The Random Forest and the MARS procedure are both designed to find optimal transformations of predictor variables. This minimized the need to perform variable transformations manually. Nonetheless a number of variable transformations were applied to a number of variables. In particular, character variables were transformed to dummy variable. A variable for creatinine clearance was also created. Initial random forest runs found some categorical variables that had natural ordinal interpretations: these variables were replaced with sets of ordinal dummy variables in the MARS and final regression analyses.