Heavy alcohol consumption during pregnancy has long been recognised as potentially harmful to the fetus. However, most women either stop drinking altogether once they know they are pregnant or, if they continue, consume only low-to-moderate amounts. Therefore, from a public health perspective, it is important to know the extent to which low-to-moderate consumption may be harmful (or not) to the developing fetus.
In 2007, we published a systematic review in this journal on the effects on the fetus of low-to-moderate alcohol consumption during pregnancy. At that time, we considered that there was no convincing evidence of adverse effects of prenatal alcohol exposure at low-to-moderate levels of consumption. Nevertheless, we cautioned that weaknesses in the evidence precluded the conclusion that drinking at these levels during pregnancy is safe. Since that time, a number of well-conducted analyses using large prospective cohort studies have been published, including the study of Skogerbø et al., which appears in this edition of the journal.
Skogerbø et al. followed up a sample of 1628 mothers and their children participating in the Danish National Birth Cohort Study when the children were aged between 60 and 64 months. Women were asked about their average weekly alcohol consumption and about the number of single occasions on which they consumed five or more drinks. The outcome measures used were scores of the child's behaviour on parent- and teacher-rated versions of a modified Strengths and Difficulties Questionnaire. No effects of either low-to-moderate alcohol consumption or of binge drinking were found. Similar findings on low-to-moderate drinking and child development have emerged from other observational studies (see, for example, Kelly et al., who followed up mothers and their children from the Millennium Cohort Study, and findings from the Western Australian Pregnancy Cohort Study).
Curiously, however, findings from other high-quality observational studies suggest adverse effects of low-to-moderate drinking on child development (for example, Streissguth et al. and Willford et al.), and so it is important to ask why some studies show these effects, whereas others do not. It is of course possible to find methodological shortcomings in both positive and negative studies.[8, 9] To some extent, there is always the potential for bias created by errors in the measurement of exposure and by confounding in cohort studies.
Nevertheless, one particular finding in studies of low-to-moderate drinking during pregnancy seems to be reported fairly consistently. In contrast with abstainers, women who drink low-to-moderate amounts during their pregnancies tend, on average, to be more affluent, better educated, better off financially, have better diets and are less likely to smoke. It is probably no surprise that they are more likely, on average, to have brighter, better behaved children who perform well at school. Thus, confounding caused by these social and lifestyle factors could go a long way to explain why any effects of low-to-moderate prenatal alcohol consumption are negated, or even why, in some studies, beneficial effects of alcohol are seen on child development.
The standard way to deal with this confounding is to use statistical modelling to adjust for the effect. This is, unfortunately, sometimes referred to as ‘controlling’ for confounding, but, in reality, it almost always falls short of this ideal. Operationalising and measuring a complex multidimensional construct, such as social position or lifestyle, is only ever achieved to a limited extent, and therefore one is always left with some residual confounding after the adjustment is made. The only way to avoid confounding is to study a population with little variation in the confounders – quite a tall order when it comes to social and lifestyle factors. Thus, one explanation of the apparent negative findings in some studies would be residual confounding caused by such factors. If such bias is common, as many suspect, and considering the importance of observational studies in this area, one way forward would be to carry out quantitative analyses of the extent to which such biases could affect study findings. Such analyses would be much more informative than the current situation in which most authors only present a qualitative assessment of the effects of bias in the limitations section of their discussion, which is very difficult to evaluate. Bias analysis would be a useful, but no doubt a difficult and time-consuming, exercise.
Is there another way to conduct studies which avoids or minimises confounding? Ideally, one would want to have experimental evidence from randomised trials. However, these are impractical and probably unethical for the study of the effects of prenatal alcohol exposure. A relatively new study design, called ‘Mendelian randomisation’, mimics trial design by making use of the random assignment of genes from parents to offspring that occurs during gamete formation and conception as a means of addressing confounding. Recently, Lewis et al. used this design to demonstrate an effect of moderate prenatal alcohol exposure on childhood IQ at 8 years of age in the Avon Longitudinal Study of Parents and Children (ALSPAC). ALSPAC is a birth cohort study involving around 15 000 women in the Avon area of England who were enrolled during their pregnancy (between April 1991 and December 1992). The continuing health of the women and the development of their children have been studied intensively since enrolment, with both genotypic and phenotypic information available. The rationale for the study of Lewis et al. was as follows. First, individuals are known to vary in the rapidity with which they metabolise alcohol: so-called ‘slow’ and ‘fast’ metabolisers. Second, this variation in metabolism is determined by genetic variation in a number of alcohol-metabolising genes (fetal and maternal). Third, the genetic variation associated with these genes is unrelated to social and lifestyle factors in the mother, i.e. the effect of the genetic variation will not be confounded. Fourth, the degree of harm caused by alcohol is known to be dose dependent: ‘slow’ metabolisers will attain higher peak blood alcohol levels and the fetus will be exposed for longer. Lewis et al. demonstrated that four genetic variants in alcohol-metabolising genes among a sample of 4167 ALSPAC children were strongly related to lower IQ at the age of 8 years. There was a dose–response effect: the more genetic variants the child possessed, the lower the child's IQ. Crucially, this difference was only seen among the children of women who were moderate drinkers during their pregnancy. There was no effect of genetic variation on IQ in the children whose mothers abstained during pregnancy. Furthermore, the variation was not related to social or lifestyle factors. This is strong and unconfounded evidence that intrauterine exposure to moderate maternal alcohol consumption is associated with a difference in child IQ. This study needs to be replicated and Mendelian randomisation itself has limitations. Nevertheless, this provides the strongest evidence of a causal effect to date.
The findings of Lewis et al. should be viewed alongside a growing body of work showing that the genetic variation of both mother and fetus is an important moderator of the fetal effects of alcohol. There will be considerable variation between women (and fetuses) in how they metabolise alcohol, and hence the level of prenatal exposure at which damaging effects may start to occur. Therefore, the determination of a ‘safe’ level which can be uniformly recommended for all pregnant women seems unrealistic. Under these circumstances, a precautionary approach suggesting that women may choose to avoid alcohol during pregnancy could be the best option.
Disclosure of interest
Contribution to authorship
Details of ethics approval