Risk Adjustment Based Upon Resource Utilization - Text Mining Techniques for Healthcare Provider Quality Determination - page 241

As a special case of resource utilization, we can also examine the relationship between the cost to the
patient and the overall costs of the procedures. To investigate the relationship of costs to procedures, we
can restrict our attention to a primary procedure. However, because of privacy issues, the specific proce-
dure is not fully identified in the MEPS dataset. Therefore, we will have to approximate the relationship
of diagnoses and procedures to costs. We will use the procedure of 36, or Operations on vessels of heart.
This procedure of 36 includes all of the procedures of 36.1 that we have used previously. Therefore, we
first look to the distribution of costs for inpatient procedures to see if there is a difference in the group of
procedures in code 36 that can be deduced from the costs. In addition, since the MEPS datafile follows
a cohort of individuals and households, there will be few occurrences of any one condition in any given
year. For 2005, there are 9 occurrences of procedure 36 in the inpatient datafile and 45 occurrences in
the outpatient datafile. While heart procedures are relatively common, the small number of procedures
indicates that in a given cohort of patients in a one year period, the likelihood of requiring such surgery
is relatively small. To find a lifetime risk for requiring such surgery, a cohort would need to be followed
for a prolonged period of time.
Therefore, in order to get a good indication of resource utilization, we will need to use multiple
years. Since it is a good indication that outpatients will use fewer resources compared to inpatients, and
that only inpatients have bypass surgery, we will follow the inpatient datafile for multiple years. For the
years 1996-2005 combined, there were 148 instances of procedure 36. The SAS code needed to create
this dataset is given below:
data meps.inpatient_2005;
set meps.h94d;
self_pay=ipfsf05x;
self_pay_physician=ipdsf05x;
private=ipfpv05x+ipfor05x;
private_physician=ipdpv05x+ipdor05x;
public=ipfmr05x+ipfmd05x+ipfva05x+ipftr05x+ipfof05x+ipfsl05x+ipfwc05x+ipfou05x;
public_physician=ipdmr05x+ipdmd05x+ipdva05x+ipdtr05x+ipdof05x+ipdsl05x+ipdwc05x+ipdou05
x;
totalfacilities=ipfxp05x;
totaldoctor=ipdxp05x;
facilitycharge=iptc05x;
physiciancharge=ipdtc05x;
totalreimbursement=ipxp05x;
run;
The above code is used to define the outcome measures for the 2005 inpatient data. Similar code must
be generated for every other year, changing the '05' in the variables to the correct year for the datafile.
Then we need to merge the datasets together.
PROC SQL;
CREATE TABLE Meps.APPEND_MEPSTABLE AS
SELECT * FROM Meps.INPATIENT_2005
OUTER UNION CORR