Tampa General Hospital’s Thoughtful Approach to Pursuing AI Advances

At the 1,350-bed, four-facility Tampa General Hospital organization, Dennis Dansby, PharmD, vice president of healthcare informatics, has been helping to lead his colleagues forward in pursing the thoughtful adoption of AI across a variety of operational areas. One key area of advance has been around clinician workflow. He spoke recently with Healthcare Innovation Editor-in-Chief Mark Hagland regarding how and in which ways he and his colleagues are making progress in the broad AI arena. Below are excerpts from their recent interview.

How are you using AI?

ONC [the office of the Assistant Secretary for Technology Policy (ATSP)/Office of the National Coordinator for Health IT] put out the regs about a year-and-a-half ago; and we stood up a governance process around AI. We have a monthly governance group. We created a library of tools with AI in them. As you know, they’ve been using AI for a while in diagnostic imaging; they’ve been using it to improve efficiencies inside the radiologists’ workflow. We categorized everything where we think we’re using AI, and anything net-new, we run through that committee—medical staff, corporate compliance, risk, to understand what we’re doing, and to poke and ask questions. And there are subcommittees that will do the work, including between cybersecurity, business intelligence, and our operational arm, and they bring it for a decision for a vote. That’s that framework.

So we have an innovation arm, a governance arm, and an application arm. And we have a separate group of people who do governance for the hospital. We run new tools through the governance arm, strategy, and then application arm.

What does utilization of this set of tools look like in practice?

It’s worth calling out the distinction between platform and point solution. Our EHR [electronic health record] vendor, is Epic. And, per that, Epic is up to about 200 different AI tools, and they’re adding more on every day. We are also a client of Palantir, which provides another tool where we have more of our data in a framework, to use that to create AI. In that regard, we’re more of a platform-play type of organization. We don’t get into a lot of development, and when we do, we work with some of the larger vendors.

What advances have clinicians made in their workflow using AI-based tools?

The easy one to mention is DAX, which is of course ambient listening. There’s another discussion to be had under AI, and the question there is what are you replacing—FTEs versus specific tasks within a role. DAX is a great example of a task within a role. It’s being used to streamline workflow.

A second example is around nursing workflow. We’ve said that DAX is interesting in nursing as well. Can we take work off the nurses using conversational AI? So we’ve been participating in a trial with Microsoft. They’re going to push that out. Workflow documentation—flowsheet row documentation. It’s conversational AI—it’s generative.

How much time nurses are being saved?

ROI for AI is a great topic. I just spoke to Microsoft yesterday about this topic. You’re in the world of minutes, not hours. They’re in a trial phase—in dev trial phase. The indirect answer is, in some organizations, you’re probably going to save an hour per shift. In areas where you don’t get that adoption, it looks like the tool’s no performing well.

Possibly up to an hour per shift per nurse?

It’s a bit early to talk about hours, but certainly numerous minutes per shift per nurse per day—that is already being validated. And another use case is the work we’ve done with Merative. That is again, this idea of laying an AI tool over a large database. Think about all of the drug information in a tool like Micromedex: being able to have a conversational agent kick back a response, and say, here’s everything we’ve found from these sources. It’s a huge time savings for both nursing and pharmacy. And maybe you’re asking a question and looking through one optic; oftentimes, AI will actually push back on you and open up different ideas.

Can you speak to the clinical decision support technology from Micromedex?

It’s about accuracy and speed to answer. Consider the old-school way that this was done: I would go to a drug monograph and would look for a pediatric dose for, for example, Diflucan. And old school, I would go to the monograph, and click through six or seven screens; now with an interactive model, it’s taking me directly to a screen. It’s more interactive, and is serving the information up to the clinician, versus the clinician having to search for it. And we do a lot of compatibility checking. So I want to know something basic such as, is IV Dilantin compatible with D5 or normal, saline? This is something you’d look up in a book. But this tool gives you the information more quickly. Eventually, over time, the information will go quickly to the clinician. And you need integration between Micromedex and your EHR, in our case, Epic.

What kinds of things will clinicians be able to do in the next few years?

One area we’re really interested in is the potential of computer vision. We use a company called Appela to improve our OR workflow. It takes images of the OR and allows you to improve throughput through your OR. Someday soon, you could improve surgical technique using computer vision. The functionality is there now to use computer vision around imaging heart rate and blood pressure. It’s not FDA-approved yet, but functionality will change clinician workflows. Do I need a UC coming in to take your blood pressure?

In two to three years, it will be totally different, and the tasks within the role—you’ll see AI replace the mundane, repetitive tasks one by one. Documentation will be lifted off the clinicians. Prior authorization is a good one: the prior authorization for medical procedures or prescriptions. Most organizations have employees chasing down that information. But future-state, some type of agentic AI, some sort of agent, will run through that. And if it’s specialty pharmacy, did they have this viral load? A specific detail from the EHR. Agents will replace some repetitive tasks. It will be a more efficient workflow for clinicians. Computer vision will be used in multiple ways. And if you have computer vision—right now, everything relies on a manual type of workflow. But computer vision is looking down to make sure we have great code or trauma technique, make sure we got the code form documented correctly, document who’s in the room, cue up the orders for the provider to move things along. A lot of interesting ideas to help streamline workflows.

How should senior health IT leaders be thinking about all of this?

There are two schools of thought in that regard. One group wants to know what they can adopt now in an organization. Another group wants to wait a while and let some of the technology develop. We’re focused on the efficiencies that we’ll gain, and those potential efficiencies are causing us to move forward more quickly. We can reduce physician, nurse, and pharmacist burnout. And it will be a better patient experience. So we’ve been aggressively pursuing AI. And we’re not just grabbing everything. I think the ROI discussion is still relatively interesting. Per DAX, there’s no hard-dollar proof of its efficacy, but looking at pajama time reduction and provider satisfaction, the tools are there.

And how should c-suite leaders be thinking about this?

I’ve mentioned point solutions and platform; we’re more of a platform shop. The applications are layering AI into their tools. And if you have applications within your matrix, you should be talking to your vendors to ask them what they’re doing to make their tools more efficient; that’s another IT-focused c-suite discussion