Introduction to Ranking Models - Text Mining Techniques for Healthcare Provider Quality Determination - page 27

quit ;
%put icd9codx_LEN=&icd9codx_LEN ;
data meps.icd9codes ;
length icd9codx $ &icd9codx_LEN ;
set work.concat ;
run ;
Negative numbers in the code columns indicate that there are no ICD9 codes in the designated columns.
The final resuls will be quite similar to those given in tables 4 and 5. Each patient will have a defined
text string, and that text string contains all diagnoses that have been assigned to that patient.
future trends
As healthcare providers and physicians implement an electronic medical record, more and larger databases
will become available for analysis. As is already done in many other businesses, data mining techniques
will be adopted to “know your customer (patient)”. Data mining will greatly enhance the use of patient
data to improve decision making. It will be extremely important to ensure that the modeling techniques
used to analyze the data will return accurate, meaningful results.
In addition, providers will become more accountable for how they treat patients. As patient sever-
ity indices improve, they will be used to decide upon provider reimbursements, which will be tied to
performance goals and quality rankings.
dIscussIon
In this chapter, we provided a basic introduction of patient severity indices as well as a brief discussion
of data mining and SAS Enterprise Miner, which is used to examine large, complex databases. We will
expand upon this information considerably in subsequent chapters.
Information codes are used in billing and administrative data to define patient conditions, and also
to define patient treatments. These codes are used to define patient severity indices. Therefore, it is
absolutely essential to both understanding the severity indices, and to defining such severity indices
to be able to work with these codes. The most difficult data to work with are contained within claims
databases where different coding methods are used by different providers; the different codes must be
reconciled in some manner.
The standard method to define a patient severity index is to use a number of patient demographics and
diagnoses in a linear or logistic regression, comparing patient information to outcomes. Unfortunately,
the only way to determine reliability is by using fresh data. There is no good way to validate the model.
We will demonstrate several types of patient severity indices to show the current state of the science.
While attempts are made to validate the indices by comparing several methods to each other; however
different methods can lead to very different outcomes.
In addition, we will propose a novel technique that depends upon predictive modeling and text min-
ing. It makes use of the text strings that were defined, with each text string containing all diagnoses
related to that individual. Unlike the standard procedure, text mining does not rely on outcomes to de-