Across the world, healthcare professionals seem to have reached an overwhelming consensus on the value and importance of breastfeeding. UNICEF, the World Health Organisation and numerous National Health bodies expend vast resources promoting the huge developmental benefits ascribed to breastfeeding. For the baby, these benefits include increased cognitive ability, improved mental health in adolescence and reduced risk of respiratory tract infections, diarrhea, asthma, food allergies, celiac disease, type 1 diabetes, obesity, infections, sudden infant death syndrome, cardiovascular disease and leukemia. For the mother, purported benefits include reduced risk of breast cancer, ovarian cancer, osteoporosis, cardiovascular disease and obesity.
On the basis of these claimed effects, parents can expect to receive regular reminders, often very forcefully worded, of the importance of breastfeeding. Marsha Walker, Executive Director of the National Alliance for Breastfeeding Advocacy, writes that even a single bottle feed before six months will cause irreparable harm to a baby, an assertion that has been parroted by a number of influential writers and bloggers. The argument underlying this rather bold claim relates to the effects of breast and bottle feeding on the bacterial content of the infant’s gut. The adult gut, and indeed the entire adult body, is filled with a delightful array of bacteria, fungi, and archaea (known collectively as the microbiota), without which many of the ordinary biological functions that keep us ticking would be impossible. Some of these microorganisms are “good” (their presence generally contributes positively to human health), whilst other are classified as “bad” (their presence may harm human health).
In contrast, the gut of a neonate is sterile, meaning that it is devoid of the bacteria typically found in the adult gut. As such, early events in the infant’s life, such as the type of birth (Caesarean or vaginal) that they experience, can have a long-lasting influence on the characteristics of their microbiota. The neonate’s gut is also “open”, which means that the small gaps (known as tight junctions) between the enterocytes (a type of epithelial cell) of the intestinal lining are larger than in adults. However, intestinal permeability (the extent to which the tight junctions are “open”) drops rapidly in the first week of life before undergoing a more gradual decline thereafter.
According to some breastfeeding advocates, the high intestinal permeability of the infant gut means that harmful bacteria and/or substances that stimulate the growth thereof can pass through to the bloodstream when the infant consumes formula milk. Breast milk, on the other hand, has been evolutionarily endowed with the perfect mixture of enzymes, antibodies and nutrients to cultivate the most awesome microbiota you could ever imagine. So nothing in breast milk would be expected to cause problems in the context of high intestinal permeability. 
It is claimed that the undesirable bacteria that formula milk allows to grow within the infant’s gut lead to a long-lasting change in the constituency of their microbiota, resulting in a variety of problems that would not arise in the presence of an optimally cultivated (through exclusive breastfeeding) microbiota. Hence, it is concluded that even a single bottle feed will cause lasting harm to an infant.
As you might have guessed, I am a little bit sceptical of the remarkable properties ascribed to breast milk. But before considering the evidence for the above claims, it is important to clarify a few issues that are often neglected or assigned insufficient attention in the breastfeeding literature.
First, and most obviously, the “correlation does not equal causation” principle is extremely pertinent when evaluating claims made about the incredible powers of breast milk. Almost all of the health benefits ascribed to breastfeeding have been inferred on the basis of correlational data – there are very, very few randomised control trials in which breastfeeding status has been experimentally manipulated. This is clearly problematic – correlations between breastfeeding and, for example, child leukaemia do not necessarily imply a causal influence of breastfeeding on leukaemia risk. It may be that factors correlated with breastfeeding, such as maternal education, have a causal impact on leukaemia risk, with no causal role played by breastfeeding itself.
This is particularly important given that there are number of factors that are linked to breastfeeding and that are very likely to have a significant impact on the health of the mother and the child. These include maternal socio-economic status, maternal age, maternal education14 immigrant status single parent status and maternal smoking behaviour. Associations found between breastfeeding and any given outcome could therefore plausibly reflect the causal influence of any of these factors, which can therefore be labelled “confounding variables”.
Researchers sometimes attempt to account the effects of confounding variables by measuring them and statistically controlling for their impact. For instance, if you have a rough of idea the impact of maternal smoking on child IQ, you could assess maternal smoking behaviour and then control for its impact on IQ in a given sample of children. In other words, after measuring each child’s IQ, you would use your knowledge of the statistical association between maternal smoking and IQ to estimate how your sample of children would vary in terms of IQ if their mothers all had equivalent smoking behaviour. If the estimates of each child’s IQ derived thusly still correlated with breastfeeding status, then it could be inferred that the correlation between IQ and breastfeeding in your sample cannot be attributed to the confound between breastfeeding and maternal smoking.
However, even when researchers take care to control for the effects of numerous confounding variables in this way, inference of causality on the basis of correlational data is still problematic for a number of reasons:
- There may be additional confounding factors for which the researchers were unable to control.
- It may be difficult to measure the confounding factors precisely, making it impossible to fully control for their impact on any observed correlations.
- Controlling for a given confounding variable requires estimating the effect of said variable on the outcome variable (e.g. IQ). If this estimate is not perfectly accurate (which it never can be), then the statistical control procedure will not completely account for the confound.
Hence, a major issue in the interpretation of research pertaining to the benefits of breastfeeding is the “correlation does not equal causation” principle: the presence of a correlation between breastfeeding and a particular outcome cannot be used to infer a causal influence of breastfeeding on the outcome in question.
A second issue is the fact that evidence for a plausible mechanism does not constitute evidence for an effect. Permit me to explain this principle with an example. Zhang et al. (2012) cultivated samples of different types of bacteria in breast milk, formula milk and bovine (cow) milk. Simply put, they found that the bacteria in the breast milk grew in a manner that would be more conducive to infant health relative to the bacteria in the other milk solutions. Essentially, the bacteria in the breast milk clustered together to form a biofilm that would serve as a barrier against infection; the bacteria in the other milk solutions failed to do so or did it so a lesser extent.
Some people have taken this finding as evidence that breastfeeding promotes resistance to infection. But herein lies “evidence for a mechanism does not constitute evidence for an effect” issue. The fact that breastmilk is shown to have a particular biological property in the laboratory does not necessarily guarantee that it will have corresponding effects in the body. The human body is extremely complex and reacts to incoming substances in a sophisticated and often unexpected manner. It may compensate for the absence of a particular nutrient by synthesising more of said nutrient by itself; it may break down incoming substances before they reach the region where they would be able to exert any effect; it may be able to initiate certain processes by itself when nutritional inputs fail to do so. There are feedback mechanisms and downstream effects, and the capacity to effect a change in one part of a biological process may be constrained by the factors that appear to be totally unrelated thereto. Moreover, a given substance may have numerous different effects across the body, meaning that isolating a single effect of a substance in a petri dish may not give an accurate understanding of its total impact on human health.
Take the example of anti-oxidants. It is well established that anti-oxidants reduce the activity of free-radicals, which are atoms or molecules that can impair cell-functioning. It follows that dietary supplementation with anti-oxidants should reduce improve cell functioning by removing free-radicals, whereas research shows that this is not in fact the case. Thus, the actual effect of a particular substance may not conform with what would be expected on the basis of the known biological function and laboratory behaviour of the substance.
Therefore, the effects of breast milk on human health cannot be discerned on the basis of studies that examine potential mechanisms for its biological impact. For instance, to determine the effect of breastfeeding on IQ, it is not sufficient to conduct laboratory studies in which the impact of substances contained in breastmilk on biological functions known to be related to IQ are investigated. The only definitive way to address the question is to measure IQ in samples of infants who are randomly assigned to receive more or less breastfeeding. We cannot determine the impact of breastfeeding on susceptibility to infection by examining the behaviour of breast milk in the laboratory. Likewise, we cannot draw conclusions about the impact of breastfeeding on Leukaemia, infection, cardiovascular disease or mental health simply by assessing the influence of breastfeeding on biological processes that are related to these disorders. Breastfeeding status must be manipulated, and the effect of the manipulation on the outcome variable of interest must be observed. Any study that focuses solely on biological markers or laboratory findings that are believed to be important for an outcome variable of interest cannot provide a definitive understanding of the health properties of breast milk.
Having established two pitfalls that need to be avoided when interpreting evidence relating to the effects of breastfeeding, we can now explore the relevant evidence. There is of course extensive correlational evidence linking breastfeeding to all of the aforementioned beneficial medical outcomes, and then some. However, as noted above, the utility of this research is limited given the fundamental “correlation does not necessarily imply causation” problem. Nevertheless, researchers have employed a variety of strategies to circumvent this problem in their attempts to ascertain the causal influence of breastfeeding on the outcome variables that it is commonly purported to affect.
Brion et al., 2011
Brion et al. (2011) examined correlations between receipt of breastfeeding, Body Mass Index (BMI), Blood Pressure (BP) and IQ in samples of participants taken from high and middle income countries. The confounding structures of these two types of countries differ – in other words, the variables that correlate with breastfeeding and that could plausibly give rise to the effects typically attributed thereto are different in high and middle income countries. For instance, Brion et al. (2011) breastfeeding is strongly correlated with socio-economic status in Britain, but not in Brazil. Brion et al. (2011) reasoned that any causal effects of breastfeeding on BMI, BP or IQ would be expected to give rise to consistent correlations between breastfeeding and the outcome in question across countries of various income levels. Conversely, any correlations between breastfeeding and a given outcome variable that were due to confounding factors would only be expected to emerge in the countries in which the relevant confounding structure was present.
Brion et al. (2011) found that IQ was consistently correlated with receipt of breastfeeding across countries of different income levels, but that BP and BMI were only correlated with IQ in high income countries. This is the pattern that would be expected given a causal effect of breastfeeding on IQ and a breastfeeding-BMI and breastfeeding-BP correlation attributable to confounding with socio-economic status. Thus, Brion et al.’s (2011) results indicated that breastfeeding has a causal role in enhancing IQ, but has no effect on BMI or BP.
Caspi et al. (2007)
Caspi et al. (2007) assessed the extent to which the relationship between breastfeeding and IQ varies as a function of a person’s genetic characteristics. They examined a single nucleotide polymorphism (a single DNA unit in the genome of which the molecular identity varies across people) on a gene called FADS2. FASD2 codes for a protein that is involved in the metabolism of fatty acids. Since breast milk contains a number of fatty acids that are absent in formula milk and that are believed to be important for neurocognitive development, Caspi et al. (2007) deemed the possibility of this gene’s modulation of the link between breastfeeding and IQ to be worthy of investigation. Specifically, they examined the IQ difference between breastfed and non-breastfed individuals for three different genetic variants of the SNP of interest on the FASD2 gene. They found that the relationship between breastfeeding and IQ differed depending on the version of the FASD2 SNP that a person had. For most of their sample, breastfeeding was associated with a significantly higher IQ. However, among people with a specific variant of the SNP under investigation, there was no relationship between breastfeeding and IQ.
This finding supports the presence of a causal link between breastfeeding and IQ because it reveals the existence of a statistical interaction that is consistent with a likely plausible mechanism of this effect. If the commonly observed IQ – breastfeeding correlation is simply due to some form of confounding, then there would be no reason to expect the magnitude thereof to vary as a function of genotype. However, if receipt breastfeeding influences IQ via the delivery of fatty acids, then it would make sense that genes involved in the metabolism of fatty acids would modulate the effects of breastfeeding on IQ. Thus, although Caspi et al.’s (2007) result do not definitively confirm the they existence of a causal link between breastfeeding and IQ, they do suggest that such a link is likely to be present.
Randomised Control Trials
Randomised Control Trials represent the Gold Standard for assessing the presence of a causal relationship between two variables. It is therefore surprising that the number of RCTs addressing a practice as highly promoted as breastfeeding is less than a handful (for a comparable example, see the case of flossing). The most highly cited of these was conducted by Kramer et al. (2001), who studied mothers attending 31 maternal hospitals and their polyclinics in the Republic of Belarus. Each site was randomly assigned to receive a breastfeeding promotion or to continue with their existing breastfeeding communication practices. The authors then compared mothers who had been seen at the sites with breastfeeding campaigns to the mothers at the sites wherein there had been no additional breastfeeding promotion. As would be expected, there was clear evidence that the breastfeeding promotion campaign had been successful: mothers attending the sites with breastfeeding promotion campaigns were significantly more likely to breastfeed – and did so for longer – than mothers attending the control sites.
Kramer et al. (2001) found that performance on IQ tests and teachers’ reports of academic performance at 6.5 years of age were significantly higher for children whose mothers had attended the sites with breastfeeding promotion compared to the children of mothers who had attended the control sites. This is the strongest piece of evidence for the beneficial effects of breastfeeding on IQ that anyone has heretofore produced.
Other studies taking measurements from the same participants have found breastfeeding to have a positive effect on attitudes towards food and healthy eating, but no effect was observed on cariometabolic risk factors such as blood pressure and obesity.
However, a separate RCT on a different sample found that preterm infants who were fed banked breast milk developed more favourable cholesterol profiles as adolescents relative to infants who were fed banked formula milk.
There have also been two RCTs in Honduras comparing infants who were exclusively breastfeed for 4 – 6 months with infants who were not exclusively breast fed. These studies did not find any particularly noteworthy differences between the two conditions.
Finally, an RCT conducted by Lindfors and Enocksson (1988) found that exposure to a small amount of formula milk prior to initiation of breastfeeding reduced the risk of subsequent allergy development relative to a control group of exclusively breastfed infants. This clearly contradicts the “even a single bottle feed will permanently harm your baby” claim, although it does not provide a clear indication of the effects of long term exposure to formula milk. Moreover, it should be noted that an RCT conducted by De Jong et al. (2002) found no significant effects in an attempted replication of Lindfors and Enocksson (1998).
Intra-sibling pair comparison
A number of studies have sought to disentangle the effects of breastfeeding from the effects of the numerous variables confounded therewith by examining sibling pairs who were discordant in their receipt of breastfeeding. This involves identifying sibling pairs in which one sibling was breastfed whilst the other was not. Since both siblings have the same parents and family background, any differences between them cannot be attributed thereto. For instance, the IQ – breastfeeding correlation is often suspected of being due to confounding with maternal IQ, since more intelligent mothers are more likely to breast feed. Comparing siblings discordant for breastfeeding eliminates this possibility, because a difference between children of the same mother clearly cannot be due to the differences in maternal characteristics.
Using this model, Metzger and McDade (2010) found that bottle fed members of siblings pairs were at significantly greater risk of obesity in adolescence than their breastfed siblings. However, other studies have found null effects using this design for a range of variables, including asthma, hyperactivity, attachment, compliance, academic achievement and IQ.
Nevertheless, it is important to note that a straight-forward interpretation of sibling comparisons may be problematic. It is possible that mothers worry more about infants that show initial signs of developmental problems (e.g. low birthweight; unstable temperament etc.) and are consequently more motivated to breastfeed such infants in order to compensate for their apparent difficulties. Such a selection effect would mask the beneficial effects of breastfeeding, because it would mean that breastfed infants within sibling pairs would start at a higher level of developmental risk relative to their bottle fed sibling.
It is also possible that mothers who do not breastfeed all of their infants are not representative of the general breastfeeding population. It may be that such mothers are relatively under-motivated to breastfeed relative to other breastfeeding mothers (hence their decision to bottle feed their other child), and they may consequently fail to develop the ability to breastfeed as effectively as more motivated mothers. The effects of effective, “textbook” breastfeeding may therefore be underestimated by examining the offspring of these mothers, who could have been breastfed in a relatively “ineffective” manner. Thus, null effects observed in sibling studies cannot be taken as definitive evidence for the absence of a beneficial causal effect of breastfeeding on child development.
The existing body of literature presents quite a mix message about the effects of breastfeeding, and it is difficult to summarise the overall gist of this research in simple terms. My interpretation is that breastfeeding is very likely to have a small beneficial effect on some developmental outcomes, and is very, very unlikely to have a negative effect relative to bottle feeding.
American Academy of Pediatrics Section on Breastfeeding. (March 2012). “Breastfeeding and the use of human milk”. Pediatrics. 129 (3): 827–841.
Bartels, M., Van Beijsterveldt, C. E. M., & Boomsma, D. I. (2009). Breastfeeding, maternal education and cognitive function: a prospective study in twins. Behavior genetics, 39(6), 616-622.
Brion, M. J. A., Lawlor, D. A., Matijasevich, A., Horta, B., Anselmi, L., Araújo, C. L., … & Smith, G. D. (2011). What are the causal effects of breastfeeding on IQ, obesity and blood pressure? Evidence from comparing high-income with middle-income cohorts. International Journal of Epidemiology, dyr020
Catassi, C., Bonucci, A., Coppa, G. V., Carlucci, A. & Giorgi, P. L. Intestinal permeability changes during the first month: effect of natural versus artificial feeding. J. Pediatr. Gastroenterol. Nutr. 21, 383–386 (1995).
Colen, C. G., & Ramey, D. M. (2014). Is breast truly best? Estimating the effects of breastfeeding on long-term child health and wellbeing in the United States using sibling comparisons. Social Science & Medicine, 109, 55-65
De Jong, M. H., Scharp-Van Der Linden, V. T. M., Aalberse, R., Heymans, H. S. A., & Brunekreef, B. (2002). The effect of brief neonatal exposure to cows’ milk on atopic symptoms up to age 5. Archives of disease in childhood, 86(5), 365-369.
Der, G., Batty, G. D., & Deary, I. J. (2006). Effect of breast feeding on intelligence in children: prospective study, sibling pairs analysis, and meta-analysis. Bmj, 333(7575), 945.
Donath, S., & Amir, L. H. (2000). Rates of breastfeeding in Australia by State and socio‐economic status: Evidence from the 1995 National Health Survey. Journal of Paediatrics and Child Health, 36(2), 164-168.
Duan, X., Wang, J., & Jiang, X. (2016). A meta-analysis of breastfeeding and osteoporotic fracture risk in the females. Osteoporosis International, 1-9.
Dubois, L., & Girard, M. (2003). Social determinants of initiation, duration and exclusivity of breastfeeding at the population level: the results of the Longitudinal Study of Child Development in Quebec (ELDEQ 1998-2002). Canadian Journal of Public Health/Revue Canadienne de Sante’e Publique, 300-305.
Goldacre, B. (2014). Bad pharma: how drug companies mislead doctors and harm patients. Macmillan.
Henderson, J., & Redshaw, M. (2011). Midwifery factors associated with successful breastfeeding. Child: care, health and development, 37(5), 744-753.
Jiménez, E., Marín, M. L., Martín, R., Odriozola, J. M., Olivares, M., Xaus, J., … & Rodríguez, J. M. (2008). Is meconium from healthy newborns actually sterile?. Research in microbiology, 159(3), 187-193.
Kalach, N., Rocchiccioli, F., Boissieu, D., Benhamou, P.-H. & Dupont, C. Intestinal permeability in children: variation with age and reliability in the diagnosis of cow’s milk allergy. Acta Paediatr. 90, 499–504 (2001).
Kramer, M. S., Chalmers, B., Hodnett, E. D., Sevkovskaya, Z., Dzikovich, I., Shapiro, S., … & Shishko, G. (2001). Promotion of Breastfeeding Intervention Trial (PROBIT): a randomized trial in the Republic of Belarus. Jama, 285(4), 413-420.
Kramer, M. S., & Kakuma, R. (2012). Optimal duration of exclusive breastfeeding. The Cochrane Library.
Kwan, M. L., Buffler, P. A., Abrams, B., & Kiley, V. A. (2004). Breastfeeding and the risk of childhood leukemia: a meta-analysis. Public health reports, 119(6), 521.
Lamberti, L. M., Zakarija-Grković, I., Walker, C. L. F., Theodoratou, E., Nair, H., Campbell, H., & Black, R. E. (2013). Breastfeeding for reducing the risk of pneumonia morbidity and mortality in children under two: a systematic literature review and meta-analysis. BMC public health, 13(3), 1.
Lesa, G. M., Palfreyman, M., Hall, D. H., Clandinin, M. T., Rudolph, C., Jorgensen, E. M., & Schiavo, G. (2003). Long chain polyunsaturated fatty acids are required for efficient neurotransmission in C. elegans. Journal of cell science, 116(24), 4965-4975.
Lindfors, A., & Enocksson, E. (1988). Development of atopic disease after early administration of cow milk formula. Allergy, 43(1), 11-16.
Marchesi, J. R. (2010). “Prokaryotic and Eukaryotic Diversity of the Human Gut”. Advances in Applied Microbiology Volume 72. Advances in Applied Microbiology. 72. pp. 43–62.
Marques, T. M., Wall, R., Ross, R. P., Fitzgerald, G. F., Ryan, C. A., & Stanton, C. (2010). Programming infant gut microbiota: influence of dietary and environmental factors. Current opinion in biotechnology, 21(2), 149-156.
Martin, R. M., Patel, R., Kramer, M. S., Vilchuck, K., Bogdanovich, N., Sergeichick, N., … & Gillman, M. W. (2013). Effects of promoting longer term and exclusive breastfeeding on cardiometabolic risk factors at age 11.5 years: a cluster-randomized, controlled trial. Circulation, CIRCULATIONAHA-113.
Metzger, M. W., & McDade, T. W. (2010). Breastfeeding as obesity prevention in the United States: a sibling difference model. American journal of human biology, 22(3), 291-296.
Najdawi, F., & Faouri, M. (1999). Maternal smoking and breastfeeding.
Noone, C., Menzies, I. S., Banatvala, J. E. & Scopes, J. W. Intestinal permeability and lactose hydrolysis in human rotaviral gastroenteritis assessed simultaneously by non-invasive differential sugar permeation. Eur. J. Clin. Invest. 16, 217–225 (1986).
Oddy, W. H., Kendall, G. E., Li, J., Jacoby, P., Robinson, M., de Klerk, N. H., … & Stanley, F. J. (2010). The long-term effects of breastfeeding on child and adolescent mental health: a pregnancy cohort study followed for 14 years. The Journal of pediatrics, 156(4), 568-574.
Quinn, P. J., O’Callaghan, M., Williams, G. M., Najman, J. M., Andersen, M. J., & Bor, W. (2001). The effect of breastfeeding on child development at 5 years: a cohort study. Journal of paediatrics and child health, 37(5), 465-469.
Rittershaus, L., & Halling, T. (2013). “Breast is best”—Infant-feeding, infant mortality and infant welfare in Germany during the late nineteenth and twentieth centuries. Health, 2013.
Singh, G. K., Kogan, M. D., & Dee, D. L. (2007). Nativity/immigrant status, race/ethnicity, and socioeconomic determinants of breastfeeding initiation and duration in the United States, 2003. Pediatrics, 119(Supplement 1), S38-S46.
Singhal, A., Cole, T. J., Fewtrell, M., & Lucas, A. (2004). Breastmilk feeding and lipoprotein profile in adolescents born preterm: follow-up of a prospective randomised study. The Lancet, 363(9421), 1571-1578
Skugarevsky, O., Wade, K. H., Richmond, R. C., Martin, R. M., Tilling, K., Patel, R., … & Gillman, M. W. (2014). Effects of promoting longer-term and exclusive breastfeeding on childhood eating attitudes: a cluster-randomized trial. International journal of epidemiology, dyu072
Vallès, Y., Gosalbes, M. J., de Vries, L. E., Abellán, J. J., & Francino, M. P. (2012). Metagenomics and development of the gut microbiota in infants. Clinical Microbiology and Infection, 18(s4), 21-26.
Walker, M. (2004). “Just One Bottle Won’t Hurt”–or Will It?. Supplementation of the breastfed baby. Accessed online at http://www. health-e-learning. com/articles/JustOneBottle. pdf.
Q Zhang, A., Y Ryan Lee, S., Truneh, M., L Everett, M., & Parker, W. (2012). Human whey promotes sessile bacterial growth, whereas alternative sources of infant nutrition promote planktonic growth. Current Nutrition & Food Science, 8(3), 168-176.
 Rittershaus & Halling, 2013
 Quinn et al., 2001
 Oddy et al., 2010
 American Academy of Pediatrics Section on Breastfeeding, 2012;
NHS choices http://www.nhs.uk/conditions/pregnancy-and-baby/Pages/benefits-breastfeeding.aspx
 NHS choices
 Walker, 2004
 Marchesi, 2010
 Recently, it has been found that the infant gut is not totally sterile – some bacteria transfer from the mother to the foetus in utero (Jimenes et al., 2008). However, the constituency of the neonate microbiota is markedly different from that of an adult, and only reaches an adult-like composition at around 12 months of age (Valles et al., 2012).
 Catassi et al., 1995; Noone et al., 1986; Kalach et al., 2001. The Alpha Parent blog, in referring to the “open gut” concept to justify the superiority of breastfeeding, mistakenly claims that the gut closure occurs at 6 months: http://www.thealphaparent.com/2011/07/virgin-gut-note-for-parents.html
Science of Mom directly addresses error: https://scienceofmom.com/2016/05/03/whats-up-with-the-virgin-gut-do-babies-really-have-an-open-gut-until-6-months-of-age/
 Marques et al., 2010
 Kwan et al., 2004
 Bartels, 2009
 Donath & Amir, 2000
 Dubois & Girard, 2003
 Singh et al., 2007
 Henderson & Redshaw, 2011
 Najdawi & Faouri, 1999
 Goldacre, 2014
 Such studies may certainly be useful, but in isolation they cannot provide a complete picture of the health consequences of breastfeeding on mother and infant health.
 Given that depriving an infant of breast milk may be harmful, ethical constraints demand that such studies are conducted by randomly assigning infants to intervention conditions (wherein mothers are actively encouraged to breastfeed to a greater extent than usual) or control conditions (wherein the researchers make no attempts to sway mothers’ breastfeeding decisions). Studies that have used this strategy are discussed below.
 Duang, Wang and Jiang, 2016; Lamberti et al., 2013
 Lesa et al., 2003
 This design is known as “cluster randomisation” – rather than assigning individual babies to be breast fed or bottle fed, the researchers assigned large groups of mothers to be exposed or to not be exposed to a breastfeeding intervention. This avoided the obvious ethical issues that would be involved in assigning a child to be bottle fed. Note that it would not have been sufficient to run a breastfeeding promotion at one site and compare it to one other control site, because doing so would mean that any observed differences between the sites could be attributable to the features of the individual sites themselves, rather than to the breastfeeding promotion.
 Skugarevsky et al. (2014)
 Martin et al. (2013)
 Singhal et al. (2005)
 Cohen (1994); Dewey (1999)
 For a good summary of their findings, see Kramer and Kakuma (2012).
 Caspi et al. (2007)
 Colen and Ramey (2014)
 Der et al. (2006)
 For anyone who is not aware: breastfeeding is not easy.