|
It is important to objectively validate how well the
Clinical Risk Group (CRG) algorithm works in predicting future costs. One of the best ways to illustrate this is
to graph the relationship between true positive and false positive
results. The data points create a Receiver Operator Curve (ROC). In an
excellent test, there are a high percent of true positives and a low percent of
false positives for the indicator, at values the application uses to
distinguish positive from negative results.
The shape of the curve would ideally ascend rapidly toward
the top of the Y axis (True Positives) at relatively low values along the X
axis (False Positives). An overall sign of an indicator’s effectiveness is the measurement of the area under the ROC
curve (AUC). The closer this is to 1.0,
the more perfect the test. A test result with an AUC of 0.5 would demonstrate that the indicator was no better
than random in finding the most expensive members, and the ROC curve would be a
perfect diagonal.
We tested the ability of CRG’s to predict the most expensive
1% of members by comparing this group with those who were the top 1% most
expensive. This group generated 20% of all costs in our test population of over 500,000 members.
The results yielded an AUC of 0.84. This means that CRGs are 84% more likely to identify those in the top 1%
than not.
These results provide confirmation of the power of CRGs to
guide aid those using the algorithm to confirm pricing decisions about individual
groups and product lines and to support budget planning.
| 
