Text Mining and Patient Severity Clusters - Text Mining Techniques for Healthcare Provider Quality Determination - page 302

Figure 7. Changing default options in Text Miner
Diagnosisclusters=_CLUSTER_;
Run;
Then we can run Text Miner on the new dataset using the procedure clusters. We can use both diag-
nosis and procedure clusters to define a severity index.
comParIson to cHarlson Index
We next want to compare these results to those of Chapter 5 with the Charlson Index. As shown in the
previous figures, the procedure and diagnosis clusters are not ordered numerically. Therefore, we con-
sider first the comparison of Charlson score to cluster number. To make this comparison, we used a 10%
subsample. Figure 8 has the proportion of each Charlson Index by each diagnosis cluster, restricted to the
first 5 Index values; Figure 9 has the proportion for the remaining Index values. We will also examine
the Charlson Index in relation to the procedure clusters in the same manner.
Note that there seems to be some order to each of the clusters. For example, for cluster 9, Charlson
index 0 has fewer than 10% classified. This increases to almost 35% for Index 1 followed by Index val-
ues 4 and 5. Index value 2 has a rate that is the same as for index value 4. In contrast, only the Charlson
Index value of 0 is present in clusters 1 and 3 with positive probability.
It appears from Figure 8 that clusters 7 and 9 have the highest risk in relationship to the Charlson Index,
followed by clusters 3 and 6. That is relatively similar to the ordering defined by using patient outcomes.