Statistical Methods - Text Mining Techniques for Healthcare Provider Quality Determination - page 141

Figure 33. Model comparison to predict mortality
Figure 34. Datasets generated by score node
We will use a different set of hospitals from the ones in the COPD example since not all of those
hospitals perform bypass surgery. Then we will examine the ranking that the model gives to the hospitals.
Cardiovascular bypass (or CABG) is assigned an ICD9 procedure code of 36.1. We will restrict atten-
tion to patients for whom 36.1 is the primary procedure. In this example, we will use the list of patient
conditions as given in Table 3 to define a patient severity level. We will use a stratified sample to define
the predicted value of mortality. Then, we will compare the predicted results to the actual results by
patient and by hospital. Note that the list in Table 3 contains a condition for congestive heart failure and
for myocardial infarction. However, it does not include a code for congested arteries. As we will show
in subsequent chapters, patients undergoing bypass surgery already have considerable severity in their
conditions. However, not all severity indices will identify that severity level.
This example differs from the previous example because the patient condition will be considered in
defining the predicted value. This example was previously discussed in Chapter 2, and the code to ex-
tract the data was listed in that chapter. Figure 37 gives the results of the predictive model, with hospital
included as one of the input variables. The best misclassification rate is 26%.