The All Patient Refined Diagnosis Related Group - Text Mining Techniques for Healthcare Provider Quality Determination - page 209

Table 4. Classification table for logistic regression

Classification Table
Prob
Level

Correct Incorrect Percentages

Non-
Event

Non-
Event

Sensi-
tivity

Speci-
ficity

False
POS

False
NEG

Event Event Correct

0.500 556E4 0 124E3 0 97.8 100.0 0.0 2.2 .

0.520 556E4 1 124E3 0 97.8 100.0 0.0 2.2 0.0

0.540 556E4 18 124E3 31 97.8 100.0 0.0 2.2 63.3

0.560 556E4 384 123E3 490 97.8 100.0 0.3 2.2 56.1

0.580 555E4 2669 121E3 3426 97.8 99.9 2.2 2.1 56.2

0.600 554E4 9086 115E3 13074 97.8 99.8 7.4 2.0 59.0

0.620 553E4 18925 105E3 29564 97.6 99.5 15.3 1.9 61.0

0.640 551E4 28151 95451 47410 97.5 99.1 22.8 1.7 62.7

0.660 55E5 34456 89146 60902 97.4 98.9 27.9 1.6 63.9

0.680 549E4 39410 84192 70966 97.3 98.7 31.9 1.5 64.3

0.700

548E4 43484 80118 79038 97.2 98.6 35.2 1.4 64.5

0.720 547E4 47262 76340 87519 97.1 98.4 38.2 1.4 64.9

0.740 546E4 50537 73065 96728 97.0 98.3 40.9 1.3 65.7

0.760 545E4 52947 70655 105E3 96.9 98.1 42.8 1.3 66.5

0.780 545E4 54430 69172 11E4 96.8 98.0 44.0 1.3 66.9

0.800 544E4 55584 68018 113E3 96.8 98.0 45.0 1.2 67.0

0.820 544E4 56788 66814 115E3 96.8 97.9 45.9 1.2 67.0

0.840 544E4 57433 66169 117E3 96.8 97.9 46.5 1.2 67.1

0.860 543E4 59906 63696 128E3 96.6 97.7 48.5 1.2 68.2

0.880 541E4 63404 60198 15E4 96.3 97.3 51.3 1.1 70.3

0.900 538E4 67450 56152 18E4 95.9 96.8 54.6 1.0 72.7

0.920

522E4 83230 40372 335E3 93.4 94.0 67.3 0.8 80.1

0.940 504E4 95693 27909 516E3 90.4 90.7 77.4 0.6 84.4

0.960 497E4 99497 24105 583E3 89.3 89.5 80.5 0.5 85.4

0.980 481E4 105E3 18986 749E3 86.5 86.5 84.6 0.4 87.7

1.000 0 124E3 0 556E4 2.2 0.0 100.0 . 97.8

more accurately. Figure 2 indicates that the optimal model is a rule induction, but also that the decision
tree and the Dmine regression are nearly identical.
We next look at the decision tree (Figure 3). It shows that the APRDRG is of first importance when
predicting mortality. The patient's age is important as well.
Because the Dmine regression had a similar outcome, we look at its results as shown in Figure 4.
This result shows that the APRDRG mortality index is of first importance followed by the severity index.
Patient demographics, including age, race, and sex, follow. The patient's income quartile is below the
threshold and is not included in the model. Note that the mortality index has an r 2 value of 58%. The
ROC curve is given in Figure 5, showing that the fit is reasonable for the training, validation, and testing