Blog 5: Navigating the regulatory and reimbursement pathways to get to market

Until now, this series has covered the journey from data, to algorithm development, and to validation of an AI medical device – all aspects driven by the organization manufacturing the product. In this installment, we turn our focus to some of the external drivers and necessary steps we must incorporate in bringing an AI product to market, and into users' hands in hospitals and labs.  

By the time an AI diagnostic reaches the clinical market, it has been through a lengthy process of approval. This process involves multiple stakeholders and has different requirements in different countries. While the ultimate goal is to improve patient care and outcomes, it is physicians who will use the technology, so both of their needs must be considered in the approval of any clinical product. Regulatory requirements ensure patient privacy, safety, and efficacy. Reimbursement is often the best way to open market access for a product and payers and healthcare funders usually expect further evidence of the clinical utility of the product, as well as the economic and patient benefit. All stakeholders require organizations and products to follow specific practices, and provide evidence that proves the product performs as it should.

While much of this evidence cannot be gathered until the AI diagnostic is developed, the specific requirements outlined by regulatory bodies must be baked into the early conception of the product. Here we overview the regulatory landscape, the winding path to clinical use, and eventually reimbursement.

Regulatory compliance in healthcare has a broad scope

The wild-west frontier of AI development has caused many countries to evaluate and attempt to mitigate risks and ethical challenges with new regulations. Laws such as GDPR in the EU and HIPAA in the US aim to protect the data privacy of citizens. HIPAA protects personal health information, while GDPR is broader and protects any personally identifiable information. Organizations in these respective regions must have adequate security measures to protect personal data. Any data collected and used by an AI diagnostic must meet these regulations.

In addition, the AI Act came into force in the EU in August 2024, providing regulation with a framework for AI systems across all industries. AI-based medical software is considered a high-risk AI system and must comply with strict requirements for risk mitigation, robust datasets, and clear user guidance. This is a fast-evolving regulatory space and manufacturers will need to adapt to answer specific requirements in the coming years.

Why is regulatory compliance important when developing medical devices?

Medical devices are further regulated to ensure patient safety and efficacy. Manufacturers of software as a medical device (SaMD) must ensure appropriate authorization before placing them in the market– CE marking in the EU, Food and Drug Administration (FDA) 510(k) clearance or approval in the US. To try to achieve the greatest impact for patients and pathologists, Owkin pursued CE marking for MSIntuit CRC, and achieved the CE-IVD mark for diagnostic use in 2022. It is the first CE-marked AI diagnostic for colorectal cancer.  

The FDA categorizes medical devices based on levels of risk to patients and the level of regulatory control needed to ensure safety and effectiveness. Class I devices pose the lowest risk to patients and have the least regulatory control. Class II devices pose a moderate risk and require more regulatory controls. Most Class II devices go through a 510(k) clearance process to demonstrate that they are substantially equivalent to legally marketed devices. Class III devices pose the highest risk and have the most stringent regulatory controls.

In the EU, manufacturers must obtain a CE mark for medical devices. Similar to the FDA, regulatory controls and marketing pathways are based on risk. From 2022, medical devices must obtain their CE mark under a new regulation (IVDR). It is more stringent than the previous invitro diagnostic directive, in terms of risk classes and oversight provided by the notified bodies, especially for software.

Launching an AI-based tool for clinical use across multiple markets presents significant challenges, as each country may require separate regulatory clearance or approval. Many countries also expect that the product validation and generated evidence will incorporate patient populations represented within their borders. An investment in multiple studies, for multiple products, across many regions is required.

This journey has highlighted the significant time and effort required to achieve clinical acceptance of new technologies. For instance, it took us nearly a decade to establish the routine use of TILs in clinical settings, illustrating the complexities involved in convincing healthcare providers, laboratories, and regulatory bodies to embrace new biomarkers.

— Doctor Roberto Salgado, pathologist at ZAS, Belgium.

Due to the long and complex regulatory process, it is essential to plan for this at the outset of developing an AI diagnostic. The requirements in different countries must be evaluated to ensure proper design, validation, and documentation. If overlooked early on, this could lead to costly redesigns, delays, or prevent the device from being approved for clinical use. 

Regulatory guidelines also clarify the necessary steps for clinical testing, data handling, risk management, and post-market monitoring, all of which are essential for building a trustworthy, effective diagnostic.

Reimbursement hurdles to distribution

Regulatory requirements aren’t the only barrier to getting an AI diagnostic into clinical use. Reimbursement is typically covered by payers who can be public and/or private health insurance programs, but it is a complex and challenging issue and also varies by country and device classification.

AI diagnostics can be reimbursed if they are approved by the regulatory body in a country and meet the coverage criteria of payers. Payers require a sufficient amount of clinical evidence of effectiveness and safety before covering a medical device. In some markets, this means that a solution must perform significantly better than the current gold standard, while in others, it is sufficient to show that the product is efficacious and could save the payer or healthcare system money.

In the US, services must have a billing code in order to get reimbursement from insurance companies and Medicare. CPT codes for digital pathology became available in 2023. However, these are for digital pathology only, not AI. Many in the industry see this as a positive sign towards a future for reimbursement for AI medical devices, but further progress is needed to advocate for the many use cases of AI diagnostics. Reimbursement in Europe varies by country, as different healthcare systems have different expectations and funding requirements, with Germany and France leading the way with options for innovation funding and AI solution task forces.

Owkin is generating evidence through workflow studies, evaluating performance with clinical studies, and conducting external validations across multiple centers for several products. We are also engaging with payors and healthcare authorities to understand evidence requirements for access pathways. These interactions introduce our capabilities to relevant authorities, encourage consideration of novel AI medical devices, and help shape our product-market fit and access strategies.

Summary

As the development of an AI diagnostic is long and complex, the path to market needs to be considered at the outset so that sufficient clinical evidence can be collected. These requirements also influence data curation needs even before an AI model can be trained.

Clearly, there are still hurdles to regulatory approval and reimbursement in many countries. Overcoming these barriers will take a united effort by all stakeholders.

Owkin is a part of multiple industry work groups like France Biotech’s Task Force for Digital Pathology and the Digital Pathology Association in the US, plus regional consortia like Pathlake in the UK, EMPAIA in Germany, and the PortrAIt consortia that we lead in France. These activities keep us abreast of the latest developments and enable us to work across the industry to better shape the market and regulatory needs for the short and long term.

In the next and final blog, we’ll look at how AI diagnostics are integrated into research and clinical pathology workflows, plus summarize what we’ve covered in this blog series.