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
Insights on implementing an AI program at a health system from an early adopter. Vincent Liu, MD, MS, a senior research scientist and regional medical director of Augmented Clinical Intelligence at Kaiser Permanente, shares how AI is helping save physicians’ time and patients’ lives and how health systems can prioritize their AI initiatives. AMA Chief Experience Officer Todd Unger hosts.
Speaker
- Vincent Liu, MD, MS, senior research scientist and regional medical director, Augmented Clinical Intelligence, Kaiser Permanente
Transcript
Unger: Hello and welcome to the AMA Update video and podcast. It's been a lot of discussion this year about the potential benefits of AI in medicine. And today we're talking about how one health system put AI into practice and some of the early results that they're seeing.
I'm joined today by Dr. Vincent Liu, a senior research scientist and the regional medical director of Augmented Clinical Intelligence at Kaiser Permanente in Santa Clara, California. I'm Todd Unger, AMA's chief experience officer in Chicago. Dr. Liu, thanks so much for joining us today.
Dr Liu: Yeah, it's a pleasure to be with you, Todd. Thank you.
Unger: Now, before we get into the discussion, I'm very curious about your title. Of course, it says, not artificial intelligence but augmented clinical intelligence. And I'm going to bet that's intentional. Would you just comment on those choice of words?
Dr Liu: Yeah. Oh, absolutely. When I think of AI and when we think of it in our medical group, the Permanente Medical Group, we think of it as augmented rather than artificial intelligence. And that's because augmented intelligence places people, patients, communities and clinicians at its center rather than algorithms. So it's an important focus for us that, really, this is technology that augments the capability of our physicians rather than just focused on the development of algorithms.
Unger: And a great focus that is. Despite the fact that it feels like we're hearing about AI all the time right now, obviously, AI technology has been in place for some time. And Kaiser Permanente has been leveraging it for years.
One example of that is your Advanced Alert Monitor program. Let's start by telling the audience out there a little bit about the program and the AI technology that makes it possible.
Dr Liu: Certainly. Advanced Alert Monitor, or AAM, is designed to identify high-risk patients in the hospital who are at risk for adverse events like impending admission to the ICU or unexpected death. So it's been known for some time that these patients are at risk. And the question was, could we leverage AI algorithms and machine learning in order to really prevent and respond to these patients?
So over the past several years, we've used millions—hundreds of millions of data points from our hospitalized patients and granular EHR data, things like lab values, vital signs, other key clinical data to develop a machine learning algorithm that worked with good accuracy to predict patients at risk for deterioration in the next 12 hours, so early enough to actually intervene and hopefully prevent. While that was an exciting achievement, the development of an algorithm, again, akin to the augmented intelligence was not enough. It had to then be paired with a robust workflow.
So we worked with extraordinary clinicians and teams across our 21 hospitals here in Northern California to develop workflows that made sense in response to an AAM alert. We also worked with palliative care, social work, regional health education to understand how do you inform patients about algorithms which are looking at their data in real time and trying to help them. And also, making sure that patient's goals of care are aligned with the response because when someone's deteriorating, not every patient wants more invasive or aggressive maneuvers to be done. We worked with many technologists on this as well.
Our results were published a couple of years ago in the New England Journal of Medicine, which showed that the implementation of AAM within this entire program, working closely with clinicians, reduced mortality, reduced the rate of ICU transfers and is estimated to save as much as 500 lives per year across our hospitals. So that was a really exciting object lesson for us in how we could leverage AI and machine learning to improve the care of our patients. But again, making sure that it was closely tied with what made sense for our clinicians and patients.
Unger: What a great example. And your point that the technology, just one piece of the overall operation, workflow and the human element both form the physician and, of course, the patient objectives like you pointed out. That's just one way that you've taken advantage of AI. What are some of the other ways that this technology is starting to play a key role in your strategy?
Dr Liu: Yeah, there are different aspects. As you know, AI can become a specialist in many different domains. So one example would be using natural language processing to examine the contents of messages between patients and physicians. As we all know, physicians are increasingly being overwhelmed by the volume of these messages and finding it challenging to really understand how do we sort them, the ones that are most urgent, the ones which really don't need a physician's attention, and then probably the ones in between where it's not clear and it needs to initiate a more complex conversation.
So over the past several years, a team here at KP has been working on using natural language processing to actually examine these notes and begin to put them into buckets that have a very, again, strongly paired workflow. So whether it's workflow related to COVID vaccination or paxlovid or when are these facilities—what are the hours of these facilities or routine prescription refills, making sure that those get sorted appropriately, passed on to the right people and those concerns are addressed.
Of course, sometimes more emergent or concerning symptoms come through, and making sure that we can leverage technology to escalate those to appropriate parties to make sure that those are seen as timely as possible so they can be addressed. This kind of technology is in place today. It's analyzing and sorting on average a million messages per month across our system. So you just think about that type of scale, that's the opportunity we have with this technology is really to have tremendous speed and scale with deploying it for our clinicians.
I'll just cite another example, which people are really excited about, which is the use of computer vision for imaging. So what we've done, investigators here at KP have looked at thousands or tens of thousands of images like breast mammograms and found that computer vision algorithms can identify high risk features even within screening mammograms that were called normal by radiologists. And when we pair those with workflows, there's the potential to increase the identification of patients who may be at risk for breast cancer from 20% using traditional approaches to as much as 60% to 70% using, again, this computer vision augmentation.
And that unlocks a lot of opportunities, personalized screening recommendations for people at risk for breast cancer, targeted outreach for patients who are overdue for screening, or even the potential to rapidly look at images on the same day and avoid having patients come back for a second visit, escalate those images for immediate review and have the patient stay there to get their care in one visit. So there's opportunities that we're looking at and implementing. Again, the hard work is often that integration between our clinicians and workflows and this technology, making sure that they work closely together.
Unger: Those are three great examples, again, the one you just talked about on the diagnostic augmentation side. And then I love where you started out too because I see that coming through as a theme in a lot of the discussions that I'm having with AI experts, which is using the technology to reduce the burdens on physicians. And I think for anybody that's not familiar with what probably does feel like a million messages coming through from patients to doctors through the portal, significant level of burden.
And we know that things like that combined with other things that folks face are definite indicators of burnout. So it's good to see that technology get put to work that way.
I'm curious. I know it's kind of early in the process for initiatives like the ones that you're describing, but are you seeing results that you can quantify? Are you getting the kind of feedback that you want to hear both from your physicians and from patients?
Dr Liu: Yeah, I mean, I think there's overwhelming pride when we show that the implementation of AAM reduces death among our hospitalized patients, as many as 500 per year, and that our readmission risk score reduces high-risk readmissions by as much as 10%. There's a lot of pride in that across the institution that we're leveraging our patients technology, really to help them, and to help our clinicians identify the highest risk patients.
But it is a very complex process because, I'm going to be honest, we hear the good and the bad. I think our clinicians are extremely overburdened, as they are everywhere. And so I think of them as running at 100% full or sometimes even more.
So as we think about integrating these technologies, we need to think about a remove or replace. If we're going to add something, it has to either replace a task that they do and make it more efficient—that's a win for everyone—or it has to remove something else because I think there are some out there who say, let's put this alert and then this technology and these alerts. And they all seem like great things when they're considered independently.
But again, they begin to produce alert fatigue, distraction and other concerns. So it's something we are approaching very cautiously and making sure that our clinicians are bought in, and then ultimately testing it, right? Because if they stand the test of time, which is if they produce better patient outcomes, it really strengthens the level of support for these types of tools, even when, occasionally, some of them may produce some excess burden or work for our clinicians, but really to prevent adverse outcomes among our patients is the goal.
Unger: I really love and appreciate that remove or replace paradigm that you're talking about because so much of what we've seen in technology and medicine has been an add that has added to that burden.
Dr Liu: Right.
Stay up to date on AI
Follow the latest news on AI and its applications and effects for health care—delivered to your inbox.
Unger: And so paying attention to that and being able to demonstrate the kind of results that you're talking about to build momentum toward just an unburdened future has got to be pretty exciting for folks there. Now, physicians aren't the only ones that are thinking about the benefits of AI. They're also thinking about the risks, as you pointed out, and the need for new regulations and guidelines.
When you implemented these AI-powered programs, what kind of policies did you update or put in place to deal with that in going, and then as you explore things, like you said, along the way that pop up?
Dr Liu: Yeah, I think this is an area of growth right now for a lot of health systems. We have partnerships and collaborations and informal conversations with a lot of our health system partners. And this is where there's a lot of attention and resources going right now.
So I will say that I think even though AI is new and exciting, there's elements of it that are very much aligned with what's happened before, whether that's the use of technology or algorithms, right? We already use hundreds or thousands of algorithms in our practice today. And those may be as simple as if the patient has this—if the level is above this, do this, if not, do that, right?
And so that's something that can be written on a paper. Anybody can understand that. But that's still an algorithm and somebody had to approve it. It had to go through the stakeholders. It has to be maintained. And at some regular interval, it has to be reassessed whether that decision point makes sense.
So I think there are lots of forums in a health system in place today, whether those are stakeholder groups, domain experts, technology governance and oversight groups that maintain the process of oversight to make sure that all stakeholders are engaged and the decision to move forward and endorse something is there. And so we utilize a similar framework to that.
I think where there's a lot of new growth that needs to happen is we need to be training up our physicians and our clinicians to understand, what is AI? What is data science? How do I enter a room or an industry conversation and understand whether or not the technology they're speaking about is really—is there something there that we care about or is it kind of a flash in the pan?
What is reinforcement learning? What are large language models? How do I need to think about bias and fairness evaluations?
So, there's a lot of training that needs to happen, specialization and development of workforce that kind of serves that translator function. They both understand the clinical deployment of these tools. But they also understand something about that technology.
And I think that's something we, as an entire field, need to be invested in. Not only that governance and oversight process within an individual health system or a practice, but then building up a workforce that's capable of making those informed decisions and then ultimately guiding the deployment of this technology into practice.
Unger: So I'm curious what's next on the agenda in terms of new capabilities or new tools that you're hoping to implement over the next couple of years.
Dr Liu: Yeah, I mean, there's so much exciting stuff on the frontier. I can speak to what I think is almost mature or really entering into workflow. Computer vision, whether that's radiology, dermatology, pathology, EKGs or EEGs, we've seen just a large expansion of these technologies and then their integration into products. And so I think that's a very tangible next step. And a lot of those conversations are happening in partnership with our industry and other medical technology companies.
I think there's been huge excitement about large language models. Is there a way that we can leverage these very, very complex and large statistical models to improve communication with our patients? Can they auto draft our notes or our secure messages? I think there's potential there. But we have to use it cautiously.
I think what is really exciting is that they can also help with information retrieval. A burden clinicians face is, well, this patient is a complex patient. They may have been seen at different places. How do I synthesize everything that's happened over the past six months when they've been in and out of the hospital and been to surgery and actually really make a decision today based on the best information that I have?
And so I think there's a tremendous opportunity for these large language models to improve our capability to glean that information from the chart. And so we're pretty excited about that. But again, something we approach with caution because we still very much feel that the human needs to be in the loop. It needs a human handler who is able to shut it down if it's not working the way that we want it to be working.
Risk prediction models continue to be built and implemented. I think trying to make the process of finding the needle in the haystack better, right? And so the better the technologies get, the more granular information, I think the more capabilities we'll have there.
And then I think into the future, it's treatment recommender systems and precision medicine with omics. And there's going to be robotics and augmented reality. So the technology is advancing at an incredible pace. We're going to see a plethora of products. It's a matter of, again, solving real actual problems for our patients and clinicians and doing it in a way that's safe and sustainable.
Unger: Now that is a pretty exciting future. And of course, the core of that is innovation, something that you are working to do and encourage on the outside of your health system too. You've got a new grant program for AI and machine learning in health care. Tell us a little bit more about how you're trying to drive innovation inside and outside of Permanente.
Dr Liu: Yeah. So this is called the AIM-HI Program, which is Augmented Intelligence in Medicine and Healthcare Initiative. It was funded through a generous grant from the Gordon and Betty Moore Foundation. And so they looked at the work that we were doing in integrating and proving the effectiveness of AI and machine learning. And what they wanted us to do was to identify three to five health systems who submitted proposals to fund them to actually do rigorous tests of the impact of AI on their patient outcomes.
So what we have is a plethora of papers and technologies that say, we perform beautifully well, and we do this and that. What we don't have is actual, real-world evidence that those claims can actually be justified based on this kind of rigorous study design. So this is really where this grant mechanism comes in. We are offering as much as $750,000 for each of three to five health systems.
We've received over 120 applications thus far and are going through the process of identifying, I think, the most promising ones. And what we want this work to do is develop best practices, shared understanding for scalability, and hopefully, prove that AI, in the right context with the right level of integration, really has an important role to play in making our health care better.
Unger: Well, last question for our conversation. Obviously, if you're paying attention, there's a lot of scrutiny in Washington about the use of AI in health care, just big meeting last week with, of course, President Biden on setting out guidelines in AI. What's one thing that you want policy-makers to keep in mind or get done in this space?
Dr Liu: Yeah, I mean, regulations are incredibly important. It's what's going to make sure that we are putting these appropriate guardrails in place. And I think getting some voluntary endorsement from these companies to examine their practices and add transparency to the process is going to be essential.
I do think there is a stage at which regulations can stifle some of the innovation. For example, the kind that we've seen in our own health system. Our mission is to use our patients' data most effectively to improve their outcomes and to improve our clinicians manner of care. It's been less about commercializing those products, right? So it's developed on the data in which it's applied. And it's really tied to health system workflows.
And I think in those cases when it's done within a specific medical practice or within a specific health system, there is a role for providing a safe harbor in some ways from the same regulations so that we can innovate, so that we can use our best data to improve our patients' care rather than kind of saying, well, to deploy a technology like this, we'd have to go through a bunch of fairly challenging regulatory steps. And so our patients wouldn't actually benefit from that. And so I think that is a fine line to walk, but I think would be an important one because it maintains the spirit of innovation in the application of data right to the patients in whom it can benefit.
And the other concern that I have is that if the regulatory mandates become extremely strong, health systems are not prepared or practices are not prepared to do that. We're not companies that typically have FDA or other types of regulatory arms. It will produce a pretty large burden. And I think the byproduct of that will actually strengthen the control of technology companies over patients' data and the products.
Essentially, if we're unable to keep up with the regulatory overhead, it will mean that companies will purchase all that data, make the products, and sell all of it back. And I think that reduces the autonomy of health systems and practices, really, again, to use their own patients' data to produce better outcomes in patient care, and ultimately, in clinician practice.
Unger: Dr. Liu, it has been such a pleasure to have you today and hear your perspective and the progress that is being made already. Hearing these kind of stories and where you've come in the past few years, it really is an antidote to a lot of hysteria that we see out there in regard to AI. I can't wait to hear back from you as you continue your progress and you can update us on the results.
That's it for today's AMA Update. And we'll be back soon with another segment. In the meantime, you can 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.
