Evaluating synthetic intelligence options from distributors is likely one of the greatest challenges informatics leaders face at present. The Well being AI Partnership (HAIP), a multi-stakeholder collaborative, has printed in NEJM AI an outline of its AI Vendor Disclosure Framework, a instrument designed to help accountable AI system procurement.
First some background on HAIP: It seeks to be a useful resource for steerage for healthcare professionals utilizing AI and associated rising applied sciences, and a platform for community-generated, expert-curated steerage, sources, and requirements for accountable AI adoption in healthcare. The group has developed a community that creates a protected house for peer recommendation and collaboration to handle points well being system leaders face whereas adopting AI in healthcare settings. Its Coordinating Heart staff, situated on the Duke Institute for Well being Innovation (DIHI), manages and coordinates the partnership’s actions. HAIP acquired preliminary funding in 2022 from the Gordon and Betty Moore Basis to determine this neighborhood useful resource.
Based on HAIP, the AI Vendor Disclosure Framework, which is publicly out there and free to make use of, identifies important data throughout 5 core domains that well being techniques ought to request — and distributors ought to disclose — to successfully consider vendor-developed AI techniques:
- System Capabilities and Meant Use establishes foundational data concerning the AI system’s functionalities, use, and affected stakeholders.
- System Efficiency and Compliance establishes the AI system’s operational metrics, potential biases, related dangers, and regulatory standing.
- Knowledge Stewardship outlines the method to knowledge governance, together with safety measures, high quality assurance processes, secondary use, and retention insurance policies.
- Integration Necessities consider the overall price of possession, together with technical conditions, useful resource necessities, and implementation timelines.
- Lifecycle Administration defines vendor duties for ongoing help, monitoring, and upkeep after implementation.
By standardizing expectations for the data wanted in procurement decision-making, the framework goals to reinforce transparency and promote safer well being care AI adoption. It serves each as a best-practice information and a customizable useful resource to help healthcare supply organizations in procuring vendor-developed AI techniques.
Vega Well being Co-founder and CEO Mark Sendak, M.D., M.P.P, was a part of the framework’s improvement staff. On LinkedIn, he defined the importance of the brand new useful resource. He wrote that “Mannequin details labels/mannequin playing cards are nice sources for front-line clinicians who want a high-level synthesis of what an AI resolution is, the way it was constructed, and the way it needs to be used.” However he added that the mannequin details label/mannequin card isn’t enough data to information procurement and implementation selections. “The knowledge wanted for these stakeholders is way more complete throughout domains and way more detailed,” Sendak wrote. “Most people do not admire the distinction. Therefore, by way of Well being AI Partnership we pulled collectively a gaggle of leaders throughout a number of establishments who had been already constructing out these vendor assessments exactly as a result of the data supplied by distributors was inadequate for procurement selections.”
Earlier than founding Vega, which seeks to curate a market of healthcare AI options confirmed protected and efficient in real-world settings, Sendak was a inhabitants well being and knowledge science lead on the Duke Institute for Well being Innovation.
