Background There are challenges in studying the effects of partner exposures around pregnancy on child health. We explored potential sources of bias in the effects of parental prenatal health behaviours on child health, to describe and quantify some of these challenges, as well as suggest ways in which they might be mitigated.
Methods First, we characterised the availability of data on partner and mother health behaviours in the prenatal period from three UK cohort studies: the Avon Longitudinal Study of Parents and Children (ALSPAC), Born in Bradford (BiB), and the Millennium Cohort Study (MCS). Second, we assessed the potential for sample selection in these cohorts by comparing characteristics of families where the partner did and did not participate. Third, using parental smoking during pregnancy and child birthweight as an example, we ran simulation studies of several DAGs to explore the extent that missing partner data and selection can affect estimates. We then explored the ‘real life’ impact of partner sample selection on estimates of maternal effect.
Results In all cohorts, data on partner prenatal health behaviours was less detailed and collected less frequently than maternal prenatal health behaviours. Partners participated in ALSPAC and MCS for the majority of pregnancies. Of 14,472 pregnancies in ALSPAC, and 18,241 pregnancies in MCS, 12,997 (78%) and 13,145 (71%) had a cohort participating partner, respectively. However, partner participation was much lower in BiB: 3131/11,538 pregnancies (27%) had a cohort participating partner. Consistently across all cohorts, in pregnancies with cohort participating partners, mothers were more likely to be living with their partner before birth of the child, be white and have a university degree. They were consistently less likely to have no qualifications and their babies were less likely to be born preterm or have a low birthweight. We saw relatively stable effect estimates within cohorts for associations between maternal smoking and offspring birthweight regardless of whether we used the full sample, the sample where fathers participated (selected either through stratification or adjusting for partner smoking), or the sample where fathers did not participate.
Conclusion Overall, these results show partner selection is unlikely to cause strong selection bias in estimates of maternal effects. This suggests that, although inclusion of partner data with high levels of non-random missingness has the potential to introduce selection bias, in practice, the biasing effect appears to be small. This also has implications for studies of partner effects in their own right.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.