Statistical Methods - Text Mining Techniques for Healthcare Provider Quality Determination - page 131

Figure 24. Decision tree for length of stay using 1000 observations
Figure 25. Results for COPD and length of stay
The tree has a larger error, but it is reasonably close to regression so that we can examine it and com-
pare it to Dmine regression and standard regression. Figure 27 gives the summary table for the standard
regression, with a 28% r 2 value. The standard regression indicates that age, race, and income quartile
are statistically significant. In addition, there are procedures, including ventilation, the use of a venous
catheter, ultrasound, and antibiotic use that are statistically significant for length of stay. There are also
patient diagnoses that are significant, including congestive heart failure, cerebral vascular event, and
dementia. However, diabetes, HIV, and peripheral vascular disease are not significant.
Figure 26 gives the decision tree; Figure 28 gives the results for the Dmine regression, which does
a stepwise procedure as well.
Many of the patient conditions are statistically significant, including age, race, and income quartile
as well as the use of antibiotics, congestive heart failure, and dementia. Diabetes is not significant. The
decision tree indicates that prolonged use of a ventilator is of first importance in making the split.