Page 84

84

solely for PACE organizations and will continue to use the historical linear slope methodology and average risk scores from 2016 through 2020 as proposed. Comment: Multiple commenters supported continuing to apply separate normalization factors for PDPs and MA-PD plans but opposed the proposed multiple linear regression methodology for calculating the factors. Commenters suggested that CMS phase in the proposed methodology or adopt alternative methodologies that they believe are more supportable or that maintain a similar degree of difference between the PDP and MA-PD normalization factors as exists under the methodology adopted for CY 2025 (i.e., approximately 12 percent). Response: As with the CMS-HCC model normalization factors, the RxHCC model normalization factors are technical adjustments applied to risk scores in the payment year and are intended to maintain an average risk score at 1.0. The average Part D risk score changes each year due to underlying trends that reflect changes in the health status and demographic characteristics of the population, and coding practices, compared to the denominator year. Therefore, when a risk adjustment model predicts expenditures in years other than the denominator year, the average Part D risk score may no longer be 1.0, as it was in the denominator year. Accordingly, a technical adjustment must be applied to risk scores to account for the risk score changes between the denominator year and payment year that are attributable to changes in demographic characteristics of the population, and reported health status, in order to maintain the average Part D risk score at a 1.0. Maintaining an average risk score of 1.0 across the Part D program in the payment year is essential to help keep the beneficiary premium at the appropriate proportion of aggregate plan payment and is a longstanding goal of Part D normalization.18 The most recent risk score data (provided in Table III-12 in the CY 2026 Advance Notice) suggests that continuing to use the historical linear slope method or another method with similar results would lead to normalization factors that underestimate the average MA-PD risk score and overestimate the average PDP risk score in 2026. This under- and over-normalization would reduce the accuracy of Part D payments and be inconsistent with the long-standing goal of Part D normalization. We are finalizing the multiple linear regression methodology (described in detail in Attachment III, Section G of the CY 2026 Advance Notice) because we believe it results in a more reasonable prediction of PDP and MA-PD risk scores in the payment year, making it more likely that the average payment year risk score across the Part D program will be 1.0. As commenters noted, the historical linear slope methodology did not account for risk score differences between PDPs and MA-PD plans in the denominator year because the calculation applies the trend to a denominator year risk score of 1.0. The multiple linear regression methodology more reasonably reflects the differences between PDP and MA-PD risk scores in the denominator year, projecting

18 See for example the CY 2010 Advance Notice p 7: https://www.cms.gov/medicare/health- plans/medicareadvtgspecratestats/downloads/advance2010.pdf.