Review and practical excursus on the comparison between traditional statics methods and Classification And Regression Tree (CART) in real-life data: Low protein diet compared to Mediterranean diet in patients with chronic kidney disease

Abstract

The Classification and Regression Tree (CART) is a supervised learning approach useful used to segment the space of and the predictors/features space into smaller homogeneous regions that are, represented in a decision tree. It computes the selection of features automatically, in contrast to traditional statistic methods. In this review, we compared CART to traditional statistics in patients who did not attempt a diet to patients who followed a low-protein diet (LPD) or the Mediterranean diet in chronic kidney disease (CKD) patients, and we analyzed them using linear regression and CART methods. In our example, diet adherence proved to be the factor with the greatest impact on renal function decline, but CART failed to detect significant differences between LPD and Mediterranean diet. Similar results were found using traditional statics, but CART gave a model with the proportion of the explained outcome by the model (R²) higher by about 20%, thus a stronger model. In addition, CART allows quick and easy identification of the variables affecting the outcome with a simple visual representation through the decision tree, increasing the interpretability of the results. In summary, no difference in the impact of variables has been detected with the two methods, but CART gave us a more detailed model with a faster and easier interpretation.