What CMS Got Right with ACCESS, and What Still Needs Work

IMPORTANT NOTE: We are partnering with Maverick Health Policy to launch a CMS ACCESS Model Collaborative and hosting a kick-off webinar on July 22nd. You can sign up for the Webinar Here.

When CMMI released the Advancing Chronic Care with Effective, Scalable Solutions (ACCESS) Model, it did something the Innovation Center rarely does this narrowly. It built a payment experiment around specific measured outcomes. CMMI usually builds population-based payment incentives and bundled payments, with specific components around highly prevalent chronic conditions layered in rather than made the target.

That alone makes ACCESS worth your attention, because it could point toward broader fee-for-service reform. Can you pay specifically for moving a clinical measurement, and let the market figure out how to do it using scalable, low-variable-cost technologies?

The model gets a surprising amount right at the level of design intent. It also carries a handful of practical assumptions that, left unaddressed, will quietly decide whether anyone but the smallest digital-health startups can actually run it.

The incentive is pointed at the right target. Outcome Aligned Payments reward improvement in a clinical measure rather than the delivery of a billable encounter. CMS is trying to pay for real, measured improvement in high-impact risk factors and chronic conditions. In the long run, these conditions contribute heavily to spending.

The target conditions are the right ones. Hypertension, diabetes, chronic musculoskeletal pain, and depression are high-prevalence, collectively touching well over half of the Medicare population, and they sit upstream of enormous downstream spending.

Uncontrolled hypertension and diabetes feed cardiovascular and renal complications, untreated depression worsens self-management across every other condition, and chronic pain drives utilization, opioid exposure, and functional decline.

The model is not prescriptive about clinical approach and technology. It sets the goal and lets the market innovate toward it. This is the single most important design decision in the program. Prescriptive models lock in whatever the state of the art was on the day the rule was written, while open models let the field discover what works. By defining success as an outcome and leaving the how open, CMS has built real room for new workflows, new engagement mechanics, and new technologies, including AI-enabled approaches, to compete on results rather than on compliance.

Here is what CMS says about the models:

“ACCESS care organizations are expected to offer integrated, technology-supported care that may include:

• Clinician consultations

• Lifestyle and behavioral support (nutrition, exercise, smoking cessation)

• Therapy and counseling

• Patient education and care coordination

• Medication management

• Ordering and interpreting diagnostic tests and imaging

• Use or monitoring of Food and Drug Administration (FDA)-authorized devices, including devices or software, or devices that are subject to FDA enforcement discretion

Care may be provided in-person, virtually, asynchronously, or through other technology-enabled methods as clinically appropriate.”

The eligibility and evaluation design is unusually rigorous. API-based eligibility checking turns enrollment into something programmatic and auditable rather than manual and gameable. And randomizing roughly ten percent of otherwise-eligible beneficiaries to an ineligible status creates a built-in control group. Embedding randomization in the eligibility layer gives evaluators a clean counterfactual without bolting one on after the fact. That should make whatever ACCESS concludes far more credible.

The bet on real-time patient-generated data is sound. The model treats continuous data and engagement not as an end in itself but as a signal for clinical intervention, something that should prompt a clinical change when a number drifts.

Here are my biggest concerns with making ACCESS work for CMS, beneficiaries, providers, and model participants:

The payment rates ignore the real cost structure of acquiring and keeping patients. The baseline cost considers hard, direct costs of software and hardware (in some cases), but not costs associated with patient acquisition, churn, and loss-to-follow-up. In health tech, these are the most expensive components. Acquiring a Medicare beneficiary, onboarding them onto a device and an app, and keeping them engaged long enough to generate a measurable outcome is expensive and leaky.

A nontrivial share will disengage before reconciliation, and with half the payment withheld pending outcomes, participants carry that attrition risk on a thin margin. The likely result is adverse.

The large, well-capitalized players who could actually scale this may calculate the unit economics, conclude the juice isn’t worth the squeeze, and sit it out, leaving ACCESS to smaller entrants with less staying power. A model that can’t attract durable operators won’t produce a durable answer. If a business model is not possible given rates, then participants will drop out. This outcome is yet to be seen, but it is a real possibility.

The model probably underestimates selection bias. The beneficiaries most likely to enroll, stay engaged, and improve are disproportionately the ones already inclined to manage their conditions, because the people who adopt a health-technology solution are, almost by definition, more activated than average. The randomized ineligible arm helps enormously with internal validity, but it does not fully resolve who selects in. If the enrolled population skews toward patients already on an improving trajectory, the model risks rewarding regression to a better mean rather than true causal lift, and overstating how well the approach would work in a less motivated population that actually costs money for Medicare in the long run.

Excluding traditional health care organizations through billing rules may invite fragmentation. By fencing out established providers, ACCESS risks creating a parallel track of chronic-care management that runs alongside, rather than through, the patient’s existing care relationships. Chronic disease is precisely where continuity matters most. Routing hypertension or depression management through an entity disconnected from the primary care chart doesn’t eliminate fragmentation, it can make it worse. The patient ends up with one more silo, and the PCP ends up with one more data source they didn’t ask for. The provider incentives to review data are weak, and an increase in those rates might be beneficial.

Getting clinicians to act on patient-generated data is harder than the design assumes. The hardest unsolved problem in remote monitoring, as an example, is not data collection, it is getting busy physicians to actually review patient-generated reports and act on them. This is true whether the data lands in the EMR or a third-party dashboard, and arguably the EMR case is worse, because it adds to inbox burden clinicians are already drowning in. ACCESS leans on this behavior without a credible mechanism to compel or compensate it.

None of this is fatal, and ACCESS is a beautiful framework for modernizing payment policy. The conditions are right, the outcome orientation is right, the openness to innovation is right, and the evaluation design is genuinely better than what came before. The problems are not in the concept, they are in the economics and calibration and in the connective tissue between the model and the rest of the care system.

CMS got the hard, conceptual part right. What remains is the unglamorous work of right-sizing the incentives, and connecting the system enough, for the good idea to actually run.