Introduction
This blog caps off our series that reviews key provisions in the 21st Century Cures Act (CCA) that are related to Medicare Advantage (MA) risk adjustment. In this installment, we discuss a provision that would expand the amount of data that could be used to account for the differences in health status when determining payment levels for each MA enrollee. More specifically, the provision would expand the time period—which is currently set to one year— for identifying diagnosis codes on medical records that could inform the risk adjustment calculations for each enrollee.
Prior installments in this series described provisions that have already begun to be enacted. They include additional adjustments to the MA risk adjustment model to account for: the number of conditions each member has; additional conditions specifically related to mental health disorders, substance abuse disorders and chronic kidney disease (CKD); and different types of MA enrollees eligible for Medicaid benefits. Unlike these provisions, changes to the time period for identifying diagnosis codes has yet to be implemented.
Using Diagnosis Data to Control for Health Status
In Part 1 of this series, we provided an overview of how the MA risk adjustment model works and why it is needed. For purposes of understanding the issue of expanding the data available for risk adjustment, it is important to know that Medicare payments to MA plans are enrollee specific; i.e., a specific payment is calculated for each enrollee which includes an adjustment for health status known as a risk adjustment. The adjustment is conducted by applying a risk score that, relative to an average payment amount, increases payments for enrollees expected to need higher than average medical resources (and consequently will incur higher than average medical costs), and decreases payments for enrollees expected to need less than average resources (and consequently incur lower than average costs).
The risk score is calculated using the risk adjustment model, known as the Centers for Medicare & Medicaid Services–Hierarchical Condition Category (CMS–HCC) model. The CMS-HCC model includes demographic (e.g., gender and age) and medical condition (e.g., whether an enrollee has a chronic condition such as diabetes or hypertension) characteristics to calculate the risk score. Each demographic and medical condition characteristic is associated with a risk adjustment factor. The specific risk score for an enrollee is essentially the summation of all of the applicable risk adjustment factors. For example, a male with hypertension would have a risk score that includes the summation of the male and hypertension risk adjustment factors. Thus, having more applicable medical conditions is associated with higher payments. Applicable medical conditions for each enrollee are identified by looking at the diagnoses documented on medical records for the enrollee.
Currently, CMS estimates the size of the risk adjustment factors with spending and diagnostic data from traditional Medicare fee-for-service (FFS) claims. This is because FFS claims are the only available data source with complete spending and diagnostic data for Medicare enrollees. In order to calculate risk scores for MA enrollees, CMS applies the risk adjustment factors estimated with Medicare FFS data to demographic and diagnostic information for MA enrollees. Therefore, the relative completeness and accuracy of diagnostic data in both FFS Medicare and MA affect the accuracy of risk scores and payments to MA plans.
What the CCA Allowed
As mentioned above, CMS currently uses one year of diagnostic data to estimate the size of the risk adjustment factors with FFS data and to identify diagnoses for MA enrollees. The CCA provides an option to use two or more years of historical medical record data, rather than one year, to adjust the payments to MA plans for the medical condition history of each enrollee. There is some ambiguity in the CCA regarding whether this means multiple years of data could be used for identifying MA enrollees with specific medical conditions, as well as in the calculations for estimating the risk adjustment factors. Based on interpretations from prior federal analyses on this issue, which we highlight below, we base our discussion of potential impacts to MA risk adjustment on the notion that the additional years of data would be used for both functions.
How Using Multiple Years Could Change the Risk Adjustment Model
It is possible that physicians and other health care providers do not consistently code conditions on claims from year to year. For example, a physician may indicate on a medical record that a patient has diabetes when initially diagnosed, but might not indicate it on a follow-up visit in the next year if the diabetes is well-controlled. Depending on the extent of the inconsistency in coding conditions over time, there can be instability in the risk scores and payments to plans.
Using two years of medical claims data, instead of one, has been considered as a way to make risk scores more stable. That is, the data for estimating the cost of having a given condition would be more complete because two years of diagnosis data would identify more individuals who have that condition. This is predicated on the notion that the conditions used in the CMS-HCC model to control for health status are chronic conditions, so people will generally have them year after year. As a byproduct of this, research also shows that it is likely that the risk factors associated with medical conditions that are currently underreported will decrease with a change to using more than one year of data. This is because the use of one year of data would be more likely to identify the more severe, higher cost cases for a given condition and miss the less severe, lower cost cases. When using two or more years, more of the lower cost cases would be identified; consequently, the risk factors associated with each medical condition would be “averaged” using more cases that are relatively lower cost. It has been estimated that the use of two years of diagnosis data could reduce MA risk scores by 1 to 2 percent.
Furthermore, an analysis by the Medicare Payment and Advisory Commission (MedPAC) also showed that the occurrence of having a chronic condition coded in one year and not in the subsequent year occurred more often for FFS beneficiaries than MA enrollees. This aligns with the notion that there are stronger incentives for complete coding in the MA environment since payments are directly tied to diagnosis coding. These differential coding rates between MA and the Medicare FFS program can cause Medicare payments to MA plans to be larger than the amount Medicare would have spent if the MA enrollees were enrolled in FFS Medicare. Using two or more years of diagnostic data could reduce the effects from the difference in coding patterns between MA and FFS Medicare, as more enrollees would be identified with chronic conditions in the FFS data.
Additional Considerations from Using Two or More Years of Diagnosis Data
CMS has so far not decided to increase the number of years of FFS and MA diagnostic data available to estimate the CMS-HCC model and calculate MA risk scores. Even if it did, there are still some issues to be determined regarding how to implement the provision. For example, because two years of diagnosis data would not be available for MA enrollees in their first or second year of Medicare eligibility, an alternative approach would be needed for them. It may be possible to estimate a risk score only using demographic information, which is currently done for enrollees in their first year (since they will not have a full year of diagnosis information available). It may also be possible to use one year of diagnostic data for enrollees in their second year of Medicare eligibility.
As discussed above, presumably, if CMS were to move forward with this provision, two or more years of diagnosis data would also be used to identify which MA enrollees have specific chronic conditions, as well as to calculate the level of risk adjustment factors. If this were the case, it is likely that MA plans would have to reconsider how they review their medical records. For example, there would be more medical record information that would need to be reviewed each year to ensure that the correct diagnosis codes are associated with each member and that they are all supported in the documentation. However, because the number of clinical encounters required to document the condition could be stretched across 24 (or more) months rather than 12 months, it is not clear how this change would actually impact the administrative burden of data collection. Any potential efficiencies in beneficiary interactions would need to be balanced with the desire to ensure adequate disease management efforts (e.g., a health plan may still conduct annual assessments even if all of the diagnosis codes have been identified and validated).
Conclusion
This concludes our review of key provisions in the 21st Century Cures Act that effect MA risk adjustment and payment amounts. All of these provisions will require further monitoring by plans, as the impact to a specific plan will depend on the case mix of enrollees for each plan, specifically the distribution of the medical conditions within their enrollee population. It can be expected that CMS will also be evaluating the impacts from these provisions, to determine if any adjustments need to be made over time. In the case of expanding the time period for including diagnosis codes in the CMS-HCC model and risk adjustment factor calculations, CMS may decide to revisit this issue in the future.