Information Technology Reference
In-Depth Information
Table 8. APRDRG mortality index compared to patient diagnoses
ICD9 Code
APRDRG 0
APRDRG 1
APRDRG 2
APRDRG 3
APRDRG 4
250
11.34
24.62
37.67
30.96
19.16
272
7.62
29.53
25.90
14.85
6.64
276
45.65
17.94
31.44
46.04
56.21
585
1.76
0.43
3.96
9.22
9.67
491
1.17
4.35
7.19
9.05
8.03
486
4.99
5.45
9.64
21.06
22.95
599
8.02
7.19
16.42
21.39
23.09
584
3.62
0.05
4.03
16.60
48.62
5308
13.29
18.27
12.93
7.68
3.20
5939
1.66
2.65
7.08
5.83
2.80
2859
6.26
8.48
13.88
12.99
6.95
2449
6.16
11.01
12.44
8.53
4.13
3051
29.72
23.54
8.85
5.50
3.03
If ((rxmatch('2859',diagnoses4digits)>0) then code2859=1 ;
If ((rxmatch('2449',diagnoses4digits)>0) then code2449=1 ;
If ((rxmatch('3051',diagnoses4digits)>0) then code3051=1 ;
If (code250=1 or code272=1 or code276=1 or code585=1
Or code491=1 or code486=1 or code584=1 or code599=1 or code5939=1
Or code2859 or code2449=1 or code3051=1) then code=1;
Data nis.aprdrgbycodes;
Set work.aprdrgcodes;
Where code=1;
Run;
It is clear from Table 8 that some of the codes are used to discriminate in APRDRG indices while
others are not. Since 46% of patients with APRDRG code 0 have condition 272, it is not a good clas-
sifier; in contrast, code 3051 has decreasing probability of appearing as the class level increases; it is a
much better classifier and is likely used to define APRDRG levels. If we substitute the severity index in
the place of the mortality index, the proportions of patients with the diagnosis codes is the exact same
as those for the mortality index.
PredIctIve modelIng of ProvIder qualIty
We again consider the relationship of predicted and actual mortality, with the predicted mortality defined
by the APRDRG mortality index. We first look at the example of COPD patients discussed previously
in Chapters 3 and 5. Table 8 gives the relationship of APRDRG mortality index by hospital; Table 9
gives the relationship of the APRDRG severity index. Recall that hospital #8 has no patients diagnosed
with COPD.
Search WWH ::
Custom Search