of the present study. On the same matter, the authors argue in the Discussion section that “most of the previous reports have traditionally dichotomised breastfeeding duration as yes/no or with a temporal cut-off [….]. Conversely, our study uses several categories of duration, which helps to explore this relationship in a more nuanced way”. While this is true overall, there are a couple of studies which used breastfeeding duration in an even more nuanced way, by using it as a continuous variable. See for example Tozzi et al. Dev Med Child Neurol 2012, Bernard et al. J Pediatr 2013, Bernard et al. J Pediatr 2017, and Belfort et al. JAMA Pediatr 20213. On top of the retrospective assessment, categorizing the breastfeeding variable is another limitation since it removes information. - We agree with Reviewer 3. The retrospective nature of the exposure could carry information bias, however, we consider that this is minimal, as breastfeeding duration was evaluated at the age of 9 months for the majority of the cohort. It was only asked again at the age of 5 years, for only less than 10% of participants who were breastfed for longer than 9 months. We have added this information in the Methods section. The categorisation of this variable is likely to be less informative than its continuous form. However, we decided to categorise it because, as it can be seen in Figures 4 and 5, not all the associations are linear (this is especially true for spatial scores). We consider that the categorisation of breastfeeding duration facilitates the interpretation of the results for a wider audience. The categorisation is also fine enough to provide information about potential patterns in the association, such as dose-response relationships. We would also like to thank the Reviewer for suggesting the studies from Tozzi et al. and Bernard et al. We have added them to our references. 3) Methods/Statistical analysis: The selection process of the confounders described on lines 167-170 is pretty contradictory with an a priori approach using DAGs, especially that the authors use a large dataset that does not suffer from a lack of statistical power. I would support forcing variables which are described as confounders in the literature, even when not associated with the outcome in the employed dataset. - We understand the concerns of the reviewer. We followed a mixed approach for variable selection, based both on epidemiological and statistical criteria. The inclusion of our main confounders (markers of socioeconomic position and maternal cognitive scores) and other important confounding variables (such as gestational age at birth, ethnicity and language) was decided a priori. While we have a considerably large sample size, we also had a comprehensive list of variables available to us that could behave as potential confounders according to the literature. Therefore, in order to select a more parsimonious final model that would control for the effect of these potentially confounding factors, we decided to use statistical criteria for the selection of these variables. 4) Methods/Ethical approval: Unless it is among PLoS One requirements, this section would read better right after the Study population section. - We agree with Reviewer 3. We have relocated the section “Ethical approval”. 5) Results/Breastfeeding duration: using Table 1, I found that 27.7% of participants were never breastfed and 25.7% were breastfed for six months or longer. Not 33.9 and 23.0, respectively. Please explain or correct discrepancies. - These discrepancies exist because the 27.7% calculated from Table 1 (2,179/7,855) does not consider the complex sample design. Design and non-response weights developed by the MCS team were used to account for the complex sampling strategy and the impact of attrition on the estimates of prevalence and the standard errors (and therefore confidence intervals) for the measures of effect. The correct (weighted) percentages are the “33.9%” and “23.0%” found in the text. This is explained in the Methods and also in the Table 1’s footnotes. 6) Results: I am not clear with how the authors consider the interactions between breastfeeding duration and the age at which the cognitive scores were evaluated. In my view, the interaction term is necessary, by default, in all repeated-outcome models. If it is significant, it means that the effect of breastfeeding on the outcome varies by age. If not significant, it means that the effect is quite the same across ages. I do not understand why one would remove the interaction term if not significant. - In Comment 4, Reviewer 1 requests the inclusion of overall results; while Reviewer 3 in Comment 5, argues that interactions should always be included (which would require us to report separately for each group). Given that the positions of the reviewers seem to differ, we have written a common response for these two comments, explaining our decisions. We are not completely sure about what Reviewer 1 means by “overall results”. If they are requesting the mean cognitive scores by breastfeeding duration group and in the overall sample, these can be found in Table 1. If they are requesting the results for all ages combined, we decided not to show the overall results for verbal or spatial abilities because there was evidence of a statistical interaction between breastfeeding duration and age at which the outcome was assessed. Reviewer 3 also asks about the inclusion of interaction terms in the models, whereby the effect of breastfeeding on cognitive scores would differ by the age at which the outcome was assessed. We have included interaction terms between breastfeeding and age of outcome assessment in all of our models and all these interaction terms were statistically significant. This means that there is evidence that the effect of breastfeeding duration on cognitive outcomes differed according to the age at which the outcome was measured. We have kept all the interaction terms in the models, and we have therefore presented parameters for each age separately (5, 7, 11 and 14) and no overall estimates. 7) Discussion: The authors seem to sit on the fence regarding the relevance of the effect size which is observed (0.10 to 0.26 SD). They sometimes write it is modest, they sometimes write it should not be underestimated. In my view, it is modest at the clinical level, but it is huge at the population level. I made quick calculations using the presented data (based on the BF prevalence shown in Table 1 and the effect sizes from Model 4 at age 14 y in Fig 4). I transformed SD units into IQ points, because I find it more striking for public health messages. Were all children breastfed 12 months or longer, the average gain (provided the effect is causal) for the overall population would be around 3.2 IQ points. Let’s imagine you and I gain 3.2 IQ points starting tomorrow, it would change very little to our cognitive performance and everyday life. Let’s imagine the whole UK population gains 3.2 IQ points tomorrow, the benefit for the country is hard to believe. - We believe this is a fair point. An IQ gain of 3.2 points could have important at the population level. We have added a sentence in the Conclusions section. We also believe that there is reasonable uncertainty regarding the true size of this estimate. As we point out, there might be residual confounding (for example, the effect of paternal intelligence). While we believe that it is unlikely that the effect would be completely explained after further controlling for this residual confounding (as suggested by the E-values), the final effect sizes would probably be smaller than the current estimates (which range from 0.08 to 0.26). In that sense, while we believe that there is an effect, this is most likely modest. Therefore, we preferred to be cautious in our interpretation of the results in the Discussion section and throughout the paper. 8) Discussion: the authors are quite short on the potential underlying mechanisms of a causal effect of breastfeeding on cognitive development. They are quite a few studies on PUFAs. More development would be welcome. They could also discuss the fact that breastfeeding is compared to non-breastfeeding, i.e., infant formula, whose nutritional composition has largely changed over the last decades. In some way, we usually compare something to another thing that we wrongly consider stable and comparable across study settings. This could explain why findings vary across studies. - We thank the reviewer for these comments. We have expanded on potential underlying mechanisms that explain the association between breastfeeding and cognitive development. We also discussed comparability between study settings. Please see the paragraph that precedes the Limitations subsection. Thanks again for the comments. Kind regards, Reneé Pereyra-Elías On behalf of the authors 7 Apr 2022 To what extent does confounding explain the association between breastfeeding duration and cognitive development up to age 14?  Findings from the UK Millennium Cohort Study PONE-D-22-03469R1 Dear Dr. Pereyra-Elías, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org. Kind regards, Emma K. Kalk Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 3 May 2022 PONE-D-22-03469R1 To what extent does confounding explain the association between breastfeeding duration and cognitive development up to age 14?  Findings from the UK Millennium Cohort Study Dear Dr. Pereyra-Elías: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org. If we can help with anything else, please email us at plosone@plos.org. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Emma K. Kalk Academic Editor PLOS ONE
Table 1

Characteristics of the participants according to BF duration, UK Millennium Cohort Study (n = 7,855).

DescriptiveAny BF duration
CharacteristicsWhole sample(n = 7,855)Never BF(n = 2,179)<2 months(n = 2,056)≥2 to <4 mo(n = 819)≥4 to <6 mo(n = 784)≥6 to <12 mo(n = 1,175)≥12 months(n = 842)
n(%)n(%)n(%)n(%)n(%)n(%)n(%)P *
Pregnancy and child-related
Gestational age at birth39.8(1.3)39.8(1.2)39.8(1.4)39.8(1.3)39.9(1.3)39.9(1.4)39.9(1.3)0.163
Female4004(49.8)1158(51.9)989(46.9)421(51.2)383(45.3)601(48.5)452(54.0)<0.001
Age at cognitive test (in months, at age 5)§63.3(3.1)63.3(2.8)63.3(3.2)63.3(3.1)63.3(3.1)63.3(3.2)63.2(3.3)0.845
No older siblings3873(48.7)1175(51.9)842(40.5)359(44.2)372(45.8)646(54.7)479(56.6)<0.001
Smoked during pregnancy
    Never5439(64.3)1201(50.7)1341(60.6)573(65.2)613(75.9)973(81.3)738(86.0)<0.001
    Gave up943(13.2)266(13.1)301(16.1)118(16.6)93(12.1)99(9.3)66(8.9)
    Kept smoking1473(22.5)712(36.2)414(23.3)128(18.2)78(12.0)103(9.4)38(5.1)
Mod/heavy alcohol in pregnancy561(7.3)167(7.5)131(6.9)55(6.3)57(7.1)93(8.5)58(7.5)0.690
Sociodemographic
Maternal age (years)§29.9(6.1)27.7(5.6)29.5(6.1)30.4(5.9)31.7(5.5)32.3(5.5)32.8(5.7)<0.001
Maternal education
    Higher2572(25.9)279(8.6)602(23.9)280(28.9)379(41.6)603(46.4)429(45.5)<0.001
    Medium864(9.5)162(5.6)249(9.9)109(12.4)98(11.7)144(12.1)102(13.8)
    Lower3395(47.5)1236(5.7)969(51.4)359(49.8)249(35.8)349(34.3)233(30.0)
    Other149(1.9)37(1.4)24(1.4)20(2.5)17(2.6)27(2.1)24(3.4)
    None875(15.2)465(27.2)212(13.4)59(6.4)41(6.4)52(5.1)54(7.3)
Highest social class
    Managerial/professional3945(44.2)534(25.8)807(44.8)377(53.4)456(63.7)711(68.7)489(65.3)<0.001
    Intermediate2151(28.0)572(31.5)516(32.8)184(27.2)149(23.6)187(21.4)149(22.8)
    Semi-routine/routine1478(22.8)520(35.2)291(18.9)99(16.8)53(9.6)69(8.0)58(9.3)
    Not applicable281(5.0)91(7.5)47(3.5)15(2.6)12(3.1)17(1.9)18(2.6)

§mean (SD) for numerical variables.

*chi2 / F test.

These estimates consider the complex sampling design.

The frequencies are unweighted counts, and the percentages are weighted using design and non-response weights.

  49 in total

Review 1.  Breastfeeding and intelligence: a systematic review and meta-analysis.

Authors:  Bernardo L Horta; Christian Loret de Mola; Cesar G Victora
Journal:  Acta Paediatr       Date:  2015-12       Impact factor: 2.299

Review 2.  The validity and reliability of maternal recall of breastfeeding practice.

Authors:  Ruowei Li; Kelley S Scanlon; Mary K Serdula
Journal:  Nutr Rev       Date:  2005-04       Impact factor: 7.110

3.  Sensitivity Analysis in Observational Research: Introducing the E-Value.

Authors:  Tyler J VanderWeele; Peng Ding
Journal:  Ann Intern Med       Date:  2017-07-11       Impact factor: 25.391

4.  Association of Gestational Age With Verbal Ability and Spatial Working Memory at Age 11.

Authors:  Alyssa Fitzpatrick; Jennifer Carter; Maria A Quigley
Journal:  Pediatrics       Date:  2016-11-03       Impact factor: 7.124

5.  Gestational age and cognitive ability in early childhood: a population-based cohort study.

Authors:  Gry Poulsen; Dieter Wolke; Jennifer J Kurinczuk; Elaine M Boyle; David Field; Zarko Alfirevic; Maria A Quigley
Journal:  Paediatr Perinat Epidemiol       Date:  2013-05-28       Impact factor: 3.980

6.  Validation Study of Maternal Recall on Breastfeeding Duration 6 Years After Childbirth.

Authors:  Emma Ayorkor Amissah; Vijaya Kancherla; Yi-An Ko; Ruowei Li
Journal:  J Hum Lact       Date:  2017-03-23       Impact factor: 2.665

7.  Association between breastfeeding and intelligence, educational attainment, and income at 30 years of age: a prospective birth cohort study from Brazil.

Authors:  Cesar G Victora; Bernardo Lessa Horta; Christian Loret de Mola; Luciana Quevedo; Ricardo Tavares Pinheiro; Denise P Gigante; Helen Gonçalves; Fernando C Barros
Journal:  Lancet Glob Health       Date:  2015-04       Impact factor: 26.763

8.  Developmental assessments during the first 5 years of life in infants fed breast milk, cow's milk formula, or soy formula.

Authors:  Jayne Bellando; Ginger McCorkle; Beverly Spray; Clark R Sims; Thomas M Badger; Patrick H Casey; Holly Scott; Sarah R Beall; Seth T Sorensen; Aline Andres
Journal:  Food Sci Nutr       Date:  2020-05-13       Impact factor: 2.863

Review 9.  Prognostic Factors for Poor Cognitive Development in Children Born Very Preterm or With Very Low Birth Weight: A Systematic Review.

Authors:  Louise Linsell; Reem Malouf; Joan Morris; Jennifer J Kurinczuk; Neil Marlow
Journal:  JAMA Pediatr       Date:  2015-12       Impact factor: 16.193
View more
  1 in total

1.  Development of Spatial Orientation in Two-to-Three-Year-Old Children in Relation to Lifestyle Factors.

Authors:  Claudia van Dun; Ilaria Lisi; Janna van Diepen; Gabriele Gross; Gabriele Janzen; Esther Aarts
Journal:  Nutrients       Date:  2022-08-13       Impact factor: 6.706
  1 in total

北京卡尤迪生物科技股份有限公司 © 2022-2023.