Working from Claims Data - Text Mining Techniques for Healthcare Provider Quality Determination - page 351

147 modifier_1_3 $2.
149 modifier_2_3 $2.
151 units_of_service_3 $7.
164 outpatient_total_charges_3 $6.
174 outpatient_non_covered_3 $10.
184 date_of_service $8.
;
output Medpar.outpatient;
end;
end;
run;
Dataset '61' gives outpatient information, and again, shifts to HCPCS codes. There can be multiple
'61' sheets for each inpatient encounter. Therefore, as in more standard claims data, we have to concern
ourselves with two or more different coding schemes. This provides little problem if the severity index
relies only upon a small subset of patient conditions while ignoring the rest.
Once the data have been separated, they need to be re-merged depending upon the data examination.
Dataset '70' contains information related to diagnosis and procedure codes. We can use the CATX state-
ment defined previously to combine them into text strings for analysis purposes. We will want to merge
the dataset information containing the episode date, the diagnosis and procedure codes from '70' and
then we want to collect summary statistics concerning costs, and merge the two pieces of information
together on a one-on-one basis. We use the following code to combine the data:
Data medpar.newdata;
Set medpar.diagnosis;
Procedures=catx(' ',principal.procedure, other.procedure_1, …,other_procedure_5);
Diagnoses=catx (' ', principal_diagnosis, code2 code3,…,code9);
This code is followed by
Proc sort data=medpar.newdata ;
by patient_control_no;
proc sort data=medpar.outpatient;
by patient_control_no;
proc merge;
set medpar.newdata medpar.outpatient;
by patient_control_no;
Once we have these datasets merged, we investigate them in the same way as we have in previous
chapters.