The 2026 RxHCC model cut the heart failure coefficient 26 percent, because negotiated prices lowered the drug spending it predicts. In the same benefit year, the Part D redesign moved catastrophic-phase liability to 60 percent of cost. Two correct policies produced a combined signal pointing away from guideline therapy.
Key Takeaways
- RxHCC 186 fell from 0.183 to 0.135, a 26 percent decline. Atrial arrhythmias fell 58 percent. Both track the maximum fair prices that took effect for initial price applicability year 2026.
- Three of the first ten selected drugs are heart failure therapies. Entresto, Farxiga and Jardiance together absorbed a disproportionate share of the recalibration.
- In the same year, plan liability in the catastrophic phase moved from 20 percent to 60 percent while federal reinsurance for applicable brand drugs fell from 80 percent to 20 percent.
- The two facts are almost never presented together. Combined, the payment signal for managing heart failure with drugs got smaller in the year the cost of doing so shifted onto the plan.
- Risk adjustment does not keep up with new drugs. Finerenone gained its heart failure indication on July 11, 2025 and will not reach the model until roughly the 2029 or 2030 calibration cycle.
- The coefficient is calibrated on undertreatment. Real-world SGLT2 inhibitor use runs between 12 and 28 percent of eligible patients, so the model prices a benefit almost nobody is delivering.
- The economics are almost entirely HFrEF. Preserved ejection fraction is more than half of all heart failure and carries the larger remaining treatment gap, while the cost literature has modeled it least.
- The 2026 HFpEF consensus pathway writes affordability into the treatment algorithm, naming finerenone the preferred agent and offering spironolactone where cost is prohibitive.
- Quality measurement is under construction. PQA convened a heart failure technical expert panel in March 2026, and three of the four pillars remain unmeasured in Star Ratings today.
What changed, in one paragraph
CMS recalibrated the RxHCC model for payment year 2026 on 2022 diagnoses and 2023 prescription drug event data, with gross drug costs re-mapped to reflect the first ten negotiated prices. Because the model predicts plan liability for drugs, and the drugs got cheaper, the coefficients attached to the conditions those drugs treat went down. Heart failure was among the most affected conditions, for the straightforward reason that three of the ten selected drugs treat it.
This is a companion to why Medicare risk adjustment is built for reactive care across disjointed medical and pharmacy benefits, which sets out the structural relationship between the two models. This piece is about a single year, and about what happens when a system built to pay for disease that has arrived meets a year in which the price of preventing its consequences changed.
Why did RxHCC coefficients decrease in 2026?
Because the model is doing its job. The mechanism is not controversial. Milliman documented it in a March 2025 analysis of the proposed model changes, offering an illustrative worked example in which average monthly gross cost per heart failure patient falls from $514 to $348 once Entresto is repriced, and attributing the observed risk score movement for heart failure members to the negotiated prices.
The prices themselves are substantial.
| Drug | List price, 2023 | Maximum fair price | Reduction | Part D gross, 2023 |
|---|---|---|---|---|
| Entresto sacubitril and valsartan | $628 | $295 | 53 percent | $3.43 billion |
| Farxiga dapagliflozin | $556 | $178.50 | 68 percent | $4.34 billion |
| Jardiance empagliflozin | $573 | $197 | 66 percent | $8.84 billion |
Exhibit 1. Source. CMS fact sheet on negotiated prices for initial price applicability year 2026. List prices reflect wholesale acquisition cost on 2022 prescription fills, for a 30-day supply. Entresto has since been removed from the selected drug list effective January 1, 2027 following generic entry, so its 2026 maximum fair price applies to one year only.
CMS itself acknowledged the directional risk in the CY2026 Rate Announcement, noting that without substituting the negotiated prices the model would overestimate expected plan liability for conditions treated with those drugs, and would underestimate relative costs for conditions treated with drugs outside the program. The substitution was the right call. Its consequence for a single condition is the subject here.
The other half of the year
While the coefficient fell, the structure of Part D plan liability changed underneath it.
The signal got smaller
The exposure got bigger
Exhibit 2. Source. Coefficients from CMS CY2025 Rate Announcement Table VIII-1 (2021/2022 calibration, p. 169) and CY2026 Rate Announcement Table VIII-1 (2022/2023 calibration, p. 129), community non-low-income aged 65 and over. Benefit parameters from the Final CY2026 Part D Redesign Program Instructions. Plan liability is 60 percent across all covered drugs. Reinsurance is 20 percent for applicable brand drugs and 40 percent for non-applicable and selected drugs. The annual out-of-pocket threshold moved from $2,000 to $2,100.
Kazi and colleagues modeled the shift across four cardiovascular regimens, one of which is quadruple therapy for reduced ejection fraction heart failure, and found the plan share of total annual drug cost rising from a range of 16 to 53 percent in 2022 to a range of 60 to 78 percent in 2025. The plan now holds most of the cost of a regimen whose associated risk weight declined.
Both movements were internally correct. Neither model nor benefit rule was built to notice the other, which is the general pattern described in the companion piece as a calibration moat.
A calibration moat exists wherever two payment models are fitted against disjoint cost pools for the same population. Both models can be individually well specified and accurate within their own pool. The limitation appears only in the space between them, where a substitution from one pool to the other remains invisible in both dependent variables.
Does risk adjustment keep up with new drugs?
Not quickly. Avalere describes the mechanism directly. In a September 2025 note they write that "the RxHCC model is calibrated on historic drug costs and is typically updated only every two to three years," and in the sentence following, that "due to the infrequent updates, the RxHCC model may be slow to incorporate the expected cost of new drug therapies." Costs enter only after a drug is covered, prescribed, and captured in the calibration window.
Finerenone is the test currently running. Kerendia received its heart failure indication on July 11, 2025, covering ejection fraction at or above 40 percent, which opens roughly half of the prevalent heart failure population. CMS dashboard data already shows average Part D spending of $4,492.82 per beneficiary in 2024 across 88,208 beneficiaries, before the heart failure indication existed. The drug is not eligible for negotiation until roughly 2030 or 2031 under the seven-year rule, and it appears on neither the IPAY 2027 nor the IPAY 2028 selected drug lists.
The 2026 coefficient was fitted on 2023 prescription drug event data. A plan building a finerenone program for preserved ejection fraction in 2026 carries roughly $4,500 per patient in annual Part D liability against $294 in heart failure risk revenue, and the model will not observe that spending until the 2029 or 2030 calibration cycle at the earliest. The plan absorbs four calibration years of exposure before the payment system acknowledges the therapy exists.
The coefficient is calibrated on undertreatment
Every per-patient drug cost in this piece, including the Milliman example above, carries a limitation worth stating outright. Milliman labels that table illustrative, built on an assumed split between treated and untreated patients rather than measured directly. The deeper issue applies to the risk model itself. RxHCC coefficients are fitted on what plans actually paid, and plans actually paid for a population that largely does not receive guideline therapy.
The utilization evidence is consistent and sobering. In the Get With The Guidelines registry, roughly 20 percent of eligible patients hospitalized with reduced ejection fraction heart failure left on an SGLT2 inhibitor, and only 4.1 percent of participating hospitals discharged half or more of their eligible patients on one. A Veterans Affairs analysis found 14.6 percent. An analysis across 28 health systems found roughly 12 percent, with no health system exceeding about 25 percent. Ambulatory prescribing reached 28.5 percent in reduced ejection fraction by mid-2023. Summarizing the same picture, the 2026 prior authorization study reports that over 80 percent of eligible patients are never prescribed these medications, and that among those who are, only 30 to 50 percent take them regularly.
A coefficient fitted on that population prices a benefit nobody is delivering. Should a plan succeed in closing the gap, its per-patient drug cost would rise toward the guideline-concordant figure while its risk weight stayed anchored to the undertreated one. The model does not merely lag new drugs. It lags adoption of drugs it already knows about.
The anchoring runs in a second direction as well. Sacubitril and valsartan is indicated in reduced ejection fraction, which skews any Entresto-driven estimate toward later-stage disease. The economic literature carries the same tilt. I develop this further in an earlier analysis of the utilization gap, where access friction turns out to explain less of the shortfall than identification and inertia do.
Preserved ejection fraction is missing from the economics
Nearly every economic estimate cited in this space describes heart failure with reduced ejection fraction, abbreviated HFrEF. The hospitalization savings figure, the quadruple therapy cost models, the out-of-pocket analyses all draw on HFrEF populations.
Heart failure with preserved ejection fraction, or HFpEF, accounts for more than half of all heart failure cases, with outcomes comparable to HFrEF, and the 2026 ACC expert consensus pathway describes it as often underrecognized and a driver of substantial resource utilization. It is largely absent from the cost framing.
That absence is becoming expensive quickly, because the therapies arriving in HFpEF are the expensive ones. The consensus pathway names finerenone the mineralocorticoid receptor antagonist of choice in HFpEF on the strength of FINEARTS-HF. It brings incretin therapies into the algorithm, with semaglutide at ejection fraction of 45 percent and above and tirzepatide at 50 percent and above, in patients with body mass index at or above 30. Registry data shows SGLT2 inhibitor prescribing in mid-range and preserved ejection fraction rising from 4.2 percent to 23.5 percent between late 2021 and late 2023.
When the guideline writes price into the algorithm
One detail in that pathway is worth flagging here and is treated at length in a companion piece. Having named finerenone the preferred agent, the document continues that "if cost or tolerance are prohibitive, spironolactone is a reasonable alternative." A clinical guideline has written affordability into the decision tree as a branch criterion sitting alongside contraindication. What the evidence says patients actually do when price decides, which is not what the algorithm assumes, is the subject of a separate analysis.
The 2026 cardiovascular-kidney-metabolic guideline attaches economic value statements to individual recommendations at 2025 US prices, with a footnote that these "inform population- and health system-level decisions." For symptomatic heart failure with ejection fraction at or below 40 percent, adding an SGLT2 inhibitor is rated cost-effective with high certainty. For symptomatic preserved ejection fraction, the same drug class is rated indeterminate, on conflicting evidence.
Read that pair carefully, because it is easy to misuse. The verdicts are indication-specific, and they price the incremental value of adding the drug for heart failure on top of what the patient is already receiving. They do not say an SGLT2 inhibitor is hard to justify in a patient with preserved ejection fraction. The same guideline rates the drug cost-effective with high certainty in chronic kidney disease with or without type 2 diabetes.
That distinction matters because of who these patients are. Chronic kidney disease affects up to 60 percent of people with preserved ejection fraction heart failure and type 2 diabetes is present in up to half. For most of them the SGLT2 inhibitor is already justified on the kidney or metabolic indication at a cost-effective rating, and the heart failure diagnosis strengthens the case rather than weakening it. Anyone citing the indeterminate verdict as grounds to restrict the class has read it backwards.
The same guideline rates finerenone in chronic kidney disease with type 2 diabetes and albuminuria as not cost-effective at 2025 prices, which is the more consequential finding for a plan, since finerenone has no comorbidity indication to fall back on. There is a coverage gap alongside the pricing one. The consensus pathway notes that insurance coverage for cardiac rehabilitation and structured exercise therapy is unavailable in the United States for people with HFpEF, and calls that lack of reimbursement a significant care gap.
There is a risk adjustment consequence in that comorbidity picture, and it runs against the clinical logic. On the drug side, diabetes with complications carries an RxHCC coefficient of 0.567. Heart failure carries 0.135. Chronic kidney disease at stage 4 or 5 carries 0.009, and below stage 4 carries nothing. So a single drug prescribed for three overlapping conditions draws its drug-side risk revenue overwhelmingly from the metabolic diagnosis, at roughly four times what the heart failure diagnosis contributes, even though heart failure is the indication the guidelines press hardest.
So the population with the largest remaining treatment gap is the population the economic literature has modeled least, and the population whose preferred therapy the pathway itself flags as cost-constrained. For a plan building a 2027 view, that is where drug liability grows fastest and where the offset evidence is thinnest.
Quality measurement is being built, and it is not here yet
A reasonable objection holds that if payment cannot align the two sides, measurement might. The architecture is under construction.
On March 4, 2026 the Pharmacy Quality Alliance announced a technical expert panel to develop a health plan measure titled Evidence of Guideline-Directed Medical Therapy in Persons with Heart Failure, assessing prescription claims across all four pillars. The panel first met on April 22, 2026. The path from measure development through testing, endorsement and rulemaking into Star Ratings typically runs three to five years.
What is live today is the medication adherence measure for hypertension built on renin-angiotensin system antagonists, which captures sacubitril and valsartan by virtue of the valsartan component. Beta blockers, mineralocorticoid receptor antagonists and SGLT2 inhibitors sit outside it. Three of the four pillars are unmeasured in Star Ratings today. The pillar that is measured is captured incidentally, through a hypertension measure, rather than as heart failure management. Until the PQA measure arrives, a plan optimizing to the measure set it is actually scored on has little reason to build the program the guidelines describe. Plans that anticipate the measurement environment rather than waiting for it will be better positioned, which is the argument I made at greater length in the analysis of the utilization gap.
What the therapy is worth to the plan that cannot book it
Keykhaei, Fonarow and colleagues linked Get With The Guidelines heart failure records to Medicare Part A and Part B claims across 50,598 patients aged 65 and over hospitalized with reduced ejection fraction heart failure. Against partial guideline therapy, quadruple therapy lowered hospitalization expenditures by $9,780 per patient per year, alongside an 87 percent relative reduction in heart failure hospitalizations and a 61 percent reduction in all-cause hospitalizations. Across drug costs spanning $1,223 to $16,136 depending on regimen, the authors report net results from $8,556 in savings to $6,347 in net cost, with most regimens producing net savings.
| Flow | Lands in | Approximate annual amount |
|---|---|---|
| Part D plan liability, after negotiated prices, at 60 to 78 percent plan share | Part D | $3,600 to $4,700 |
| RxHCC 186 risk revenue | Part D | $294 |
| Avoided hospitalization spending | Part A and Part B | $9,780 |
| HCC 226 risk revenue, paid without regard to management quality | Part C | $5,160 |
Exhibit 3. Source. Hospitalization savings from Keykhaei et al., JAMA Cardiology, 2026, Get With The Guidelines linked to Medicare claims, 50,598 patients aged 65 and over hospitalized with reduced ejection fraction 2016 to 2020, measured against partial guideline therapy. Plan share range from Kazi et al., an ACC.23 moderated poster abstract published in the JACC supplement, 2023, volume 81, supplement 8, page 1671. Drug liability estimated from CY2026 maximum fair prices applied to a four-pillar regimen and is illustrative. Every figure describes reduced ejection fraction. No equivalent published estimate exists for preserved ejection fraction.
For an integrated plan the arithmetic is favorable, which is why the persistence of the divide is interesting. The $9,780 arrives in a different bid, in a later year, credited to a different actuary, and returns to the pharmacy side only through a capped rebate channel that competition spends on premium buydowns rather than on the clinical program that generated it. For a standalone drug plan alongside fee-for-service Medicare, the saving accrues to the Trust Fund and the plan receives nothing.
Erik's Hot Take
The Inflation Reduction Act lowered heart failure drug prices, which is the outcome the policy sought. The risk model correctly followed those prices downward. The benefit redesign shifted liability toward plans in the same year, which was also deliberate. Three correct policies produced a combined signal that points away from the therapy the guidelines recommend.
Worth sitting with, the model priced heart failure off a population that mostly is not on guideline therapy. A plan that fixes that is rewarded with higher drug cost and the same risk weight.
The gray space is that nobody is responsible for the combined signal, because no forum computes it. The white space is that the plans and vendors who model 2027 bids against the joint effect rather than against each piece separately will price heart failure populations more accurately than competitors still reading the two changes in isolation.
The structural relationship behind this year's numbers is set out in what a heart failure diagnosis is worth in Medicare risk adjustment. The upstream version of the problem, where neither model can represent the patient at all, is covered in whether Stage B heart failure risk adjusts.
Frequently asked questions
Why did the RxHCC heart failure coefficient fall in 2026?
CMS recalibrated the model on expenditures re-mapped to reflect the first ten negotiated drug prices. Three of those ten are heart failure therapies. Because the model predicts plan liability for drugs and that liability fell, the coefficient fell with it, from 0.183 to 0.135 in the community non-low-income aged segment.
How did the Part D redesign change plan liability in 2026?
The redesign eliminated the coverage gap and moved plan liability in the catastrophic phase to 60 percent, with federal reinsurance falling to 20 percent. Against an annual out-of-pocket threshold of $2,100, plans now hold a substantially larger share of the cost of expensive chronic regimens than they did before 2025.
Does risk adjustment keep up with new drugs?
Not quickly. The RxHCC model is calibrated on historic drug costs and recalibrated every two to three years, so a new therapy enters the model only after it has been covered, prescribed, and captured in a calibration window. Finerenone gained its heart failure indication in July 2025 and will not reach the model until roughly 2029 or 2030.
Are heart failure medications measured in Star Ratings?
Only indirectly today. The live medication adherence measure for hypertension, built on renin-angiotensin system antagonists, captures sacubitril and valsartan through its valsartan component. Beta blockers, mineralocorticoid receptor antagonists and SGLT2 inhibitors are not captured. The Pharmacy Quality Alliance convened a technical expert panel in March 2026 to develop a four-pillar guideline-directed medical therapy measure, which typically takes three to five years to reach Star Ratings.
Does the RxHCC coefficient reflect what guideline therapy actually costs?
No. The model is calibrated on what plans paid, and real-world SGLT2 inhibitor use runs between roughly 12 and 28 percent of eligible patients depending on setting. The coefficient therefore reflects a largely untreated population. A plan that closes the treatment gap incurs guideline-concordant drug cost against a risk weight anchored to undertreatment.
What is a calibration moat?
A calibration moat exists wherever two payment models are fitted against disjoint cost pools for the same population. Each model predicts its own pool well. Neither can price a substitution between them, because the substitution does not appear in either dependent variable. Better data integration does not resolve it, because the constraint is the model specification.
Method and sources
Coefficients and model parameters come from the CMS CY2026 Rate Announcement and CY2026 Advance Notice, with prior-year comparison from the CY2025 Rate Announcement. Coefficients reflect the community non-low-income aged 65 and over segment, and segment selection materially changes every figure shown.
Negotiated prices come from the CMS fact sheet on negotiated prices for initial price applicability year 2026, with list prices reflecting wholesale acquisition cost on 2022 prescription fills. Benefit redesign parameters come from the Final CY2026 Part D Redesign Program Instructions. Finerenone spending figures come from the CMS Part D Spending by Drug dashboard, which reports net of point-of-sale concessions through 2024. Commentary on the recalibration mechanism comes from Milliman's March 2025 analysis, whose cost table is labelled illustrative, and commentary on calibration lag from Avalere's September 2025 note. Both are vendor-published and flagged as such.
Prevalence, therapy selection and the cost branch points in preserved ejection fraction come from Kittleson et al., the 2026 ACC Expert Consensus Decision Pathway on management of heart failure with preserved ejection fraction, published in the Journal of the American College of Cardiology. Economic value statements come from the 2026 AHA, ACC, ADA and ASN guideline for the prevention, detection, evaluation and management of cardiovascular-kidney-metabolic syndrome, published in Circulation. Clinical and economic estimates come from Keykhaei et al. in JAMA Cardiology (2026) and from Kazi et al., a conference abstract published in the JACC supplement in 2023 rather than a full peer-reviewed paper. The Part C dollar figure in Exhibit 3 is derived rather than published, combining MedPAC's March 2026 average plan bid and average rebate, and should be treated as approximate.
This piece is analysis and commentary based on public sources and professional experience as of the date noted, and nothing in it constitutes legal, clinical, or financial advice. Descriptions of risk adjustment, benefit and formulary practices are general characterizations of common industry structures and are not directed at any specific company or plan. Interpretive conclusions are the author's opinion, offered for analytical discussion. All views reflect independent professional judgment and do not represent the views or positions of any current or former employer or affiliated organization.