Robustness of Multiple Imputation Methods for Missing Risk Factor Data from Electronic Medical Records for Observational Studies