Modern Evidence Generation on Medication Effectiveness and Safety During Pregnancy: Study Design Considerations

Abstract

Non-randomized studies will remain the mainstay for evidence on medications' effects in pregnancy since the number of pregnant participants in randomized clinical trials is insufficient to evaluate uncommon but serious pregnancy outcomes. There has been a growing interest in conceptualizing causal inference based on observational data as an attempt to emulate a hypothetical randomized trial: the target trial. This approach can help identify design flaws and ensuing biases and can point toward potential solutions. Adoption of the target trial emulation framework in perinatal studies raises unique challenges due to the distinct role of gestational time. Challenges include, among others, identifying the timing of conception, pregnancy losses as competing events for later outcomes, different etiologically relevant time windows depending on the outcome, and time-varying outcome risks. We discuss various considerations in developing a protocol for a target trial evaluating drug effects in pregnancy and its observational emulation in databases and registries. While not a panacea, the framework offers a valuable tool to guide us through the specification of the causal questions, the study population and the treatment strategies to be compared and helps to identify avoidable biases as well as unavoidable deviations from the optimal protocol. Making these deviations explicit elucidates the assumptions we make when drawing causal conclusions, and the types of analyses that can be undertaken to quantify the potential magnitude of such biases. Such discipline in the design, conduct, and reporting of pregnancy studies will ultimately lead to the best information possible to inform treatment decisions during pregnancy.