The next 5 years of AI in healthcare

By Laura Dyrda for Becker’s Hospital Reviews

Hospitals have begun incorporating artificial intelligence into their operational and clinical workflows to identify areas where the technology boosts clinicians and improves efficiency. But there are risks and executive teams need a strong plan to sustainably leverage AI for high quality, low cost care.

During the Becker’s 14th Annual Meeting, April 8-11 in Chicago, a panel of leaders from across the U.S. gathered to share their expertise on AI use cases, building an effective governance model, and what’s next. To close the session, each speaker answered the question: What will health system leaders need to set themselves up for success with AI over the next five years?

Note: responses are lightly edited for clarity.

Deepa Kumaraiah, MD, Senior Vice President and System Chief Medical Officer, NewYork Presbyterian: Systems are going to get bombarded, if they aren’t already, with inputs and inbounds. They need to ask themselves this question: What are we going to go deep on? We’re going to invest in the people, process and technology in the space because there’s a lot of point solutions right now, and at some point, [that changes]. Take radiology, as an example. There’s pulmonary embolism prediction, aneurysm prediction, stroke prediction and they are all different. At some point these are going to have to roll up to be operationalized in a workflow.

What has been valuable for us as a system, and is my advice in the space, is identify the problem you want to solve and where you’re going to go deep because we are all learning and we’re learning to ride the bike as we’re building it. Then we have to share with each other because it’s changing so quickly.

Michael Pfeffer, MD, CIO of Stanford (Calif.) Health Care and Associate Dean of Stanford School of Medicine: Sharing is going to be key. We all have to learn together on this. I am really excited about the potential. There’s coalitions to share and learn from each other about the key problems we’re trying to solve. I also think about the value equation: quality over cost. We have to do both. Cost can’t go up with quality going up; quality has to go up and costs have to go down.

We actually have to move the needle on value. As I’m looking at how we are going to implement technologies over the next five years, it’s got to be better quality and lower costs. We have to reduce the cost of healthcare. We can’t just keep adding. Point solutions all cost money. So if we keep buying multiple products, costs are going to go up.

Think about a platform-based approach from a technology standpoint where you can embed different models, outputs, predictions, generative text, and more into the current platform so they’re seamless in the workflow, but also don’t raise your costs.

One more piece is the environmental and sustainability piece. These models take a lot of compute to run, and it’s not an unlimited resource. We have to learn how to not use trillion parameter models like GPT for everything we’re doing. We have to figure out how to get to smaller models that are much more use case focused if we’re going to be sustainable. We are learning more and more that social determinants of health and the environment make up more [of outcomes] than anything in terms of genetics than we could ever possibly imagine. We can’t further make things unsustainable through this process. We have to keep that in mind as well.

Brian Miller, PhD, Executive Vice President and Chief Digital Officer of Intuitive: I would also like to build on Deepa’s remarks with regard to being really clear about the problem you’re trying to solve, and then how you’re measuring it, and whether or not you solve it. As a technologist, I can also geek out and develop a lot of technology and applications, and whatnot, and you can find yourself straying pretty far away from the problem you’re actually trying to solve.

Jared Antczak, Chief Digital Officer of Sanford Health (Sioux Falls, S.D.): There was a book I read a while back before generative AI / ChatGPT transformer model revolution called AI 2041: Ten Visions for Our Future by Kai-Fu Lee and Chen Qiufan. They made the declaration that in the year 2041 we’ll look back and healthcare will be the most transformed industry by AI. The timing might be a little bit sooner now, but I happen to agree with him. To Michael’s point, we’re on an unsustainable path from a cost standpoint within the industry, and AI really holds a lot of potential to help us bend the cost curve, and transform and disrupt the way we deliver care while also improving quality at a lower cost.

We’re just scratching the surface. We’re still at the beginnings of this whole generative AI revolution, and I think the next step is really looking for those opportunities to build more fit-for-purpose models that are suited for specific use cases and tapping into the incredible potential that these models have to transform healthcare.

Short of that, we can’t predict what’s going to happen in five years, and I won’t even try because I’ll get it wrong. ChatGPT wasn’t on anybody’s mind two years ago. We have to do our best to position our organizations with a culture of agility and nimbleness so we can actually respond to these merging technologies as they become available.