Reducing Hospital Readmissions: Final Presentation

Matthew Trowbridge, MD, MPH [00:00:00] So what's AI in healthcare? What should we do next? The whole world's trying to figure that out. I'm excited to see how a team breaks that down, comes up with opportunities, and then builds some prototypes that people can actually get their hands on as quickly as possible.

Conner Brew [00:00:15] Innovation gets stopped in the ideation phase, because we come up with all these ideas and then we want to figure out how do we know if these are real, what's good, what's not good. And really the way that you get past that is by just making something and allowing people to react to it. So that's what we did in the "Attack the Concept," where we put those mockups in front of real experts, real people, who could provide really excellent feedback on the direction of the idea.

Sydnor Gammon [00:00:38] We had the benefit of having an actual emergency medicine physician on the "Attack the Concept" panel working session with industry experts from a technical lens, from a data science lens, from a care provision clinical lens, to think about: how do we make this better? How do we think about how we can actually deliver it and bring it into the forefront?

Conner Brew [00:01:07] This is the final presentation of an expedited project where we've really challenged ourselves to reimagine what the patient experience could look like both within the stay of a hospital and beyond.

Kristen Duke [00:01:18] What we found interesting when we did this exercise is that at the end, the discharge experience was really disjointed. We asked people if they looked for more information after their visit, and 92% of the people who took the survey said that they did look for additional information, and about 70% of them were just turning to Google. So they're just using Dr. Google instead of actual doctor-backed references.

Kristen Duke [00:01:48] So what this showed us is that we really need to focus on what happened after they were discharged and integrate these ideas into that experience. Out of all of that research, we really narrowed in on two promising use cases: 1) In-Room Transcripts and AI Note-Taking. And that is recording the doctor or maybe even nurses and providing a transcript for the patient to be able to reference after the fact. And the second one is a Conversational AI Assistant. So giving the patient the ability to reference doctor-backed information about their condition — or related to their condition — so we can hopefully mitigate some of the readmission process that happens today.

Ryan Davis [00:02:40] And so the tool we've come up with here is the AI Healthcare Assistant. This is a tool that faces both patients and physicians. When we think about the physician experience, some of the key features we want to highlight are the ability to record. Because we imagine that someone who is in the middle of a lot of pain, we don't want to have that burden to be on the patient. As well as a summary of the discussion that's had, as well as any action items that need to be followed up on. And of course, for the physician and for their team, the ability edit all the information that comes through it to make sure that it's accurate. For patients, this is going to be their place where they go to basically get all their information about the discharge. For the features there, we've got summaries, any follow-ups and action items, and then also the key feature, which is the Chat.

Matthew Trowbridge, MD, MPH [00:03:33] We get what we call zebras, it's usually like someone coming in with a rare variant of something. In that context, you have to make a decision about managing an acute complaint and trying your best to understand the context of what else is happening in this patient's life. If you were a resident physician in the same ER, it would make me feel comfortable as an attending to know the resident was getting a lot more handholding. Then the nurse comes in, spends a long time. That actually might be the moment where you push Recording. I want it documented correctly, coded correctly. And then the patient, though, probably now wants an opportunity to have a two-hour conversation. But Dr. Edesina has to move on.

Reza Mousavi [00:04:20] I think the most unique aspect of this application is the unique data that you are capturing, which is going to connect the patient's conditions to his or her concerns after the treatment. So that's the unique value I see.

Margo Bulka [00:04:35] This dialogue has really revealed that the more discreet, the more narrow that we focus our energy on, the more value it can provide. And so I think as we all are so ambitious to apply AI in varied ways, just keeping our attention on the very, very narrow problems as the places that we'll find the most impact.

Sydnor Gammon [00:04:58] The idea that it actually is more behind the scenes than this early concept is really exciting. The fact that there could be a capture of information that doesn't require as much physician focused in a screen, but rather it's riding alongside of a care team as they're thinking about providing care. And it's provided in a form factor that's most organic to the patient. Maybe it's a text message or an email to their inbox, or maybe they love a mobile app and that's what they want to see. But I'm excited about the bones of it because I think where we could take it — in a way that feels as organic and seamless in care delivery as possible — is super exciting.

Matthew Trowbridge, MD, MPH [00:05:36] When somebody's trying to decide, does it have to be through the emergency department? But what if we had tools that might make that decision easier and have more confidence of when you need to go, but also when you don't need to go. I think that could be really powerful.

Reza Mousavi [00:05:50] Given the amount of time that the team just had, I was expecting something more generic, very high-level, not as detailed as we saw today. So that was pretty impressive.

Conner Brew [00:06:02] We received a lot of great feedback from the experts about the overall design of the experience, both from a specific usability perspective, as well as the way that the doctor-side of the experience and the patient-side of the experience interacted with one another. And so what we would really like to do is take a researcher and a team of designers and to continue conducting rapid iterations to improve that user experience. In parallel with that, we would set up a cross-functional team consisting of software engineers, AI engineers, and data scientists who could begin laying the groundwork for that agentic AI system.

Matthew Trowbridge, MD, MPH [00:06:33] It's an exciting time. It's daunting, but as you saw and I saw today, it's the process and the whole stack of the team that can not just be there for the actual end coding, but they have world-class talent helping you think at the beginning, and the strategy, and the design, and the user testing. We could have much better utilization of our existing resources, which includes doctor's attention, time, nursing, even just the physical space of emergency departments. Everything could run a lot better. And I really do think if we stay focused on: what is it we want out of the human experience of healthcare that we can't have today, and do these new tools offer something maybe new and more powerful? That's very exciting.