The story on the BBC website says:
Drinking one or two units of alcohol a week during pregnancy does not raise the risk of developmental problems in the child, a study has suggested.
Official advice remains that women abstain completely during pregnancy.
A study of more than 11,000 five-year-olds published in the Journal of Epidemiology and Community Health found no evidence of harm.
There were more behavioural and emotional problems among the children of heavy-drinking women.Mothers were first interviewed when their infant was 9 months of age (sweep 1).
Questions were asked about mothers' drinking during pregnancy, other health-related behaviours, socioeconomic circumstances and household composition. Sweeps two and three of the survey took place when cohort members were aged approximately 3 and 5 years. At the age 5 years home visit cognitive assessments were carried out by trained interviewers and questions were asked about the cohort members' social and emotional behaviour, socioeconomic factors and the psychosocial environment of the family.Now, this raises a number of alarm bells for anyone used to thinking about study design issues. For one, sensitive subjects are hard to measure accurately, especially by recall interviews. Sexual behavior, alcohol consumption, and so on are notorious for that problem. Doctors, for example, routinely double the number their patients give them when they're asked how many drinks they consume per day. And alcohol consumption during pregnancy is particularly sensitive. So whether or not the data are reliable in this study, especially given that it was recall data -- mothers were asked to remember how many drinks they had per week during each trimester of their pregnancy when their infant was already 9 months old -- is one possible problem here.
But perhaps that's not the most important issue here. Confounding is an important potential study killer any time, but this study seem particularly fraught. The authors do recognize that confounding variables could be a problem, that many other factors could influence socioemotional behavior, not just alcohol consumption during pregnancy, and they try to control for some of these. But, given how many different things can affect a child's behavior between pregnancy and age 5 (that is, in fact, everything), it's very hard for us to believe, no matter how well possible confounders are controlled for, that it's possible -- or even sensible -- to try to boil down the explanation for behavior differences between 5 year olds to the difference between 3 and 5 drinks a week. Especially given the fragility of the data on alcohol consumption. And the idea has been prevalent that alcohol can pose a fetal danger even so early that the mother doesn't yet realize she's pregnant; so if she quits or cuts back when she does learn, reliance on recall may lose some accuracy.
That said, let's turn to the results. Here's the link (if you can get access) to the table that interests us most, the prevalence of socioemotional problems according to whether the mother never drank, didn't drink during pregnancy, was a light, moderate or heavy drinker. Note that in every problem category -- every category -- the prevalence of the problem under study (conduct problems, hyperactivity, emotional problems, etc.), and the odds ratio, are higher in children of mothers who never drank than in either all children, or all children except for those whose mothers drank heavily during pregnancy. That's in all the models they tested, from controlling for just one variable, age of child, to controlling for many different variables. Not drinking is a stronger risk factor than almost any amount of alcohol.
Interestingly, and curiously, the authors use not-in-pregnancy as their reference, rather than not drinking at all. Clearly they can be missing some sociocultural or other environmental confounders by doing that. Also, the data are reported in terms of 5% significance, with little if any mention of the correction for the huge number of tests they have done. They did find some sorts of trends, which mollifies this concern, but only somewhat. They report this as a confirmation of earlier findings, but that, too, is a bit uncalled for since this is an extension of, and hence not independent from, the earlier phases of the same study (it includes the earlier results, essentially).
When there are huge numbers of covariates, and potential confounders, and clearly countless other factors could, in principle, be unmeasured confounders, one has to be circumspect about this study. Even if the results are correct as reported and interpreted, the net impact of moderate drinking only applies after regressing out the other measured factors. They may, in aggregate, have a much greater effect on childhood behavior risks than the net effect of pregnancy-imbibing, so that the absolute effect of drinking (even if alcohol is the actual cause itself) is going to be small.
If we can believe the data, and that never drinkers really never drank and so on, this certainly makes one wonder about the effects of unmeasured confounding variables, and raises the question of how much of the reported effect due to heavy drinking really is due to alcohol.
And yet again, about publishing obviously inconclusive studies.... Such studies are very costly, and likely to change years hence as envirionmental exposures and confounders and ways of measuring things change. It is not a shock to learn that a touch of a drink now and again is not a particular problem, especially if it relaxes the mother, say. The most important question is not that, but whether very early drinking can lead to birth defects of various sorts. The reason is that the very early embryo is only a few cells with most of its differentiation yet to come, so a damage to any cell can have proliferative effects.
But later, during the period of organogenesis and then mainly growth, fetuses are generally much more robust to small external exposure effects. There are many more problems about pregnancy and child health than the rather trivial effects that this very large study, even if all its results are true, deals with. And the solution to many of the real problems is: moderation, improvement in socioeconomic differences. But those are deeper problems that society doesn't want to deal with. It's safer to look at a blizzard of statistical data and talk, with serious demeanor, about the minor things that might be guessed at and perhaps even changed.