“…If left to ''grow'' unchecked, these trees become ''bushy,'' with nodes having few cases, and may not generalize well; several ''pruning'' techniques are available [21]. We employed an implementation [22] of the CART (Classification and Regression Trees) algorithm [12,23], previously applied to trauma data [24][25][26][27][28][29][30]. When employed with categoric outcomes, CART by default minimizes the Gini impurity criterion [23], a measure of variability within the subgroups created at each stage of the tree.…”