We implemented propensity score-based methods in both the empirical example and the simulation studies to account for measured confounders. The propensity score was estimated as the predicted probability of statin exposure during the first trimester using logistic regression models. In the empirical study, all the confounders described above (under Empirical example ) were included in the propensity-score model, while in the simulation studies the ten confounders, c1-c10, were included in the propensity-score models. After propensity-score estimation, the following three approaches were used to derive adjusted associations between statin exposure and congenital malformations.
We excluded the observations from the non-overlapping regions of the propensity-score distributions among exposed and unexposed populations before conducting propensity-score full-matching, stratification and SMR weighting. This step, also referred to as ‘trimming’, ensures exclusion of patients who will always or never receive therapy because of indications or contraindications and focuses the estimation of treatment effects in a population with clinical equipoise.22 (link) The ability of propensity score-based approaches to allow researchers to measure treatment effects in a population with clinical equipoise through trimming is one of its great strengths over traditional multivariable outcome regression models.
We excluded the observations from the non-overlapping regions of the propensity-score distributions among exposed and unexposed populations before conducting propensity-score full-matching, stratification and SMR weighting. This step, also referred to as ‘trimming’, ensures exclusion of patients who will always or never receive therapy because of indications or contraindications and focuses the estimation of treatment effects in a population with clinical equipoise.22 (link) The ability of propensity score-based approaches to allow researchers to measure treatment effects in a population with clinical equipoise through trimming is one of its great strengths over traditional multivariable outcome regression models.
