Information Technology Reference
In-Depth Information
Table 11. Patients with immune disorder by hospital
Table of DSHOSPID by immune
Hospital Code
immune
Total
Frequency
Row Pct
Col Pct
0
1
1
2198
76.37
8.84
680
23.63
11.05
2878
2
1164
77.29
4.68
342
22.71
5.56
1506
3
638
70.34
2.57
269
29.66
4.37
907
4
6324
80.80
25.44
1503
19.20
24.42
7827
5
1928
79.67
7.76
492
20.33
7.99
2420
6
3599
66.55
14.48
1809
33.45
29.40
5408
7
1362
90.50
5.48
143
9.50
2.32
1505
8
878
93.60
3.53
60
6.40
0.97
938
9
4868
89.40
19.58
577
10.60
9.38
5445
10
1901
87.20
7.65
279
12.80
4.53
2180
Total
24860
6154
31014
The ROC curves indicate that accuracy decreases considerably on the test data compared to the train-
ing data. Table 13 gives the actual and predicted mortality levels by hospital.
The provider that has the largest difference between actual and predicted mortality is #7. The overall
ranking is 1>3>9>2>4>6>5>1>10; again, a hospital with zero mortality is penalized using this system.
Usually, zero mortality would be considered good. In fact, regardless of the actual mortality, a hospital
with zero predicted morality will rank low in comparison to other providers.
In a third example, we will restrict attention to ten hospitals, and examine patients undergoing just
one procedure, that of cardiovascular bypass surgery. We will compare actual mortality rates across these
hospitals, and look at the relationship of patient diagnosis to prediction of mortality. We discussed this
example briefly in Chapter 3. We will discuss it in more detail here.
 
Search WWH ::




Custom Search