Confounding and Missing Data

The protocol will not automatically use multiple imputation for missing data. DAGs will be used to identify confounders for exposure-outcome, exposure-mediator, and mediator-outcome relationships. The confounding strategy focuses on baseline confounders that precede or coincide with birth and avoids adjusting for mediators. Missingness-directed DAGs will characterize missingness mechanisms and assess whether estimands can be recovered from observed data. The protocol will assess unmeasured confounding using E-values, quantitative bias analysis, negative control outcomes, and sensitivity analyses.