Introduction to Ranking Models - Text Mining Techniques for Healthcare Provider Quality Determination - page 25

Table 6. Concatenated diagnoses

Patient Concatenated Diagnoses

1 486 4932 7863 2780 486 493 786 278

2 486 4661 4659 486 466 465

3 486 4932 2720 486 493 272

4 486 486

5 7298 4273 4534 4598 4019 412 729 427

6 7802 2865 4375 7803 40199 412 729

7 486 4031 2852 2720 3860 5198 7805

8 6826 486 4534 4939 0411 2500 7823

9 4010 2500 2780 7823 401 250 278 782

10 486 2765 4939 5110 4019 412 486 276

11 2765 4241 4280 1122 4589 2500 v586

12 3310 2765 2941 2888 5751 4019 4140

13 V571 486 5849 4280 v586 4019 2449

14 7221 7419 340 7373 722 741 340 737

15 4280 5849 4039 2504 v586 4140 2859

in need of physical therapy. The last two patients in the table are extremely ill. Number 14 has both spina
bifida and multiple sclerosis; patient #15 has kidney and heart problems that are related to diabetes.

mePs data

The following SAS code is used to combine the diagnosis codes into one column. In order to perform
text analysis on HC-078, Medical Conditions data file from MEPS. The data must first be pre-processed.
In this data file, each patient condition is an observational unit, and each patient will have multiple
observations in the dataset. In order to put all of the patient conditions into one observation per patient,
and then to define a text string of patient conditions, the data must first be transposed. Proc Transpose
creates one observation per patient by creating a string of variables. The total number of variables cre-
ated is equal to the maximum number of patient conditions listed for any one patient.

libname meps 'c:\Meps';
proc Transpose data=meps.h78codes

out=work.tran (drop=_name_ _label_)

prefix=med_;

var icd9codx ;

by dupersid;

run;

The next step is to concatenate all of the variables created by Proc Transpose. The creation of the
chconcat array allows for this combination. The different codes are separated by spaces in the text
string.