A comparison of two approaches to implementing propensity score methods following multiple imputation