What the Feds Are Doing With AI—and What Patient Care Organization Leaders Could Learn

A lot of activity is taking place these days at ASTP/ONC—the Office of the Assistant Secretary for Technology Policy/Office of the National Coordinator for Health IT. Among other things, ASTP/ONC professionals are very much involved in using artificial intelligence (AI) tools themselves, as they work to optimize internal processes. As a consultant with Audacious Inquiry, a PointClickCare company, Greg Farnum, Audacious Inquiry’s general manager, is in the thick of things with that work.

Recently, Healthcare Innovation Editor-in-Chief Mark Hagland spoke with the Vermont-based Farnum (Audacious Inquiry is headquartered in Toronto). Below are excerpts from that interview.

Let’s begin with your background. You have a decades-long background of healthcare IT experience, correct?

Yes, that’s correct, I’ve been involved for many years. I’m on my thirtieth year now in healthcare IT. I started in a hospital association, and have been working in the state of Vermont my entire career. And it’s great to be in a small state, where you can literally get the providers, payers, and policy people into the same room. I was at a hospital association for ten years, then helped to create VITL [the organization that operates the Vermont Health Information Exchange] for ten years, then ONC/ASTP for four years, then Microsoft briefly, then Audacious Inquiry twelve years ago. And we’ve been working recently with ASTP, to help them with some of their challenges.

What have the leaders at ASTP been doing with AI?

ASTP has been struggling with some challenges around some of the complex data analysis they do—really some routine things like content creation like blog-posting, regulatory updates. And then they deal with thousands of public inquiries every year coming out of their role around certification. What we found was we had to spend a couple of months across ten different workgroups to document their current workflows, to figure out how to use AI to help them. They were drowning in in document generation—newsletters, blogs, etc.

This was internally inside ASTP/ONC?

Yes, around administrative workflows internal to ASTP; but a lot of this is parallel to work being done in patient care organizations. For example, at ASTP, every year, there are hundreds of health IT developers who submit their real-world testing plans, but there’s no standard format for that. They standardized it on their side; some plans can be 50 or 200 pages long. Using NLP and other tools, they pull out the answers. So in a patient care organization, you can use it to answer questions or respond to regulatory documents, where you can implement robotic processing automation and other tools to take massive regulatory or other types of documents and internal data, and use it to auto-complete regulatory documents, to save providers a pile of time. We’ve also been working with them to achieve better document creation and better communication with patients, and automatic scheduling, things that are low-hanging fruit for healthcare organizations.

Speaking of staffing shortages and related issues, their ticketing system holds thousands of answers to questions previously asked. So the tickets are coming in, and the subject-matter experts inside ASTP are answering questions. And we’re using the AI to draft an answer for human review, to a question, and it can route it to the right person. They’re thinking they’ll save hundreds of work-hours a year for their staff, because this information is instantaneously available, and generating fairly simple low-hanging fruit generates an answer and makes sure it’s in alignment with their processes.

We’ve now got algorithmic, generative, and agentic AI, all being developed in healthcare: where will patient care organizations be able to move forward most quickly?

Part of it involves convincing people it’s safe to move forward. In the next couple of years, we’ll see a tremendous amount of movement across the board on clinical-use cases. But the smart money will be on the low-hanging fruit, the administrative tasks that aren’t super-high-risk. Until we settle in and build trust around different architectures, and making sure AI is best-trained, the lower-risk administrative processes will be easy buy-in. They’re time-savers and money-savers.

Do you think areas around coding/revenue cycle management/physician documentation, will be fertile for development?

I do. It’s not dissimilar from other use cases we’ve tackled before. There’s a lot of money involved, so it’s easy to document potential ROI. And money often drives change in physician behavior. I think those bridge topics make a lot of sense, and the ability to do pattern recognition, per clean claims and dirty claims, there’s potential there. That can help the physician to record and code more accurately their patient visits. That’s really set up for it. And all these things come together inside a nice business case, so you can get the attention of leadership.

What kind of advice might you like to share with patient care organization leaders? The technology is moving fast, but it’s hard to know when to leap in, correct?

At ASTP, we actually spent a lot of time on low-cost technology and figuring where to plug AI in to save money. And there are the governance issues, to make sure there’s transparency, monitoring, accountability; the folks who are sprinting to get something started may skip a few steps and end up having to backtrack a little bit. There’s a low-cost way of doing this: per administrative workflows, you can take one area, start small, and build on your wins. Depending on what kind of money you have to spend, let things settle out a bit; go as fast as you can, but make sure you lay a nice foundation.

What’s your sense of the pace of development of usable technology for patient care organizations in the next few years?

I think it will depend. There’s a question mark as to how much will end up becoming agentic AI. So it’s hard to say. I agree that you have check in almost daily to understand where things are, how the big players are competing with each other to capture the biggest market share. I certainly don’t think we’ll be dealing with the same tools several years from now. In HC, we’ll see tremendous progress in the automation of the clinical workflow and the administrative workflow. So you need to lay a good foundation to make sure you’re not ahead of your clinicians and staff and patients. 

It can be a bit like what consumers have to figure out in terms of, for example, whether they should leap into the latest version of the iPhone, or wait a bit, correct?

Yes, that's absolutely correct. Is the iPhone 16’s camera that much better than the 8’s? Buy wisely. Buy within your budget; spend what you can, but don’t stop progress because you can’t afford the latest, greatest tool, when there are some things out there that work great. I think the efficiencies will far outweigh the cost, if we can get the foundation right. And as you start to recoup savings of time and money, you can reinvest in technology. Everybody’s going to want the shiniest object, but it’s more important to move strategically, move carefully, but yes, move. Move responsibly, but move.