Statistical Methods - Text Mining Techniques for Healthcare Provider Quality Determination - page 146

Figure 39. Decision tree to predict mortality for patients undergoing cardiovascular bypass surgery

Table 16. Percent of predicted versus actual mortality by procedure

Procedure Actual Mortality Predicted Mortality

3610 2.78 42.59

3611 2.78 42.36

3612 2.49 39.55

3613 1.14 39.02

3614 1.07 40.53

3615 0 32.14

3616 0 32.14

3617 0 0

3619 0 100

univar totchg/gridl= 20000 gridu= 100000 out=nis.kdecardchg bwm= .9 ;
by pr1;
run ;

Figure 40 shows the length of stay by hospital. As shown in Table 18, stay differed considerably by
procedure. It also differs considerably by hospital. Hospital #6 has the greatest probability of a shorter
length of stay compared to the other hospitals. Hospital #2 has the highest probability of a longer length
of stay. Hospital #7 tends to be in the middle in probability for both a high and low length of stay, as
does hospital #1.
Figure 41 shows the total charges compared to hospital. There is a definite shift in the curves, indi-
cating that some hospitals charge far more compared to other hospitals, especially hospitals #3 and #9.
Hospital #1 has the least charges, reinforcing the fact that it more generally performs a procedure that