Statistical Methods - Text Mining Techniques for Healthcare Provider Quality Determination - page 83

Table 3. Summary statistics for septicemia and pneumonia

pneumonia septicemia N Obs Variable Mean Std Dev Minimum Maximum N

0 0 7411485 LOS

TOTCHG

4.3245137
20759.23

6.2727093
34663.43

0
25.0000000

365.0000000
999926.00

7411108
7291776

1 155063 LOS

TOTCHG

12.6039550
70837.14

15.9763921
104672.44

0
29.0000000

361.0000000
998554.00

155043
151724

1 0 390336 LOS

TOTCHG

6.6659886
29305.00

7.7732323
47163.21

0
35.0000000

356.0000000
997627.00

390301
385543

1 35123 LOS

TOTCHG

12.0056663
68425.38

14.2393374
95437.36

0
35.0000000

308.0000000
998514.00

35120
34618

Table 4. Quartiles for pneumonia and septicemia

pneumonia septicemia N Obs Variable N Lower Quartile Median Upper Quartile

0 0 7411485 LOS

TOTCHG

7411108
7291776

2.0000000
5501.00

3.0000000
11198.00

5.0000000
23168.00

1 155063 LOS

TOTCHG

155043
151724

4.0000000
15748.00

8.0000000
34162.00

15.0000000
79140.00

1 0 390336 LOS

TOTCHG

390301
385543

3.0000000
8379.00

5.0000000
15545.00

8.0000000
31209.00

1 35123 LOS

TOTCHG

35120
34618

4.0000000
16441.00

8.0000000
35630.50

15.0000000
79799.00

use median rather than mean. Septicemia with or without pneumonia has a median stay of 8 days and
approximately $23,000 more in cost. We then add Immune Disorder (Table 5).
Again, the greatest costs and length of stay occur for the condition of septicemia with or without the
presence of the other two conditions. The linear model for length of stay increases the r 2 value to 0.049;
it is 0.045 for total charges. Table 6 gives the corresponding quartiles. Again, it appears that septicemia
by itself is extremely costly with 15 days and more than $75,000 in total charges. The median value for
septicemia is eight days compared to a mean of 17 days.

example restricting Patients to one diagnosis 2

We next want to consider the procedures that are most closely related to a diagnosis of COPD, and then
we want to see if a linear regression with these procedures as explanatory variables can be used to predict
length of stay and total charges. We use some SAS code to combine the 15 procedure columns and then
find those procedures that occur most frequently with a patient diagnosis of COPD:

data work.charlsoncopd;
set nis.charlson;
if (rxmatch('486',diagnoses3digits)> 0 ) then code= 1 ;
else code= 0 ;
data nis.charlsoncopd;
set work.charlsoncopd;