Safe, effective, and trustworthy AI for healthcare delivery organizations and populations across the United States - Project Summary Artificial intelligence (AI) is being developed and deployed across the U.S. healthcare system at a rapid rate with the goal of improving quality and efficiency. AI applications vary dramatically - from predicting risk of sepsis and type 2 diabetes to managing appointment scheduling and billing. Healthcare delivery organizations are therefore faced with the challenge of how to select, implement, and monitor a broad set of AI tools across clinical areas while maximizing value and minimizing risk to patients. Recent data suggest that most healthcare delivery organizations are not succeeding in meeting this challenge, as more than half of US hospitals that have deployed AI are not consistently conducting robust local evaluation of the tools they use. This indicates that organizations are not yet ‘AI capable’ - that is, they do not have the organizational routines, standards, and resources required to successfully navigate the complex and shifting requirements of AI governance. The reasons for the current lag in AI capability are likely variable by the technological capacity, system characteristics, financial resources, patient populations, data practices, and electronic health record infrastructures of different health care organizations (e.g., large health systems, small community health centers). Without evidence-based tools for AI capability tailored to these organizational differences, a digital divide between organizations in their ability to use AI safely and effectively will negatively impact patient care and safety. The proposed project will use mixed methods to identify barriers and mitigation strategies for AI capability across organizational contexts that will ensure AI is safe, effective, and trustworthy throughout the U.S. healthcare system. In Aim 1, we will analyze high-impact AI use cases to identify barriers to and best practices for local AI evaluation routines across heterogeneous organization types (large health system and community health center). In Aim 2, we will identify key attributes of patient-centered AI that promote safety (e.g., oversight), effectiveness (e.g., quality), and trustworthiness (e.g., transparency) through a national survey of the public that multiple common AI applications representing different points along the care continuum (e.g., diagnosing, scheduling) to inform patient-centered AI standards. In Aim 3, we will co-design a modular AI capability toolkit with patients, experts, and healthcare leaders that will adapt to different contexts across the country to ensure organizations can access and safely apply the right AI resources at the right time. To ensure that all organizations are AI capable and narrow the digital divide between them, the proposed project will identify and operationalize key components of AI capability, tailored to varied healthcare delivery organizations.
