As stakeholders and legislators examined how to redesign the Medicare Part D program, a principal priority was protecting seniors financially by providing an annual $2,000 out-of-pocket maximum. Of equal importance was reducing the government reinsurance payments which had increased dramatically since the inception of the program in 2006. These two primary changes, in tandem with other smaller regulatory tweaks, would result in a sizeable increase in health plan liability. To avoid significant premium increases, and by virtue of the basic payment mechanics of the program, the offset to this increase in plan liability would be a substantial increase in the risk-adjusted direct subsidy payments. With the signing of the Inflation Reduction Act (IRA), health plans and risk-bearing providers are now focusing intently on preparing for this exact outcome in 2025. The considerable increase in direct subsidy payments and insurance risk that this represents for plans makes accurate Part D risk scores more critical than ever.

On January 31, 2024, the Centers for Medicare and Medicaid Services (CMS), released a proposed 2025 Part D risk adjustment model alongside the Advance Notice of Methodological Changes for Calendar Year 2025. This paper provides insights into the proposed model updates, comparisons to the CMS-HCC model used for Part C and provides health plans and Part D sponsors a framework to approach a holistic (Part C and D) risk score strategy. Some noteworthy considerations include:

1. A handful of disease states continue to drive Part D risk scores with diabetes and cancer contributing materially more to the risk score under this new model while heart, lung and psychiatric conditions contribute less.
2. The proposed model significantly shifts the risk score from the demographic to condition-based component.
3. When comparing the CMS-HCC and RxHCC models, there are some clear gaps that should not be overlooked. A risk adjustment strategy focused purely on Part C risk scores will almost certainly result in blind spots for Part D risk revenue optimization.

The remainder of this paper builds on these key pillars, highlighting specifics by disease category and member cohort, ending with key considerations for health plans into 2025 and beyond.

The most notable changes in the 2025 Advance Notice

The Advance Notice detailed the proposed changes to the Part D risk adjustment model to account for the upcoming benefit redesign. The model coefficients were significantly adjusted to reflect the post-IRA plan liability and the calibration/denominator years were updated, but essentially all other aspects of the model remained untouched (for instance, condition categories did not change in number or classification). Notably this means that the same model limitations exist as in prior years, but levers to control costs and maintain coding accuracy are far more impactful. These limitations include:

1. The model is calibrated on 2021 diagnoses and 2022 expenditure data from prescription drug event (PDE) records. Predicted costs will not account for any changes seen after 2022, including drugs new to the market, utilization for different medical conditions, patent losses, or other changes in the market.
2. The model still predicts expenditures prior to rebates. Although bid mechanics attempt to account for this in aggregate, it could still impact the appropriateness of reimbursements for some conditions.
3. The model continues to predict expenditures using only medical diagnoses. It is often the case that the conditions that risk adjust on Part D, but not on Part C, can be managed through prescription drugs alone and are not necessarily accompanied with regular doctor visits. This increases the likelihood that these conditions will not be coded every year and will fail to be reflected in the risk score. Pharmacy claims can be reviewed on a regular basis to detect gaps in coding of this nature.

A novel feature proposed by CMS for 2025 is the separate development and application of normalization factors for MAPD and PDP plans. To support this approach, CMS points to the diverging coding and cost trends of the two populations, and the resulting model overprediction (MAPD) and underprediction (PDP) of costs.  These normalization factors are materially different, and the methodological change represents a headwind for MAPD plans and a net positive for PDP plans.

This paper focuses on the proposed RxHCC model, but CMS also released the coefficients for an alternative model calibrated on 2018 diagnoses and 2019 expenditures with a 2020 denominator for reference. This alternative model generally has a lower proportion of the score contributed by heart conditions and diabetes and a higher proportion contributed by neurological and psychiatric conditions. Of course, with a larger gap between the denominator and payment year, the normalization factors for this model are larger adjustments and it will not account for changes in the market after the COVID-19 pandemic.  

Drivers of Part D reimbursements

In the past, changes to the RxHCC model were not accompanied by major overhauls of the benefit design which made changes in risk score simpler to interpret. With this RxHCC model change, the average risk score for non-low income beneficiaries decreases, while the average risk score for low income beneficiaries increases. This is driven by the increase in plan liability for low income beneficiaries being relatively larger than the increase for non-low income beneficiaries. This distinction does not mean that plans will be paid less for non-low income beneficiaries. While direct subsidy payments will increase for all members, payments for low income members will increase by more than those for non-low income members, consistent with the additional liability plans are taking on for those members. This is due to a shift in low income subsidies to plan liability with the elimination of the coverage gap. While health plans will need to prepare 2025 bids considering their own specific situations to understand the financial implications, there are definitive conclusions regarding what will be important for risk scores going forward that are broadly applicable to all MAPD and PDP plans.

A handful of disease categories continue to drive Part D risk scores under the new model. Among the aged population, regardless of a member’s low income status, four major disease categories along with demographics account for close to 75% of the risk score: diabetes, heart, cancer, and lung. Diabetes will represent nearly the same cohort on Part C as Part D and cancer will be more restricted on Part D. Therefore, longstanding efforts to accurately code diabetes and cancer should translate over well to Part D risk scores. This is good news for plans because the proposed model sees a substantial increase in the proportion of the score contributed by these two disease categories. Conversely, heart, lung and psychiatric conditions see a decrease on average. Hypertension, asthma and supraventricular tachycardiam are highly prevalent conditions that should be evaluated for inaccurate coding to ensure that risk scores on Part D are appropriately compensating costs.    

A further four disease categories account for nearly 90% of the score: musculoskeletal, neurological, metabolic, and infections. Notably, disorders of lipoid metabolism are highly prevalent, but only impact low income scores because CMS has set the coefficients to 0 for non-low income members.  Kidney and cognitive RxHCCs now have coefficients of 0 in the proposed model as well.

Finally, the 2025 model sees a material shift away from the demographic portion of the score and toward the condition-based component. This underscores the heightened importance of accurately capturing Part D risk in medical records going forward.

How Risk Adjustment on Part D relates to Part C

In the past it has been conventional wisdom that having a well-rounded and optimized risk adjustment program on Part C would produce adequate Part D risk scores as a byproduct. This sentiment is both due to a large overlap in ICD-10 codes that flow to payment condition categories in both models and the historically relatively low financial impact of Part D risk scores as compared to Part C risk scores. Given the recent clinical reclassification on Part C with the release of the 2024 v28 model and the anticipated unprecedented increase in Part D direct subsidy payments in 2025, it is worth closely re-examining this notion.

The relationship between the 2024 HCC and 2025 RxHCC models is complex, but a simple categorization of the 84 RxHCCs can be revealing.

As the table illustrates, 23 of the 84 RxHCCs consist of nearly all non-payment ICD-10s on Part C or have significant overlap, but with some of the most prevalent codes not contributing to Part C payments. This suggests that a sophisticated risk adjustment program solely focused on Part C may have substantial blind spots. Some notable examples are consolidated in Exhibit 2.

Although more than a third of the Part C model HCCs consist of ICD-10 codes that do not contribute to payments on the Part D side, these are more concentrated among less prevalent conditions. In general, the Part C model tends to be more selective and focused on higher severity/specificity codes. Despite what the count of ICD-10 codes might suggest, it is more likely for a code that contributes to Part C payments to also contribute to Part D payments than the other way around. A risk adjustment program focused primarily on Part C will assist in accurately coding many conditions contributing to Part D risk scores. However, this type of program may fail to identify and rectify the inaccurate coding of highly prevalent and financially impactful conditions on the pharmacy side in the coming years.      

Next steps for enhanced risk adjustment

There is robust evidence that uncoded patients tend to be, on average, sicker than the coded ones with materially higher ED and inpatient utilization, longer stays, and lower PCP/outpatient utilization. Furthermore, coded patients show high rates and levels of clinical improvement as well. Health plans need to be appropriately reimbursed for healthcare costs to offer richer benefits and better access to care.

A preliminary review of the proposed 2025 Part D risk adjustment model reveals two key areas for plans and risk-bearing providers to focus on to achieve these goals:

1. Ensure that the conditions that contribute the most to Part D risk scores are being coded appropriately: diabetes, hypertension and heart failure, lung conditions, cancer, and atrial arrhythmias. PDE records may be particularly useful in identifying diseases that aren’t being coded for hypertension, lung, and arrhythmias.
2. Put further particular focus on the prevalent RxHCCs and ICD-10 codes in Exhibit 2 that risk adjust on Part D, but not on Part C. Members with these conditions may be well-managed with prescriptions and fail to have seen a physician recently. This could result in inaccurate coding for these and other conditions since the risk adjustment models are based only on medical diagnoses.