All Times EDT
Keywords: propensity score, propensity score matching, stratification, covariate adjustment, monte carlo simulations
The propensity score has become an acceptable and powerful tool to eliminate imbalance in the distribution of confounding variables between treatment groups and quite often applied for observational studies or real world data analysis these days. We conducted Monte Carlo simulation for continuous outcomes to investigate the performance of matching on the propensity score, stratification on the propensity score and covariate adjustment using the propensity score for continuous outcomes by 4 different set of variables for the propensity score model. Similar to the previous work by Austin et al.(2007) for binary outcomes, we generated 9 independent variables with different association strength with the treatment and the continuous outcome. The simulation studies explored 4 conventional propensity score models by different choices of variables: 1) all variables associated with treatment allocation 2) all variables associated with Outcome 3) true confounder 4) all variables. Each propensity score model was assessed by standardized mean difference and mean squared error (MSE). In addition, we computed a two-step bayesian propensity score and compared with the conventional approach. We found that covariate adjustment using the propensity score facilitating dimension reduction presented the smallest standardized mean difference and MSE across the different sets of variables included in the propensity score model.