Text this: Random effects modelling versus logistic regression for the inclusion of cluster-level covariates in propensity score estimation: A Monte Carlo simulation and registry cohort analysis