Fig. 7

Classification tree considering age, gender, and diagnosis and drug codes for CRGs 1000-5192-6144-7071. The LDA plot at the top of this figure is shown to understand why clinicians chose the diagnosis code ‘250’ as the condition in the root node. The plot shows that samples of CRGs 1000-5192 can be easily separated from those of CRGs 6144-7071 by means of the longest axis vector, representing the diagnosis code ‘250’ (DM). This tree is similar to the one constructed for CRGs 5192-6144-7071 (see Figure 6). The main difference is found in the left branch of the root node (non diabetic patients), which now includes healthy patients (CRG-1000). The most discriminative feature between CRG-1000 and CRG-5192 is the diagnosis code ‘401’ (hypertension), which is associated with that chronic condition