The observation that the country of origin seemed to influence the result of the study is intriguing.1 Since it is less probable that the teratogenetic potential of alcohol differs with geographical position at consumption, one explanation may be that the heterogeneity in risk estimates found in well-conducted epidemiological studies arises from uncontrolled confounding. The list of potential confounders is long and no study can control for all. Contextual influences, such as the social acceptance (or the opposite) of alcohol drinking during pregnancy, may account for some of the differences found in risk estimates. Those who drink even moderately in a society with a strong social injunction against alcohol use during pregnancy are simply different and may take other risks as well. The finding of the high risk estimates for alcohol related fetal loss in our study, conducted in a society with a, relatively, liberal attitude towards moderate alcohol drinking in pregnancy, speaks against this source of confounding.
Yet another source of confounding could arise from undetected differences in health-related behaviour as a result of the reproductive experience of the women. Women with a history of fetal loss (or other adverse reproductive outcomes) may have a higher baseline propensity of fetal loss, but they may on the other hand change behaviour in subsequent pregnancies as a result of their consciousness of that. We found no indication of such confounding of the association when we analysed the risk of fetal death in the subgroup of the 31 790 women who had never been pregnant before or when we restricted to primigravid women who became pregnant within 6 months of trying, i.e. low-risk women.