Figure 2.

A comparison of the area under curve (AUC) distributions for the random forest model and all baseline approaches. A. For diabetes, the AUC of the random forest model improved upon the best baseline approach (UK diabetes–retrained) by 0.239 (0.910 vs 0.671, P<.001). B. For hypertension, the AUC of the random forest model improved upon the best baseline approach (Framingham-retrained) by 0.095 (0.792 vs 0.697, P<.001).