Dear Principal Deputy Posnack:
On behalf of Paragon Health Institute, we respectfully submit these comments in response to the Department of Health and Human Services’ Request for Information: Accelerating the Adoption and Use of Artificial Intelligence as Part of Clinical Care. This RFI has explicitly sought public comment on what measures the agency can implement “to establish a forward-leaning, industry-supportive, and secure approach to accelerate the adoption and use of AI as part of clinical care.” Our response reflects not only Paragon’s research and publications on AI but also our conviction that this important technology has significant potential to reduce American health care expenditures.
HHS should declare increased AI deployment in U.S. health care a national priority. However, AI adoption itself is affected by many conditions. Paragon has organized our feedback around five:
- Health expenditures, deficits, and the national debt
- Reimbursement
- Complementary research
- Regulatory evolution
- Increasing AI autonomy
Health Expenditures, Deficits, and the National Debt
AI-enabled medical devices have attracted considerable press regarding feats such as early cancer detection and predictions of illness years before symptoms manifest, but their promise to reduce American health care expenditures is perhaps an even greater priority if the nation is to sustain health care access and affordability for our citizens. The United States’ national health expenditure (NHE) is enormous, and this spending represents a considerable burden for families and American businesses as well as a fiscal threat to competing non-health care expenditures within the federal budget.
In 2024, the most recent year of reporting, NHE reached $5.3 trillion, which represented 18 percent of Gross Domestic Product (GDP). The Organisation for Economic Co-operation and Development (OECD) has compared the United States’ NHE as a share of GDP to more than three dozen countries across the world.
“In 2024, the United States spent by far the most on health…well above Germany, the next highest spender (at 12.3%). These are followed by a group of around 15 countries that all allocated 10‑12% of their GDP to healthcare. In many of the Central and Eastern European OECD countries, as well as in the newer OECD Member countries in Latin America, spending on health represented between 6‑9% of their GDP.”
Recent government research projects that, during the 10-year period from 2024 to 2033, our nation’s average NHE growth (5.8 percent) will outpace average GDP growth (4.3 percent), raising health care spending as a percentage of GDP from 18 percent to a projected 20.3 percent. Moreover, the estimate of NHE spending growth in the next decade is made all the more concerning by the trillions in increased national debt projected for the coming decade. Congressional Budget Office (CBO) Director Phillip Swagel, commenting on the deficits and national debt during this period, stated, “Our budget projections continue to indicate that the fiscal trajectory is not sustainable.” Likewise, back in 2009, when NHE as a share of GDP was 17.2 percent, the Social Security Advisory Board warned that the nation’s health care cost trajectory was “unsustainable” and “perhaps the most significant threat to the long-term economic security of workers and retirees.”
It is within the context of these NHE sustainability concerns that AI emerges as a promising means to lower health care spending. The probabilistic nature of (most) AI combined with the processing of large data resources has opened the door to predicting disease and beginning medical interventions that are significantly less expensive than would be the case when disease is detected at a later stage. While not its only way to lower health care costs, “predict and prevent” could be transformative for American health care. By positioning AI as a means to make health care more affordable, its adoption and use will accelerate.
Reimbursement
AI’s “predict and prevent” paradigm is dissimilar to the dominant “detect and treat” approach to health care. A major concern associated with this difference is the possibility that predictive AI technology will encounter insurer resistance with respect to reimbursement. Without strong prospects for reimbursement, device manufacturers will be cautious about investing in new AI solutions given ROI worries. Conversely, positive reimbursement prospects can accelerate AI adoption across the American health care system. HHS should continue its efforts to eliminate government obstacles and disincentives for the use of AI for which there is a market demand from patients, providers, and related parties. This endeavor necessarily includes consideration of how current CMS Medicare policies can be clarified and improved in ways that facilitate AI adoption and its related savings potential while discouraging any net increase in federal outlays. These policies include New Technology Add-On Payments and Transitional Pass-Through Payments, which provide transitional supplemental hospital payments for new technologies; the HCPCS coding process run by CMS, which includes the AMA’s CPT codes; and coverage processes, including local and national coverage decisions as well as the Transitional Coverage for Emerging Technologies pathway. Any new HHS policies relating to AI payment, coding, and coverage should reflect a commitment to using AI innovation to bolster fiscal responsibility and health care affordability.
Complementary Research
HHS’ promotion of AI should include funding for research studies quantifying the economic value of AI technologies that predict disease before the manifestation of symptoms. Insurers, for their part, are more likely to expand AI reimbursement practices in the presence of compelling economic arguments for AI utilization. Complementary HHS research initiatives should explore how clinical AI tools can empower lower-cost medical staff (nurse practitioners and physician assistants) to perform more services safely at a lower expense for patients. Most importantly, these collective learnings on AI-facilitated cost reductions should be instantiated in a real-world research pilot whose exclusive aim is to deliver health care at 80 percent of the cost without lowering quality. If AI is not approached with such an explicit government aspiration for cost savings and a corresponding project to validate this vision, the technology is more likely to become one more incremental expense within an increasingly unaffordable health care system.
Regulatory Evolution
With respect to regulation, HHS should consider state lawmakers’ concerns about health care AI and collaborate on solutions that will prevent a patchwork of differing state rules for AI-enabled medical devices from emerging. Such an inconsistent patchwork of rules would inflate AI medical costs and potentially delay patient access to life-saving technology.
HHS should also find ways to support the FDA as it adapts its medical device approval processes to the realities of AI, including articulating the principles that should be reflected in FDA rulemaking. The historic FDA categories governing AI-enabled medical devices, Software as a Medical Device (SaMD) and Software in a Medical Device (SiMD), were developed during a period characterized by deterministic software. Probabilistic reasoning models, continuous post-deployment system learning, generalization uncertainty, and emergent capabilities require ongoing FDA oversight to evolve, particularly if the U.S. is to maintain its international leadership in health care AI.
Increasing AI Autonomy
Of particular importance in the consideration of AI regulation is the prospect of AI-enabled medical devices becoming more autonomous over time. It is reasonable to assume some AI will offer self-service to a patient (without any health care provider assistance) while maintaining safety equal to or surpassing that of health care professionals providing the same services. In instances where a medical service can be performed independently of a clinician, a sizable expense within health care delivery is eliminated. Because medical devices aren’t licensed by individual states, there is also the future possibility of a subset of AI-enabled health care devices (like medical image interpretation devices) competing online for patients across the nation, giving alternatives to regions dominated by a consolidated hospital system and lowering prices through market competition. Such online competition bodes well for consumer utilization of health care AI.
Autonomous care—the delivery of a medical service via a self-service system a consumer uses without clinician assistance—is not unique to AI. For years, pharmacies have offered simplistic blood pressure machines for unassisted consumer use, with some systems having additional functionality such as body mass index calculation or vision evaluation. AI’s differentiation is the sophistication of its software and the additional services it enables. Compared to productivity gains and quality improvements, gains from autonomous AI medical services may have the best long-term potential for material reductions in health care expenses, including labor costs.
HHS should formally state it is supportive of autonomous AI where such solutions can be empirically demonstrated to match or exceed both the quality and accuracy observed for human physicians. Additionally, given the continual evolution of AI, HHS should encourage the FDA to provide an economical pathway for innovators to reapply for FDA approval on their devices where the functionality remains the same, but system autonomy increases over time. On this front, the FDA could leverage work already performed by the U.S. Department of Transportation for self-driving vehicles with differing levels of system autonomy (e.g., driver-assistance vs self-driving). Such efforts will ultimately expand AI adoption from exclusively clinical contexts to the wider ecosystem of American health care.
Conclusion
The pace of AI adoption and use will significantly benefit from leadership on the part of HHS. Paragon Health Institute hopes to support the agency in its efforts on this front.
Sincerely,
Kev Coleman
Director of the Health Care AI Initiative at Paragon Health Institute
Brian Blase
President, Paragon Health Institute
