Risk Adjustment Based Upon Resource Utilization - Text Mining Techniques for Healthcare Provider Quality Determination - page 270

AND Median household income quartile for patient's ZIP Code IS ONE OF: 1 2
AND Disease Staging: Length of Stay Level < 1.5
THEN
NODE: 55
N: 2670
AVE: 1.27978
SD: 0.74301
IF Age in years at admission < 65.5
AND 0.5 <= Disease Staging: Mortality Level < 3.5
AND 2.5 <= Disease Staging: Resource Demand Level
AND 2.5 <= Disease Staging: Length of Stay Level < 3.5
THEN
NODE: 60
N: 139408
AVE: 3.54198
SD: 3.18064
The last split shown separates an average stay of 1.2 days and 3.5 days. There is a strong possibility
that this tree is too extensive and should be pruned. Such pruning is beyond the scope of this text, and
we refer the reader to more detailed information on decision trees. (Ville, 2006) We also consider the
outcome of total charges. The decision tree is again the best model. Since it is equally as complicated as
the one for length of stay, we provide the beginning English rules. This tree does differ in that income
quartile and race are prominently featured in the tree, although the disease staging levels form the first
branches.
IF Disease Staging: Length of Stay Level < 4.5
AND Admission type < 1.5
AND 4.5 <= Disease Staging: Mortality Level
AND 4.5 <= Disease Staging: Resource Demand Level
THEN
NODE: 42
N: 1522
AVE: 74756.2
SD: 68010.7
This first level depends upon the disease staging, separating level 5 from the remaining levels.
IF Admission type IS MISSING
AND 4.5 <= Disease Staging: Mortality Level
AND 4.5 <= Disease Staging: Resource Demand Level
THEN
NODE: 45
N: 1437
AVE: 208045