A note on Breiman's random forest data mining technique and conventional cox modeling of survival statistics: The case of the phantom "induct" covariate in the Ohio State University kidney transplant database

This note represents a portion of the research that has been conducted using the Ohio State University's large kidney transplant database. Our latest results into understanding the impact of covariates on renal graft success including the impact of drug therapy are discussed. The major result here is that using both the Cox model and Breiman's Random Forest Data Mining techniques has helped to unravel the mystery of the "induct" immunosuppressant covariate slipping in and out of the list of important variables. We also make the interesting observation that the Random Forest method seems to mimic the clinician's (R.P.) understanding of the importance of variables.