To illustrate this, we built a sample ACO using the Medicare Synthetic Public Use Files that are freely downloadable from CMS. We first removed all beneficiaries who were under age 65 (indicating individuals who were disabled or had ESRD) so that we were only dealing with an aged population for simplicity. We subtracted fourteen years from the ages of all of the patients (this data is from 2008) in order to get a population that is not overly weighted towards older beneficiaries and in line with the original data. We then utilized the table referenced above to calculate the demographic relative factors for each beneficiary. We made a proxy for risk scores by assigning weights to the chronic condition flags in the beneficiary summary file and summed up those scores for each beneficiary. Finally, we multiplied the scores for the non-dual eligible beneficiaries by an adjustment factor to ensure that their risk scores averaged out to 0.949 (the national average risk score for a non-dual eligible beneficiary) and similarly adjusted the scores for the dual-eligible beneficiaries to ensure their average was 1.495. Note: we could have used the claims files provided with the Medicare Synthetic Public Use Files to mimic the CMS HCC v.24 algorithm, but for the purposes of this example risk score precision at a beneficiary level was not critical.
For year one of our sample ACO, we randomly selected 10,000 beneficiaries (out of the over 93,000 in the sample set), including 550 dual eligible beneficiaries. A dual-eligible population of 5.5% is well below the national average of 19%. The average risk score for this ACO is 0.981, which includes an average demographic relative factor score of 0.464 and an average disease relative factor score of 0.517.
Scenario 1: Dual-Focused Growth
Our sample ACO has decided that their growth strategy is to add only dual-eligible beneficiaries to their ACO in year 2. In year 2, our sample ACO added 3,000 additional beneficiaries, all of whom are dually eligible. These 3,000 new beneficiaries have an average risk score of 1.495, which includes an average demographic relative factor of 0.635.
Our sample ACO was extremely fortunate to not lose one single beneficiary between Year 1 and Year 2 – not to death, moving out of the area, or choosing a different provider. And, in some sort of miracle, all of our initial beneficiaries retained the exact same health profile with the exact same disease relevant factors component of their risk score. All 10,000 initial beneficiaries remain with our sample ACO, but of course they are now one year older which has increased their average demographic relative factor score to 0.478 (from 0.464).
After Year 2, the ACO now has 13,000 beneficiaries with an average risk score of 1.111, an increase of 13% over year 1’s average risk score of 0.981 – well over the 3% cap. The breakdown of year 2’s 1.111 risk score includes a demographic relative factor of 0.514 and a disease relative factor of 0.596.
Since ACO REACH caps the risk score growth relative to the demographic score growth, we need to look at the percent risk score growth in that light. The demographic score from year 1 to year 2 grew from 0.464 to 0.514, which represents a 10.8% increase. The 3% cap is applied on top of the 10.8%, meaning the overall HCC score for our sample ACO can increase 13.8% before hitting this cap, and so our sample ACO is within these bounds for year 2 having only increased its overall average risk score 13.25%, for a relative growth rate of 2.47%.
Of course, the assumptions that we made to illustrate this point are not realistic: patients will die, patients will get additional chronic conditions as they age, no ACO will be able to add only dual eligible beneficiaries. But taking this somewhat extreme example serves as a good illustration of the risk score growth cap relative to demographic factors.
Scenario 2: Cap Casualty
In our second ACO example, we will use the same base ACO. However, in year 2, rather than adding 3,000 beneficiaries, they have added 5,006 dual-eligible beneficiaries (2,006 more than our first scenario, and a group that represents the sickest). These 5,006 beneficiaries have an average risk score of 1.679, which includes an average demographic relative factor of 0.652.
Looking at the percent risk score growth of the demographic score, we see an increase from 0.464 to 0.536, representing a 15.5% increase. The 3% cap is applied on top of this 15.5%, bringing us up to 18.5% which, in our example results in a cap casualty. The overall average risk score increased from 0.981 to 1.224, which represents an increase of 24.77% or a growth rate of 9.25% relative to the demographic score growth. Since the ACO is only allowed to grow at a 3% rate relative to the demographic factors, the risk score will cap out at 1.162 (before any coding intensity factor is applied). This is the risk score that will be used to calculate your benchmark and could put you at a significant disadvantage if your beneficiaries are sicker than that average risk score represents.
What the risk score cap in ACO REACH aims to control is disease relative factor score creep over the years. While it is expected that a beneficiary base that doesn’t change over time will have a risk score that grows as beneficiaries age (and therefore their demographic relative factor score increases) and they get sicker (increasing their disease relative factor score), the overall increase for an ACO is expected to be counterbalanced with younger beneficiaries joining the ACO.
We have heard concerns regarding this 3% cap on risk score growth relative to a static year. While this cap will certainly be a factor that REACH ACOs need to be cognizant of, the impacts on an ACO will be less than feared due to the allowance for unhindered demographic risk factor score growth. Although we shared an example of a cap casualty ACO above, to see the population grow in that way is not likely and would only occur if that ACO seeks out those beneficiaries and is not cognizant of the cap. With that in mind, we encourage each REACH ACO to ensure they understand the specific mechanisms of the model, including risk scoring.