Assessing the merits and limitations of using fathers as a negative control exposure to test DOHaD using a Swedish case-study

摘要

Background: Developmental Origins of Health and Disease Hypothesis (DOHaD) studies are often observational in nature and are therefore prone to biases from loss to follow-up and residual confounding. Register-based studies can reduce these issues since they allow almost complete follow-up and provide information on fathers that can be used in a negative control analysis to assess the impact of unmeasured confounding. The aim was to propose a causal model for testing DOHaD using paternal exposure as a negative control, and to assess its merits and limitations using an application from Swedish register data: fetal exposure to distress and the risk of asthma. Methods: A causal diagram including shared (L1 and U1) and parent-specific (L2, U2, L3, U3) measured (L) and unmeasured (U) confounders for maternal (fetal) and paternal exposures is proposed (see Figure). The case-study consisted of all children born in Sweden from July 2006 to December 2008 (n = 254 150). Information about childhood asthma, parental distress and covariates was obtained from the Swedish National health registers. Associations between maternal and paternal distress during pregnancy and offspring asthma at age 5 years were assessed separately and with mutual adjustment for the other parent's distress measure, as well as for shared confounders. Results: Negative control exposure studies are useful for eliminating shared residual confounding but may continue to be biased by unshared confounding. The beauty of using fathers as a negative control exposure is that in most cases the parent-specific unmeasured confounders are likely to be of a similar type and risk for both parents eg common genetic pathways, disease status. Therefore U2 and U3 could be considered to act like shared confounders. In addition, if there is a robust association between maternal and paternal exposures it is likely that both maternal and paternal causal pathways share all unmeasured confounders U1-U3, therefore the paternal model can act as a sufficient negative control. However, if parent-specific unmeasured confounders are not of similar risk or direction then interpretation is more difficult. Our case-study found that maternal distress during pregnancy was associated with offspring asthma risk; adjusted Odds Ratio (OR) 1.32 (95% CI 1.23, 1.43). The mutually adjusted paternal distress-offspring asthma analysis (OR 1.05 95% CI 0.97, 1.13) indicated no evidence for unmeasured confounding. One important limitation highlighted by the case-study was maternal exposure misclassification bias, as we found that the rate of maternal distress during pregnancy was lower than expected, possibly due to mother's not wishing to take anxiolytic or antidepressant drugs when pregnant. This not only means the effect size will be pushed towards the null but the association between paternal distress on maternal distress will also be subject to misclassification. Conclusions: Using paternal exposure in a negative control model to test the robustness of fetal programming hypotheses can be a relatively simple extension of conventional observational studies. However, although a null finding for the paternal exposure association likely suggests a lack of residual confounding in the maternal exposure association, a positive finding is harder to interpret.