Data Visualization and Data Summary - Text Mining Techniques for Healthcare Provider Quality Determination - page 43

univar totchg/gridl= 0 gridu= 50000 out=nis.kdechgpneumonia method=srot;
by pneumonia;
There is considerably more variability in the length of stay for patients with pneumonia compared to
patients without pneumonia (Figure 9).A patient with pneumonia, however, has a much higher probability
of staying four or more days compared to patients with pneumonia, who are very likely going to stay
less than four days. Figure 10 shows the total charges. The peak without pneumonia occurs at $4500;
with pneumonia, it occurs at $9125. Patients with pneumonia have a higher probability of $11,000 or
more in total charges. However, because of the overlap between the two curves, it is difficult to use just
the occurrence or non-occurrence of pneumonia to predict patient outcomes.
We next look at the condition of septicemia. Notice that patients without septicemia have a peak
length of stay of 2 days; this increases to a peak of 6.6 days with septicemia (Figure 11). The variability
with septicemia is considerable, so that after 5.45 days, the probability is far greater that the patient will
have septicemia compared to patients who stay less than 5.45 days. Similarly, the peak at $5500 total
charges without septicemia is much less than the peak at $12,500 with septicemia (Figure 12). Moreover,
the patient with charges of $20,000 or more has a higher probability of having septicemia.
We want to look at one more diagnosis, that of immune disorder. Figure 13 shows that there is little
separation between those with the disorder (peak of 2.9 days) and those without (peak of 2 days). The
crossover point occurs at 3.8 days when the probability of having the immune disorder exceeds the
probability of not having the disorder. Figure 14 shows the crossover point at $11,500 with the peak of
$4625 without the immune disorder, and a peak of $7000 with it. The kernel density shows the added
cost when a patient has a diagnosis of immune disorder.
We can also examine two diagnoses simultaneously using kernel density estimation. The following
code will divide the patients into 4 groups: those with neither pneumonia nor septicemia, those with
one but not the other, and those with both conditions. The following code is used for pneumonia and
septicemia.
Figure 9. Probability density of length of stay with and without pneumonia