The All Patient Refined Diagnosis Related Group - Text Mining Techniques for Healthcare Provider Quality Determination - page 228

Table 9. APRDRG mortality index and hospital for patients with COPD

Table of DSHOSPID by APRDRG_Risk_Mortality
Hospital APRDRG_Risk_Mortality(All Patient Refined DRG: Risk of Mortality Subclass) Total

Frequency
Row Pct
Col Pct

1 2 3 4

1 39

30.71
8.57

63
49.61
12.43

21
16.54
12.50

4
3.15
13.33

127

2 14

22.22
3.08

34
53.97
6.71

12
19.05
7.14

3
4.76
10.00

63

3

13
27.66
2.86

25
53.19
4.93

9
19.15
5.36

0
0.00
0.00

47

4 104

45.22
22.86

94
40.87
18.54

28
12.17
16.67

4
1.74
13.33

230

5 35

40.70
7.69

36
41.86
7.10

14
16.28
8.33

1
1.16
3.33

86

6 60

27.52
13.19

89
40.83
17.55

57
26.15
33.93

12
5.50
40.00

218

7

13
25.00
2.86

36
69.23
7.10

2
3.85
1.19

1
1.92
3.33

52

9 63

28.77
13.85

126
57.53
24.85

25
11.42
14.88

5
2.28
16.67

219

10 114

96.61
25.05

4
3.39
0.79

0
0.00
0.00

0
0.00
0.00

118

Total 455 507 168 30 1160

Very few of the hospitals have patients in APRDRG category 4. Hospital #6 has the highest percent-
age of patients in category 3; hospital #7 has the lowest, and the differential between the two hospitals
is considerable. Note that hospital #10 has no patients in category 3 at all, and almost 97% of its patients
are in category 1. We have filtered out all of the APRDRG code values of 0 (Table 10).
The probability of class 4 in the severity index is higher than in the mortality index. However, it is
still hospital #6 that has a higher proportion in the higher classes compared to the other hospitals, and
hospital #10 has the lowest. It suggests that hospital #6 is very good at shifting its patients into a higher
severity level OR it tends to treat sicker patients.
We next look at the APRDRG index restricted to the one procedure, 36.1. Table 6 shows the rela-
tionship between the APRDRG mortality index and procedure; Table 7 shows the relationship with the
APRDRG severity index. Using both APRDRG indices, the optimal model is the rule induction with
28% misclassification (Figure 24).
The decision tree (Figure 25) has but one branch based upon the mortality index, with a dividing