CMS Innovation Center value-based model data requirements
The CMS Innovation Center has launched over 50 value-based care models since 2010, and every one of them requires participating organizations to submit clinical, financial, and operational data that meets specific quality thresholds. Most participants underestimate what those data requirements actually demand, and the gap between what CMMI expects and what health systems can deliver is where model performance collapses.
The CMS Innovation Center has tested more than 50 payment and service delivery models since its creation under the Affordable Care Act in 2010. Each model carries data submission requirements that go far beyond standard claims reporting. Participating organizations must deliver clinical quality measures, patient-reported outcomes, cost benchmarks, social determinant indicators, and operational metrics on timelines that leave little room for data remediation after the fact.
When a model fails to demonstrate savings or quality improvement, CMS does not always blame the intervention design. Increasingly, the agency points to data quality as a root cause. If your organization is participating in or evaluating a CMMI model, the data requirements are the first thing you need to understand, not the payment methodology.
What the CMS Innovation Center actually requires
CMMI models are not uniform in their data demands, but they share a common architecture. Every model requires some combination of the following: claims and encounter data, electronic clinical quality measures (eCQMs), patient experience surveys (typically CAHPS), beneficiary-level risk adjustment data, and increasingly, social determinant of health (SDOH) screening results.
The ACO REACH model, for example, requires participants to submit beneficiary-level demographic data, claims-linked quality measures, and prospective beneficiary alignment files. The Making Care Primary (MCP) model, launched in 2024, adds requirements for care management documentation, behavioral health integration metrics, and SDOH screening completion rates.
The Kidney Care Choices (KCC) model requires transplant referral tracking, home dialysis utilization data, and patient-reported outcome measures. As MedPageToday coverage noted in its analysis of the kidney transplant system, the data infrastructure required to track patients across the transplant continuum is fragmented across multiple organizations, making longitudinal data assembly a significant operational challenge.
The gap between what CMS expects and what organizations submit
CMS expects data that is complete, timely, standardized, and attributable to specific beneficiaries. What it receives is often none of those things.
A 2024 House panel heard testimony from multiple experts arguing that CMS needs to do a better job designing value-based care models that account for the real-world data capacity of participating practices. MedPageToday reported that witnesses told the panel value-based care is a sound concept undermined by reporting complexity and inconsistent data infrastructure across practice sizes.
The problem is structural. Small and mid-size practices participating in CMMI models often lack the EHR configuration, data engineering staff, and quality reporting workflows to meet submission requirements without significant manual effort. A June 2026 bill passed by the House aimed to ease quality reporting requirements for Medicare ACOs, acknowledging that current reporting burdens are driving some practices out of value-based arrangements entirely.
This is not a technology problem alone. It is a data trust problem. The data that reaches CMS is only as reliable as the systems that generate, transform, and transmit it.
Why CMMI data trust matters more than CMMI data volume
CMS does not just want data. It wants data it can trust enough to make payment decisions on.
Consider what happens when CMS calculates shared savings for an ACO. The agency compares actual spending against a benchmark, risk-adjusts for patient complexity, and then determines whether the ACO has generated enough savings to earn a payment or incurred enough losses to owe money back. Every variable in that calculation depends on data: diagnostic coding accuracy, claims completeness, quality measure reliability, and beneficiary attribution precision.
If the underlying data has provenance gaps, meaning CMS cannot verify where it came from or how it was transformed, the entire calculation becomes suspect. If consent records are incomplete, patient-reported outcomes become unusable. If recency is poor, meaning data arrives 60 to 90 days after the measurement period, CMS cannot use it for prospective adjustments.
A September 2025 study covered by MedPageToday found that all clinical quality outcomes tested favored value-based payment models over fee-for-service in Medicare Advantage. But the study's validity rested entirely on the quality of the underlying data. When the data feeding value-based models is untrustworthy, even models that genuinely improve care cannot demonstrate that improvement.
Key statistics
CMMI has launched more than 50 models since 2010, each with distinct data submission requirements covering claims, quality measures, patient experience, and SDOH indicators.
The Making Care Primary model requires SDOH screening completion data, behavioral health integration metrics, and care management documentation, expanding data requirements beyond traditional claims-based reporting.
CMS risk adjustment calculations use HCC (Hierarchical Condition Category) coding, where a single missed or inaccurate diagnosis code can shift a beneficiary's risk score by 10 to 30%, directly affecting shared savings calculations.
SuperTruth's work with imaware demonstrated that standardizing 105,000 diagnostic records reduced processing time from 3 weeks to 2 hours, a 95% reduction, illustrating the scale of data remediation required when records lack baseline trust scoring.
Claims data lag of 30 to 90 days, common across Medicare reporting, means that CMMI model evaluations often rely on data that is already stale by the time it reaches CMS.
The eight dimensions of data trust applied to CMMI requirements
The Data Trust Index (DTI) scores health data across eight dimensions. Each one maps directly to a specific CMMI data failure mode.
Provenance (25% weight): CMS needs to know where data originated. When a quality measure passes through an EHR, a registry, a health information exchange, and a reporting intermediary before reaching CMS, provenance degrades at each handoff. Without chain-of-custody documentation, CMS cannot distinguish a validated clinical measure from a manufactured one.
Consent (20% weight): Patient-reported outcome measures (PROMs) and SDOH screening data require explicit consent frameworks. CMMI models increasingly require this data, but most practices collect it without structured consent records that would survive an audit.
Recency (15% weight): CMS measurement periods are specific. A blood pressure reading from 14 months ago does not satisfy a 12-month measurement window. Recency failures are among the most common reasons quality measures are excluded from CMMI calculations.
Quality (10% weight): Data completeness, field-level accuracy, and coding precision all affect whether CMS accepts submitted data. ICD-10 coding errors alone can invalidate risk adjustment, and eCQM numerator/denominator mismatches can disqualify entire measure sets.
Concordance (10% weight): When claims data says one thing and clinical data says another, CMS flags the discrepancy. A patient attributed to an ACO in claims but absent from the ACO's clinical registry creates an attribution dispute that can take months to resolve.
Validation (10% weight): CMS runs its own validation checks on submitted data. Organizations that do not pre-validate their submissions face rejection rates that delay payment and erode model performance metrics.
Breadth (5% weight): CMMI models increasingly require multi-domain data: clinical, financial, social, and experiential. Organizations that can only submit claims data without clinical context cannot meet the full reporting requirements of models like MCP or ACO REACH.
Stability (5% weight): Data definitions and measure specifications change across model performance years. Organizations whose data infrastructure cannot absorb specification changes without manual rework face cumulative reporting failures over multi-year model participation.
SDOH data requirements are expanding faster than infrastructure
The newest CMMI models treat SDOH data as a core requirement, not an optional supplement. The Making Care Primary model requires practices to screen for food insecurity, housing instability, transportation barriers, and interpersonal safety using validated instruments like the AHC-HRSN.
But collecting SDOH data is not the same as collecting trustworthy SDOH data. A January 2025 MedPageToday opinion piece argued that value-based care models could improve long-term care outcomes and efficiency, but only if the data infrastructure supporting those models captures the social context that drives utilization.
The challenge is that SDOH data often comes from community-based organizations (CBOs) that operate outside the health system's data governance framework. When a food bank reports a referral completion or a housing authority confirms a placement, that data needs to flow back into the CMMI reporting pipeline with provenance, consent, and recency intact. Most health systems have no mechanism for this.
How claims data lag undermines model evaluation
CMS evaluates CMMI models using claims data that arrives 30 to 90 days after services are rendered. For models with annual performance periods, this means that final evaluation data is not complete until three to six months after the performance year ends.
This lag creates several problems. Risk adjustment relies on complete claims histories. If a significant diagnosis is coded in a claim that has not yet been processed, the beneficiary's risk score is artificially low, making the ACO appear to have generated savings it did not actually achieve. When the late claim arrives and the risk score is recalculated, the savings disappear.
The lag also affects quality measure calculations. A patient who received a screening in December may not have the corresponding claim processed until February. If CMS locks the measurement period in January, that screening is excluded from the quality score.
Organizations that pre-validate and score their data before submission can identify these timing gaps and either accelerate claims processing or supplement claims data with clinical data that provides the same information without the lag.
The attribution problem is a data trust problem
Beneficiary attribution, the process of determining which patients belong to which provider organization for purposes of accountability, is one of the most contentious aspects of CMMI model participation. Attribution disputes account for a significant share of model complaints and appeals.
Attribution typically relies on claims data: whichever primary care provider billed the plurality of evaluation and management services for a beneficiary gets credit. But claims data is noisy. A beneficiary who sees three different providers at two different organizations in a six-month period may be attributed to the wrong entity entirely.
The fix is not better algorithms. It is better data. When clinical encounter data, scheduling data, and claims data are concordant, attribution becomes deterministic rather than probabilistic. When they disagree, attribution becomes a coin flip with financial consequences.
What the legislative response tells us about the current state
The House bill passed in June 2026 to ease quality reporting requirements for Medicare ACOs was not a policy preference. It was an acknowledgment that current requirements exceed the data capacity of most participating organizations.
This legislative signal matters. It tells us that the gap between CMMI data expectations and organizational data capabilities is wide enough to threaten the viability of value-based care models at scale. If practices cannot report accurately, CMS cannot evaluate models accurately. If CMS cannot evaluate models accurately, it cannot determine which models to scale and which to retire.
The answer is not lower standards. It is better data infrastructure. Organizations that invest in data trust scoring before submission will meet current requirements and will be prepared for whatever requirements come next.
What this means for organizations in CMMI models today
If your organization participates in ACO REACH, Making Care Primary, Kidney Care Choices, or any other CMMI model, your data submission is not just a compliance exercise. It is the mechanism by which CMS decides whether your model works, whether you earn shared savings, and whether you face financial penalties.
Every record you submit carries implicit claims about its provenance, recency, completeness, and accuracy. CMS tests those claims. When records fail, the consequences are financial and reputational.
Scoring your data before it leaves your organization, not after CMS rejects it, is the only way to control this risk. The organizations that will succeed in CMMI models over the next five years are the ones that treat data trust as infrastructure, not as an afterthought.
The DTI Engine scores every health data record 0 to 100 across 8 trust dimensions before your AI model or your CMS submission package sees it. If your team is evaluating data for CMMI model reporting, risk adjustment accuracy, or quality measure submission, contact Louis Simeonidis at louis@supertruth.ai or (215) 918-4140.
Further reading:
Jason Alan Snyder
Co-founder of SuperTruth and Artists & Robots, and an inventor on the Data Trust Index patents. Twenty-plus years building technology inside Interpublic Group. He writes here nearly every day on data trust, provenance, and what AI should be allowed to act on, and publishes essays on his Substack.
About SuperTruth · LinkedIn · Substack · jasonalansnyder.com
See it in practice
DTI scores the record, not the patient.
8 dimensions. 0–100. Travels with every record permanently.