Methodological Considerations for Studies of Preterm Birth After Vaccination during Pregnancy: A Quantitative Bias Analysis on the COVID-19 Vaccine

Purpose of ReviewTo discuss methods that address temporal and confounding biases impacting estimated associations between mRNA COVID-19 vaccination during pregnancy and preterm birth in observational studies.Recent FindingsCOVID-19 vaccination safety studies, using observational data, faced methodological challenges due to the time-dependent nature of vaccination, temporal variability in vaccine access during the pandemic, and differences in baseline characteristics. Analytic approaches to address these potential sources of bias were critical for valid estimation of vaccination and birth outcome associations.SummaryWe reviewed approaches and conducted a quantitative bias analysis using data from a population-based retrospective cohort of live births in Ontario, Canada from December 2020 to December 2021. We compared four analytic techniques to a 'na & iuml;ve' setting (time-fixed exposure; unadjusted): (i) modelled time-varying vaccination status (immortal time bias); (ii) restricted by conception date (cohort truncation bias); (iii) excluded pregnancies without access to vaccination (calendar time bias); and (iv) accounted for differences in baseline characteristics by vaccination status via propensity scores (confounding bias). We calculated the relative difference between each correction and the na & iuml;ve estimate, individually and in varying combinations, to determine bias magnitude and direction. Among 143,331 live births, 49,318 (34.4%) were to individuals who received at least one mRNA COVID-19 vaccination during pregnancy. The initial na & iuml;ve hazard ratio (HR) estimate for preterm birth was 1.06 (95% confidence interval (CI): 1.02-1.10). Corrections (i-iv) produced varying estimates. Combining all four produced an adjusted HR of 0.98 (95% CI: 0.94-1.03), highlighting the importance of implementing appropriate analytical approaches to minimize potential biases.