Exclusive Breastfeeding and Cognition, Executive Function, and Behavioural Disorders in Primary School-Aged Children in Rural South Africa: A Cohort Analysis

Background Exclusive breastfeeding (EBF) is associated with early child health; its longer-term benefits for child development remain inconclusive. We examine the associations between EBF, HIV exposure, and other maternal/child factors and the cognitive and emotional-behavioural development of children aged 7–11 y. Methods and Findings The Vertical Transmission Study (VTS) supported EBF in HIV-positive and HIV-negative women; between 2012 and 2014, HIV-negative VTS children (332 HIV exposed, 574 HIV unexposed) were assessed in terms of cognition (Kaufman Assessment Battery for Children Second Edition [KABC-II]), executive function (Developmental Neuropsychological Assessment Second Edition [NEPSY-II]), and emotional-behavioural functioning (parent-reported Child Behaviour Checklist, [CBCL]). We developed population means by combining the VTS sample with 629 same-aged HIV-negative children from the local demographic platform. For each outcome, we split the VTS sample into scores above or at/below each population mean and modelled each outcome using logistic regression analyses, overall and stratified by child sex. There was no demonstrated effect of EBF on overall cognitive functioning. EBF was associated with fewer conduct disorders overall (adjusted odds ratio [aOR] 0.44 [95% CI 0.3–0.7], p ≤ 0.01), and there was weak evidence of better cognition in boys who had been exclusively breastfed for 2–5 mo versus ≤1 mo (Learning subscale aOR 2.07 [95% CI 1.0–4.3], p = 0.05). Other factors associated with better child cognition were higher maternal cognitive ability (aOR 1.43 [95% CI 1.1–1.9], p = 0.02, Sequential; aOR 1.74 [95% CI 1.3–2.4], p < 0.001, Planning subscales) and crèche attendance (aOR 1.96 [95% CI 1.1–3.5], p = 0.02, Sequential subscale). Factors positively associated with executive function were home stimulation (aOR 1.36 [95% CI 1.0–1.8], p = 0.04, Auditory Attention; aOR 1.35 [95% CI 1.0–1.8], p = 0.05, Response Set) and crèche (aOR 1.74 [95% CI 1.0–3.0], p = 0.05, Animal Sorting). Maternal mental health problems and parenting stress were associated with increased emotional-behavioural problems on the total CBCL (aOR 2.44 [95% CI 1.3–4.6], p = 0.01; aOR 7.04 [95% CI 4.2–11.9], p < 0.001, respectively). Maternal HIV status was not associated with any outcomes in the overall cohort. Limitations include the nonrandomised study design and lack of maternal mental health assessment at the child’s birth. Conclusions EBF was associated with fewer than average conduct disorders and weakly associated with improved cognitive development in boys. Efforts to improve stimulation at home, reduce maternal stress, and enable crèche attendance are likely to improve executive function and emotional-behavioural development of children.


Introduction
There is strong evidence that exclusive breastfeeding (EBF) for 6 mo, as recommended by the World Health Organization (WHO) [1], optimises infant nutrition and substantially decreases mortality and morbidity from infectious diseases [2,3]. The relationship between EBF and cognitive development is less clear [4,5], although studies in high-income settings [6][7][8], a randomised trial from Belarus [9], and a recent study from Brazil [10] have shown positive associations. A large systematic review showed conflicting results depending on the study design and rigour, as well as the number of factors adjusted for [4]. The few studies from resource-limited settings were almost twice as likely to find no association. This suggests that confounding variables, including socioeconomic status and maternal cognitive ability, affect the choice to breastfeed and the positive effects found. Measuring the duration of EBF accurately is challenging because of factors related to definition, timing, and duration of recall [11,12], and many of the studies were limited by poor documentation of breastfeeding patterns [13,14]. Further limitations included small sample sizes [15,16] and predominantly Caucasian populations, with only one small study from Africa [16], which found no association with cognitive development but some advantages for child behaviour in breastfed infants. There was no evidence from HIV-prevalent areas where the long-term effect of EBF on child development remains unquantified.
Studies exploring the link between EBF and later development have focused on core cognitive development, sometimes termed the intelligence quotient (IQ). However, higher-order cognitive function, termed executive function, is critical for later development, particularly the ability to function in society [17]. Executive function coordinates and controls information processing, which is important for a child's ability to manage emotions and behaviour, to follow rules, to concentrate, and to form friendships. Thus, executive function influences educational and social success [18]. Executive function is susceptible to environmental influences and therefore an important intervention target [19]. In addition, few breastfeeding studies have examined emotional-behavioural development, an important outcome affected by early life factors, which predicts later educational achievement.
The Vertical Transmission Study (VTS) (2001)(2002)(2003)(2004)(2005)(2006) supported HIV-positive and HIVnegative women to practice EBF in a rural area of South Africa before antiretroviral treatment became available [20], providing the first evidence that EBF reduced the risk of postnatal HIV transmission [21] and was associated with significant benefits for children's health and growth until up to 2 y of age [22,23] (Registration: NCT01948557, National Institute of Health, Clini-calTrials.gov). Here we investigate the association between EBF, HIV exposure, and other early and current life factors and later cognitive development, executive function, and emotionalbehavioural development in VTS children now aged 7 to 11 y. We accounted for maternal cognitive function, home stimulation, crèche attendance, and maternal/caregiver stress and mental health and hypothesized that EBF would result in improved longer-term development in children, despite exposure to HIV and poverty.

Methods
Ethics permission for this study was granted by the Biomedical Research Ethics Committee (BREC), University of KwaZulu-Natal, South Africa (BF184/12). Women were contacted by telephone or a home visit to ask if they would be interested in this study. Those who agreed were then visited by a field worker who explained, and provided written details of, the study and obtained written informed consent from the mother and primary caregiver (if this was not the child's mother).
The VTS, a nonrandomised, prospective, intervention cohort study, was implemented between 2001 and 2006 from the Africa Centre for Population Health, which also hosts a Demographic Surveillance System (DSS) platform [24]. This lay-counsellor, home-based intervention was designed to support mothers to practice EBF for the first 180 d of life [25]. Between 2012 and 2014, we re-enrolled HIV-negative children (aged 7-11 y) born to HIV-positive ("exposed") and HIV-negative ("unexposed") mothers from the VTS cohort; HIV-infected children were not re-enrolled because they have different developmental trajectories [26].
To establish a comparative population mean for the developmental outcomes (in the absence of appropriate normative data for validated cognitive assessments), we assessed 630 (485 unexposed; 145 exposed) same-aged HIV-negative children from the DSS, not included in the VTS, and combined these with the VTS sample. Mothers in the DSS group were exposed to the same antenatal care at local clinics, including messages regarding HIV and infant feeding, but not to the home-based intervention to support EBF. We aimed to rerecruit all available 1,289 VTS children meeting the inclusion criteria, of whom 935 (75%) were enrolled and 906 (70%) fully assessed. To establish a robust population mean, we used the population platform of the Africa Centre for Population Health surveillance to identify all 1,226 resident children who were matched for age and HIV exposure to the VTS children but had not been exposed to the VTS intervention. Of these, 844 children met eligibility criteria, 657 (77%) enrolled, and 630 (75%) completed assessments. This analysis includes the developmental outcomes of the VTS children, for all of whom we have accurate data on infant feeding and HIV exposure; their outcomes are related to the population means. Children were enrolled if both the mother and child were alive, the child was a resident in the research area, the mother and child's current HIV status was known, and, for the DSS children, if the HIV status during pregnancy was known, the mother received antenatal services in the study community, and the maternal-held child Road-to-Health Card was available.

Outcome Measures
Child cognition. To assess the cognitive development, we used the full Kaufman Assessment Battery for Children (KABC-II) [27]. This has four subscales, each with subtests, which measure audio and visual memory and memory span ("Sequential Processing"); spatial and visual perception, reasoning, and maths ability ("Simultaneous Processing"); focused and selective attention and the ability to store auditory and visual stimuli simultaneously ("Learning Ability"); and decision-making ability ("Planning"). There is an additional subtest to assess reasoning and language development ("Riddles"). After discussion with experts, including the authors of the KABC-II, to be more culturally appropriate, we substituted a subtest in the Learning Ability subscale ("Atlantis"/ "Atlantis delayed" replaced "Rebus"/ "Rebus delayed"). The KABC-II has been used in low-middle-income countries and validated in Africa [28] (Cronbach's α = 0.75).
Child executive function. Three subtests from the executive function domain of NEP-SY-II [29] were used: Animal Sorting (inhibition, planning, and cognitive flexibility), Auditory Attention (vigilance and selective/ sustained auditory attention), and Response Set (inhibition of previously learned stimuli).
All subtest raw scores for both NEPSY-II and KABC-II were transformed to scaled scores, according to the child's age, using standardised tables published by the test developers.
Child emotional and behavioural functioning. We used the parent-reported Child Behaviour Checklist (CBCL) [30], which has been validated across multiple cultural settings [31]. This includes 120-items in two subscales-"Internalising disorders" and "Externalising disorders"-and a composite Total score. A high score indicates more problems. Scores were normed using multicultural Rating-to-Score norming software to produce normed t-tests for the Total score, the two subscales, and the six Diagnostic and Statistical Manual (DSM) disorders: affective, anxious, somatic, attention deficit hyperactivity, oppositional, and conduct disorders (Cronbach's α = 0.94).

Maternal Mental Health Measures
All psychometric measures had been previously used in the population; clinical algorithms for depression and anxiety were used.
• Alcohol: the WHO Alcohol Use Disorders Identification Test (AUDIT-6) assessed alcohol use and severity [34].
• Parenting stress: measured using the Parenting Stress Index Short Form (PSI-36), a 36-item scale measuring stress related to the parental role, the parent-child relationship, and the degree to which the parent finds the child difficult [35].
The home environment was assessed using a locally adapted version of the Home Observation for Measurement of the Environment (HOME) inventory [36]. Maternal cognitive ability was assessed using the Standard Raven's Progressive Matrix [37]. (Fig 1). Compared to HIV-unexposed children, exposed children were more likely to have a mother who was older and the main income provider, less likely to have been exclusively breastfed until 6 mo of age and to have attended crèche, and more likely to have a primary caregiver with a current mental health disorder (Table 1). Compared to children included in the current analysis, the 383 VTS children excluded were more likely to have a younger HIV-uninfected mother with more years of education at the time of pregnancy and were less likely to have a low birthweight (Table 1).

Cognitive Outcomes
None of the cognitive development measurements were significantly associated with EBF or maternal HIV status in adjusted analyses (Table 2). In multivariable analyses, the only variable significantly positively associated with performance on all cognitive subscales was maternal cognitive ability (measured using the Standard Raven's Progressive Matrix) (  Footnotes apply to Tables 1-9. Bold numbers indicate a p-value < 0.05 a "Birth" indicates that the variable relates to when the child was born as opposed to the time of follow-up at 7-11 y; "Current" indicates that the variable relates to the time of follow-up in this study when the child was aged between 7-11 y. b Low birthweight was defined as <2.5 kg. c Exclusive breastfeeding (EBF) for this analysis was defined as the number of days when the child only received breastmilk and no other fluids or solids; the total number of days was divided by 30 to categorise the number of months of EBF. d Crèche is a noncompulsory, nongovernmental preschool; children start primary school in South Africa at the age of 7 y. e A provisional depression diagnosis determined by the PHQ-9 diagnostic algorithm required at least one Criteria A (mood and loss of interest) and 2-4 Criteria B (weight, sleep, agitation/retardation, fatigue, guilt, concentration, and suicidality) for more than half the days. A provisional anxiety diagnosis required meeting Criteria A (anxiety) and 3 Criteria B (worry, restlessness, fatigue, concentration, irritability, and sleep) for more than half the days. f Child age and birth order are not shown for the excluded children, some of whom died or were lost to follow-up. Other variables omitted for the excluded children are factors measured since the VTS cohort, including current perception of wealth, attendance at crèche, caregiver mental health, and parenting stress.   Income Provider (at Birth) Other

Executive Function
None of the executive function measures were significantly associated with EBF duration, maternal HIV, or child sex (Table 3). Compared to children whose mothers were aged less than 20 y, those with mothers aged 20-29 y at their birth were almost twice as likely to score above the mean on the Animal Sorting subtest (aOR 1.82 [95% CI 1.2-2.8], p = 0.01), as were children whose mother was the main income provider during their infancy (aOR 1.81 [95% CI 1.0-3.1], p = 0.03) and those who had attended crèche (aOR 1.74 [95% CI 1.0-3.0], p = 0.05).
For the Auditory Attention subtest, compared to children aged 8 y, those aged 9 and 10 y were over three (aOR 3.38 [95% CI 1.6-7.3], p = 0.01) and four times (aOR 4.56 [95% CI 2.1-9.9], p < 0.001) more likely to perform above average, respectively. Children with better stimulation at home (i.e., a HOME score above the median) were more likely to perform above the mean in the Auditory

Emotional and Behavioural Problems
Being born in an urban environment and having a primary caregiver with high parenting stress were associated with more emotional and behavioural problems (higher scores on the Internalising and Externalising subscales and the Total score) (urban  (Table 5), EBF was significantly associated with lower scores (fewer problems) for conduct disorders. Those who were exclusively breastfed for 6 mo compared to 1 mo or less were approximately half as likely to score above the mean for conduct disorders (aOR 0.44 [95% CI 0.3-0.7], p < 0.01). Caregiver mental health and stress were associated with increases in all six disorders. Urban residence was associated with increases in somatic, attention deficit hyperactivity disorder (ADHD), and oppositional problems. Boys were less likely to be anxious (aOR 0.64 [95% CI 0.5-0.9], p < 0.01) but more likely to have somatic (aOR 1. 34

Outcomes Stratified by Sex (Tables 6-9)
Contrary to the finding in the overall cohort, boys, but not girls, who were exclusively breastfed for more than 1 mo were twice as likely as those who were exclusively breastfed for a very short period to score above the mean for Learning Ability (aOR 2.07 [95% CI 1.0-4.3], p = 0.05) and half as likely to score above the mean for Externalising problems (aOR 0.48 [95% CI 0.2-1.0], p = 0.05) ( Tables 7 and 9). However, girls who were exclusively breastfed for less than 1 mo were more likely to score above average on Auditory Attention compared to those who were exclusively breastfed longer. The finding of an association between maternal cognitive ability and improved performance on all four cognitive subscales in the overall cohort held for boys but not for girls. Boys whose mother's cognitive ability was above the median score (Standard Raven's Progressive Matrix) were twice as likely to score above average for the cognitive  In this stratified analysis, the associations between maternal education and cognitive outcomes and HOME scores and executive function held for girls but not for boys. Likewise, girls were approximately three times more likely to score above the median for higher emotional-behavioural problems if their mothers had a mental health problem (Total CBCL aOR 3.28 [95% CI 1. In the full model, interactions between sex and EBF were significant for Conduct (p = 0.02) and CBCL Internalising (p = 0.02) outcomes and marginally significant for Learning (p = 0.07) and Auditory Attention (p = 0.08).
In summary, while duration of EBF was not associated with child cognitive development or executive function in the overall sample (including girls and boys), it was associated with fewer conduct disorders, and, when stratified by sex, there was weak evidence of improved cognitive development in boys.

Discussion
To our knowledge, this is the first study examining EBF, HIV-exposure, and child cognition, executive function, and emotional-behavioural outcomes at primary school age in Africa.
Our finding that duration of EBF was associated with fewer conduct disorders has significant implications. Conduct disorders lead to aggressive or disruptive behaviours that interfere When the association was significant, the OR and aOR are in bold (because of rounding, the confidence intervals sometimes overlap with 1.0). ORs are based on bivariate logistic regression of the outcome on the covariate. AORs are based on multivariate logistic regression on the outcomes including all covariates The CBCL includes 120-items in two subscales: "Internalising disorders" and "Externalising disorders," and a composite Total score. A high score indicates more problems. MC-HOME: Middle Childhood HOME a High/low based on splitting the sample on the median. b High/low based on splitting the sample on the median.    with learning and peer relationships, in turn leading to low self-esteem and further behavioural problems [39]. Conduct disorders in childhood are associated with an increase in violent criminal behaviour and poor long-term mental health and academic achievement in later life [40]. While to our knowledge there have been no formal analyses of the economic costs of conduct disorders in low-middle-income countries, the evidence from carefully conducted studies in high-income countries is that the costs are enormous [41]. For example, a report from the United Kingdom stated that the total cost of crime attributable to people who had a conduct disorder in childhood was estimated to be £60 billion per annum [42]. Given these costs to individuals, families, and society, it is highly relevant that this study has shown EBF to be associated with reduced likelihood of conduct disorders at this critical stage of development. Further, for boys, a longer EBF duration was weakly associated with a doubling of the odds of better cognitive development on the Learning subscale, which assesses ability for maintaining focused attention while coding and storing complex auditory and visual stimuli simultaneously and generating strategies to learn efficiently. Identifying potential strategies to improve the life chances of boys is important, and, in our context, EBF appears to be associated with a longerterm advantage for boys. There is increasing evidence that some groups of children are more  Owns Fridge (at Birth) (Continued) Table 6.      susceptible to the effects of negative rearing and, importantly, that they may benefit more from the effects of positive rearing and interventions. This is known as the "differential susceptibility hypothesis" [42]. Boys may be more susceptible in a number of domains of development to the effects of negative rearing-for example, in language development [43,44]-and therefore, in line with our findings, may benefit more from early EBF. The finding that EBF did not have an effect on children's cognitive development overall is in accord with studies from other resource-limited settings that also found no association [4], including studies in India (n = 514, aged 9-10 y), China (n = 442, aged 3 y), Malaysia (n = 1,394, aged 9 y), Chile (n = 784, aged 5.5 y), and the Philippines (n = 1,984, aged 8.5 and 11.5 y), although these studies did not stratify by sex.
An important strength of this study is the investigation of a wide range of key determinants of child development. In our cohort, maternal cognitive ability was strongly positively associated with children's performance across all cognitive subscales, but not with executive function, although higher maternal cognitive ability was associated with improved performance on the Planning subscale, which assesses ability to focus attention, make decisions, and apply working memory, abilities considered to reflect executive function [27]. Interestingly, while having attended a crèche was associated with improvement in children's cognitive development (sequential scores that depend on practice, on which boys scored particularly low) at the primary school age, it was independently associated with poorer behaviour, with similar associations found for children born in urban and rural settings. This apparent paradox in relation to crèche exposure is consistent with data from high-income countries [45].
Executive function is a high-order cognitive function that coordinates and controls information processing [18]. It is critical in enabling a child to successfully integrate into environments such as school [46]. We found a positive association between executive function and better stimulation at home, older maternal age, and a mother who was the main income provider at the time of the child's birth. These findings are in accord with evidence suggesting that executive function is susceptible to environmental factors [47]. Executive function is increasingly thought of as the core ingredients (creativity, flexibility, self-control, and discipline) that will determine a child's success in life. There is evidence that executive function is negatively affected by stress, emotions, poor physical fitness, and childhood obesity [39]. Encouragingly, a growing body of evidence suggests that simple, low-cost interventions such as noncomputerized games, aerobics, martial arts, and mindfulness may support improvements in executive function [19]. Our finding that poor maternal mental health and high parenting stress, measured at the time of the child assessments, were associated with increased emotional-behavioural problems in children was unsurprising but useful to quantify in an African setting. The link between parental mental health and child behaviour outcomes is well-established, with children of depressed, anxious, or highly stressed mothers known to be at increased risk of psychological and behavioural problems [48,49].
There is a dearth of evidence examining the developmental trajectories of HIV-exposed children, in particular longitudinal studies including HIV-unexposed controls. A recent systematic review [26] examining HIV exposure and child development found only 11 studies (1,591 children aged 0-18 y) and showed that HIV-exposed children are disadvantaged compared to their HIV-unexposed peers, in particular in emotional-behavioural and, to a lesser extent, cognitive development. However, results are not consistent across research settings or age groups, with most of the current evidence being based on small samples with wide When the association was significant, the OR and aOR are in bold (because of rounding, the confidence intervals sometimes overlap with 1.0). AORs are based on multivariate logistic regression on the outcomes including all covariates. The CBCL includes 120-items in two subscales: "Internalising Disorders" and "Externalising Disorders," and a composite Total score. A high score indicates more problems. MC-HOME: Middle Childhood HOME a High/low based on splitting the sample on the median. b High/low based on splitting the sample on the median. heterogeneity in outcome measures [26]. Our results, based on a large sample of children including HIV-unexposed controls suggest that, overall, HIV-exposed children performed as well as HIV-unexposed children in the domains examined and that the findings of other small studies may be overstated. Limitations include the nonrandomised study design of the original intervention, the lack of maternal mental health measurement at the child's birth, and no assessment of cognition or emotional/behavioural problems at earlier time points in the children's lives. Infections of the developing brain and childhood malnutrition also affect later cognitive ability but were not included in the analyses. In addition, caution must be taken in interpreting the sex-stratified models, as examining subgroup effects increases uncertainty and is more likely to produce larger effect estimates. Further, because of the complexities in measuring cognitive and executive function and the need to model each of the outcomes separately, multiple tests were used, and there may be some false positive significant results. However, our results are in line with findings from other studies.
Strengths include a large cohort of HIV-exposed and HIV-unexposed children, population normal values, adjustment for a wide range of confounders including current maternal IQ, and a battery of culturally appropriate developmental assessments including executive function and behavioural outcomes. A unique strength in this study is the accurate documentation of daily EBF data. Both HIV-positive and HIV-negative women had high rates of EBF, with few breast health problems, likely due to the quality of lay breastfeeding counselling [25,50]. Given that all women received intensive support, it is possible that this may have limited our ability to detect differences between EBF and non-EBF children.
In conclusion, while EBF was not significantly associated with cognitive development at the primary-school age, there was an association between EBF and a reduction in conduct disorders and, for boys, weak evidence of positive associations both in aspects of cognitive development and behavioural problems more generally. Given the number of adverse factors in these families' environments, including poverty and high HIV prevalence, the fact that these benefits were evident into the crucial early school years is important. Child outcomes were associated with a range of other key factors. While core cognitive development was principally associated with maternal cognitive abilities, executive function was associated with a number of modifiable environmental factors including home stimulation and crèche attendance. Child's emotional development was largely associated with caregiver mental health. These findings highlight a number of avenues for potential interventions.
Supporting Information S1 STROBE Checklist. Checklist of items that should be included in reports of observational studies. (DOCX) S1 Table. Factors associated with children's cognitive and executive function continuous outcomes measured by the KABC-II and NEPSY-II. (DOCX) Community Liaison Office of the Africa Centre and Community Advisory Board for their guidance throughout the study.