The Advanced Medical Technology Association (AdvaMed) has published a policy roadmap for artificial intelligence (AI), identifying the tremendous benefits AI has to offer patients. However, AdvaMed also brought attention to several policy problems that Washington will have to overcome if AI is to fulfill its potential for patient care.
Scott Whitaker, AdvaMed president and CEO, said the future of AI in medical technology “is vast and bright. It’s also mostly to be determined.” He stated that AI will rapidly evolve and that policymaking on the part of the executive and legislative branches will have to stay abreast of that potential if patients are to enjoy the benefits of AI.
The roadmap cited privacy as a policy issue that will have to be managed to properly balance the need for innovation and the imperative of protecting the individual’s sensitive health data. One recommendation is that the Department of Health and Human Services take steps to ensure that the Health Insurance Portability and Accountability Act (HIPAA) permits sharing of data sets needed to train, test and validate AI models without violating privacy. This would include data needed to retrain an existing AI model. AdvaMed stated that the de-identification methods currently allowed by HIPAA “stifle the high-volume data usage and sharing that can optimize the development of safe and accurate AI models,” a roadblock that must be better managed to avoid unnecessarily impeding AI development.
AdvaMed stated that Medicare payment for AI is hampered by a framework that lacks the specificity and clarity to provide appropriate coverage and payment for AI and digital health technologies of all kinds. The Centers for Medicare & Medicaid Services (CMS) typically provides payment via incremental policy changes directed toward specific types of technologies, but AdvaMed acknowledged that there is no single, overarching policy that could address all the technologies in their various contexts of use.
One solution would be for Congress to pass legislation that allows CMS to provide coverage for AI and digital health technologies without the constraint of budget neutrality requirements. AdvaMed emphasized the need for a formalized payment pathway for algorithm-based health care services (ABHS), which could be handled by CMS without new statutory authorities. CMS could adjust its payment policies for software as a service to “provide stability and certainty moving forward,” the roadmap document states.
The paper states also that CMS may consider a modification of the Ambulatory Payment Classification (APC) policy for new technologies in a way that provides APC code stability. These APC codes could also be modified to provide payment rates that reflect the value of ABHS-based services. Other Medicare-related recommendations include improved clarity of policies for digital therapeutics and the development of payment pilot programs that test the benefits of AI technologies in a value-based payment model.
AdvaMed made an implicit argument against horizontal regulation of AI by stating that the FDA should remain the sole regulatory agency with oversight of AI-enabled medical devices. However, the roadmap also indicated that the FDA guidance for predetermined change control plans (PCCPs) fails to incorporate the breadth of the statutory authority provided by Congress to allow the use of PCCPs. AdvaMed stated that the guidance inappropriately restricts the types of premarket submissions that are eligible for use in a PCCP regulatory submission given what the statute would seem to permit.
Additionally, AdvaMed recommended that the FDA put forth more effort to aid in the development of international consensus standards for AI and to recognize those standards when appropriate. In a similar vein, the roadmap proposed that the agency adopt policies that harmonize with those of other nations when harmonization is practical. The roadmap gives the example of the policy collaboration between the FDA, Health Canada, and the U.K. Medicines and Healthcare Products Regulatory Agency that produced a set of good machine learning practices.
While the roadmap outlines several legislative and regulatory deficiencies for AI, AdvaMed stated that Congress “already has been thoughtful on considerations to serve patients with AI-enabled health tech.” Both the House of Representatives and the Senate have convened task forces to deal with AI policymaking, and Whitaker said the device industry “looks forward to continuing to work with Congress on these exciting advancements” in AI-based medical technology.