The Advanced Research Projects Agency for Health is embracing generative artificial intelligence as part of its internal operations, while also considering how the technology could be used across the broader health ecosystem.
Launched in March 2022, ARPA-H is an independent agency within the National Institutes of Health. The organization is modeled after the Defense Advanced Research Projects Agency, with the goal of supporting “transformative research to drive biomedical and health breakthroughs.”
Susan Monarez, deputy director at ARPA-H, said officials are already using generative AI technologies to help the new research organization “get our houses in order.” She said the agency processes large volumes of data ranging from personnel information to contracting data to information from partner agencies that help ARPA-H shape its programs.
“We’re leveraging the power of generative AI to be able to learn from that, not just harness it as data, not just information, but translating that into knowledge to improve how we’re recruiting, how we’re retaining highly talented individuals, but also how we’re positioning our programs to be unique, and have the largest potential impact to optimize our resources,” Monarez said during a Nov. 15 webinar hosted by the Association for Federal Information Resources Management.
As ARPA-H officials consider the specific healthcare challenges they want to address, Monarez said generative AI can also help process the “zettabytes” of data related to potential program areas.
“We have to make sure that we are gaining knowledge from all of the activities that are currently ongoing across the health ecosystem, and pulling that in a meaningful way, so that we are creating a unique niche, a unique opportunity for ARPA-H to really accelerate these very high risk, high reward programs and initiatives across the healthy ecosystem,” she said.
Earlier this year, ARPA-H issued its first funding opportunities through an agency-wide Open Broad Agency Announcement. It has also started launching research programs, including one aimed at developing technologies for more precise cancer tumor removal, and another with the goal of designing vaccines that can target multiple viruses at once.
Monarez said those programs are generating reams of data as well.
“We want to be able to pull that data in as quickly as possible to figure out what’s working, what’s not working, where the opportunity is to modify those programs for the highest impact,” she said.
Beyond its internal operations, ARPA-H also sees opportunities to apply generative AI to healthcare challenges.
“We fully anticipate having a number of program managers who are coming in to launch really big programs in this space that will leverage generative AI to help enable capabilities across the patient, payer and provider sectors,” Monarez said.
She said ARPA-H specifically thinks generative AI could help with the challenge of integrating patient data.
That information is often found in different locations, ranging from a patient’s electronic health records to a personal health device. The data usually isn’t integrated “in a way that allows us to gain greater insights into patient populations representative across this country,” Monarez said.
“How could we actually help improve those patient outcomes through improved access to high quality primary care, help access screening diagnostics, help accessing monitoring for someone who has a chronic disease,” she continued. “So we are looking at that particular space. We don’t have any programs yet, but it is something that we’re super excited about.”
ARPA-H also wants to help build assurance in AI-based healthcare tools. As vendors potentially integrate AI into their digital health technologies, Monarez said healthcare providers will only use them if they trust the output, especially regarding decisions on patient care.
“If that technology is not sufficiently trained, or we don’t have sufficient assurances in place when the training data set deviates from the patient population that it’s currently helping to support, there is a real risk that any of the information that a digital health technology or an enabling service will actually cause more harm than potentially good,” she said.
Monarez said creating AI assurance as an “overlay” for digital health technologies could help address patient health and safety risks.
“Is there something more systematic that we can do that would reduce the burden on digital technology performer-developers, so that they don’t have to worry about that,” she said. “They can worry about their algorithm, and we can put in place this assurance. That alone is sort of ARPA-H challenge . . . worthy.”
The research agency is also considering how AI could help streamline administrative services, which are time consuming activities that represent upwards of 30% of all healthcare costs. Monarez said any resources saved on those administrative activities could be funneled back into patient care and provider services.
“Looking across that triumvirate of patients, payers and providers and thinking about holistically, how can we actually use generative AI,” Monarez said. “Again, no programs launched yet. This is a little teaser, but we are thinking very much about how we could help contribute to that space.”
ARPA-H is not the only agency considering the risks and opportunities associated with using AI for healthcare. The NIH, the Centers for Disease Control and Prevention, the Department of Veterans Affairs, and the Food and Drug Administration are also working on AI-related issues.
Monarez said officials are exchanging lessons and information about progress on discrete programs through a federal interagency coordination committee. The goal is to ensure agencies share information about their progress across different research and development activities without duplicating efforts.
“Our goal is to make sure that we have consistent and complementary outreach with those federal partners,” she said.