Risk Adjustment
LEAD applies risk adjustment to account for differences in the expected costliness of each ACO’s aligned population. The risk adjustment methodology differs by beneficiary category, and LEAD introduces several innovations relative to ACO REACH and MSSP. (RFA §VIII.D)
Risk Adjustment Models by Beneficiary Category
| Category | LEAD Model | What Changed from REACH |
|---|---|---|
| Aged & Disabled (A&D) | CMMI HCC Prospective Risk Adjustment Model V1 (a modified version of the 2024 CMS-HCC V28 model, recalibrated to exclude High Needs populations) | REACH uses the standard CMS-HCC prospective model (Standard/NE ACOs) or CMMI-HCC concurrent V1 (High Needs ACOs only) |
| ESRD | 2023 CMS-HCC ESRD Risk Adjustment Model | Same as REACH |
| High Needs | CMMI HCC Concurrent Risk Adjustment Model V2 (updated from REACH’s V1 to reflect V28 architecture, recalibrated for the High Needs population only) | REACH’s concurrent model was only available to High Needs Population ACOs (a separate participant type). In LEAD, all ACOs get it for qualifying beneficiaries |
(RFA §VIII.D)
Granular model specifications (for example, coefficient updates, calibration detail, and year-by-year technical parameters) will be spelled out in LEAD methodology publications as CMS finalizes them; the RFA establishes the framework above.
New in LEAD Universal High Needs: In ACO REACH, concurrent risk adjustment was only available to High Needs Population ACOs (a distinct ACO type that had to meet separate eligibility criteria). In LEAD, the High Needs beneficiary category applies across ALL ACOs: any beneficiary meeting the criteria receives a separate benchmark using the Concurrent V2 model, regardless of the ACO’s type. This eliminates the structural disadvantage Standard ACOs faced with complex patients. (RFA §VIII.D)
High Needs Eligibility Criteria (RFA §VII.A)
A beneficiary qualifies as High Needs if they meet at least one of the following:
- Mobility impairment (identified via specific ICD-10 codes)
- Frailty evidence: claim for hospital bed or transfer equipment for home use
- Kim Claims-based Frailty Index score ≥ 0.35 (moderate to severely frail)
- Risk score ≥ 3.0 for A&D beneficiaries (CMS-HCC V28 or CMMI-HCC Concurrent V1)
- Risk score ≥ 0.35 for ESRD beneficiaries (2023 CMS-HCC ESRD model)
- Risk score 2.0–3.0 for A&D (or 0.24–0.35 for ESRD) AND 2+ unplanned hospital admissions in prior 12 months
- 45+ Medicare-covered SNF days in the past 12 months
New in LEAD Universal High Needs: In ACO REACH, concurrent risk adjustment was only available to High Needs Population ACOs (a distinct ACO type that had to meet separate eligibility criteria). In LEAD, the High Needs beneficiary category applies across ALL ACOs: any beneficiary meeting the criteria receives a separate benchmark using the Concurrent V2 model, regardless of the ACO’s type. This eliminates the structural disadvantage Standard ACOs faced with complex patients. (RFA §VIII.D)
Coding Intensity Caps (RFA §VIII.D)
| Category | Cap | Reference Year | REACH Comparison |
|---|---|---|---|
| A&D / ESRD | 3% each | Static BY3 (CY 2026) | REACH: symmetric 3% cap vs. a fixed reference year (CY 2022) |
| High Needs | 3–8% (TBD) | Static BY3 (CY 2026) | 10% (High Needs ACOs only) |
Key Insight: The squeeze under LEAD is less about REACH using a “rolling” year and LEAD using a fixed year (both programs anchor the cap to a stated calendar reference) than about how risk scores evolve against that anchor over many performance years. LEAD’s split between the recalibrated A&D model and the concurrent High Needs track changes how scores flow into benchmarks relative to the cap; as demographic and coding-related score drift accumulates against static BY3, the headroom under the 3% A&D/ESRD cap can tighten faster than many ACOs experienced under REACH’s prior calibration.
AI-Inferred Risk Adjustment (RFA §VIII.D)
LEAD introduces a phased transition to AI-inferred risk scores for A&D beneficiaries:
| Year | AI Weight | HCC Model Weight |
|---|---|---|
| PY 2028 | Shadow test only (0%) | 100% |
| PY 2029 | 33% | 67% |
| PY 2030 | 67% | 33% |
| PY 2031+ | 100% | 0% |
Key Insight: AI-inferred risk scores could fundamentally change the risk adjustment landscape. If validated, they may be less responsive to coding intensity efforts and more predictive of actual spending. Organizations that have invested heavily in HCC coding programs should plan for the possibility that this investment yields diminishing returns as AI scores phase in.
Frequently Asked Questions
The A&D prospective model is trained on a population that excludes High Needs beneficiaries. This means the model’s coefficients are tuned for the non-complex A&D population, improving prediction accuracy for that group. It also means A&D risk scores will be slightly lower than under the standard CMS-HCC model that includes the full population.Will AI risk scores completely replace HCC coding?How does the coding intensity cap work?
Have questions? Get in touch with Brad via LinkedIn. Contact Brad →Generated by AI — may contain errors© 2026 LEAD Model Knowledge Center. All Rights Reserved. Contact Us
Not immediately. CMS has indicated the AI phase-in is subject to testing and validation and may be extended. For PY 2027–2028, there is no AI impact. Even at full adoption, beneficiaries will still have HCC codes; the AI model will use them as inputs alongside other data sources.
The cap limits how much an ACO’s average risk score can grow relative to the BY3 reference year. If an ACO’s PY risk score exceeds the BY3 risk score by more than 3% (for A&D/ESRD), the excess is capped. The cap is applied symmetrically in ACO REACH (both up and down), but LEAD’s cap methodology and symmetry details will be specified in methodology papers.
