AMA Update covers a range of health care topics affecting the lives of physicians, residents, medical students and patients. From private practice and health system leaders to scientists and public health officials, hear from the experts in medicine on COVID-19, medical education, advocacy issues, burnout, vaccines and more.
Featured topic and speakers
How is AI used in health care data? What does AI help with in health care? How is AI used in emergency rooms? What are the benefits of AI in health care?
How is AI used in medical documentation? Learn about using AI in electronic health records to improve patient care and the future of AI in health care with guest Dana Sax, MD, research scientist at Kaiser Permanente Division of Research and emergency physician at The Permanente Medical Group. AMA Chief Experience Officer Todd Unger hosts.
- Watch AMA Update for health care news for physicians, residents and medical students.
- Not a physician? Join the Patients Action Network.
- Learn more about the AMA Health System Program.
- The AMA is your powerful ally, focused on addressing the issues important to you, so you can focus on what matters most—patients. We will meet this challenge together. Join us.
- Learn more about our AMA advocacy priorities, including:
- Reforming Medicare payment
- Fighting scope creep
- Fixing prior authorization
- Reducing physician burnout
- Making technology work for physicians
Speaker
- Dana Sax, MD, research scientist, Kaiser Permanente Division of Research; emergency physician, The Permanente Medical Group
Transcript
Dr. Sax: We sought to develop a novel risk tool to risk stratify these patients. The tool works in real time at the point of care, and it uses about 60 different variables from our electronic health record. And it uses machine learning model to predict risk of an adverse event.
Unger: Hello and welcome to the AMA Update video and podcast. Today, we're talking about how one health system is using AI in the emergency room for patients with heart failure. Our guest today is Dr. Dana Sax, research scientist at the Kaiser Permanente Division of Research and an emergency physician with the Permanente Medical Group in Pleasanton, California. I'm Todd Unger, AMA's chief experience officer in Chicago. Dr. Sax, welcome.
Dr. Sax: Hi. Thank you so much for having me. Happy to be here.
Unger: Well, it's good to talk to you. And I have to say, your organization is using AI in so many ways right now. So I'm eager to hear more background on why you created this tool for the emergency department and exactly how it works.
Dr. Sax: Sure. So I have been working in the emergency department here, actually, in Oakland and Richmond, California about 10 years and have been studying the care of heart failure patients. And something that we have learned is that these are very complicated patients, often with multiple medical problems on multiple medications.
And they are often very difficult to risk stratify when we take care of them. In the emergency department. So many of the patients are admitted to the hospital, and we found that some high-risk patients were going home, and on the other hand, some low-risk patients were being admitted to the hospital and then quickly discharged.
So we sought to develop a novel risk tool to risk stratify these patients. The tool works in real time at the point of care, and it uses about 60 different variables from our electronic health record, and it uses machine learning model to predict risk of an adverse event. And the scores are presented to the treating provider at the point of care, streamlined into the workflow of the emergency provider.
Unger: So the inputs into this, you mentioned they come from EHR. They're 60 different inputs. Is this kind of historical data or is it stuff that you're collecting on the spot when the patient presents?
Dr. Sax: It's both of those. So it's historical data, like their problem list, their comorbidities, their medications that they're taking, their most recent ejection fraction, their recent emergency department hospital and intensive care unit utilization, and then also their acute physiology—so their vital signs when they come to the emergency department, the lab values, the renal function, the troponin, the BNP, the weight today and compared to their baseline weight, their blood pressure compared to their baseline blood pressure. And it pulls in all of these and then calculates a risk class in real time.
Unger: Now you've been working on developing and testing this AI model for a couple of years now, and I'm very eager to hear about its impact on patient outcomes so far.
Dr. Sax: So like you mentioned, we spent a couple of years deriving and validating the risk model. And then we built the model into our electronic health record to make sure that it was calculating correctly for the right patients at the right time, presenting to the provider at the right time. We did a lot of usability testing with the providers to find out what they wanted to see and where. And then we did a pilot study in two of our emergency departments for a year to see the impact of the tool and to make sure that the recommendations were safe.
And so that pilot study, we found that among patients that the risk tool identified as low risk and the patients were discharged home, there were no 30-day adverse events, suggesting that the risk estimates and the guidance to consider discharge was a safe recommendation.
We also found the tool also gives some guidance on medical management for these patients. And we found that the proportion of patients who are on guideline-directed medical therapy, the optimal therapies increased as a result of use of the tool.
We've since, in January this year, we deployed the tool across all 21 medical centers in our health system. And we are currently studying over—it'll be a year trial—the impact more broadly across the region.
Unger: Well, really interesting and encouraging results on the patient side. But it also strikes me this is a kind of clear augmented intelligence tool for physicians that's going to help them to talk a little bit more about how it helps physicians.
Dr. Sax: Sure. So it was really important, as we were designing this, that we wanted to help patients and families and also help our providers to try to help with decision-making in terms of admission or discharge or follow-up, but also to help pull in a lot of clinical data from the electronic health record in one place. So the risk score sits within a operational report in our electronic health record.
And that report also collates lots of useful clinical data for the provider at the point of care, to help streamline their work, to not have to look in multiple places. So it includes all of their recent echos, EKGs, stress test, event monitor, data, medications, blood pressures, vital signs, all of this information in one place to help make this process more streamlined and efficient for the provider.
Unger: So I can see where just the aggregation of all of that data and all of those tests in one place to make it very convenient. And then there's just the broad number of factors that you're kind of bringing together for presentation. What just kind of typically does that allow a physician to see that they might not see without the use of AI?
Dr. Sax: We think that it can help and give providers one additional tool at the point of care, especially for these very complicated patients. There's a wide range of data that the AI tool pulls in. And it would often be very difficult in a very busy emergency department to look at all of that information in an efficient manner. So we think that by pulling in things like medication change recently or a weight change, increase in weight, or a slight change in their renal function, having all of that available and pulled into the risk model can help make better decisions and support the providers.
Unger: Well, great. Important, too, because it really reflects how physicians are feeling about AI right now. In a new AMA survey about this topic, a top priority for physicians was having it well-integrated into their workflows. And what you just said about how that brings it all together and makes it easy—just a perfect example about this. Talk to us a little bit more about how you integrated this AI tool into your workflow in the emergency department, where, as you said, things are moving fast.
Dr. Sax: Yeah, so we did a lot of work before we deployed the tool surveying providers, emergency providers, interviewing providers, watching providers interact with the electronic health record, asking them specifically what information would you want to see, when, where do you want it to appear? What wording, what text do you want? What colors do you want? At what point in your workflow should the risk score be available?
And we integrated all that information to develop the tool. And the tool appears as an alert that's noninterruptive. Many providers don't want their workflows interrupted, so it's a noninterruptive message that a risk score is available. And you can click on that to access all of the information as part of your standard workup of these patients.
Unger: What's neat is it really brings the premise of AI, and also, the promise of EHRs to life and really making them work together. I have to imagine you've got a roadmap about what you want to see improve about your current approach. What do you see coming up in the next couple of years?
Dr. Sax: Yeah, so we are conducting this regional trial across 21 hospitals currently, and I'm sure that we're going to learn a lot about how we can further improve on this risk tool and decision support. We also would like to potentially spread this to other care settings, so the inpatient setting potentially or arranging transitions of care for patients going back home. So I think that the lessons that we'll learn from this trial will help inform how to further spread this and then iteratively improve it for the next round.
Unger: Well, speaking of spreading, just one final question—in addition to the use case that we have focused on in this conversation, what are the other ways that you view machine learning could really help in the emergency department?
Dr. Sax: Yeah, so I think we hope to spread this heart failure tool, which is kind of our first use case, to other settings like I was mentioning. But I think there's additional use cases for risk stratification, for identifying optimal next therapies and treatment venues. And so I think it can be one additional tool that providers can use at the point of care to help streamline care and to improve the safety and quality of care.
Unger: Well, Dr. Sax, you'll have to come back and tell us more about your progress sometime soon. Thanks so much for joining us, and we hope to talk with you again.
Dr. Sax: Thank you so much for having me.
Unger: One of the AMA's top priorities is all about what we just talked about today, and that is making technology work for physicians. To support that work, we encourage you to become an AMA member at ama-assn.org/joinnow. That wraps up today's episode. We'll be back soon with another AMA Update. Be sure to subscribe for new episodes and find all our videos and podcasts at ama-assn.org/podcasts. Thanks for joining us today. Please take care.
Disclaimer: The viewpoints expressed in this video are those of the participants and/or do not necessarily reflect the views and policies of the AMA.
Subscribe to AMA Update
Get videos with expert opinions from the AMA on the most important health care topics affecting physicians, residents, medical students and patients—delivered to your inbox.
