select search filters
briefings
roundups & rapid reactions
before the headlines
Fiona fox's blog
A study published in Archives of Disease in Childhood looks at breastfeeding duration and educational achievement in England.
Prof Kevin McConway, Emeritus Professor of Applied Statistics, The Open University, said:
“This is a careful piece of research on about 5,000 children born in 2000-2002. All are from the UK, and in fact the study used only data from those in England to avoid having to compare exam results across the four UK countries with their different examination systems. The children have reached an age where they can be compared on standard school-level exams, and in this case the research used GCSE results. The research used data from two major sources, the Millennium Cohort Study (MCS) and the National Pupil Dataset (which covers only state schools).
“Though the results are certainly interesting, you have to bear in mind the limitations that inevitably arise in research using observational data from major cohort studies like the MCS, valuable though the data are. Also, the press release is right to say that the differences in exam performance are ‘modest’. But in my view, some of the actual numerical results that it gives are misleading large. (I explain this in more detail, with numbers, near the end of my comment.)
“First, the research is observational. So all that can really be found is associations, that is, correlations between different quantities, and it’s not possible to be certain about what’s causing what. The researchers found that, on average, children who had been breastfed for different lengths of time (or not at all) had different levels of performance in their GCSEs, with (roughly speaking) better exam results for longer durations of breastfeeding. This could be because breastfeeding causes the child to be better in school exams at age 16, or it could be because some other factor (or factors) is independently influencing the chance that a child is breastfed and the chance that they will do well in their GCSEs. These so-called potential confounding factors need to be considered in all observational research.
“For instance, as the researchers point out, it’s well-known that in countries like the UK, mothers who have a higher socioeconomic position (i.e. are better off) are more likely to breastfeed their children, and their children are also more likely to do well at school. That doesn’t mean that it’s the breastfeeding that causes the children to do well at school – obviously it could be some other aspect of the fact that their family is relatively well off.
“Naturally, the researchers are perfectly well aware of these possibilities. It’s possible to make statistical adjustments to try to allow for potential confounding factors, and the researchers in this new study did just that, for a range of possible confounders, including the socioeconomic position of the family, the mother’s verbal cognitive ability (measured on quite a short vocabulary test), and many more factors. The associations between breastfeeding duration and exam performance were still there, albeit not so strongly, after these adjustments. (My concern about the percentages in the press release is that the first figures they give are unadjusted, and hence show rather misleadingly large associations.)
“But anyway, doing the adjustments does not entirely get the researchers off the hook of not being able to be sure what causes what. You can’t adjust for factors on which you don’t have data. For instance, the researchers did adjust for the mother’s cognitive ability, but not for the father’s cognitive ability, despite suggesting in their research paper (Figure 1) that the father’s cognitive ability could have independent effects on the family’s socioeconomic position, the duration of breastfeeding, and the child’s exam results. Now maybe, in fact, the father’s cognitive ability could have turned out not to be an important confounding factor – but the researchers couldn’t tell whether it is or not, because they had no data.
“Another issue is that, as in most cohort studies, most of the factors that were adjusted for were measured only when families were first enrolled in the MCS. Some factors wouldn’t have changed over time. But others (like the status of the parents’ relationship, and whether the mother worked outside the home) could well have changed over time, and that might possibly have mattered. Without major cohort studies like the MCS, that follow up people for long periods of time, we couldn’t know much at all about long-term associations, but that doesn’t mean that cohort studies can be problem-free.
“So the results of this new study could have arisen because being breastfed for a longer time does cause a child to do better in their GCSEs 16 years later, or they could have arisen in some other way, and that’s why the researchers are calling for further research. Also, as they rightly point out, the possibility that your child might do better in school exams at age 16 is far from being the only reason why breastfeeding can be important.
“If in fact the association between breastfeeding and exam performance is one of cause and effect, this doesn’t at all clear up the question of how the cause and effect might work. One might assume at first that it has something to do with improved brain development at the start of life, but there are many other possibilities. The researchers mention this in their paper, pointing out the possibility of what’s known as mediation, where the better average exam scores aren’t directly caused by the breastfeeding, but by something else a bit later in the child’s life that is somehow caused by breastfeeding.
“I’ll mention a few points about the press release.
“Firstly, it says that the MCS enrolled nearly 19,000 children across the UK. That’s true, but, I’d say, not directly relevant to this particular new research. That’s for several reasons. First, the new study excluded children from Scotland, Wales and Northern Ireland because of their different school exam systems. That seems fair enough, but that means only 11,700 children were initially considered. But about 6,700 of those, well over half the number originally considered, couldn’t be included, for a range of reasons (lost to follow-up from the MCS, didn’t consent to linkage of the MCS data to the exam results data, plus a whole range of other reasons (twins and other multiple births were excluded, premature births were excluded, mother did not speak English on enrolment, and more). That’s how the researchers ended up with a much smaller sample to analyse, of 4,940 children, as the release does make clear later. The release is correct to quote that the researchers consider their findings to be nationally representative for children in state schools in England, and maybe that is indeed the case, but with so many reasons for exclusion, I’m not sure how the researchers know that for sure.
“Second, though the press release does, rightly, state that the ‘gains in academic achievement’ were ‘modest’, I think one needs to be careful in reading the numbers given in the body of the release. The percentages given for the those who failed their English or Maths GCSEs, or got high passes for those who weren’t breastfed compared to those who were breastfed for 12 months or longer, do appear to show differences between those groups that look a bit big to describe as ‘modest’, I’d say. For example, the fail rate for English GCSE in the group who weren’t ever breastfed, over two-fifths (41.7%) is more than double the fail rate of about a fifth (19.2%) for those breastfed for at least two months. Those figures are indeed in the research paper (in Table 1), but I’d argue they are misleading. That’s because these are the figures before any adjustment for potential confounding factors, and, as I’ve described, those factors (particularly how well-off the family is) are pretty important in this context. Also, those percentages don’t take into account the inevitable statistical margin of uncertainty in these estimates from a sample of children.
“It’s much less misleading, in my view, to use the adjusted results. These are also mentioned in the press release, but again the statistical margin of uncertainty isn’t mentioned. So, in a group of 100 children, like those in this study, who were never breastfed, one might expect about 42 to fail GCSE English. But if you had another group of 100, who had the same characteristics in terms of the factors that were adjusted for, you’d expect about 31 to fail GCSE English – but with a margin of uncertainty that goes from 23 fails to 42 fails. So the difference in the number of fails is most likely to be somewhere not too far from 11 out of 100, but it could plausibly be as big as 19, or it could plausibly be no difference at all. And, I repeat, the study can’t tell us that any difference is caused by the breastfeeding.
“For failing Maths GCSE, the data indicate that there could well be no difference at all in the fail rates, allowing for the potential confounding factors. Again you’d expect about 42 of 100 children who weren’t breastfed to fail Maths GCSE, and the best estimate for the number out of 100 who were breastfed for 12 months or more is also 42, but it could plausibly be anywhere between 31 and 57.
“The press release says that “compared with children never breastfed, children breastfed for at least 12 months were 39% more likely to have a high pass for both exams.” This needs careful reading. It sounds as if the statement is about children who, on an individual basis, got a high pass in both English and Maths. But the researchers didn’t look at numbers of individuals who did well on both exams. It just happens that, coincidentally, the relevant percentages for both English GCSE and Maths GCSE are both about 39%.
“Looking again at 100 children, not breastfed, about 10 would get a high (A* or A) pass in GCSE English. The central estimate for what would happen in a comparable group with at least 12 months breastfeeding is about 13, with a margin of uncertainty from 10 to 18, so again probably not a vast difference, and possibly effectively no difference. For GCSE Mathematics, about 11 of a not-breastfed group of 100 children would get a high pass, and for a comparable group of 100 who were breastfed for at least 12 months, the central estimate would be about 15 high passes, with a margin of uncertainty from 11 to 21. And there are no results from this study on the numbers who got an A* or A in both maths and English.
“I’d definitely say that these adjusted results are indicating ‘modest’ gains (at best), and I’ll repeat myself that, even if there are gains, this study can’t tell us that they are caused by the breastfeeding.
“Third, the press release says that the intelligence of the children’s parents was taken into account. As I mentioned above, that seems not to be true. The research paper says that there was no data on the father’s cognitive ability, and it makes clear that the measurement of the mother’s cognitive ability was limited because it looked only at a verbal aspect of cognitive ability. That’s why the researchers ask for further research that looks more widely at the mother’s cognitive ability and should “adjust for […] maternal general intelligence.”
Prof Andrew Whitelaw, Emeritus Professor of Neonatal Medicine, University of Bristol, said:
“It has been shown repeatedly that, in developed countries, breastfed infants have significantly better scores on a variety of developmental tests and educational outcomes. This finding is confounded by mothers who breastfeed having consistently better education and socio-economic status, both factors which are associated with improved educational outcomes in their children. It is an important question as to whether this is a biological effect of substances in breast milk or whether intelligent mothers tend to have smarter children anyway by a combination of good genes and stimulating home environment.
“A strength of the study is the large number of children studied, 4940, and this included a good spread of socio-economic and ethnic backgrounds.
“The investigators tested the mothers´ cognition when the children were 14 years old, by asking them the meaning of 20 chosen words. This is recognised as a simple method of roughly assessing verbal intelligence but does not measure mathematical ability.
“Educational achievement was based on the grades achieved by the children in the national examinations which nearly all take at school.
“A true biological effect of substances in breast milk might be expected to show a “dose effect” and another strength of the study is that the investigators were able to examine different durations of breastfeeding from zero to over 12 months.
“The investigators were able to adjust the findings to exclude the influence of maternal verbal intelligence but they could not exclude the influence of maternal mathematical intelligence. They were able to adjust for socio-economic status. Their conclusion was that there was a significant but modest benefit of breast milk on educational achievement and that there was a “dose effect”. Since it is not possible to do a randomised trial of breastfeeding, using statistical techniques to eliminate the influence of important factors such as maternal intelligence and socio-economic status is the nearest an observational study can get to “proving cause and effect. Thus, they went a little further than previous studies in identifying a pure biological effect of breastmilk on child development. This is one of many important benefits of breastfeeding on the child and the mother.”
‘Association between breastfeeding duration and educational achievement in England: results from the Millennium Cohort Study’ by Reneé Pereyra-Elías et al. was published in Archives of Disease in Childhood at 23:30 UK Time Monday 5 June 2023.
DOI: 10.1136/archdischild-2022-325148
Declared interests
Prof Andrew Whitelaw: “I have no conflict of interest.”
Prof Kevin McConway: “I am a Trustee of the SMC and a member of its Advisory Committee. My quote above is in my capacity as an independent professional statistician.”