AI as a Catalyst for Value-Based Care

 A conversation with Robert Jarrin, Managing Member, The Omega Concern, and Mario Romao, Global Director of Health Policy, Intel Corporation 

The Association of Medical Colleges published a report that projected “the United States will face a shortage of between 54,100 and 139,000 physicians by 2033.” This combined with the growing aging population and high levels of chronic diseases is setting up the U.S. healthcare system for further strain. 

Companies across the globe are racing to find a solution. Creative technologies are surfacing, one of them being advances in artificial intelligence (AI) that present a significant opportunity to support and compliment workflows across the board.  

The potential for AI is enormous when it comes to optimizing workflows with limited resources, and the results are promising. A study investigating the performance of AI convolutional neural networks in detecting lung disease showed that experts aided by AI reduced their average assessment per case from almost three minutes to over 30 seconds. 

With results like these, the power of AI to revolutionize the healthcare industry can’t be ignored. But as the number of medical devices enabled by AI has increased, challenges on how to pay for these tools, and the services derived from them, have arisen.  

To explain what’s at stake, Mario Romao, Global Director of Health Policy, Intel Corporation sat down to speak with Robert Jarrin, Managing Member at The Omega Concern, a consultancy that advises medical associations on Digital Health and Medicine. Here are the highlights of their conversation. 

Mario Romao: The way to pay for medical procedures is complex. Can you guide us through the main steps? 

Robert Jarrin: In the United States, the largest public payer for medical procedures is the Center for Medicare and Medicaid Services (CMS). The CMS annual budget is second only to Social Security. Its annual budget goes to pay for Medicare, Medicaid, the Children’s Health Insurance Program (CHIP), and other administrative expenses.  

Professional provider services within the Medicare program use "coding" to describe medical procedures. This process is fundamental in determining coverage and payment. 

Every aspect of medicine and healthcare has an associated nomenclature or taxonomy by which things are described. Such terminology is produced by entities dedicated to developing terms or language and is maintained by the American Medical Association (AMA) through a copyrighted code set called Current Procedural Terminology (CPT). Decisions on whether a payor, public or private, will cover and pay for medical services, are guided by such coding.  

Because Medicare and Medicaid account for such a large part of the most vulnerable older population, CMS is fundamental in establishing coverage and payment policies. Their decisions are often mirrored by private insurers. Each public and private payor may elect to cover and pay for codes. 

MR: What makes AI different from other medical technologies? 

RJ: AI represents an evolution of medical device technology that will no longer be premised on a specific form of hardware, nor on one device used by a patient or their provider.  

AI will increasingly be distributed, and agnostic of platform. It will most certainly leverage real-world performance data, and evidence to learn and improve itself with the aid of machine learning and deep learning technologies. Because of this, it's important to understand payment and reimbursement regulations.  

AI and software do not neatly fit into existing medical business models or coverage and payment frameworks that are decades old. Take for example, the practice expense (PE) methodology used by CMS, that is composed of indirect and direct practice expenses.  

PE currently classifies “computer software” as an indirect practice expense, meaning it receives little to no payment. However, in my opinion, Software as a Medical Device (SaMD) should be categorized as a direct practice expense, which is paid for. SaMD should qualify as “medical equipment,” which is also paid for while updates, patches, and security improvements are analogous to “medical supplies” that are typically also paid for. 

In instances where autonomous AI software complement some forms of clinical staff time and associated work, this should also be considered direct PE and paid for. Thus, there’s a strong case for AI SaMD to be a direct practice expense, and therefore, reimbursable.  

Currently, the agency is being forced to shoehorn these technologies into an antiquated regulatory structure under a benefit category and a PE methodology. CMS is aware of the limitations of these coverage and payment issues and has done a remarkable amount of work to highlight the need to better understand the associated technologies (see the 2021 and 2022 physician fee schedule final rules). 

The Agency is constrained by regulations based on arcane statutory language. There’s only so much the Agency can do within their existing authorities unless Congress decides to act and modernize the payment framework.  

MR: As far as I know, there is work on AI coding and some reimbursement by CMS: so, what’s the problem? 

 RJ: We should differentiate coding from coverage and payment. AMA and CPT have been proactive on integrating AI in various ways, including through policy recommendations and through the creation of an AI taxonomy to provide guidance for classifying those applications.  

Yet, when it comes to coverage and payment, this is where things get much more complicated. In the 2022 Physician Fee Schedule Proposed Rule, CMS employed a “crosswalk” approach to define costs. 

When using a “crosswalk”, the valuation for a given service is substituted by a valuation assigned to other existing services that have similar costs. But how can you do that when no such services exist or have been recognized in the past?  

Thus, even though the agency has decided to extend coverage and payment for these services by cross walking them to existing non-AI/SaMD services, we are being faced with a lack of clarity or predictability for coverage and potential payment of computer software AI services and solutions. That makes it harder for providers and results in ambiguity, which, in coverage and payment is anathema to adoption. 

MR: Can AI be a catalyst for value-based care and payment modernization? 

 RJ: AI will be a catalyst for modernizing many aspects of healthcare and medical practice. But its practical implementation will take time.  

As technology evolves, the way healthcare is being delivered will change with it. So too must cultural attitudes within healthcare and providers to understand these new technologies.  

As the adoption of AI continues and more of these solutions find their way into modern day practices, we’ll see payment policies evolve.  

This will take time, but if the past is prologue, we find ourselves currently in a very exciting place when payment policies for non-face-to-face services, such as remote monitoring, virtual care, and telehealth, are beginning to find broader utilization.  

In the past five years alone we've seen many new codes created in digital health, more than ever before – some of which have gone on to receive coverage and payment by CMS and other private insurers. These are significant developments!  

Even CMS has signalled a heavy commitment to evolving through the creation of the Technology, Coding, and Pricing Group within the Office of the CMS Deputy Administrator, which bodes very well for the future of payment modernization and evolving programs such as value-based care. 

MR: What’s the way forward? 

 RJ: I predict that over the coming years we will see a dynamic coding environment where new services and procedures based on AI will be introduced. CMS, within their regulatory authorities, will do everything possible to evolve programs to better align with current medical technologies (such as AI and SaMD). Congress may need to intervene. We’re beginning to see aspects of that through recently introduced legislation related to digital therapeutics.  

It’s also important to mention other work by other agencies, such as the Food and Drug Administration, who continue to refine their approach towards the regulation of AI and medical devices. I hope we see a closer harmonization between those stakeholder entities. As with any nascent area, it's always slow at first but before you know it, Moore's Law takes over. 

Closing 

The prevalence of AI is set to increase as it finds new ways to improve the performance of healthcare workflows and improve patient outcomes. As Jarrin explained, long established payment methodologies need to better reflect novel software reliant services, including AI, and the value they bring to clinical practice. This issue is being recognized by the largest public healthcare payor, CMS, and we look forward to further developments in the upcoming 2023 Physician Fee Schedule Final Rule.  

We encourage CMS to leverage the recently developed AMA’s AI taxonomy and to expand appropriate payment for AI tools by recognizing SaMD as any other medical equipment. 

CMS has recently published the 2023 Proposed Physician Fee Schedule where it solicits comments on current and evolving trends in health care business arrangements, the use of technologies, such as AI and machine learning, and how those costs are evolving medical practice. 

_________________________________________________________________________ 

Robert Jarrin's areas of expertise include the coding of healthcare services and procedures in federal programs and the coverage and payment of digital medical services. He is also a current serving member of the American Medical Association (AMA) Digital Medicine Payment Advisory Group (DMPAG).  

Mario Romao works at Intel’s Business and Technical Affairs team. Mario oversees global digital health policy and coordinates Intel’s Artificial Intelligence policy interactions. Mario works with governments on policies to promote the adoption of health information technologies to address some of the current challenges healthcare systems are facing.