Challenges in using data on fathers/partners to study prenatal exposures and offspring health

Abstract

AbstractIntroductionPaternal exposures (and other non-maternal factors) around pregnancy could have important effects on offspring health. One challenge in research of paternal effects is that study samples with data on partners are usually a subgroup of those with data on mothers, which could introduce selection bias and limit generalisability of the findings. Here, we use maternal and father/partner data on prenatal behaviours to explore the extent to which selection in partner analyses might bias findings.MethodsWe characterise the availability of data on father/partner and mother health behaviours (smoking, alcohol consumption, caffeine consumption and physical activity) 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). We assess the extent of sample selection in these cohorts by comparing the characteristics of families where the father/partner does and does not participate. Using the association of parental smoking during pregnancy and child birthweight as an exemplar, we used simulations to explore the extent to which missing father/partner data may bias estimates.ResultsIn all three cohorts, data on prenatal health behaviours of fathers/partners were less detailed and collected at fewer timepoints than data on prenatal health behaviours of mothers. Partners of mothers who had a lower socioeconomic position were less likely to participate. Estimates of the association between maternal smoking and offspring birthweight were similar in samples with and without participating partners in all three cohorts. In simulations based on ALSPAC data, there was little evidence of selection bias in associations of maternal smoking with birthweight, and although bias was observed for father/partner smoking, its magnitude was relatively small.DiscussionUsing real and simulated data, we show that bias due to selected recruitment of partners into ALSPAC, BiB and MCS had a relatively small impact on estimates of the effects of maternal and partner smoking on offspring birthweight. In other applications, the impact of selection bias will depend on both the analysis model and the selection mechanism. We have shown how to use a simulation study to assess that and recommend that applied researchers working with partner data use simulations and other sensitivity analyses to assess the robustness of their results.