Live Birth Bias In Epidemiological Study Of Timing Specific Exposure Effect During Pregnancy And Child Health: A Simulation Study

In reproductive and perinatal epidemiological studies, measurement of child health outcomes that can only be ascertained in live born children may be incomplete since only 60 – 70% of fertilized eggs result in live births and early pregnancy loss is often undetected. Studies assessing outcomes among live born children are subject to live birth bias, a phenomenon previously proposed as a form of collider-bias in which conditioning on live-birth status induces a non-causal association between exposure and outcome. In this study, we expanded a previously proposed common structure of this bias to evaluate its impact on the estimation of time-specific prenatal exposure effects on child health outcome, using causal diagrams. We used Monte Carlo simulation techniques to investigate two scenarios in which prenatal exposures led to pregnancy loss. Our findings confirmed previous simulation findings showing biased estimates of prenatal exposure effects on child outcome risk, assuming a true null association between each exposure and the outcome and using trimesters to characterize the exposure timing. We observed larger bias sizes when the effect size of the exposure-fetal survival relations increased and/or when other unmeasured and uncontrolled risk factors had stronger effect on both fetal survival and the outcome. Our study underlines the needs for the development of analytic methods that adjust for live birth bias in scenarios accounting for time-specific exposure effects and time-specific selections