Skip Internal Navigation
EB Research - Methodological Research in Epidemiology
Causal Inference in Reproductive and Perinatal Epidemiology
Here we extend the methodological framework for causal inference to reproductive and perinatal epidemiology. The objective of this research is to develop methods using causal inference tools, specifically as they improve researchers' understanding of confounding and colliders, and as applied to the birth weight paradox and the role of birth weight in analysis of perinatal data. In addition, our objective is to apply the same tools to better understand the role of history of prior outcomes in appropriate modeling. Our team of researchers has made significant contributions to this literature in the areas of:
- The Birth Weight Paradox: Utilizing analytical methods as well as directed acyclic graphs (DAGs) to graphically evaluate bias, confounding, and possible explanations for the birth weight paradox. In addition, a combination of DAGs and simulation studies were utilized to quantify bias and evaluate the proposed solution of utilizing birth weight z-scores.
- Overadjustment: Using causal diagrams, analytical, proofs, and an empirical example estimating the total effect of maternal smoking on neonatal mortality, researchers at NICHD illustrated and clarified the definition of overadjustment bias, distinguished overadjustment bias from unwarranted adjustment, and quantified the amount of bias and loss of precision associated with overadjustment and unwarranted adjustment.
- Role of Prior Outcomes: Pregnancy outcomes, such as spontaneous abortion and preterm birth, are often predictive of future pregnancy outcomes. As a result, many researchers adjust for reproductive history. Research here using DAGs illustrates that this may not always be the correct approach. In fact, there is no single answer as to whether reproductive history should be included in the model; the decision depends on the research question and the underlying DAG.
- Exposure-enriched designs: Designs in which a cohort study is enriched for a primary exposure of interest to improve cost-effectiveness. We have shown that caution should be employed when conducting secondary analyses in studies that have already been enriched, intentionally or unintentionally, for a primary exposure of interest. Specifically, causal diagrams can help identify scenarios in which secondary analyses may be biased, and specific analytical methods can be used to remove the bias (e.g., inverse probability weights).
- Person-time at risk: Although commonly excluded, we have shown that there are scenarios where person-time not at risk should be included. When interested in estimating treatment effects that allow and account for potential noncompliance, or where the exposure may be associated with the time at risk, we argue that person-time not at risk should be included. In the case of time to pregnancy, although the ITT-type analysis may underestimate the biological fecundity of the population, it may also yield an answer to a question that is of more interest to couples trying to become pregnant.
Enrique F. Schisterman, Ph.D.
- Ahrens KA, Cole SR, Westreich D, Platt RW, Schisterman EF. A cautionary note about estimating effect of secondary exposures in cohort studies. American Journal of Epidemiology 2014; (In Press).
- Mumford SL, Schisterman EF, Cole SR, Westreich D, Platt RW. Time at risk and intention to treat analyses: parallels and implications for inference. Epidemiology 2014; (In Press).
- Schisterman EF, Cole S, Ye A and Platt R. (2013).Accuracy loss due to fixed or random left truncation in cohort studies. Journal of Paediatric Perinatal Epidemiology. 27(5):491-502. PMID: 23930785
- VanderWeele TJ, Mumford SL, Schisterman EF. (2012). Conditioning on intermediates in perinatal epidemiology. Epidemiology, 23(1):1-9. With discussion. PMID: 22157298
- Westreich D, Cole SR, Schisterman EF, Platt RW. (2012). A simulation study of finite-sample properties of marginal structural Cox proportional hazards models. Stat Med., Apr 11. doi: 10.1002/sim.5317. PMID: 22492660