How Federal Health Agencies Are Scaling AI Pilots

Federal health IT leaders are finding practical ways to put AI into mission-critical workflows, emphasizing open-source platforms, workflow redesign and workforce enablement as keys to scaling adoption. 

At the National Institute of Allergy and Infectious Diseases, the team recognized early that the speed of AI innovation could outpace scalability and built a platform allowing users to develop and deploy AI tools to avoid a bottleneck.

The approach has helped NIAID deploy about two dozen applications, including tools for medical image analysis and research-grant coding. The strategy is particularly important as “citizen developers” become more common across the scientific workforce. 

“One of the things we’re focusing on a lot more is embedded AI,” said Associate Director of Informatics Joe Croghan at the Health IT Summit Wednesday. “I tell people Anthropic and Microsoft are spending more in one day than we’ll spend in a whole year on those things, so we’re going to use their chatbots, and we’re focused on embedding our AI systems in our current business practices and so forth.” 

Croghan said his team is also becoming more comfortable deploying AI before a system is fully mature, then improving it based on real-world use. 

“We’re saying fine, get it out there, and let people start using it, and we’ll add. Whereas typically, as IT people, we want to make sure the system has got all the bells and whistles and does everything it needs to do. So we’re doing much more of this kind of minimally viable product early on, getting something out there,” he said.  

Those efforts come as NIH is taking a broader approach to AI, including a new initiative aimed at using advanced computing and AI to accelerate the pace of biomedical discovery.

Chris Kinsinger, assistant director for catalytic data resources at NIH, said the newly announced BioGenesis Mission, part of an interagency effort led by the Energy Department’s Genesis Mission, aims to use AI to accelerate scientific discovery at twice the current pace. 

Bio Genesis Mission is focused on six challenges: biomedical foundation models, trusted AI access to protected and biomedical data, autonomous laboratories, identifying root causes of chronic disease, accelerating drug discovery and using AI to unlock cures for pediatric cancer. 

“We’re looking to have Bio Genesis serve that coordinating function across the NIH for how we implement AI in various areas, and so there are some things that’ll be coming out later this fall that are aiming to do that through these challenges, but then also broadening out across NIH,” he said.

For NIH, expanding AI’s impact will also depend on making valuable health data easier to use across organizational boundaries.  The agency is working to better connect its real-world data resources while recognizing the security and governance requirements surrounding them.

“We have a lot of great NIH resources that many of them house real-world data, but they’re not very connected,” Kinsinger said. “Could we use AI or some kind of federated learning solution to be able to annotate the research data with richer clinical data? That’s a big challenge. I think there’s a huge opportunity there and would love to see us move in that direction.” 

That emphasis on solving a specific mission problem is also driving AI adoption at Maximus. 

Managing Director of Federal Health Corinna Dan said the company has deployed an AI-powered system to support veterans’ disability evaluations, where clinicians previously had to review 2,000 to 3,500 pages of unstructured medical records. The system processes and structures the information before it reaches the clinician. The system has produced a ninefold increase in processing efficiency, with eight to 10 million pages processed per day, Dan said.

“We’ve taken, you know, activities that could be done by a human. We’ve pushed those to AI, and now we’ve got our clinicians spending their time using their judgment for the betterment of veterans who are awaiting these decisions,” she said.