Emulating a Target Trial of Interventions Initiated During Pregnancy with Healthcare Databases: The Example of COVID-19 Vaccination. Issue 2 (11th November 2022)

• Record Type: Journal Article
• Authors: Hernández-Díaz, Sonia; Huybrechts, Krista F.; Chiu, Yu-Han; Yland, Jennifer J.; Bateman, Brian T.; Hernán, Miguel A.
• Is Part Of: Epidemiology. Volume 34:Issue 2(2023)
• Page Start: 238
• Page End: 246
• Publication Date: 2022-11-11
• Subjects: Target trial -- pregnancy -- Healthcare databases: Immortal time bias -- Methods -- COVID-19 -- Vaccines -- Drugs -- safety -- Effectiveness; Epidemiology -- Periodicals; Epidemiology -- Environmental aspects -- Periodicals; Epidemiology -- Periodicals; 614.405
• DOI: 10.1097/EDE.0000000000001562
• Languages: English
• ISSNs: 1044-3983

Abstract

Abstract : Background: Observational studies are often the only option to estimate effects of interventions during pregnancy. Causal inference from observational data can be conceptualized as an attempt to emulate a hypothetical pragmatic randomized trial: the target trial. Objective: To provide a step-by-step description of how to use healthcare databases to estimate the effects of interventions initiated during pregnancy. As an example, we describe how to specify and emulate a target trial of COVID-19 vaccination during pregnancy, but the framework can be generally applied to point and sustained strategies involving both pharmacologic and non-pharmacologic interventions. Methods: First, we specify the protocol of a target trial to evaluate the safety and effectiveness of vaccination during pregnancy. Second, we describe how to use observational data to emulate each component of the protocol of the target trial. We propose different target trials for different gestational periods because the outcomes of interest vary by gestational age at exposure. We identify challenges that affect (i) the target trial and thus its observational emulation (censoring and competing events), and (ii) mostly the observational emulation (confounding, immortal time, and measurement biases). Conclusion: Some biases may be unavoidable in observational emulations, but others are avoidable. For instance, immortal time bias can be avoided by aligning the start of follow-up with the gestational age at the time of the intervention, as we would do in the target trial. Explicitly emulating target trials at different gestational ages can help reduce bias and improve the interpretability of effect estimates for interventions during pregnancy.