Statistics from Altmetric.com
Breast feeding has many important benefits in children and in mothers. These benefits are greatest in infants in developing countries in whom breast feeding, particularly when initiated early and practiced exclusively, protects against mortality and severe infectious morbidity. Studies from developed countries tend to show smaller health benefits than those observed in developing countries. Nevertheless, there is good quality evidence from the UK and other developed countries that breast feeding protects against diarrhoeal disease and respiratory infection in infancy, including hospitalisation for these infections. In addition, preterm infants who do not receive breast milk are at an increased risk of developing necrotising enterocolitis. A recent report, which was based on evidence from UK studies, estimated that a moderate increase in breastfeeding rates in the UK could save over £17 million annually as a result of the reduced costs of treating infants for diarrhoeal disease, respiratory infection, otitis media and necrotising enterocolitis.1
The evidence on the long-term effects of breast feeding is less clear and is the subject of much debate. A systematic review of the effect of breast feeding on five long-term outcomes (blood pressure; diabetes and related indicators; serum cholesterol; overweight and obesity; and intellectual performance) concluded that the evidence was either inconclusive or suggested no effect for most of these long-term outcomes. However, there was strong evidence of a causal effect of breast feeding on IQ although the magnitude of this effect seems to be modest.2 Most of the studies in the review were observational; hence, it is difficult to infer causality. As an example, in studies which observe a higher child IQ in breastfed children compared with formula-fed children, it is difficult to infer whether this is a direct causal effect of breast feeding rather than a result of other factors which are associated with both breast feeding and child cognitive development. Even adjusting for key confounders—such as factors related to socio-economic status, parental education and IQ, and parenting characteristics—does not rule out the possibility of residual confounding.
The gold standard for inferring causality is the randomised controlled trial. Given that it is neither feasible nor ethical to randomise mothers to breast feed or formula feed, most studies of the effects of breast feeding are observational. Such studies tend to use standard statistical models such as multivariable linear or logistic regression to adjust for other factors in the hope that they are allowing for the inherent differences between the breastfed and formula-fed groups. Several other methods for the analysis of observational data have been proposed as providing stronger evidence than simply adding these confounders to regression models. The paper by Sacker et al3 used two of these methods to assess the effect of breast feeding on a long-term outcome, that of intergenerational social mobility. Using data from two historical British cohorts from 1958 and 1970, they observed that breast feeding was associated with individuals having a higher social class at age 33–34 years than their social class at age 10–11 years, and that this was likely to be mediated through improved cognitive development and reduced emotional stress.
The first method used for assessing causality between breast feeding and social mobility was propensity score matching. This is an alternative to adding a large number of confounders to a regression model. First, a standard statistical model is used to estimate the probability that an individual in the study was breast fed based on a large number of their characteristics.3 These conditional probabilities are called propensity scores—the larger the score, the higher the probability that the child was breast fed. If both a breastfed and non-breastfed child have the same propensity score, then they are comparable in all characteristics which predict breast feeding and therefore the distribution of these characteristics should be similar. Matching on the propensity score in the analysis automatically adjusts for all of the characteristics included in the propensity score, thus allowing an undistorted assessment of outcomes such as cognitive development and social mobility. This method is more efficient than adding a large number of confounders to standard regression models, particularly when the number of confounders is large relative to the number of outcome events. However, it only controls for the confounders which were measured and included in the analysis, so residual confounding cannot be ruled out.
The second method used for assessing causality was to compare the results in two different cohorts, in which confounding structures are known to differ, a method which was previously used to compare the effect of breast feeding on long-term outcomes in cohorts from the UK and Brazil.4 In the paper by Sacker et al,3 socio-economic status was a key confounder when assessing the effect of breast feeding on intergenerational social mobility, although breast feeding was much more socially patterned in the 1970 cohort than the 1958 cohort. The fact that the association between breast feeding and social mobility was the same in these two cohorts supports—though does not prove—a causal effect of breast feeding on social mobility.
Some scientists will be sceptical about breast feeding being causally related to intergenerational social mobility, particularly while there is still debate about the causal effect of breast feeding on the proposed mediating outcomes, namely, improved child cognitive development and reduced emotional stress. Nevertheless, it is an interesting hypothesis and should be explored in other studies. These could include assessing whether there is a dose–response effect over the duration of breast feeding or exclusive breast feeding, or including a measure of maternal IQ in the propensity score, or measuring the effect in populations where there is no association between breast feeding and socio-economic status.
How relevant are the data from these historical cohorts today? A comparison of the 1958 and 1970 cohorts with more recent UK data highlights two important trends in breastfeeding rates which merit further comment. First, breastfeeding rates in the UK declined during the 20th century and reached a low at around 1970. For example, breastfeeding initiation rates were 68% in the 1958 cohort and 36% in the 1970 cohort.3 Breastfeeding initiation rates increased thereafter and remained above 60% in the 1980s and 1990s, since when they have continued to increase, from 69% in 2000 to 81% in 2010.5 Second, the social inequalities in breast feeding which were striking in the 1970 cohort3 are still evident today, 40 years later. Data from the last three national UK Infant Feeding Surveys (in 2000, 2005, 2010) show that breastfeeding initiation rates are still highest in women with managerial and professional occupations and lowest in women with routine and manual occupations (figure 1). There is a suggestion that the gap may be closing in the 2010 survey although there is still a large difference between the highest and lowest socio-economic groups (90% compared with 74%). Rates of breast feeding at 6 weeks show the same striking association with women's socio-economic group (figure 1).
What are the implications of these findings? It is already recognised that breast feeding is an important determinant of inequalities in health in the UK and other developed countries. If breast feeding has long-term effects on health and development, then these inequalities will be greater and will continue into adulthood. Even if the long-term benefits of breast feeding for any given individual are small, the impact at the population level would be much larger, with large costs to the health service and society.1 Further research should explore novel methods which aim to obtain unbiased estimates of the effect of breast feeding on lifelong health and well-being.
I am grateful to Claire Carson and Laura Oakley for helpful comments on an earlier version of this manuscript.
Competing interests None.
Provenance and peer review Commissioned; internally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.