A Method/Macro Based on Propensity Score and Mahalanobis Distance to Reduce Bias in Treatment Comparisonin Observational Study

In observational studies, investigators usually do not have the same control over the treatment assignment as they do with randomized controlled studies. As a result, the treatment and control groups may have a large difference on their observed covariates. These differences could lead to bias in estimating treatment effects. There are several propensity score based methods that could reduce the bias caused by these differences and make the two groups comparable. One method is the nearest available Mahalanobis metric matching within the calipers defined by the propensity score. This paper will present and demonstrate the matching algorithm based on this method. A macro is developed to implement the matching algorithm; a growth hormone observational study is used as an example to demonstrate the bias before the match and percentage of bias reduction after the match.