Use of Risk Adjustment Models for Provider Reimbursements - Text Mining Techniques for Healthcare Provider Quality Determination - page 363

Figure 4. Kernel density of physician charges by medication
are substantially under-paid for the proportion of outlier patients encountered. In addition, the outlier
cost can be exponentially higher compared to the more typical patient, but the outlier reimbursement is
just a linear increase over standard payments.
We consider again our example with osteoporosis and medications. As we saw in Chapter 9, there
is a considerable difference in the use of diagnostic tests given the different medications. We again
examine the difference by medication to examine the physician reimbursements. Table 1 gives the aver-
age reimbursement by medication as well as the total charges; Figures 3 and 4 give the kernel density
distributions.
It appears that patients prescribed Boniva® yield lower average reimbursements compared to the
other three drugs, with Fosamax® having the highest reimbursement. Actonel® has the highest level of
variabililty. However, the kernel density estimations show something slightly different.
In both cases, patients prescribed Boniva® have higher charges and higher reimbursements compared
to the other three medications, even though patients prescribed Actonel® have more diagnostic tests.
Because the users of Boniva® have a lower probability of extreme outliers, the average cost is actually
less compared to the other three medications.
examInatIon of tHe numBer and ImPortance of dIagnosIs codes
One of the things that is important to payers and to patients is to be able to determine which providers
are gaming the system. In order to do so, we need to examine the severity index of a provider in rela-
tionship to the codes used to define the patient's condition. In particular, we start with examining which
providers use the maximum number of codes to identify patient co-morbidities. Certainly, a provider who
identifies 9 or 10 conditions will tend to have a sicker patient compared to a provider who identifies 4 or
5 conditions, and it is to the provider's advantage to increase the number of diagnoses. More diagnoses
will have a higher probability of including those that are used to define the severity index.