 This paper compared five different regression methods for imputing missing co-variates in the context of predicting a time-to-event outcome. Lassie was found to have the lowest bias, mean square error, mean square prediction error, and median absolute deviation, mad. SVM also performed well, followed by GLM and Mars, with the latter two having the highest relative performance. This article was authored by Nicole Solomon, Yulia Lechnigina, and Susan Hallaby.