Working from Claims Data - Text Mining Techniques for Healthcare Provider Quality Determination - page 353

If (last.identifier and numdays=1) then delete;
Run;
We also want to eliminate repeat observations:
Data sasuser.norepeats;
Set multipletreatments;
By identifier;
If (not first.identifier and differenceindays=0) then delete;
Run;
Then we filter out the first observation so that we can use it to determine the number of episodes per
patient, and look at the recurrence in episodes.
Data sasuser.episodes;
Set sasuser.norepeats;
By identifier data;
If first.identifier;
Run;
We can then append the dataset created above (called sasuser.episodes) to the dataset with only one
entry (sasuser.onetreatment). In this way, we can compare patients with and without repeat episodes in
relationship to the ICD9 (or CPT) codes identifying treatment to see if some decisions are more likely to
be related to particular outcomes. In this respect, we are going beyond looking at actual versus predicted
mortality to examining real outcomes. For example, we can look at bypass surgery versus angioplasty
with a stent, and looking at the values of a drug-eluting stent. We can also examine the relationship
between medications prescribed for diabetes and time to disease progression to see if some medications
prolong the time to disease progress over others. We can also take into consideration patient compliance
with medications by using pharmacy claims data.
future trends
Insurers will hold providers more and more accountable with how healthcare dollars are spent. More-
over, insurers will be less willing to pay for errors and for poor treatment. They will be relying on their
own claims databases to investigate the quality of care, and will start disallowing claims from providers
residing at the bottom of the quality list. For this reason alone, providers need to understand how such
claims information is used to define the quality of care.
dIscussIon
While more difficult to work with, claims data are routinely available. Because such data are proprietary,
most health insurers are reluctant to make such information available. The Medpar claims are available