What You Should Know:
– Today the Coalition for Health AI (CHAI) released a draft framework for responsible health AI with an invitation for public review and comment.
– The framework, consisting of an Assurance Standards Guide, provides considerations to ensure standards are met in the deployment of AI in healthcare. This draft framework was created with a consensus-based approach, drawing upon the expertise and knowledge of multiple, diverse stakeholders from across the healthcare ecosystem.
Assurance Reporting Checklists for Health AI: Building Consensus and Ensuring Equity
A set of draft companion documents, called The Assurance Reporting Checklists, has been developed to provide criteria for evaluating standards across the AI lifecycle, from identifying a use case and developing a product to deployment and monitoring. The principles underlying these documents align with several authoritative sources, including the National Academy of Medicine’s AI Code of Conduct, the White House Blueprint for an AI Bill of Rights, various frameworks from the National Institute of Standards and Technology, and the Cybersecurity Framework from the Department of Health and Human Services Administration for Strategic Preparedness & Responses.
Multiple, diverse stakeholders are involved in the selection, development, deployment, and use of AI solutions intended for patient care and related health system processes. This includes clinicians, nurses, AI technology developers, data scientists, bioethicists, regulators, patients, and their caregivers. The Guide aims to build consensus among these stakeholders by providing a common language and understanding of the lifecycle of health AI solutions, highlighting best practices for designing, developing, and deploying AI within healthcare workflows. This consensus will help ensure that AI technologies provide effective, useful, safe, secure, fair, and equitable care.
“We reached an important milestone today with the open and public release of our draft assurance standards guide and reporting tools,” said Dr. Brian Anderson, CHAI’s chief executive officer. “This step will demonstrate that a consensus-based approach across the health ecosystem can both support innovation in healthcare and build trust that AI can serve all of us.”
The Assurance Reporting Checklists translate these consensus considerations into actionable evaluation criteria to assist in the independent review of health AI solutions throughout their lifecycle. They ensure that AI solutions are effective, valid, secure, and minimize bias. Independent reviewers and organizations evaluating AI solutions, as well as individuals involved in the AI lifecycle, can use these checklists to review their work. Public reporting of the results from applying the Checklists ensures transparency regarding the risks and benefits of AI solutions, aiding organizational leadership in making informed decisions about AI technology development and deployment.
To demonstrate the practical application of these guidelines, the Guide describes six diverse examples showcasing variations in considerations and best practices in real-world scenarios: a predictive EHR risk use case for pediatric asthma exacerbation, an imaging diagnostic use case for mammography, a generative AI use case for EHR query and extraction, a claims-based outpatient use case for care management, a clinical operations and administration use case for prior authorization with medical coding, and a genomics use case for precision oncology with genomic markers.
In April 2023, CHAI released the “Blueprint for Trustworthy AI Implementation Guidance and Assurance for Healthcare,” the first broad consensus-based effort among subject matter experts from leading academic medical centers, regional health systems, patient advocates, and a range of healthcare and technology stakeholders in collaboration with federal agencies. The Guide combines principles from the Blueprint with guidance from federal agencies, while the Checklists provide actionable steps for applying assurance standards in day-to-day operational processes.
The Guide and Checklists were reviewed by CHAI’s editorial board and presented during its May community convening at Stanford University, which included patient advocates, regional and local health system leaders, technology developers, regulators, industry standard groups, and other stakeholders. Moving forward, CHAI will proactively engage stakeholders across the healthcare ecosystem, including patient advocates, under-resourced local health systems, and start-ups, to gather additional feedback and finalize the Guide. This ongoing input will ensure the Guide remains updated and relevant, addressing the urgent need for consensus standards and practical guidance to ensure that AI in healthcare benefits all populations, including underserved and underrepresented communities.