The All Patient Refined Diagnosis Related Group - Text Mining Techniques for Healthcare Provider Quality Determination - page 237

Figure 29. Dmine regression results
classification of 1-4 that has been identified for each primary patient diagnosis, although with this pro-
cess, some patients cannot be classified. The proportion of patients that cannot be classified is small,
approximately 0.05%.
These classifications were defined using expert panels. For this reason, there will remain a percentage
of patients who are assigned a code of 0 because they cannot be placed in any of the levels 1-4. Those
who cannot be assigned are not then used to determine the quality of providers.
In this chapter, also, the Charlson Index was compared directly to the APRDRG index values. It
shows that there are a number of patient conditions used to define the Charlson Index that are related to
high severity levels in the APRDRG classification. However, the comparison also shows that patients
can be in the least severe Charlson level while also in the most severe APRDRG. These different results
need to be examined carefully to find an explanation as to why a patient can be simultaneously very
severe and not at all severe.
references
Anonymous-3M. (2008). 3M™ APR-DRG Expert Software .
Antioch, K. M., & Ellis, R. P. (2007). Risk adjustment policy options for casemix funding: international
lessons in financing reform. The European Journal of Health Economics , 8 (3), 195-212.
Barbash, G., & Safran, C. (1987). Need for better severity indixes of acute myocardial infarction under
diagnosis-related groups. The American Journal of Cardiology , 59 (12), 1052-1056.
Cerrito, P. (2007). Text Mining Coded Information. In H. A. D. Prado & E. Ferneda (Eds.), Emerging
Technologies of Text Mining: Techniques and Applications. New York: IGI Global.