Risk Profiles Help Break Cycle of High-Cost, Unplanned Utilization

Friday, November 5th, 2010
This post was written by Jessica Fornarotto

Predictive modeling can help healthcare organizations break the cycle of unplanned utilization, explains Rebecca Ramsay, senior manager of care support and clinical programs at CareOregon.

Recently, CareOregon found that 12 percent of its membership was utilizing 60 percent of their resources. This is a very familiar pattern that doesn’t change from year to year.

Those 12 percent are medically complex with multiple comorbid conditions. We did a simple calculation and discovered that if we focused on improving the health and thus reducing the cost of care for this subpopulation, it would have a significant fiscal impact for us. We chose to start with the most costly and often the sickest members. We’ve since been able to intervene with a larger and more varied population as we have built capacity in the program.

The ROI for us often comes from those that are currently costing the health plan a lot of money because of the frequent and unplanned utilization. One of our early tasks was to figure out how to identify this high-risk, high-cost population as early as possible — ideally, before they reach that equity tipping point, because at that point it becomes even harder to improve their health, We needed tools to help us do this.

One of the first things we tried was the Adjusted Clinical Group (ACG®) predictive model designed at Johns Hopkins University. The idea behind a predictive model is that it uses information that is submitted on claims to create a risk profile for each member. The risk is usually of future utilization; our model looks at the constellation of diagnoses and medications that a member has on the claims that are submitted and creates a risk score.

We have been using this predictive model for a number of years and it has been successful in helping us identify members to reach out to what we may not have known about without the predictive model. But we have also learned about the limitations of predictive models; a predictive model can only predict risk based on information that is typically found on a claim. And a claim cannot identify when a member, for instance, is homeless or socially isolated. There is no diagnosis code for self-management deficits or for an unsafe living environment. Because of these limits, we have had to fill in the gaps and build additional reports and referral patterns. Our goal was to cast as broad a net as possible to decrease the likelihood that our high-risk patients would fall through the cracks.

Related Posts:

Comments are closed.