The Charlson Comorbidity Index - Text Mining Techniques for Healthcare Provider Quality Determination - page 191

Table 26. D'Hoore Charlson index using 3 digits of ICD9 codes

Condition ICD9 Codes Weights

Myocardial infarction 410,411 1

Congestive heart failure 398,402,428 1

Peripheral vascular disease 440-447 1

Dementia 290,291,294 1

Cerebrovascular disease 430-433,435 1

Chronic Pulmonary disease 491-493 1

Connective tissue disorder 710,714,725 1

Ulcer disease 531-534 1

Mild Liver disease 571,573 1

Diabetes 250 2

Hemiplegia 342,434,436,437 2

Moderate or severe renal disease 403,404,580-586 2

Any tumor 140-195 2

Leukemia 204-208 2

Lymphoma 200,202,203 2

Moderate or severe liver disease 070,570,572 3

Metastatic solid tumor 196-199 6

Figure 21. Probability density for total charges for the D'Hoore Charlson index

Using the Charlson Index along with patient demographics, we have the results in Figure 23. The
smallest misclassification rate is equal to 21% from a neural network model. The lift indicates that the
model is better than chance through the 6 th decile in the training set, but only through the third decile
in the testing set (Figure 24). The high misclassification rate demonstrates that the model has difficulty
in actual prediction.