The Charlson Comorbidity Index - Text Mining Techniques for Healthcare Provider Quality Determination - page 173

data nis.modifiedcharlson;
set nis.charlson_regression;
if (charlson< 3 ) then charlsoncode= 0 ;
if (charlson> 2 ) then charlsoncode= 1 ;
run ;

Note that there is now a considerable difference in the mortality rate between those with a code of
0 and those with a code of 1; all discrepancies have disappeared. We can just as easily make a different
cutpoint, say at the index value of 5 (Table 14). The difference is approximately 6% between mortality
for code 0 and mortality for code 1.
Similarly, if we restrict our attention to averages, we can show that patients with code 0 use fewer
resources compared to code 1 (Table 15).
There is a difference of 2 days in the length of stay between the two subgroups, and a difference of
$11,000 as well. For a cutpoint of 5 (Table 16), this difference increases slightly with a difference of
just under 3 days in length of stay and a difference of $12,000.

Table 14. Mortality by Charlson code with a cutpoint at 5

Table of charlsoncode by DIED
charlsoncode DIED Total

Frequency
Row Pct
Col Pct

0 1

0 7765995

97.96
99.25

161809
2.04
96.80

7927804

1 58862

91.68
0.75

5341
8.32
3.20

64203

Total

7824857 167150 7992007

Table 15. Use of resources for Charlson code with a cutpoint at 2

charlsoncode N Obs Variable Mean Std Dev Minimum Maximum N

0 7173172 LOS

TOTCHG

4.3975221
21082.16

6.7260265
37802.33

0
25.0000000

365.0000000
999926.00

7172794
7057513

1 818835 LOS

TOTCHG

6.6983407
33491.09

7.6227648
48127.72

0
25.0000000

335.0000000
998508.00

818778
806148

Table 16. Use of resources for Charlson code with a cutpoint at 5

charlsoncode N Obs Variable Mean Std Dev Minimum Maximum N

0 7927804 LOS

TOTCHG

4.6106097
22253.78

6.8357345
39017.86

0
25.0000000

365.0000000
999926.00

7927375
7800359

1 64203 LOS

TOTCHG

7.4293503
34737.46

8.8511050
53172.30

0
32.0000000

311.0000000
991975.00

64197
63302