LAS VEGAS—While the debate over how to ensure AI tools are safe and effective is being tackled by stakeholders across the healthcare industry, health systems are underprepared to effectively implement AI solutions, the nation’s leading AI experts said at HLTH.
One of the nation’s largest clinical AI companies, Aidoc, is teaming up with software heavyweight Nvidia to develop a set of guidelines for AI implementation called BRIDGE: the Blueprint for Resilient Integration and Deployment of Guided Excellence for AI adoption.
Demetri Giannikopoulos, chief transformation officer at Aidoc, said the companies agreed to join forces to create a free, open-source guide on the implementation of AI because both have widely deployed AI models across the country. The evidence-based and vendor-agnostic set of best practices will also include input from a variety of health systems and healthcare professionals.
The BRIDGE guide will seek to address on-the-ground issues like solution fragmentation, scalability and interoperability that health systems are likely to face when they implement AI. Giannikopoulos said that a large portion of the guide will help healthcare delivery organizations with change management and adapting their workflow to a new AI model.
It will also help systems think about AI governance from the beginning of their AI implementation journey. The guideline will be available in early January 2025.
"Our goal with the BRIDGE guideline is to create a consistent, scalable framework for AI adoption in healthcare,” Giannikopoulos said in a statement. “By partnering with NVIDIA, who has led the way in AI acceleration, and involving health systems in its development, we’re ensuring that AI can be integrated into scalable clinical practice. Together, we’re helping health systems overcome challenges like fragmentation and interoperability, so both systems and patients can benefit from AI at scale."
Giannikopoulos said the implementation guide harmonizes with the effort of the Coalition for Health AI to create a nationwide network of AI assurance labs. Aidoc is a member of CHAI, and Giannikopoulos serves on its model card working group.
While the assurance labs will validate the safety of an AI model and its accuracy across subgroups, the implementation guide will include practical considerations on how to deploy the AI solution once validated.
Some of the practical considerations will include continuous monitoring for model drift and drift of the underlying data.
The guide will encourage healthcare organizations to consider federal regulations that may impact the development or deployment of an AI solution, but it will not give detailed advice on policy. Giannikopoulos said there are many unanswered questions about which models are regulated.