Innovation interrupted: How reimbursement is stalling health AI

Current reimbursement landscape hampers adoption

While current reimbursement systems were developed far before the widespread adoption of AI in healthcare, there are some pathways to getting paid once a technology receives FDA approval.

Following approval, an AI developer must obtain a billing code for the AI medical service, should a payer decide to reimburse for it. However, this is another lengthy process, following years of back-and-forth with the FDA for device approval, Roehrenbeck explained.

Once a code is assigned, developers then pursue reimbursement depending on the intended setting of the AI medical service -- yet another hurdle providers and developers face with reimbursement as a result of the siloed nature of the healthcare system.

If the AI medical service is designed for the outpatient setting, developers can seek a new technology APC (ambulatory payment classification) reimbursement. The technology APC reimbursement is a Medicare payment rate for a new hospital outpatient service or procedure for which CMS does not have enough cost data to classify it in a standard clinical APC. Instead, CMS provides a temporary reimbursement rate that uses cost bands until it collects enough cost data to move it out of the new technology APC category into an appropriate clinical APC. This usually takes about two to three years, but reimbursement is not guaranteed.

"Other types of services also go through that process, and it doesn't lend itself toward assurance for the provider that once they bring on the technology and start to provide it to patients, that five years out, they'll still be able to do so," Roehrenbeck said.

For the inpatient setting, developers can seek a new technology add-on payment, or NTAP, through Medicare's Inpatient Prospective Payment System. NTAPs are an additional payment above the standard Medicare Severity Diagnosis-Related Group (MS-DRG) payment amount that lasts for no longer than three years for certain technologies. Per CMS, those technologies must be new, be inadequately reimbursed by the existing MS-DRG system and must demonstrate clinical improvement for specific patient populations.

"Again, the concept there is you get this payment, it's good for a certain number of years and then it expires," stated Roehrenbeck. It goes on to assume "that the cost sort of gets absorbed into the underlying DRG, but there's no way to tell us that's actually happening."

Without payment certainty in the long term, health systems have expressed reticence to adopt innovative technologies, including AI medical services, she added.

The regional payment problem

Even after AI medical services obtain FDA authorization and secure billing codes, they face a critical bottleneck: regional payment decisions by Medicare administrative contractors, or MACs.

As providers and developers wait for CMS to determine consistent national payment rates -- which can take several years, as mentioned above -- new technologies like AI medical services are priced by MACs on a regional basis through local coverage determinations. This can lead to inconsistent approaches to coverage and reimbursement across different regions.

"CMS has had a pattern of saying, okay, we understand you have a code and we could see that maybe this should receive national payment, but we're going to defer to our contractors to work with you on those payments," Roehrenbeck said.

The regional problem can drain resources as providers and developers navigate the patchwork of payments, especially if the AI technology is being deployed in different parts of the country.

"What that means as a company is, now, you're having to devote resources to working with medical directors at perhaps six different MACs to go through the literature review, the clinical evidence, explaining how this is covered," Roehrenbeck elaborated.

The inconsistency in payment also makes it difficult for large health systems to implement AI tools uniformly across their enterprises, as they must deal with differing local coverage determinations. The regional payment obstacles ultimately slow adoption, meaning patients cannot access the latest and greatest for their health, despite FDA authorization and proven clinical benefits.

The path to predictable payment

The current reimbursement system does not lend itself to predictable payment for medical AI services. That's where organizations like the AI Healthcare Coalition come in to advocate for policy and regulation that support validated innovation in healthcare. Currently, its priority is the Health Tech Investment Act.

Introduced by Sens. Mike Rounds (R-SD) and Martin Heinrich (D-NM) in April, the Health Tech Investment Act aims to deliver a predictable Medicare payment pathway for AI-enabled medical devices through a new technology APC in the Hospital Outpatient Prospective Payment System that lasts at least five years.

"Medicare patients deserve access to the life-changing care that artificial intelligence-enabled devices can offer," Rounds said at the time. "There is currently no clear Medicare payment system for these devices, meaning that it can take years to be approved and paid out by Medicare accurately. This legislation would create that system, improving diagnoses and encouraging the adoption of AI devices in clinical settings."

The AI Healthcare Coalition has also urged Congress to review outdated regulations that hinder the adoption and payment of AI-enabled healthcare services. For example, the Deficit Reduction Act has a provision that caps Medicare payments for imaging services -- a category many AI medical services fall under.

Roehrenbeck and colleagues are concerned that the regulation could impact payments for AI imaging services, which are very popular.

But greater reform is needed to ensure the "clunky, outdated and regulation-heavy healthcare system" still works in the age of AI, she said.

"That's just going to take some time," Roehrenbeck stated, but it represents the most comprehensive approach to creating predictable payment pathways for medical AI services.