Reports coming out in the media (such as Express, Mail, Mirror) are suggesting that breastfeeding for six months or more can lower the child’s risk of developing leukaemia. But where did these findings come from and how reliable are they?
In this post, we take a look at the research that led to these reports, and suggest that the results are not very robust, the conclusions are overstated and the claims likely to cause unnecessary alarm.
Greater than the sum of its parts?
The new research, published in the peer-reviewed journal JAMA Pediatrics, combines many previous studies that looked at whether children who had been breastfed (and for how long) went on to develop childhood leukaemia. The studies in the new analysis were a mixture of size and quality, so on their own couldn’t lead to any definitive conclusions. But when put together and re-analysed – known as a ‘meta-analysis’ – it is intended to boost the power of the findings.
The new study includes 18 studies that met a certain quality threshold. The researchers, based at University of Haifa in Israel, report that children who had been breastfed for six months or more had up to a 19% lower risk of developing leukaemia than children who had never been breastfed or had been breastfed for under six months.
From this, they conclude this is sufficient evidence of a protective effect to further promote the health benefits of breastfeeding and encourage greater uptake amongst new mothers.
But we are not persuaded these claims truly stack up.
Correlation is not causation
The strength of a meta-analysis is that it tries to make sense of all the best research on the subject. But all of the studies within the new analysis looked at the association between breastfeeding and leukaemia, and therefore cannot tell you about true causes. Other background factors that affect both the likelihood of breastfeeding and leukaemia risk could have been missed or overlooked.
For example, parental affluence may affect the decision to breastfeed. But affluence will also affect a host of behaviours, like attendance at nursery, exposure to infections, decision to vaccinate, time of weaning, and many, many more. We don’t know for sure what factors do influence leukaemia risk – and parental affluence is just given as an example here merely to illustrate the complexity of background factors – but simply linking breastfeeding and leukaemia risk without consideration of other potential influencing factors is far too premature.
Experts who conduct these types of studies are well aware of these issues and always attempt to account for background factors, like socio-economics, lifestyle, gender, ethnicity, and so on, but it's always hard to eliminate them altogether. Nevertheless, we were alarmed when we spoke to our statistical experts who noticed that the meta-analysis unusually relies on crude data that did not appropriately account for background factors.
They suggested that the authors’ relative inexperience with this type of analysis has led to a number of flaws.
The experts we spoke to pointed out a gross error in the data – one dataset, that had been used in two different publications, is included twice. This could distort the statistical robustness or the size of any effect, and could be serious enough to consider a correction or withdrawal of the paper.
There are also a few ways these data could have been skewed to give wrong or exaggerated results. One is a bias in those participating in these studies. Many were based on phoning mothers at home or by a self-administered questionnaire. This introduces a potential participation bias, where the people surveyed and who agreed to take part were not representative of the population as a whole. This may mean certain groups, such as more educated or time-rich parents, were overrepresented in the comparison group (children who didn’t get leukaemia), suggesting a larger proportion of children who did not get leukaemia were also breastfed.
Almost all of the studies asked mothers or parents to remember the duration of breastfeeding, sometimes many years later. This introduces a possible recall bias, where parents may not have accurately remembered what they did or their responses were affected by knowing that their child had had leukaemia. And because childhood acute leukaemia is thankfully relatively rare – only three to four children in every 100,000 are affected each year in the UK – the small numbers could have inflated these biases.
It would be far better to recruit a large random sample of people, collect data in real-time, and then look at whether children went on to develop leukaemia. This is more costly and time-consuming, but it would help diminish some of these potential biases because the particpants and research questions would be defined up-front.
The authors acknowledge these limitations in their discussion section of their paper, which is why we were surprised by the strength and certainty of their claims in the conclusions section.
What would this mean in the real world anyway?
There is some evidence that links proper immune system development to a reduced likelihood of developing leukaemia as a child and certain genetic faults present at birth can raise the risk substantially, but we are still far from truly understanding all the many different factors at play. It is an important area to be able to understand who is most at risk and what factors can alter their likelihood of developing this disease if we are to prevent some of these cases, but we are not convinced that this new analysis provides strong evidence for a significant role of breastfeeding.
We should also stress that because of the relative rarity of childhood leukaemia, even if the authors’ claims were true, it would still affect only a tiny number of children. And it would still only affect children who are already at risk because of key genetic faults that occur in the womb.
There is a lot of pressure on new mothers, some mothers cannot breastfeed and many factors can affect how long new mothers can breastfeed for. Parents of children who develop leukaemia can also feel a lot of guilt, even though we know some children will unfortunately get the disease whether they’ve been breastfed or not.
Stories based on problematic research do not help anyone.
- For some expert opinion, including our Research Director and Prof Eve Roman (an epidemiologist whose work we support), see the Science Media Centre
- Reference: Amitay & Keinan-Boker (2015). Breastfeeding and Childhood Leukemia Incidence A Meta-analysis and Systematic Review. JAMA Pediatr. 2015;169(6):e151025. doi:10.1001/jamapediatrics.2015.1025