The use of the propensity score for estimating treatment effects: administrative versus clinical data