Working from Claims Data - Text Mining Techniques for Healthcare Provider Quality Determination - page 347

H94G.VARPSU AS VARPSU1
FROM SASUSER.medications AS sasuser.medsandvisits
LEFT JOIN sasuser.H94G AS H94G ON sasuser.medications.DUPERSID = H94G.DUPERSID);
QUIT;
Once we merge the files, we can examine the relationship of physician orders to medications (Table
1).
There are considerable differences in the type of treatment given the differerent medications. Patients
taking Actonel® are much more likely to have an X-Ray or an MRI; those taking Boniva® are much
less likely. It could be that patients with more serious conditions are given Actonel® while Boniva®
is used more for prevention-or it could be that physicians prescribing Actonel® are more knowledge-
able about needed follow up to guard against side effects. Data mining can identify the issue that will
require drill down to investigate the reasons for the outcomes. However, without data mining, the dif-
ferential between medications and additional tests would not be identified, and you would have no way
of understanding that drill down is required. In other words, data examinations such as these allow us
to formulate questions to investigate.
medPar data
We next examine some claims data that are relatively public. Medpar gives the billing information for
Medicare. It is publicly available, although there can be a small cost affiliated with data acquisition. In
fact, many of the public companies that define patient severity indices rely upon Medpar because it is
so readily available. In addition, The Centers for Medicare and Medicaid maintain a data warehouse
consisting of patients with a defined chronic disease, and this information, too, is available for a fee.
There are multiple forms in Medicare data and we provide the SAS code on how to “unpack” the dif-
ferent forms for use in analysis. We are using the 2005 version of the data.
Libname Medpar “F:\Medpar Data\”;
Data Medpar.customer Medpar.accommodations Medpar.ancillary_services
Medpar.diagnosis Medpar.outpatient;
Infile 'F:\Medpar Data\Suburban-DFU-V6-Q4-2006.txt' TRUNCOVER;
Input @ 1 type $2. @;
Select;
When (type='20') do;
Input @ 5 patient_control_no $20.
55 sex $1.
56 birthdate $8.
64 marital_status $1.
65 admission_source $2.
123 start_date $8.
131 admission_hour $2.
133 START_DATE $8.
141 END_DATE $8.