AI Summit panelists offer tips for obtaining reimbursement for innovative health care technology

Victoria Wallace, partner in the Hogan Lovells health regulatory practice area, set the stage by explaining how coding, coverage, and payment are the three essential pillars of obtaining health insurance reimbursement through Medicare in the U.S. However, she explained, because the Social Security Act defines the categories of medical products and services that can be reimbursed under Medicare, and the law was enacted at a time when AI was not being contemplated, reimbursement attorneys must work to shoehorn new technology into the long-standing, legislatively-defined Medicare coverage pathways. There remains no defined benefit under the Medicare statute for AI-enabled services.

Adding to this lack of regulatory clarity, CMS currently has no guidance on categorizing autonomous technology, noted Anitra S. Graves, MD, chair of the A/B MAC AI Workgroup and executive contractor medical director for Medicare provider First Coast Service Options. In order to help AI health product sponsors obtain appropriate coverage and payment from CMS, Dr. Graves explained how her group is developing a taxonomy to perform appropriate technology assessments and to discern potential future pricing for those products. To aid the likelihood of obtaining coverage, Dr. Graves of encouraged sponsors to include in their applications information that is in the public domain, and that can be found in publicly accessible literature.

Speculating on forthcoming CMS guidance to explain how AI-enabled technologies should be priced, Dr. Graves observed uncertainty over how that system will value computational complexity as compared to the volume of data, and how it will be integrated into pricing determinations. She called this a “wide open space” for input from stakeholders. Asked about the process for creating a local coverage determination (LCD), Dr. Graves called this process a “heavy lift,” describing the significant evidence needed to inform whether coverage is “reasonable and necessary.”

Dr. Richard Frank, co-chair of the AMA/CPT Digital Medicine Coding Committee, suggested assurance labs could be a reliable standard for AI model validation. Dr. Graves agreed that it’s important that stakeholders should not have to “reinvent the wheel” when bringing their technology to market.

Continuing the conversation over medical device sponsors’ efforts to obtain Medicare coverage, Ms. Wallace shared her experience with manufacturers who have opined over a lack of adequately-defined goalposts for demonstrating that an AI-enabled product is “reasonable and necessary”: CMS’ standard evidence for permitting Medicare coverage. Indeed, clinical evidence necessary for marketing authorization from the U.S. Food and Drug Administration (FDA) is not always sufficient to gain coverage, Ms. Wallace noted. Dr. Graves explained that sponsors should have literature indicating that AI-enabled technology has the ability to produce a model that is “necessary” to inform management of a disease or condition, and that has outcomes from which Medicare beneficiaries may “benefit.” CMS’ “reasonable and necessary” standard goes beyond FDA’s “safe and effective” standard, Dr. Graves added.

Providing the perspective of a sponsor of an AI-enabled medical device, Clark Daniel, director of government relations at Heartflow, discussed his experience with evolving regulatory standards at both CMS and FDA, highlighting how specialty society engagement and buy-in are “critical” for obtaining coding for an AI-enabled device. Given that there are many settings of care, “you need to determine where your service is offered” prior to seeking Medicare reimbursement, Daniel explained. Regarding coverage, Mr. Daniel stressed the need for “substantial clinical evidence” that can assert the importance of a sponsor’s technology.

Ms. Wallace acknowledged that a pain point in the process of coding AI-enabled devices comes when obtaining a Category III code, with commercial payers treating those services as “investigational,” and not always providing coverage. Mr. Daniel noted that in his experience, smaller commercial plans may be more open than larger ones to paying for novel technology, at least initially. 

Shifting the conversation to the perspective of the health care provider, David L. Reich, MD, Chief Clinical Officer for the Mount Sinai Health System, provided his experience with AI-enabled health care technology, emphasizing how artificial intelligence is increasingly critical to health systems’ improved outcomes. He pointed out that the benefits of AI come in many different forms of return on investment (ROI), citing, for example:

• AI-enabled radiological tools permitting accelerated care for life threatened conditions.
• Strategic redirection of resources.
• Helping health systems to improve ratings and reputation.
• Improved patient experiences.

Dr. Reich also noted the importance of bringing real-world evidence (RWE) into the service of developing beneficial technology. Highlighting the need to bring novel technology into the workflow, he also explained how hospital staff are required to integrate AI technology in order for that AI to be useful. “AI is not going to replace radiologists, but radiologists who use AI will replace those who do not,” he predicted. Integration of technology seamlessly into workflow is paramount to adoption of AI, Dr. Reich commented. 

Providing an overview of updates at CPT, Dr. Frank acknowledged the struggle for sponsors of AI-enabled technology to identify a proper CPT code, emphasizing how AMA’s Digital Medicine Coding Committee (DMCC) weighs heavily the input of medical specialty advisors to CPT, who have the opportunity to weigh in on each code change application. Dr. Frank explained that the DMCC does not adjudicate codes, but merely advises on issues at the request of the Panel Chair. He spotlighted the importance of “synchronization” of evidentiary standards across private and governmental entities, discussing with the panelists how it is critical not only to harmonize our lexicons, but also to synchronize our requirements for validation of “clinical validity,” “clinical meaningfulness,” and “medical necessity,” such that the requirements for FDA authorization, CPT coding, and CMS/payer coverage, are not disparate but represent a continuum. 

Dr. Frank told the Summit that DMCC is working on a manuscript as well as an update to CPT Appendix S to provide supplementary information on coding the services and procedures performed by AI-enabled devices, and considering how an “algorithm-only” code, for the work of an algorithm which is not integral to the primary procedure (e.g., source of input data) could look in the future. To the extent that there is “separability,” i.e., a beginning (input), a middle (the work of the algorithm), and an end (output), DMCC is also mulling how a code could appear in an autonomous setting, Dr. Frank mentioned. “There should be discrete and differentiable codes to describe the work of algorithms,” he said.