How AI Is Already Changing Health Care
Artificial intelligence is reshaping health care faster than policy can keep pace. New AI-based software interprets echocardiograms with accuracy that rivals trained professionals. A bionic pancreas with embedded AI delivers better insulin control for patients. These are not future scenarios — they are happening right now.
However, the critical question is not whether AI will transform medicine. The real question is who pays for it, and how. According to LDI Senior Fellow Dr. Amol Navathe, payment policy choices will define AI’s entire future trajectory in health care.
Without smart payment reform, the U.S. risks two equally bad outcomes: driving costs sky-high with no better patient outcomes, or stifling the very innovation that could save lives.
Why Current Payment Policy Falls Short
AI Is Built Differently From Human Services
Today’s payment systems were designed around human labor. They reimburse based on time, skill, and the number of services a clinician delivers. AI, by contrast, works very differently. Consequently, forcing AI into a payment model built for humans creates serious distortions.
Dr. Navathe argues this mismatch is not minor — it is structural. He raised these concerns during his tenure as Vice Chair of MedPAC, where policymakers began grappling with how disruptive AI technologies affect Medicare pricing. The current system simply was not built with AI in mind.
The Scalability Problem With AI Tools
One Tool Can Serve Millions at Minimal Cost
One of AI’s defining features is scalability. Once an AI product is deployed, each additional patient it serves costs almost nothing extra. This stands in sharp contrast to a human clinician, whose time and effort cost more with each additional patient.
Furthermore, AI can act in ways that fall outside the patient’s direct view. It can perform tasks that human clinicians do not currently provide. Therefore, it opens entirely new categories of services — and entirely new billing challenges.
Prescription Digital Therapeutics: A Clear Example
Consider prescription digital therapeutics (PDTs). These are FDA-authorized, software-based treatments delivered via smartphone apps. They address conditions like anxiety and depression through cognitive behavioral therapy. Each new user adds nearly zero cost to the system. Yet prices reflect what the market will bear — not the actual cost of delivery.
Moreover, PDTs do not fit the traditional durable medical equipment framework used for oxygen tanks or CPAP machines. As a result, existing pricing templates simply do not apply. This gap illustrates why AI demands new payment thinking.
What Drug Pricing Teaches Us About AI
Useful Parallels — and Key Differences
Prescription drug pricing offers a helpful starting point. Like AI, once a drug molecule is developed, the marginal cost of manufacturing each additional dose is low. Similarly, AI’s marginal cost per patient is near zero. Additionally, both drug companies and AI developers set prices that reflect R&D investment rather than delivery cost alone.
Where the Comparison Breaks Down
Nevertheless, the similarities end there. Drugs are relatively fixed products. AI evolves, adapts, and changes over time. Setting stable prices for a moving target is genuinely difficult. Therefore, Dr. Navathe argues that AI technologies must meet a clinical benefit standard — enforced by regulators like the FDA or CMS — to ensure they actually improve care before payers foot the bill.
Shortcomings in Today’s Payment Systems
Existing Mechanisms Are Inadequate
Currently, hospitals receive fixed payments under the Outpatient Prospective Payment System (OPPS) and the Inpatient Prospective Payment System (IPPS). These cover AI-assisted services but do not account for AI’s unique cost structure. Supplemental payments exist — such as transitional pass-through payments and New Technology Add-On Payments (NTAP) — but these are temporary solutions.
Critically, NTAP requires proof of substantial additional costs. This actually incentivizes developers to set high prices. In addition, these designations do not weigh clinical improvement against cost — the central challenge for AI reimbursement.
The Medicare Physician Fee Schedule Problem
Another path involves securing a new code on the Medicare Physician Fee Schedule. Yet this system still anchors value to human labor inputs. Consequently, it lacks any reliable method for valuing the unique capabilities AI brings to care delivery.
How Policymakers Should Respond
Existing Tools May Not Be Enough
CMS does offer some custom coverage pathways. Coverage with Evidence Development (CED), for instance, provides conditional coverage while patients participate in approved clinical trials. This approach factors in clinical effectiveness — but it is rarely used. It demands extensive administration and takes considerable time.
Similarly, obtaining a CPT code is another mechanism. However, it does not rely on a clinical effectiveness standard. Therefore, neither path fully addresses the need for outcome-based AI payment.
Congress May Need to Act
Ultimately, Dr. Navathe concludes that Congress may need to grant CMS new authority. Specifically, federal officials need the power to factor clinical effectiveness into AI coverage and payment decisions directly. Aligning AI payment with outcomes — rather than inputs — is the essential shift. Without it, the U.S. healthcare system may pay far more for AI than the health gains justify.
