To date, the Food and Drug Administration has approved 950 artificial intelligence-enabled medical technologies. But Medicare has lagged well behind that figure, and industry lobbyists are pushing Congress to expand dedicated payment streams for these technologies. Bipartisan legislation is already in the works.
Policymakers should be cautious. Medicare spending on new health-care services is the factor most responsible for America’s projected long-term fiscal crisis. AI has the potential to reduce the cost of health care dramatically, but new expenditures should not simply be layered on top of existing payments for care. Instead, Congress should fund AI technology indirectly, paying providers to deliver quality treatment for patients and allowing them to adopt AI when they can do so more efficiently.
Artificial intelligence seems set to revolutionize the health-care sector. Drawing on a knowledge base broader than that possessed by any individual clinician, the technology has improved disease detection, reduced the potential of misdiagnosis, and prevented the prescription of possibly harmful interventions. In time, AI could enable increasingly personalized treatment, accounting for patients’ specific genetic conditions, while allowing more treatment to be delivered 24/7 in convenient settings, unconstrained by staff shortages.
It could also be a cost-saver. While a Harvard-McKinsey study notes that “historically, labor productivity growth in healthcare has been negative,” its authors estimate that adopting AI “could result in savings of 5 to 10 percent of healthcare spending.” AI not only allows lower-cost clinicians to treat conditions that previously would have required the most expensive specialists but also will increasingly permit the full automation of complex procedures.
And yet, establishing dedicated payments for AI-powered medical devices could increase, rather than reduce, health-care costs. Generally, Medicare expects medical providers to pay for high-tech equipment out of reimbursements for performing clinical procedures, but the Centers for Medicare & Medicaid Services recently began making add-on payments for AI software. In addition to the $471 fee for each outpatient MRI scan, for example, Medicare now typically pays hospitals $950 per patient for software to analyze the images generated. Since the marginal cost of using such software or subscription for each additional patient is close to zero, these reimbursement arrangements are highly profitable. That encourages providers to use and bill for the software more frequently—a familiar problem in the use of imaging services.
Establishing dedicated payments for these products would create new problems. In a fast-developing sector like medical software, where cheaper substitutes constantly appear, a fixed reimbursement would impede price competition. While software companies need to finance some ongoing updates and maintenance, they could use incremental improvements to justify perpetually raising their prices, as has happened with insulin.
As the marginal cost of selling software licenses to additional medical facilities is negligible, tech firms will raise prices according to what Medicare is willing to pay. While it’s important to encourage innovation, the development of medical software generally involves much lower fixed costs than does drug development. Though Medicare does not pay for medical software that patients use on their own devices, more than 120,000 health-related mobile apps now support the management of chronic medical conditions. Nor does AI innovation appear to be waiting on the advent of dedicated Medicare payment systems.
For software that is part of treatment delivered by hospitals or physician practices, payment should come entirely out of standard fees to medical providers. That way, if a new innovation (such as automated anesthesia) reduces the cost of treatment (subject to state regulations on the appropriate practice of medicine), providers can purchase it. If they discover a more cost-effective alternative, they can switch immediately.
In some cases, AI will make possible entirely new services and procedures, for which reimbursement was previously unavailable. While the already fiscally unsustainable Medicare program has not budgeted for them, they can be funded for beneficiaries, without further costs to taxpayers or changing federal law, through Medicare Advantage. These privately managed plans, which now enroll the majority of beneficiaries, receive fixed payments to treat enrollees according to their medical needs. The plans have an incentive to adopt the most cost-effective new methods of doing so.
Medicare’s standard payments already provide adequate incentives to fund new cost-cutting AI systems. If policymakers choose to add on dedicated payments for the use of AI software, tech firms will instead focus on developing cost-hiking innovations, which will only accelerate the upward trend of Medicare expenditures.
Photo: Kirk Sides/Houston Chronicle via Getty Images
