Normalized risk scores for a Standard REACH ACO tend to increase throughout the performance year, averaging around a 1% lift between Q1 and final settlement.

Accountable Care Organizations (ACOs) participating in the ACO Realizing Equity, Access, and Community Health (ACO REACH) model may struggle to understand all the moving parts of the financial benchmark—normalized risk scores do not have to be one of them. We recently conducted a study to better understand the development of risk scores throughout the performance year. We found that, for a Standard REACH ACO, the average raw risk score for the population decreases throughout the performance year mainly due to beneficiary mortality, but the normalized risk score tends to increase, averaging around 1% between Q1 and final settlement. This paper provides some key findings from our study.

ACO REACH normalized risk scores

In ACO REACH, the payment risk score (i.e., the risk score that will be used to adjust the final benchmark) is calculated separately for the two beneficiary categories—aged and disabled (AD) and end-stage renal disease (ESRD). A three-step process is used to convert a raw risk score produced by the Centers for Medicare and Medicaid Services Hierarchical Condition Category (CMS-HCC) model into a payment risk score: (1) normalization, (2) risk score capping, and (3) the application of the Coding Intensity Factor (CIF).¹

The normalization factor is computed as the average raw risk score of the REACH National Reference Population (Reference Population) within the same year. Dividing beneficiaries’ raw risk score by this normalization factor ensures that the average risk score of the Reference Population is 1.0.

Throughout a given performance year (PY), a REACH ACO’s emerging normalized risk score will change due to the shift in its raw risk score relative to the shift in the raw risk score of the Reference Population (which determines the normalization factor).

Our study also examined how REACH ACOs’ risk scores in PYs 2022 and 2023 change from Q1 to year-end final settlement (“S2” in REACH nomenclature). Key insights from this study are:

• The ACO’s raw risk scores may change throughout the year for a variety of reasons discussed later, but the two main levers are (1) diagnoses capture runout, and (2) beneficiary decrement (primarily due to mortality). We found that the impact of beneficiary decrement outweighs the impact of diagnoses capture runout, resulting in a net decrease in raw risk score as the PY progresses.
• Once raw risk scores are normalized, ACOs’ normalized risk scores exhibit a net increase of approximately 1% from Q1 to year-end. Higher-risk ACOs tend to show a larger increase than lower-risk ACOs. The study suggests that, on average, REACH ACOs’ raw risk scores decrease slower than that of the Reference Population (leading to an increase in normalized risk scores throughout the performance year).

In the remainder of this paper, we explain the various aspects of risk score development, both raw and normalized, in detail to help the reader understand risk score dynamics in ACO REACH.

Risk score claims and diagnoses periods

In each performance year (PY), CMS first releases a preliminary risk score report and benchmark report to each REACH ACO prior to the start of the PY, and then provides updated quarterly benchmark reports (QBRs) and two rounds of settlement reports (S1 preliminary and S2 final) in May and August in the subsequent calendar year. Specifications of these reports are summarized in Appendix 1 for PY 2023 as an example.

In each update of the risk score and benchmark report, the normalization factor is also calculated and incrementally updated by CMS, based on observed diagnoses and eligibility data submitted for the Reference Population.

ACO REACH Reference population

In ACO REACH, the normalization factor is computed as the average raw risk score of the Reference Population, which consists of the subset of Medicare fee-for-service (FFS) beneficiaries who meet the eligibility criteria for the REACH program, outlined in Appendix 2.

The Reference Population is reassessed monthly and includes newly eligible Medicare age-ins of that month. For this reason, the Reference Population is considered an “open” cohort. In a given PY, the average raw risk scores of the Reference Population (and consequently the normalization factor) would change from Q1 to S2 due to the following reasons:

• Beneficiary-level risk score change:
• The impact of additional diagnosis period runout
• Change in demographic status, such as ESRD, Medicaid, and institutional status change
• Population mix change at the aggregate level:
• The influx of newly eligible beneficiaries with only demographic risk scores (which are lower than risk scores that also include HCCs)
• Beneficiary decrements (death, enrollment in Medicare Advantage, etc.)

In aggregate, the seasonality of the ACO REACH normalization factor reflects the combined impact of these factors.

ACO REACH aligned population

If the Reference Population and the aligned population exhibit the same raw risk score seasonality, then the impact would cancel out, resulting in the aligned population demonstrating a normalized risk score seasonality of 1.0. Said differently, seasonality in the aligned population’s normalized risk score is a direct result of the differences in the raw risk score seasonality of the Reference Population and of the aligned population. This section investigates the differences between the two populations.

A beneficiary can be aligned to a REACH ACO through either claims-based alignment or voluntary alignment.² We focus on claims-aligned beneficaries for this white paper.

For claims-based alignment, a beneficiary must meet the same criteria for the Reference Population (described in Appendix 2), as well as the following criteria:

1. The beneficiary is alignment-eligible on January 1 of the performance year, meaning the beneficiary meets the Reference Population criteria and lives in a county in the ACO’s service area.
2. The beneficiary received the plurality of their Primary Care Qualified Evaluation and Management (PQEM) services during the alignment period (i.e., the two-year period ending in June prior to the PY) from the ACO’s participating providers.³
3. The beneficiary is not aligned in the Medicare Shared Savings Program (MSSP) or other Medicare value-based initiatives that take precedence over ACO REACH for purposes of beneficiary alignment.

These definitions demonstrate that the ACO REACH aligned population is a subset of the Reference Population. Key differences between the two are:

• The Reference Population includes $0 claimants in the alignment period; the aligned population excludes them by design. This, and the fact that an ACO’s alignment window overlaps somewhat with the diagnosis capture window, leads to ACOs generally having higher raw risk scores than the Reference Population.
• For an ACO in a given performance year, the pool of claims-aligned beneficiaries is a “closed” cohort whereas the Reference Population is an open cohort. The ACO’s claims-based aligned population can only decrease in size as the PY progresses (e.g., mortality).

From Q1 to S2, the factors that impact the raw risk score of the Reference Population listed in the previous section apply to the ACO’s claims-based aligned population as well, except for the impact of the newly eligible beneficiaries. Being a closed cohort means that, unlike for the Reference Population, the average risk score of the ACO will not be diluted by the influx of beneficiaries with lower (demographic only) risk scores. Everything else equal, the claims-based aligned beneficiaries’ risk scores will decrease slower compared to the Reference Population.

These differences between the Reference Population and the claims-aligned population result in the raw risk scores of the two populations developing differently throughout the PY, leading to a gradual increase in the ACO’s normalized risk score as the performance year progresses.

Conclusion: Why does this matter?

For many ACOs, monitoring and projecting performance in a risk-sharing arrangement like ACO REACH with so many moving parameters can be difficult.

ACOs rely on the quarterly risk score and benchmark reports to estimate the final performance year settlement. Thus, it is crucial to understand how normalized risk scores can be expected to evolve throughout the PY and to estimate the seasonality impact of risk scores.

Our study on ACO REACH risk scores suggests two key observations. First, the impact of beneficiary decrement outweighs the impact of diagnoses capture runout, resulting in a net decrease in an ACO’s raw risk scores from Q1 to year-end. Second, a typical ACO’s normalized risk scores exhibit a net increase from Q1 to year-end. The average increase is approximately 1%, but higher-risk ACOs may see larger increases compared to their lower-risk counterparts. We will continue this study as risk scores in PY 2024 are based on a blend of v24/v28 CMS-HCC risk score models.

Please do not hesitate to contact your Milliman consultant to see the detailed results of Milliman’s study, or to understand how your ACO’s risk may develop in relation to the Reference Population to help inform better, more stable financial projections.

Appendix 2: Eligibility Criteria for the ACO REACH National Reference Population⁵

• The beneficiary is alive on the first day of the month.
• The beneficiary is enrolled in Part A and Part B.
• The beneficiary is enrolled in traditional FFS Medicare (e.g., not enrolled in Medicare Advantage or a Medicare managed care plan).
• The beneficiary has Medicare listed as the primary payer.
• The beneficiary is a U.S. resident.

¹ For a description of each of these three steps, see https://www.milliman.com/en/insight/interactions-between-cif-3-percent-risk-score-floor-ceiling-aco-reach.

² Information for this section is based on Appendix B (Beneficiary Alignment Procedures) of https://www.cms.gov/files/document/aco-reach-py24-financial-operating-guide.pdf.

³ Appendix B of the PY2024 Financial Operating Guide also requires the beneficiary to be claims-aligned (i.e., have at least one PQEM service provided by Medicare FFS in the alignment period). Meeting requirement #2 would automatically make the beneficiary claims-aligned.

⁴ Source: Table 5.1.1 of PY2023 ACO REACH Reporting and Data Sharing Overview, Table 1.0 of PY2023 ACO REACH Quarterly Benchmark Report Information Packet, both available on 4i. See https://4innovation.cms.gov/secure/knowledge-management/view/491 and https://4innovation.cms.gov/secure/knowledge-management/view/1375.

⁵ CMS. FAQs: Benchmarking. Retrieved May 22, 2024, from https://app1.innovation.cms.gov/s/model-faq/a9Bt00000004CBgEAM/benchmarkingfinancial-methodology.