Text Mining and Patient Severity Clusters - Text Mining Techniques for Healthcare Provider Quality Determination - page 333

Figure 28. Rate of MRSA by procedure cluster
Figure 29. Scoring text clusters
Figure 28 indicates that there is a considerable difference in the rate of MRSA by procedure cluster.
There are some clusters that have extremely high rates and these high rates suggest that patients in those
clusters should be treated prophylactically to reduce the rate of resistant infection. Clusters 22, 25, and
31 have the highest rates of infection. These suggest that there is a relationship between infection and
surgical wounds with debridement, joint destruction, and wound drainage.
In contrast, procedure clusters 6, 8, and 18 have low rates of resistant infection. Clusters 6 and 18
relate to childbirth, reinforcing the discovery from the diagnosis clusters that childbirth has a low oc-
currence of MRSA. Cluster 8 has to do with hysterectomies. Patients in these clusters do not seem to
need prophylactic treatment.
We want to determine the robustness of the defined clusters of diagnoses. Figure 29 shows the model
using new data from the National Inpatient Sample. We used 2005 data to define the text clusters; we
use 2004 data to score them.
Table 23 gives the frequency count for both the 2004 and 2005 data by cluster to see if the distribu-
tion is relatively similar. In particular, we want to determine whether there are any patients in cluster 7
in the year 2004, dealing with MRSA infection from homosexual sexual practices. Approximately 447
patients are predicted into that cluster, indicating that the trend started as early as 2004.
The vast majority of patients were predicted into cluster #6, with diagnoses: Atherosclerosis of the
extremities with ulceration, Ulcer of other part of foot (toes), Diabetes with peripheral circulatory disorders
(unspecified type), Diabetes with other specified manifestations (Type II or unspecified uncontrolled),