The Charlson Comorbidity Index - Text Mining Techniques for Healthcare Provider Quality Determination - page 170

run ;
proc kde data=nis.charlsonsample;
univar los/gridl= 0 gridu= 20 out=nis.kdelos4 method=srot ;
univar totchg/gridl= 0 gridu= 50000 out=nis.kdetotchg4 method=srot ;
by charlson;
where charlson> 8 ;
run ;
Figure 5 gives the probability density graph for a Charlson Index of 0, and for indices of 1 and 2.
Note that the average length of stay for patients with a Charlson code of zero occurs at 2 days; for
an index of 1, it peaks at 2.1 days and at 2.6 days for an index of two. There is more variability in the
distribution for indices 2 and 3 compared to an index of zero. Figure 6 gives the probability density for
indices 3-12.
For values 3-8, there is a clear order to the distribution with 3>4>5>6>7>8 in terms of the prob-
ability in the peak of the curve at less than five days. There is also a cutpoint at about 6 days where the
probabilities reverse and 3<4<5<6<7<8. There is some confusion in that the peak for index 3 is at 2.85,
for 4 at 3.2, for 5 at 3.1, for 6 at 3.25, for 7 at 3.75 and for 8 at 4.15. These values indicate that there is
some ambiguity between indices 4 and 5 that could result from moving the patient from one class into
the other. However, the cutpoints are reasonably clustered compared to those for Index values 9-12.
This confusion between indices increases when we consider 9-12. For index 9, the peak occurs at
3.25 with a crossover at day 7.5. There is a second crossover at 15.6, where index 9 again has the highest
Figure 5. Probability density for a Charlson index of 0,1,2
Figure 6. Probability density for a Charlson index of 3-12