AI in Medical Devices and Healthcare: Opportunities, Challenges, and What Lies Ahead

In recent years, the digitalization of the healthcare industry has been accelerated to meet demands for smarter devices and robotics, wearable technology, AI-based data analysis, and enhanced platforms and simulations, among others. This digitization has driven an increased interest in incorporating artificial intelligence (AI) and machine learning (ML) technologies into medical devices.

Over the last decade, the US Food and Drug Administration (FDA) has reviewed and authorized a growing number of devices using its 510(k) clearance, de novo, and approved premarket (PMA) approval processes with AI/ML functionality across many different therapeutic categories—and anticipates this trend to continue. In addition, AI and ML technologies may be used to support the investigation, development, and/or production of medical devices and other FDA-regulated products. When used for medical or other healthcare-related purposes, these technologies are likely subject to FDA regulations, policies, and guidance.

In this article, we discuss the existing FDA programs and recently issued guidance impacting AI/ML technologies intended for use in healthcare, as well as what to expect from the FDA’s fiscal year (FY) 2023 priority list. We also examine the reimbursement framework for AI/ML and some challenges ahead for the medical device industry.

EXISTING FDA PROGRAMS IMPACTING AI/ML TECHNOLOGIES

FDA’s regulation and oversight of AI/ML software continues to grow, as evidenced by the online list compiled and maintained by FDA’s Center for Devices and Radiological Health (CDRH) of medical devices using AI/ML technologies that CDRH has cleared or approved. That list currently includes more than 500 devices, the vast majority of which were cleared via the 510(k) process, along with a few de novo submissions and PMA applications. In terms of an FDA review branch, a significant majority fall under radiology, followed by cardiovascular, hematology, and neurology.

However, FDA/CDRH’s regulatory priorities for AI/ML technologies expand beyond premarket review, and are led by CDRH’s Digital Health Center for Excellence.

• Launched in September 2020, the Digital Health Center for Excellence’s main purpose is to foster responsible and high-quality digital health innovation. Its three main goals are to develop and issue guidance documents, increase the number and expertise of the digital health staff, and develop the Software Precertification Pilot.
• On January 12, 2021, the FDA released its first Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) Action Plan that details a multipronged approach to advance FDA oversight of AI/ML-based medical software. This action plan is in response to stakeholder feedback that it received from an April 2019 discussion paper, Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning–Based Software as a Medical Device.
  • The five-point SaMD Action Plan outlines how to further develop the proposed regulatory framework, including through draft guidance issued on a predetermined change control plan; harmonize good ML practices to evaluate and improve ML algorithms; foster a patient-centered approach, including device transparency for users; support regulatory sciences; and advance real-world performance monitoring pilots.
• In October 2021, the FDA, Health Canada, and the UK Medicines and Healthcare products Regulatory Agency (MHRA) jointly identified 10 guiding principles that can inform the development of “Good Machine Learning Practice” for medical devices and how they can help promote safe, effective, and high-quality use of AI/ML.

Recently Issued Guidance Documents Affecting AI/ML

CDHR remains active in promulgating guidance documents impacting AI/ML technologies, including the following recently issued guidance:

• Clinical Decision Support Software – Final Guidance (September 28, 2022): This long-awaited final guidance represents a significant and more conservative shift from the prior draft guidance issued in September 2019 and could be present challenges for developers of AI/ML technologies. This guidance document describes FDA’s interpretation of the statutory exemption for clinical decision support (CDS) software functions under the Federal Food, Drug, and Cosmetic Act. Software that meets the four criteria set forth in the statute would be exempt from FDA’s medical device regulatory requirements. FDA’s interpretation of these four criteria, as described in the final guidance, will make it more challenging for software developers to fit their software products (including AI/ML software) within the scope of the CDS exemption. Further, unlike the prior draft guidance, the final guidance does not include any proposed enforcement discretion policy for software that does not fully meet all four statutory criteria.
• Computer Software Assurance for Production and Quality System Software – Draft Guidance (September 28, 2022): This new draft guidance provides recommendations for “computer software assurance” for software and automated systems used for medical device production or quality. The guidance describes “computer software assurance” as a risk-based approach to establish confidence in the automation used for production or quality systems, and identify where additional rigor may be appropriate. The guidance further includes various methods and testing activities to establish “computer software assurance” and ensure compliance with quality system regulation and other regulatory requirements.
• Cybersecurity in Medical Devices: Quality System Considerations and Content of Premarket Submissions – Draft Guidance (April 8, 2022): This new draft guidance would replace the 2014 final guidance, Content of Premarket Submissions for Management of Cybersecurity in Medical Devices. This is FDA’s second attempt at a new draft—a prior draft guidance was issued in 2018 and received significant criticism. This new draft guidance includes changes to align with the use of a Secure Product Framework, removal of risk tiers (from the prior draft), replacement of the Cybersecurity Bill of Materials with a Software Bill of Materials, additional clarification regarding premarket submission document requests throughout the draft guidance, and addition of Investigational Device Exemptions to the scope. The guidance also makes clear that cybersecurity is part of FDA’s Quality System Regulation (QSR) design control requirements.
• Digital Health Technologies (DHTs) for Remote Data Acquisition in Clinical Investigations – Draft Guidance (January 21, 2022): This draft guidance describes considerations when using DHTs in clinical investigations and applies to all types of clinical investigations (whether the investigation is for a drug, biologic, or device product) using a digital health technology for remote data acquisition.
• Assessing the Credibility of Computational Modeling and Simulation in Medical Device Submissions – Draft Guidance (December 23, 2021): This new draft guidance sets forth a proposed nine-step process to assess the credibility of computational modeling and simulation (CM&S) used to support a medical device premarket submission. CM&S can be used in a variety of ways in medical device regulatory submissions, such as to support in silico device testing or as-a-device development tools, or within the device itself as software as a medical device (SaMD) or software in a medical device (SiMD).

What’s in Store for 2023?

Toward the end of 2022, the CDRH published its annual list of intended guidance documents for FY 2023 (A-List and B-List). The following priorities from these lists are most likely to impact AI/ML technologies:

A-List

• Cybersecurity in Medical Devices: Quality System Considerations and Content of Premarket Submissions (final)
• Content of Premarket Submissions for Device Software Functions (final)
• Transition Plan for Medical Devices That Fall Within Enforcement Policies Issued During the Coronavirus Disease 2019 (COVID-19) Public Health Emergency (final)
• Transition Plan for Medical Devices Issued Emergency Use Authorizations (EUAs) During the Coronavirus Disease 2019 (COVID-19) Public Health Emergency (final)

B-List

• Marketing Submission Recommendations for a Change Control Plan for Artificial Intelligence/Machine Learning (AI/ML)-Enabled Device Software Functions (draft)

Changes are also on the horizon for AI/ML-enabled devices marketed for pandemic-related uses under an emergency use authorization (EUA) or one of FDA’s many COVID-related guidance documents describing enforcement policies. As noted above, CDRH’s A-list includes finalization of its previously issued draft guidance documents on transition plans for such devices. Under the draft guidance documents, FDA had proposed a three-phase, 180-day transition period for devices covered by either an EUA or a COVID-related enforcement policy. The final guidance documents are expected to issue this year.

REIMBURSEMENT DEVELOPMENTS FOR AI/ML TECHNOLOGIES

Above, we examined how AI/ML technologies may be approved by the FDA. But what are the benefits of developing said technologies if they cannot be implemented into medical devices and sold? How will these AI/ML technologies find their way to day-to-day use in the healthcare industry?

Limited but Growing Opportunities for Direct Reimbursement

The reimbursement framework for AI/ML technologies is not advanced, and there are currently limited opportunities to realize direct reimbursement. One of the biggest impediments lies in the fact that US healthcare reimbursement remains focused on clinicians as the “source” of a reimbursable service. Recognizing that, removing a clinician from a patient care service is in some ways antithetical to the existing payment framework. But, although human clinicians remain a fixture, AI/ML can enhance clinicians’ ability to make faster decisions based on larger sets of patient data collected, see more patients due to efficiencies in appointments and evaluations, and lead to an overall reduction in time and overhead costs.

Indeed, the American Medical Association (AMA) has already developed a conceptual framework for AI/ML in healthcare, releasing its classification system in December 2021 and updating it in August 2022. The AMA recommends classification of AI devices into three overall categories based on the “work performed by the machine” in delivering an overall service: Assistive, Augmentative, and Autonomous.

Minimal Exploration by Federal Health Agencies

There has been some minimal, but growing, exploration of AI/ML reimbursement in federal healthcare programs in recent years. The Centers for Medicare & Medicaid Services (CMS) has been exploring reimbursement for certain limited procedures utilizing AI since 2018, but recent activity demonstrates that the agency’s interest is increasing.

In 2022, CMS continued to explore payment for Current Procedural Terminology Code 92229 (described by AMA as an “autonomous” service) in both the Hospital Outpatient Prospective Payment System (OPPS) and Physician Fee Schedule (PFS), and it requested public comment about software as a service, analytics, and payment for new technologies and clinical software, not only in the context of OPPS payments but as part of future adjustments to the PFS practice expense methodology. Changes to the practice expense methodology could be a game-changer for various technologies that struggle to achieve reimbursement because the expense must be directly incurred by the physician practice.

AI/ML Impact in Value-Based Care

• While direct reimbursement of AI/ML technology remains limited, AI can nevertheless be successfully integrated into other existing payment models.
• Increased efficiencies and better outcomes that certain AI/ML technologies can foster will ultimately result in greater shared savings opportunities for healthcare providers involved in value-based care models and other alternative payment models.
• Private insurers have the flexibility to reimburse for services in a variety of ways, including through pilot programs that may attempt to test the clinical and financial return on investment of AI/ML-enabled services.

State Laws Impacting AI/ML

Irrespective of growing reimbursement opportunities, the use of AI/ML in healthcare settings quickly implicates state rules governing the practice of medicine and other licensed professions. Already, many state medical boards are assessing how the introduction of AI will reshape medical practice.

Boards are considering the impact of telehealth, AI, and the use of other technology on the standard of care and how licensees should responsibly use these tools to furnish healthcare services. For exaample, the Federation of State Medical Boards passed a resolution in 2018 to establish a workgroup on “AI and Its Potential Impact on Patient Safety and Quality of Care in Medical Practice.” Although the working group has not yet issued formal guidance, this highlights professional licensing agencies’ focus in identifying how AI can improve patient safety and care and whether a revised regulatory framework may be necessary to respond to this new reality.

Watch our on-demand Artificial Intelligence Boot Camp session for more information on AI developments in digital health.