Introduction to Ranking Models - Text Mining Techniques for Healthcare Provider Quality Determination - page 13

dataset, ICD9 and CCS codes. We will discuss the use of the ICD9 codes with an available translation
at http://icd9cm.chrisendres.com or at http://www.icd9coding1.com/flashcode/home.jsp. There are three
outcomes available for each patient observation: mortality in the hospital, total charges, and length of
stay. There is information concerning patient age, race, and gender. There are fifteen columns to record
diagnosis codes and another fifteen columns to record procedure codes. There are thousands of pos-
sible diagnoses. One of the problems with defining a risk adjustment model is to work with these codes
to define a risk score. They must be compressed in some manner, either by restricting attention to a
small subset of the diagnosis codes, or by other methods. One novel method using text analysis will be
discussed in Chapter 8.
In particular, NIS data contain patient severity indices defined using the APRDRG, a proprietary code
available through a software purchase. It contains other severity indices as well that are also proprietary.
Although proprietary, the index values are publicly available in the NIS datasets. Its date information is
more complete compared to the MEPS. Some of the variables we will work with in the NIS data include
patient demographics:

• Age (in years)

• Female (0=male, 1=female)

• Race(1=White, 2=Black, 3=Hispanic, 4=Asian/Paciic Islander, 5=Native American, 6=Other)

• DRG

• Patient diagnoses in ICD9 codes (DX1-DX15, ifteen columns)

• Patient procedures in ICD9 codes (PR1-PR15, ifteen columns)

• TOTCHG (Total Charges)

• LOS (Length of Stay)

In order to work with these variables, there are some preprocessing issues, especially to work with
15 columns of diagnoses and procedure codes. We will discuss these requirements briefly. For more
details we refer you to The Little SAS Book, available from SAS Press or to Cerrito (Cerrito, 2008b).
The severity measures available in the NIS are listed in Table 5.

Table 4. Cost variable definitions for the year 2001

Variable Name Variable Definition

RXSF01X Self-pay

RXMR01X Medicare

RXMD01X

Medicaid

RXPV01X Private insurance

RXTR01X TRICARE

RXOF01X Other federal sources

RXSL01X State and local government sources

RXWC01X Worker's compensation

RXOT01X Other private source of insurance

RXOU01X Other public source of insurance

RXXP01X Sum of payments for each prescribed medication