New AI Paper + Strong CMS Action on Fraud in California and Beyond

In this newsletter, I discuss the Trump administration’s decision to defer $1.3 billion in federal Medicaid funds to California over fraud concerns. On Monday evening, I joined Bill O’Reilly’s program to discuss hospice fraud in California, how California expanded Medicaid to unauthorized immigrants, and the broader problem of Medicaid waste, fraud, and abuse. I also discuss CMS’s nationwide moratorium on new Medicare hospice and home health agency enrollments because of widespread fraud concerns, preview the administration’s expanding TrumpRx initiative, and summarize a new Paragon paper from Kev Coleman on improving the reliability and safety of AI-enabled medical devices. I begin with our virtual event tomorrow on the hospital cost crisis.

Virtual Event on the Hospital Cost Crisis Tomorrow

On May 21 at 1:00 p.m. EDT, Paragon will host an event examining the hospital cost crisis and policy reforms to address it, based on our recent study, The Hospital Cost Crisis: How Government Policies Drive Consolidation, Undermine Competition, and Fuel Soaring Prices. If you are interested, you can register for the event here.

CMS Responsibly and Appropriately Deferring Medicaid Funds in California

On May 14, Vice President Vance and CMS Administrator Dr. Oz announced a $1.3 billion deferral of federal Medicaid funds to the state of California. Vance said the reason is simple: “The state of California has not taken fraud very seriously.”

This deferral mirrors CMS’s February deferral of $259 million from Minnesota. CMS’s deferral authority comes from Section 1903 of the Social Security Act, which authorizes the agency to adjust its normal Medicaid payments for any past overpayments. In regulations, CMS has divided that process into deferrals, which are temporary withholdings of funds, and disallowances, which permanently deny federal reimbursement but can be appealed.

This $1.3 billion deferral could ultimately become a disallowance. But the review is lengthy and dependent on multiple procedural hurdles. California will have to turn over documents to CMS, and CMS will then review them and decide whether a disallowance is appropriate. Extensions and appeals can draw the review out significantly.

The Trump administration is correct to use its deferral authority. I discussed the California action with Bill O’Reilly this week and explained why stronger federal oversight of Medicaid spending is long overdue. Waste, fraud, and abuse are rampant throughout state Medicaid programs. States have very little incentive to minimize waste, fraud, and abuse because the federal government provides open-ended reimbursement for Medicaid expenditures. Improper payment estimates understate the true extent of Medicaid waste, fraud, and abuse—and properly measured improper payments likely constitute more than $100 billion a year in federal Medicaid expenditures.

States have also developed financing schemes and accounting gimmicks—legalized Medicaid money-laundering techniques—to minimize their own spending obligations and maximize federal payments. These perverse incentives are a major reason Medicaid is so vulnerable to waste, fraud, and abuse.

To improve its enforcement tools, CMS should bolster the PERM program (read our policy brief here) to accurately measure improper payments and penalize states with persistently high improper payment rates.

Medicare Hospice Moratorium

CMS also announced a nationwide moratorium on new Medicare enrollments for home health agencies and hospices, including a six-month freeze on new enrollments and certain ownership changes for home health and hospice providers due to “systemic and deeply troubling fraud” in both sectors.

CMS stated that fraudulent providers have exploited vulnerable Medicare beneficiaries through improper billing, medically unnecessary services, sham employment contracts for kickbacks, and the creation of new entities to evade enforcement actions. During the moratorium period, CMS will intensify investigations and remove fraudulent providers from the program. The nationwide action reflects growing concern within the administration that organized fraud networks have increasingly targeted Medicare and Medicaid due to weak oversight and fragmented enforcement.

New AI Paper—Tackling Generalization Uncertainty

Kev Coleman, director of Paragon’s health care AI initiative, has authored a new paper exploring how AI-enabled medical devices can perform more reliably across real-world clinical settings. In Generalization Uncertainty in AI-Enabled Medical Devices: A Safer Way Forward, Kev explains that many AI systems perform well during development and testing but less reliably when used on patients who differ from the data used to develop and validate the model.

This challenge—known as “generalization uncertainty”—has important implications for patient safety and clinician trust. AI systems rely on probabilistic models trained on large datasets, and they may perform less reliably when encountering patients, clinical settings, or imaging equipment that differ from what was used to develop the model.

Concerns about generalization are growing as AI becomes more integrated into clinical care. At the same time, AI tools are increasingly being used for high-value clinical applications, including early disease detection and predictive analytics that may improve patient outcomes and lower costs. Poorly designed regulatory responses to these reliability concerns, Kev argues, could unintentionally slow innovation and reduce patient access to potentially life-saving technologies without adequately addressing the underlying problem.

The paper evaluates several proposed responses to generalization uncertainty, including third-party certification systems and reviews of AI training data. Kev identifies significant drawbacks in each approach, including high compliance costs, limited personalization, and incompatibility with future adaptive AI systems that evolve after deployment.

Instead, Kev proposes a novel framework called Digital Similarity Analysis (DSA). Under the DSA approach, a patient’s medical image would be compared against the training and testing data used to develop a specific AI medical device. If the patient’s image significantly differs from the data used to train the model, physicians would receive a warning that the AI tool may be less reliable for that patient. Clinicians could then decide whether to avoid using the device, seek additional validation, or place less confidence in the AI-generated recommendation.

Kev argues that DSA offers several advantages over existing proposals. Rather than relying primarily on broad demographic categories for confirming device safety, the approach focuses on individualized patient-level analysis. It also accounts for differences caused by imaging equipment and provider techniques, avoids expensive third-party consultation, and may be more compatible with future adaptive AI technologies. Most importantly, the proposal seeks to improve patient safety without slowing innovation or limiting patient access to beneficial AI tools.

Expansion of TrumpRx

TrumpRx.gov is expanding to include competitive cash prices for more than 600 generic medications from Amazon Pharmacy, Cost Plus Drugs, and GoodRx. Combined with the platform’s existing 64 “Most Favored Nation” discounts on branded drugs, this expansion represents a meaningful step toward greater transparency and competition in prescription drug pricing.

By placing both brand-name and generic drugs on a searchable platform, the update gives patients a clearer view of prices across categories that have historically lacked transparency. Although existing direct-to-consumer sites have helped surface lower prices for many generics and biosimilars, manufacturers have often been reluctant to offer comparable discounts on branded drugs because of pharmacy benefit managers’ (PBMs’) leverage over formularies. As Mark Cuban, the CEO of Cost Plus Drugs, notes, “[Brand name manufacturers] have been told if they work with [Cost Plus Drugs], which would get low prices to patients, the PBMs will remove them from formularies, costing them [tens] of billions of [dollars].”

For uninsured individuals and those in cash-pay or alternative coverage arrangements, this transparency offers immediate practical value. However, TrumpRx remains limited by existing insurance architecture. Most commercially insured patients will still find that cash purchases through these channels generally do not count toward deductibles or out-of-pocket maximums, limiting their usefulness. These Americans would benefit from reforms that better align cash-pay options with their insurance coverage.

Critics initially faulted TrumpRx for lacking generic options, even though administration officials made clear at rollout that the platform would expand over time. This week’s expansion demonstrates how transparent pricing across branded and generic drugs can help sidestep distortions created by entrenched intermediaries and give patients more tools to navigate high drug costs.