select search filters
briefings
roundups & rapid reactions
Fiona fox's blog
A study published in PLOS Medicine looks at the influence of socioeconomic status and prenatal behaviours on child health.
Professor Kevin McConway, Emeritus Professor of Applied Statistics, The Open University said:
“This is a complicated and large-scale piece of research, involving almost 600,000 statistical analyses of data on 72 child health measures and a considerable number of measures of tobacco smoking, alcohol consumption, caffeine consumption and socio-economic position in each child’s mother and her partner, that could affect risks to child health. In all, data on more than 230,000 children was used, taken from four different long-term observational cohort studies in the UK and Norway.
The research certainly does have some strengths. The authors assembled and analysed a lot of data. They are making the software for their own analyses available for other to check and use, though the actual data are controlled by the organisations responsible for the four cohorts.
It’s fair to say that the research has indicated quite strongly that it’s not enough to look at obvious behavioural aspects of what mothers do during pregnancy and in early life, in investigating risk factors for child health. Measures of what the mother’s partner does, and of both partners’ socio-economic positions, are clearly relevant too. In a way that’s no surprise, but a lot of related previous research has concentrated too much on data about the mother.
The researchers on the new study have produced a software tool called EPoCH Explorer that allows other researchers to dig further into the findings. I haven’t had time to look at that tool in much detail in preparing this comment, but I have seen enough to realise that it could be very helpful.
The heroic scale of the research means that, in communicating its findings, the researchers have had to take a broad-brush approach.
A lot of detail is given in the research paper, and yet more in the several supplemental files that are published alongside, as well as the EPoCH Explorer software. But in the research paper (and the press release), the researchers have picked out some summary features of the findings that they want to give prominence to.
For example, the top line of the press release says, “Parents’ socioeconomic status is more important than prenatal behaviors to a future child’s health”, and fairly similar wording appears in the paper.
I’m not at all sure that this makes any sense. There’s a question of what it means to say that one determinant of child health is ‘more important’ than another. For example, what’s more important might depend on how easy it is to change a potential risk factor, and changing socio-economic position requires different approaches to changing alcohol, tobacco and caffeine consumptions.
To explain more why I’m not keen on this wording will, unfortunately, require a bit of statistical jargon.
The researchers do explain how they chose to measure the importance of possible risk factors. The association between a risk factor (an exposure) and a measure of child health is considered important (and given a dark green dot in Fig 2 in the paper) if an estimate of how big the effect is (known, in the jargon, as Cohen’s D) comes out greater than a threshold of 0.2 and, crucially, if the statistical test for the existence of a non-zero effect meets a different threshold (in the jargon, the P value is less than 0.05).
The trouble with this definition, in my view, is that it doesn’t take into account how much bigger than 0.2 the estimate of the size of the effect of the risk factor on the measure of child health is. So, judging by Fig 2 in the paper, there are indeed a lot of green dots in the columns for low SEP (socio-economic position), and not as many in the smoking columns, particularly for the mother’s partner. But several of the green dots in the mother’s smoking column are a long way from the centre line to the left or to the right, indicating large, strong effects. There aren’t so many of those in the low SEP columns.
[Note that there’s a positioning and labelling mistake in the relevant diagrams in the paper in the version that was circulated under embargo – see note at the end of this comment.]
In a sense, and speaking a bit loosely, quite a lot of the analyses involving low SEP showed statistically viable evidence of what might be a relatively small risk (because an effect measure of just over 0.2 is small). Rather fewer analyses involving smoking showed statistically viable evidence of a risk, but for some of those that did, the risk was considerably larger.
I don’t think this means that socio-economic position is more important than smoking, in terms of risks to child health – at least, it doesn’t really fit in with my understanding of ‘more important’.
And I really don’t think that comparing the percentages of analyses involving socio-economic position that were considered ‘important’, 15%, with similar percentages for analyses involving smoking (6%), as appears in the press release and the paper, is at all helpful. The pattern of available measures of smoking and of socio-economic position is just too different. (There were smoking measures at various different times during pregnancy, but only one overall SEP measure for each partner, for instance.)
A further, even more technical, point is that, other things being equal, the chance that a statistical test will come up with a small P value (and hence come up as a green dot in Fig 2, provided the estimated effect size is more than 0.2) depends on the sample size involved, and that varied in some cases a great deal from one analysis to another. That’s because the four different cohorts didn’t all include every quantity of interest at all, and because the necessary data was missing for, in some cases, rather a lot of the survey participants.
In the analyses pictured in Fig 2, the sample sizes varied between 50 and over 15,000. This range of difference sample sizes could affect what is considered as an ‘important’ effect quite considerably, though I can’t be sure how much this matters in practice without spending far more time digging into the details of the paper.
I’ve got other technical statistical points, but this comment isn’t the place for them.
Now a non-technical matter. The paper and the press release contrast the research findings with work on the Developmental Origins of Health and Disease (DOHaD) hypothesis. According to the press release, this hypothesis proposes that “prenatal and early childhood environmental exposures can affect a child’s health long-term.” The research paper doesn’t use the same wording – its wording makes it clearer that the hypothesis relates also to what happens to the children after they grow up. adults too. The reference it quotes (reference 1) to explain DOHaD, from 2007 by David Barker, the key developer of the hypothesis, explains that it was introduced to try to understand patterns in coronary heart disease (CHD), which typically do not arise until later life.
The paper also quotes other references, some of them by some of the authors of the new paper, that point out (rightly in my view) that too much research on DOHaD has concentrated on the mother during pregnancy, and the new research findings certainly emphasise that one does need to look further.
However, I mention this mainly to point out that this new study can’t be the only driver of future research in this area. That’s because it doesn’t look at adult health outcomes at all, only outcomes in children up to age 11. In turn that’s not surprising, because the children recruited to three of the four cohorts involved were born between 1999 and 2011, and so are not old enough to have reached the kind of age where diseases like CHD become relevant. Even the children in the fourth cohort (ALSPAC) were born, at earliest, in 1991, so have still not really reached the age when such diseases become a big issue (and data from ALSPAC for ages over 11 were not used in this research anyway).
Of course, regardless of what DOHaD actually is about, it’s important to consider potential causes of poor health in children. Think of all the research there has been in recent years on potential causes of childhood obesity, for instance. I’m not saying that this study was pointless. It certainly wasn’t
It would be good, in future, to see research of this sort that looks at outcomes in adults – but to do that on a major scale will be very difficult, given the need to take into account factors in the adults’ lives over a long period of time.
Note about errors relating to the Figures and their legends in the version of the paper that was circulated under embargo.
Though the wordings of the legends for Figs 2 and 3 in the paper, and the text that refers to them, appears to be correct, the actual diagrams have somehow been swapped, so that the diagram above the legend for Fig 2 is actually the diagram for Fig 3, and vice versa.
There is also a mismatch between Fig 4 and its legend. The actual Figure has four panels, A to D, but the legend refers to seven panels, A to G. The legend wording for panel A is correct. The legend wording for Panel B actually refers to Panel C in the Figure itself. The legend wording for Panel C actually refers to Panel B in the Figure itself. The legend wording for Panel G actually refers to Panel D in the Figure itself. The legend wordings for panels D, E and F refer to diagrams that do not appear at all in the actual Figure.
The SMC and I reported these errors to the lead author and the journal, and it’s possible that they might be fixed before publication.”
Professor Sir Andrew Pollard, Director of the Oxford Vaccine Group, University of Oxford, said:
“This rigorously conducted observational study of parental prenatal influences on child health finds known factors, such as smoking, affect child health outcomes but importantly identifies socioeconomic status of the parents as being relatively more important than such behaviours. While cohort studies can be flawed by unseen bias in the data and causality cannot be assigned with confidence, the authors made efforts to check their findings with multiple approaches and report plausible associations that make biological sense and are consistent with current knowledge. Poverty and inequalities in our society drive poor health in our children. As shown in a series of reports on child health from the Academy of Medical Sciences (2024), Institute of Fiscal Studies (2025), The Royal College of Paediatrics and Child Health (2026), addressing services that provide early childhood support for families can transform health in childhood and beyond. A reset in Number 10 provides an opportunity for a renewed focus on reducing inequalities in society to ensure improved childhood outcomes and the future physical, mental and financial health of our nation.”
Professor Asma Khalil, Professor of Obstetrics and Maternal-Fetal Medicine, St George’s Hospital, University of London, said:
“This is an impressive and methodologically sophisticated study that brings together data from more than 230,000 participants across four large birth cohorts and applies several complementary approaches to strengthen causal inference. Rather than relying on a single observational analysis, the authors use triangulation, including parental comparisons, Mendelian randomisation-informed analyses and negative control approaches, which adds confidence to many of their conclusions.”
“However, the findings should be interpreted carefully. This is still an observational study, and although the authors have gone to considerable lengths to address confounding, residual confounding cannot be completely excluded. Importantly, the results should not be interpreted as showing that maternal health behaviours during pregnancy are unimportant. The study continues to find evidence linking maternal smoking with several adverse outcomes, including small-for-gestational-age birth, higher childhood BMI and behavioural outcomes, which is consistent with the wider body of evidence.”
“Perhaps the most important contribution of this work is that it reminds us that improving child health requires both approaches: supporting healthy behaviours during pregnancy and addressing the broader social and economic circumstances in which families live. These should be viewed as complementary, not competing, public health priorities.”
‘Exploring parental prenatal influences on child health: A multicohort study and data visualisation tool’ by Gemma C. Sharp et al. was published in PLOS Medicine at 19:00 UK time on Thursday 23rd July, which is when the embargo will lift.
DOI: doi.org/10.1371/journal.pmed.1005153
Declared interests
Professor Andrew Pollard: “Professor Pollard was a co-author of the Academy of Medical Sciences report on Child Health”
Professor Asma Khalil: “No relevant DOIs”
Professor Kevin McConway: “I have no conflicts of interest to declare.”