Risk Adjustment Based Upon Resource Utilization - Text Mining Techniques for Healthcare Provider Quality Determination - page 258

Figure 11. Decision tree branch including Charlson
The misclassification rate appears to be little changed from that defined with only the Charlson
Index and not the APRDRG. However, as shown in Figure 12, the lift function has moved to the fourth
decile. Therefore, the model can give fairly accurate predictions for approximately 40% of the patients
in the data set.
The APRDRG index is used in the first branch of the tree, with patients in classes 0 (unknown) or 1
following the path in Figure 15. The Charlson Index is little used in the predictive model.
The entire tree is given in Figure 16. It shows that for the APRDRG of 2 or larger in the first branch,
there is a further split between class 2 and classes 3-4. Note that the Charlson Index still does not appear
in the model. We can conclude from this that the APRDRG is more closely related to the disease staging
indices than is the Charlson Index.