Confounding

Exposed women were more likely than unexposed women to have CMU-C coverage, previous NSAID use, comorbidities, and residence in deprived areas. The observational design can detect associations but cannot establish causation. The analysis adjusted for maternal characteristics and health factors measured before conception. High-dimensional propensity score analyses changed the main estimate only modestly. Residual confounding may remain despite indirect adjustment attempts. The protocol collects broad baseline data and uses multivariate regression to adjust for confounding. Many birth, family, socioeconomic, and activity factors may influence neurodevelopmental outcomes. Residual confounding remains possible despite adjustment. Important confounders such as smoking and alcohol consumption were not directly measured in the administrative databases.