The Department of Health and Human Services (HHS) released a strategic plan on artificial intelligence last week, 10 days before President Joe Biden leaves office.
The strategic plan outlines short- and long-term goals for HHS in medical research and discovery, medical product development, healthcare delivery, human services delivery, public health, cybersecurity and critical infrastructure protection, and internal operations.
For healthcare delivery, HHS will focus its efforts on technical assistance programs, training healthcare providers on AI, and clarifying how existing regulation applies to AI, including on the topics of privacy, security and liability.
The strategic plan specifically calls out the Coalition for Health AI (CHAI), which is working to establish a nationwide network of quality assurance labs for AI. The strategic plan says HHS will consider supporting the external validation of algorithms by third-party, conflict-free organizations.
This follows HHS Deputy Secretary Andrea Palm’s promise for a strategic plan and support for a nationwide network of quality assurance labs for AI at the Health Datapalooza conference in September.
HHS’ five strategic goals for AI in healthcare delivery are: to catalyze health AI innovation and adoption, promote trustworthy AI development and ethical and responsible use, promote external validation, enhance healthcare infrastructure to accommodate AI and empower the healthcare workforce.
The strategic plan provides a lengthy list of AI use cases in healthcare delivery, including different interventions based on the setting, like assisting with clinical notes for providers, managing administrative tasks for healthcare organizations, and adjudicating claims for payers. In each of these categories, HHS also lays out a list of potential risks that correlate to the benefit the technology could provide.
Catalyze health AI innovation and adoption
To spur AI innovation and adoption, HHS plans to offer grants to support research on best practices, use federal programs, and publish information to help organizations decide how to invest in AI. It will also build on “challenge” initiatives like the CMS AI Health Outcomes Challenge and the NIH CRDC AI Data Readiness Challenge.
It also says it will develop guidelines on how to pilot AI within a healthcare organization before it fully implements an AI product.
To provide clarity on coverage and payment, HHS says it will create evidentiary standards for AI developers to meet to secure coverage and payment.
Promote trustworthy AI development and ethical and responsible use
To promote trustworthy AI, HHS plans to release additional, targeted guidance for how existing regulation applies to AI and technology. It plans to collaborate with other agencies, like the Federal Trade Commission, to do so.
HHS says it may require healthcare organizations to form internal committees to assess how AI is being deployed in their organization. It also may set minimum standards for AI risk assessment.
HHS will support efforts to develop risk management frameworks and AI governance within healthcare delivery organizations. The strategic plan also notes that HHS will decide what bodies could best enforce AI regulation, like through audits and corrective action plans. And, it will develop specific standards for the use of AI in hospitals.
The agency will also consider whether rulemaking could enable successful governance and oversight of AI. It also wants to motivate and support “nationwide public-private approaches to validate AI.”
Another area outlined in the strategic plan was the opportunity to work with the industry on consensus requirements for HTI-1’s AI source attribute model cards.
HHS says it will “explore the use of federal programs to … motivat[e] deployers of AI to undergo independent, external algorithmic audits conducted by certified entities free from conflict of interest.” The strategic plan specifically identifies the efforts of the CHAI and the Trustworthy & Responsible AI Network (TRAIN), headed by CHAI co-founder and board member Michael Pencina.
HHS says it will consider incentives for providers and payers to disclose their use of AI and its performance information.
Democratize AI technologies and resources
Technical assistance is the backbone of HHS’ recommendations to democratize AI technologies and resources. HHS says it will establish regional technical assistance centers to aid under-resourced care settings on the use of AI.
The agency also said it would fund and disseminate AI impact assessment templates and implementation toolkits, as well as fund research on AI best practices in under-resourced settings, including helping under-resourced organizations run pilots of AI tools.
It also calls for technical assistance to help healthcare organizations modernize their technology infrastructure to support AI, disseminate AI readiness assessments, and promote open-source AI specifications for stakeholders. Here it links CHAI’s AI model card, which was recently released in a draft form on GitHub.
Cultivate AI-empowered workforces and organization cultures
“HHS will take steps to increase AI knowledge and expertise among healthcare professionals, ensuring foundational know-how within delivery organizations that lowers the cost of implementing new tools and ensures they are applied appropriately,” the strategic plan said.
All of HHS’ goals in this section are long-term goals. The agency plans to do long-term public education initiatives, convene regular AI sessions at healthcare conferences, fund workforce training programs and share AI training resources. The agency will also study how AI impacts the healthcare workforce.