AI development seen outpacing reimbursement mechanisms
Teeing up the Summit’s discussion over the need to define categories within AI for reimbursement, Michael Abramoff, MD, PhD, founder and executive chairman at Digital Diagnostics, an autonomous AI company, and Chair of the AI Healthcare Coalition, shared concerns over how to fit the technology of AI into clinical practice, and explained how ethics have been leveraged to create AI reimbursement frameworks, providing for adequate guardrails related to patient benefit, mitigation of racial, ethnic and other bias, and cost, as has now been successfully achieved for an autonomous AI product. Stuart Langbein, partner in the Hogan Lovells health practice group, pointed out that there is a range of utilization of health care AI technologies, questioning whether we need to be thinking about the different types of AI from a reimbursement perspective, in order to match the type of AI to an appropriate reimbursement mechanism.
Nicole Sweeney, senior director of Market Access Strategy and Operations at Tempus Labs, expressed concern that AI development is moving at a faster pace than our reimbursement mechanisms allow. She explained how her company is often thinking about how they can best fit their AI into the payment systems that exist today. For example, clinical laboratory offerings are evolving to be driven by AI, but the Medicare clinical lab payment systems do not clearly account for AI in their valuation of tests.
Anitra Graves, MD, medical director at Noridian Health Care Solutions, a Medicare contractor, agreed with the concerns of the other panelists, saying that the payment systems currently in place don’t consider ensemble machine learning (ML) algorithms that have high computational expense, but instead compare them to outcomes of non-ML technologies. Challenges exist “in wide open spaces,” Dr. Graves said, where we need to figure out how to value the new systems. From a regulatory standpoint, the group agreed, we have not yet met the level of sophistication to best accommodate AI reimbursement.
Fitting new tech into old coding frameworks
Moving on to consider coding issues, Mr. Langbein explained how three CPT® (Current Procedural Terminology[1]) categories have been created to deal with AI – assistive, augmentative, and autonomous – asking the panel whether this is a start toward categorization of AI that will be useful looking at reimbursement. Mr. Langbein noted that while sometimes coding drives payment, in other cases, payment comes before coding. Generally, Ms. Sweeney responded, the U.S. may need to encourage our coding system to evolve to promote innovation through adequate payment for new technologies, which can be limited given the confines of the coding systems today. Dr Abramoff added that it has become clear that the CPT Editorial Panel process can be one of the guardrails for AI, especially with respect to patient benefit, as fine grained AI specific codes are being developed together with AMA®’s Digital Medicine Payment Advisory Group.
Dr. Graves said that although medical product sponsors with a code may get to the front of the line for that reason, coding presents challenges when it comes to health care AI because regulators are trying to fit new ways of providing services into old frameworks. “The reimbursement question when it comes to AI can be quite frustrating,” she remarked. Dr. Graves predicted that ultimately, we’ll land on a tiered system of reimbursement, because a product may not have the maturity to have explainability and mature interpretability, but that doesn’t mean that product isn’t of some value.
Medicare working to improve valuation, payment levels for AI
Turning to consider their biggest concerns with the current state of reimbursement for health care AI, Ms. Sweeney answered that from an industry perspective, there is too much pushback on value-based approaches. The costs of research for AI are not represented in the minimal payment offered by Medicare for the product or service, Ms. Sweeney said. Dr. Abramoff agreed that the current Physician Fee Schedule does not seem to accommodate that, and that therefore value based approaches should be transparently developed by AI creators, and explained how coupled with invoices in a competitive market, this has led to a workable autonomous AI reimbursement for CPT code 92229.
Dr. Graves discussed how MACs are focused on developing a consistent analysis for each of the new technologies that expect reimbursement through Medicare. She explained how the vast majority of these technologies have 510(k) approvals, meaning MACs have insufficient evidence to develop the level of trust in the medical product to certify that it has the outcomes that are expected and proven in peer-reviewed literature.
Dr. Graves also said she is working with CMS to develop a stakeholder meeting that will aim to enhance regulators’ understanding of AI health care reimbursement issues at a granular level. For example, MACs are accustomed to looking at performance metrics with traditional diagnostics, but these metrics have a unique nomenclature in the AI space. Accordingly, MACs are questioning whether they can expect medical product sponsors to provide them with additional information, which would enable CMS to have a better understanding of their accuracy and precision data. Dr. Graves told the panel that CMS may seek feedback on performance standards that regulators may rely on to evaluate AI technologies. Regulators would be influenced by sponsors and academic medical centers in terms of the information submitted surrounding how their technology fits into the patient care space, and about the patient population for which that model was vetted. Dr. Graves said that while there is no formal group at CMS taking charge of these issues, there is collaboration behind the scenes on regulatory development in this space.
Reimbursement issues with commercial payors, Medicaid
Shifting to concerns regarding commercial payment for health care AI, Ms. Sweeney commented that some private insurers are more open to payment for new technologies than others, describing how she has observed innovators having some success with pilot payment approaches from commercial plans. Dr. Abramoff explained how commercial payors can be more flexible than CMS, but they usually follow the actions of Medicare, and in fact rapidly followed CMS’ national reimbursement decision for 92229 as of Jan 2022.
Dr. Abramoff also discussed the challenges with obtaining payment from Medicaid, which he described as time-consuming due to the need to work separately with each of the 50 state’s agencies, but is crucial as these payors cover the most vulnerable populations that most urgently need access to a sight saving, equity enhancing product such as IDx-DR. As a result, he said, we need to ensure that Medicaid recipients will not lose out on access to innovative autonomous AI diagnostics. Ms. Sweeney agreed that Medicaid needs central support for reimbursement programs and pilot payment programs, because state plans lack funding.
Regarding pilot payment programs, Mr. Langbein suggested the results of those programs should be brought to the attention of MACs, which may have some leeway in the AI space as compared to reimbursement for more established technologies. Dr. Graves agreed that MACs would benefit from knowing the results of pilot programs, while noting that commercial payors have greater leeway in reimbursement for new technologies.
Dr. Graves concluded the panel discussion by spotlighting how MACs are interested in partnering with stakeholders to make sure that patients have access to innovative therapies. While, Medicare “may deserve some of [the] criticism” for preventing access to therapies, the agency is razor-focused on patient outcomes and collaborating with industry, Dr. Graves emphasized. Ms. Sweeney agreed that open conversation is critical in ensuring that health care technologies are being developed in the best interests of patients.
