Risk Adjustment Based Upon Resource Utilization - Text Mining Techniques for Healthcare Provider Quality Determination - page 266

Figure 24. Kernel density for length of stay by resource demand level
Figure 25. Kernel density for total charges by resource demand level
If we perform predictive modeling for length of stay, the optimal model is a decision tree. The tree is
extensive with many different branches. Figure 28 shows the left-hand part of the tree; Figure 29 shows
the center, and Figure 30 shows the right hand side. The left side predicts a shorter length of stay; the right
side predicts a much higher length of stay. All three disease staging measures were used in the model.
Because of its extensiveness, we give some of the English Rules below:
IF Admission type < 3.5
AND 1.5 <= Disease Staging: Length of Stay Level < 2.5