We agree that our study had the limitation of selective attrition. We were open about this issue in the paper, describing the cohort attrition in both the text and a full-page table (Table 1), and acknowledging it as a limitation in our discussion. We also drew attention to a recent paper from a similar cohort, the Avon Longitudinal Study of Parents and Children (ALSPAC),3 which found that, although attrition in longitudinal cohort studies is likely to be nonrandom, this attrition does not invalidate the regression models used to predict behavioural disorders. In addition, in the survival analysis conducted, there was no significant relationship between loss to follow-up and maternal alcohol status. The suggestion by Todorow et al.1 that our sample was ‘over-represented by healthy, resilient mothers’ is not supported by the sample characteristics (Table 4).
We did not specifically adjust for maternal psychopathology, and acknowledged this as a limitation in our discussion. We do not have data from pregnancy on maternal clinical depression and anxiety in our cohort, but, as noted in the paper, we used the measure of stressful life events experienced during pregnancy as a proxy variable; although not perfect, this has been found to correlate with psychological distress in other studies.4
The measures used to assess behavioural development, the Child Behaviour Checklist (CBCL)/2–3 and CBCL/4–18, are self-reports, and we agree that, with any self-report measure, there is the potential for reporting bias. Our study followed the offspring from 2 to 14 years, and therefore the Youth Self-Report (YSR) or Teacher’s Report Form (TRF) would have been inappropriate for use at all five follow-up points; TRF cannot be used with pre-school age children, and YSR cannot be used before the age of 11 years. We remain comfortable with our use of CBCL, as the Achenbach System of Empirically Based Assessment (ASEBA) manual provides detailed information on the cross-informant validity of the CBCL, YSR and TRF instruments, and concludes that there is adequate agreement between the measures.5
Our discussion of mechanisms was speculative; we were careful to refer to ‘potential mechanisms’ and we posed these suggestions with what we believed to be appropriate caution. An observational study such as this cannot define cause and effect; to do this would require a randomised controlled trial, which would be impossible and inappropriate in the case of alcohol. In no way do we believe or did we claim that the moderate consumption of alcohol may ‘fix’ maternal anxiety.
Finally, our data were collected around 20 years ago. Another 20 years before, women were being advised in a 1967 guide to pregnancy that ‘Alcohol is not harmful, but it has a high caloric content and should be avoided for this reason’.6 The last 40 years have seen much more research into alcohol and pregnancy, and we have substantially revised our advice to pregnant women. With this paper, we have taken another step in the necessary continuing exploration of data related to threshold levels of alcohol consumption. We disagree that, by publishing our results, we have taken a blasé, permissive and potentially dangerous approach. Sometimes data will not conform to our expectations. This does not mean we should not publish them. To do this would be remiss and would be poor scientific practice, perhaps even unethical.