Statistical Methods - Text Mining Techniques for Healthcare Provider Quality Determination - page 96

Current 2008 weights have been published.(Weights, 2008) Patients designated as “outliers” can
have higher reimbursement rates:
To qualify for outlier payments, a case must have costs above a fixed-loss cost threshold amount (a dollar
amount by which the costs of a case must exceed payments in order to qualify for outliers).
An example of the computations is given in the Federal Register. (Example, 2003) The computations
are linear, making assumptions concerning the normality of the data. They are highly dependent upon
a hospital's cost-to-charge ratio, defined by hospital and adjusted on a yearly basis. (Anonymous-KY,
2008) This ratio is defined by CMS and reflects the relationship between actual costs (as determined
by an annual audit) and the charges that are billed by the hospital. Therefore, the formula to determine
outlier costs includes the value of charges times the cost-to-charge ratio. In addition, a yearly threshold
value is set by CMS ($33,560 in 2003) The outlier reimbursement is a linear function of this threshold
value and the cost-to-charges ratio.
Therefore, we can look at the outlier charges and the cost-to-charge ratio as determined by the hos-
pital. The cost-to-charge ratio by hospital as well as the wage adjustment is provided with the National
Inpatient Sample data. Figure 18 shows the variability in the cost-to-charge ratio across the different
hospitals. It clearly shows that charges are not uniformly assessed by hospitals.
Note the considerable variability from 0.1 to 0.8. Hospitals with a higher cost-to-charge ratio tend to
bill charges that are more reasonably in line with actual costs compared to hospitals with a rate of 10%.
Figure 19 shows a comparison between charges and charges x cost ratio.
The cost-to-charge ratio, then, reduces the probability in the tail, but does not reduce its size. If we
compare the kernel density graph of the cost-to-charge compared to the normal distribution assumption
(Figure 20), it is clear that the assumption of normality will under-count the outliers.
A normal assumption increases the variance, but fails to count many of the extreme outliers.
Figure 18. Cost-to-charge ratio