On this episode, Tom talks with Dr. Christopher Longhurst, Chief Medical Officer and Chief Digital Officer at UC San Diego Health. Together they discuss his organization’s use of ChatGPT, pairing physicians with data scientists to collaborate on improving patient care using data, and advancing AI to alleviate challenges facing the healthcare workforce.
On this episode, Tom talks with Dr. Christopher Longhurst, Chief Medical Officer and Chief Digital Officer at UC San Diego Health. Together they discuss his organization’s use of ChatGPT, pairing physicians with data scientists to collaborate on improving patient care using data, and advancing AI to alleviate challenges facing the healthcare workforce.
00:00
Welcome to Digital squared, a podcast that explores the implications of living in an increasingly digital world. We're on a mission to inspire our listeners to use technology and data for good. Your host Tom Andriola is the Vice Chancellor for Information Technology and data and Chief Digital Officer at the University of California at Irvine. Join us as Tom and fellow leaders discussed the technological, cultural and societal trends that are shaping our world.
00:29
On this episode, I talked with Dr. Christopher Longhurst chief medical and Digital Officer at UC San Diego health. In this capacity his unique role, combined with his varied experiences across clinical care, technology and data allow for innovation to thrive under his leadership, innovation that is a necessity in today's evolving healthcare sector. Together, we discuss his organization's use of chat GPT, pairing physicians with data scientists to collaborate on improving patient care, and the future of advancing AI to alleviate undue burdens for the healthcare workforce. Dr. Longhurst, welcome to our podcast.
01:07
Thank you, Tom. It's pleasure to be here.
01:09
I want to start our discussion today around your current roles at UC San Diego health. It's kind of a unique combination of roles and how you put that together how you thought about it, and how you bridge the gap between two what's some may say my very different worlds,
01:25
it's been quite a journey. And Tom, you and I have been working together since the beginning of this journey at the University of California. I initially left Stanford to join UC San Diego and a capacity of CIO. And then also something I was really excited about leaving our IT organization into the future. I was building on a lot of great work that had been done prior to my arrival and wanted to bring an academic bent as well as real push towards innovation. I ended up serving a CIO for a little over six years. But along the way, I assumed an additional role as the Associate Chief Medical Officer for quality and safety. And in this capacity, I was leaving our chief quality and patient safety officer and quality department around our quality improvement journey. And it turned out those two roles were really synergistic and a lot of fun. And we found ways to lift both departments through the integration. Around two and a half years ago, I stepped into a new role as chief medical officer and obviously could not continue playing the role of CIO in that CMO role. And so we've hired a CIO, but it remains part of my portfolio. And so I also have the title of Chief Digital Officer. And as you suggested, people asked, you know, how do you do both and actually say that neither one would be fun without the other. So in my role as chief medical officer, I have incredible visibility into all sorts of challenges and opportunities across the health system quality improvement, Systems Improvement, process improvement, etc. And in my role as Chief Digital Officer, I have insight into solutions that can be brought to bear for these types of problems. And in fact, if we look over the last 50 years, and healthcare chief medical officers often dealt with the same problems over and over. And the solutions are often temporary education and things that tend to have cyclic kind of impact. In the Chief Digital Officer world, we have new tools that can bring new solutions and often more sustainable hardwired solutions. And so, to me, that's really the fun part of having these two disparate hats, is figuring out how we can solve long standing problems in new and different ways.
03:29
Yeah, I love Chris, the way that you put that together, right, someone once shared with me that opportunity exists at the intersection of two spaces or domains. And that's kind of what your role represents. When I think about what you're doing. I have one follow up around this, which is, I'd love to hear your thoughts or your experience in terms of the platform that having both of these pillars create for you to drive change and transformation in an industry that is very much challenged with driving transformation.
03:58
In my role as CIO, I certainly had a seat at the table with the executive team. But oftentimes CIOs are looked at as the technology leaders rather than process improvement engineers and solution designers. While I certainly had a lot of support from my CEO and colleagues, I found that stepping into the Chief Medical Officer role gave me a different vantage point, and a different level of authority and ability to influence decisions among our physician leaders, and health system leaders. And so I've really valued that position in terms of being able to bring innovation that would have been difficult to lead from behind.
04:36
So UC San Diego health is one of the early adopters of integrating GPT into doctor patient messaging. We're very early in the game here. Tell us what you've learned so far,
04:47
as some have suggested, we're just a few steps into a long journey. And I think that's absolutely true. GPT was just released last November, so a little bit less than a year ago, and certainly after release I participated in a research study where we compared chat GPT based responses to patient messages that had been posted to a public forum and physician responses. And the headlines on this study that was published in JAMA medicine suggested that the Chatbot was potentially higher quality and more empathetic. And I can tell you, as one of the reviewers of all 200 of these questions, I took some issues with those headlines. There were certainly cases where the Chatbot responses were high quality and they were empathetic, and the physician responses were short or abbreviated. But I know that a physician who knew that they were being measured against the chat bot, given enough time, could also draft an empathetic or long response. I think that the real lesson learned from this paper was about time, our doctors are short on time, they don't have the time to spend writing these long responses, and the Chatbot can draft them very quickly. And so when I was speaking with some leaders at our electronic health record vendor partner, back in January and February of 2023, I shared with them the preliminary and not yet published results, and asked them if they were working on anything related to GPT. And the answer was, yes, we are we're integrating it on the back end. And this looks like a really great use case. And so we've partnered together to implement this as a pilot project. And we're learning all sorts of things. We started with GPT, three and a half. And we piloted that with about a dozen physicians for a couple of months, before we swapped the back end for GPT. Four, and we did see a significant improvement between three and a half and four, we found that our primary care physician draft messages are probably higher quality in general than our sub specialists, which makes a little bit of sense. But we've also learned that prompt engineering is really important, because if you prompt GPT, that you're an academic year, oh, gynecologist, it'll give a different responses than if it's functioning as a primary care physician. And so while the primary care use case is probably still the best use case, we're learning how to make this better. We've implemented this in such a way that there's only two buttons for physician in the pilot program, start with draft message or start with blank message those no opportunity to just send now. And that's really important, because we want to keep a human in the loop. And we want to make sure there's eyes on these messages before they're going back to patients. In fact, our ethics input on this project was to make sure that not only are we sending this to patients with a physician review or clinician review first, but we actually are publicly transparent with our patients with a message that's appended to the end of every patient message that or response that says, This message was automatically generated and edited by the person who's sending the message. And so that provides really maximum transparency.
07:48
I think I had many questions from patients. You know, when they seen that little disclaimer at the bottom,
07:53
you know, we've only had positive feedback from patients. And in fact, this was featured in the front page of Wall Street Journal with one of our ambulatory physician leaders, Dr. Millan, who's been leading some of this work operationally, and Dr. Millan received a message the following day from one of her patients that said, I saw that work in Wall Street Journal, really excited for you, because I know these patient messages can take a lot of time to respond, and we appreciate your response. So I'm sending you a bot generated patient message, so you can send me a bot generated reply. It was a funny joke, but that's the general sense that we're getting from many of our patients, which is, they sometimes feel badly about messaging their clinician, because they know their physicians are busy. They know that their doctors have to take time out between patients or at the end of the day to respond to these messages. And so generally, the patient response has been very positive.
08:45
So I want to build on this. It's been amazing to watch you in San Diego be an early adopter here. I have my own experiences from the past generation where I was inside the internet, tornado at the time, and was one of those whiz kids running around telling executives, you need to pay attention to this thing called the internet and hearing responses like why would I ever want my product catalog electronically? Our customers are always going to walk into our stores. And after Amazon took out the book industry, people telling me sure you can make this work for books, but you can tell me someone's gonna buy a pair of shoes online, sports jacket online, a car online, yet here we are. How idiotic? Did those responses sound now? a generation of business later. So when I hear executives today say look, we're three steps into a 10k race. I understand what that 10k race looks like haven't been able to have run the last one. Where do you think we are in this and what do you expect the road to look like going forward?
09:46
Without a doubt we're very early in this journey and the large language models are a transformative step forward in artificial general intelligence. Dr. Peter Lee has described this as an unexpected advance apps that he thought would take 40 or 50 years. And I think that's a really accurate description. Many of us envisioned this as a long term future from Isaac Asimov and science fiction writers, to AI researchers themselves. But the scale of these large language models moving from 70,000 to 7 million to 70 million nodes really is beginning to approximate, at least, if not real human intelligence, the impression of providing that level of response and text based responses. So there's no doubt that while we're early, this is going to fundamentally change medicine. I felt a lot about this metaphor, and don't put it out there without recognizing it could be perceived as hyperbole. But I personally think that this is going to be the biggest impact on healthcare since penicillin. Think about all the problems that we experienced from a healthcare standpoint, probably 30 to 40% of cost in healthcare related to administrative overhead. Every hospital and health system has dozens, if not hundreds of coders who review text data and code it for billing and compliance and regulatory and epidemiologic purposes. Could AI help there without a doubt? Think about all of the time that our physicians spend responding to insurance denials, and prior authorizations could AI help there without a doubt. Think about the scribes that we've been hiring to help enter data into the electronic health records. Could AI based scribes help without a doubt. So there's all sorts of problems that we can start to tackle in the near term. What I'm really excited about, though time is when we think about the unfulfilled promise of electronic health records. When Larry weed first described the soap note in the early 1970s. His passion was about computerizing the medical records so that we could deliver more highly reliable care with better differential diagnoses, and more consistent high quality care across all physicians. We really haven't yet delivered that if you think about some of the alerts and problems that we've had an electronic health record in the last decade, we're good at preventing overdoses of medications, because that is discrete data in the record that we can trigger alerts, we're starting to figure out how to identify sepsis early from discrete data. But we looked at our electronic health record data recently, Tom, and it turns out 99.6% of our electronic record, excluding all the billing data, and audit data, 99.6% of the data in the clinical record is text. So we've actually been trying to trigger these alerts and decision support using only point 4% of the clinical record. And suddenly, we now have the opportunity to take the other 99.6. So I am really bullish that over the next five or 10 years, we're going to see primary care physicians operating at a higher level with less variability, we're going to see mid level providers who providing frontline care across the country, nurse practitioners, physician assistants, nurse midwives, CRNAs, and others, supported by AI co pilots to provide the best possible care and that care will be as good as the best primary care physicians today. I think these large language models are going to have a huge impact in imaging based specialties as well. We're going to transform pathology, radiology, ophthalmology, and other specialties that rely heavily on images. So the transformation that's going to come about in the next two, 5, 10 and 20 years in healthcare, as a result of this specific technology is really hard to predict because it's so transformative.
13:41
I have to ask this question based on what you just said, How are you thinking about the challenges around evolving the workforce to really be able to take advantage of this? We're at a point now where the technology evolution is much faster than the human evolution to absorb and utilize the tools that we're developing? How are you thinking about that for your organization?
14:01
Tom, I'm really glad you asked. In fact, our Vice Dean of Medical Education, Dr. Michelle Daniel has been thinking a lot about this as well. She and I are involved with a national project to relate it to redefining the curriculum for undergraduate medical education for medical students, for nursing students and others, as well as for residency training, we probably can further emphasize things like not just quality improvement, but cybersecurity and data science all along the training pathway. And in fact, we've got a whole workforce of 800,000 practicing physicians, 500,000 of whom are highly active in the United States who need further information and training. As these tools roll out. Not only do we need to implement these ethically and equitably, we also need to make sure that a human is in the loop. I do not foresee a point when we're going to allow chatbots to perform surgery or make diagnoses without a physician co pilot.
14:57
This podcast is called life in an increasinly digital world, how is it that we're not going to lose the human element in this so much emphasis is put on the bedside manner for the medical professional? How do we not lose the human element in the care that we deliver for people?
15:14
You know, some like Abraham VC and Eric Topol might argue that, in some respects, we've already lost the human element. Right. And with the focus on our views, high patient turnover, and increasingly aging population demographic, and large exit workforce, there's more pressure than ever on our clinicians to seek patients quickly to move them through to bill at maximum levels. And some of that country doctor aesthetic and maintaining relationship over many years has not been preserved in these settings. I think the promise of the AI copilot is actually to restore some of that is to restore humanism in medicine, because is empathetic as a chatbot can sound. We know from studies that when patients learn that it's robot generated empathy, they find it creepy, right? Our patients, we as people, we want the hand on the shoulder, we want to be able to look somebody in the eyes when you're talking about end of life care. And that's where our physicians should be spending their time, not necessarily coding the chart, dropping bills, and ensuring you're compliant with the latest revisions of CPT codes.
16:22
So one of the things I find interesting is since the introduction of chat GPT in November of 2022, all of the other technologies that have been really transforming this, keep it at healthcare have kind of been pushed to the backburner. However, I know of working with you at San Diego, you've got a lot of other interesting innovations that now the large language models are also supporting. But can you talk about, for example, the Mission Control Center concept at UCSD that you put out there as a vision and are now putting that vision into reality?
16:58
Absolutely. As I mentioned, early time, in my role as chief medical officer, we have 150 plus physicians with medical directorship time. And one of the cultures I'm working to inculcate is that every one of those medical directors who are working on quality improvement related to their service line or their area of responsibility, could either bring the bear informatics and data science to help with that quality improvement or partner with a data scientist to innovate new methods and approaches. For example, Dr. Gabe warty, who's a critical care emergency room physician is partnered with Dr. Salim Tamati, one of our PhD data scientists and biomedical informatics. They've been working on this intractable problem of predicting sepsis early enough to intervene and make a difference. And in fact, one of their theses has been that the electronic health record data is necessary, but not sufficient for early intervention. And in fact, from data that's been published by Michigan and elsewhere, we know that the EHR based sepsis prevention models really aren't helpful. If you're predicting sepsis based on a clinician ordering a lactate, it's too late because clinicians already thought about it. And yet that data is not going into the health record frequently enough from nursing documentation and other to predict it earlier in the process. And so we've been able to bring in bedside monitoring data and other data sources like imaging to help build more predictive models. They don't have a higher sensitivity and specificity. They're earlier in the clinical process where they're actually predicting the point where it's helpful for clinicians. Another thing that they've done as a team to instill confidence and credibility in their algorithm. Is there one of the first in the world to teach their AI algorithm how to say, I don't know. So if there's a low confidence or prediction, the algorithm doesn't try to make a low confident prediction. It actually says, I don't know. And just like a physician, will gain credibility with their patient by acknowledging when they don't know something, but they'll work with patients to find the answer. This has generated more credibility of AI algorithms. And in fact, since we implemented this, around nine or 10 months ago, we've seen some of the lowest observed expected mortality ratios we've ever had at UC San Diego, in our sepsis patients. So we're working to rigorously evaluate that data and submit it for peer review and publication now. But that's a really great example of how we're pairing up our medical directors and data scientists to make progress in areas that we haven't been able to make progress for 20 or 30 years. Similarly, we're working on things around re-admissions, many of our services have a higher than expected number of patient readmissions. When COVID hit and telehealth became a routine expectation. Our chief of Hospital Medicine said you know, maybe we can do a post discharge telehealth visit. We know that post discharge visits by hospitalist have been shown to reduce re-admissions and Johns Hopkins and other data but hospitalist go into hospital medicine because they don't want to do clinic. And it's very expensive to set up new clinics because you need front office staff you need physical space you need new staffing, etc. And so this innovation of doing post discharge virtual visits was a really great idea. It turns out the patients liked it, it has some of the highest satisfaction of any clinic we have. The providers liked it, because our hospital medicine physicians can actually do this remotely. And if you're a hospitalist, you don't have other opportunities to work remotely, you're in the hospital. And then the institution, of course, values us because our pilot data and subsequent expansion demonstrated a 34% reduction in patient readmissions within a patient population. That's high acuity. And so this was a win just by thinking a little bit differently about how to solve an old and long standing problem. It's time you asked about the Mission Control. This is an area of focus for large philanthropic grant we received from Joan and Irwin Jacobs, really excited about our funded center for health innovation, and this mission control idea. So imagine, a 24/7 staffed area looks a lot like the NASA mission control, that is a command center, getting all the right information and make the right decisions around patient flow, not only in the inpatient setting, but also for our patients in the outpatient setting. And at home, right. We've been overrun with patients in our emergency department. And we're looking for opportunities to clean out and allow more discharges than we've had otherwise. And so we started a new program called acute care home where we do remote monitoring when an outpatient is discharged from the emergency department, they get a home health nursing visit, and a physician telehealth visit. And this has allowed us to reduce our admission rates for some of those soft admissions patients that could stay or could go home. Our physicians have been more comfortable with the acute care at home program in discharging them home because they know there's eyes on it. Now this 24/7 command center would allow a new level of eyes on these patients in real time, as well as patients that we would discharge as part of the hospital at Home program and other patients in a home setting who might be getting chronic disease monitoring, but have acute issues. And so I'm extremely excited about the convergence of remote patient monitoring, with all the new incredible sensors being released to consumers, with the health system taking a population health view and AI algorithms helping to surface the right information to the right people at the right time out of the count, to help care for our entire population across San Diego deliver better care and add 10 years of healthy life to every resident.
22:31
That's incredible. I love the vision. Talk a little bit more for me about the Center for Health Innovation, right? So you talked about the command center, that's probably the anchor of what the center is doing. But it's not the only thing that the innovation centers is doing. What are some of the other projects that maybe are a little farther down the pipeline that you might be able to talk to our listeners about?
22:51
Well, the Center for Health Innovation is where some of our novel and bespoke artificial intelligence efforts are being housed. Because we have resources as a result of this generous grant from Joan and Irwin Jacobs, allow us to do things that we wouldn't otherwise be able to do with operations margins, right. It is a tough time to be a hospital CEO Tom, most hospitals operate on two and 3% margins anyway. And then over the last couple of years, you've seen a 10% increase in labor costs with the large exit of the workforce post COVID. You've seen 10% Supply Chain increases because of inflation and other supply chain challenges, right. And yet, our payer reimbursement is still only increasing by two or 3%. That puts incredible pressure on hospital margins. I know Tom, your hospital, my hospital, many others are feeling that and we have to find ways to operate more efficiently AI may be part of those solutions. But in the short term, it's really hard to justify investments in things like information technology, when the entire hospitals so pressured and so this generous philanthropic endowment and gift is giving us Resources staff members, Endowed Chair of digital health from a faculty standpoint, that can focus on the opportunities to invent the future to move us forward in new and unexpected ways that we would not be able to resource and staff with the keep the lights on let's solve problems mentality that often comes from a thinly resourced Information Technology Department.
24:20
I was down at your location, it was part of the opening for the center really had a great opportunity to understand the vision of it, but also to interact with the many partners that you have been able to get connected to work with San Diego. One of the things I've always been amazed in working with you is how far your network has expanded and how you cajole all of us to kind of work collaboratively with you and your team and to work with each other. And it's the last question to you thinking about leadership, right and Impact and legacy. Talk to me a little bit about how you think about doing more than just for San Diego health and impacting healthcare writ large as well.
25:01
Yeah a great question, Tom. First of all, our job is always to help train our successors, because none of us will be in these jobs forever. And so something I'm always thinking about in every role is how to develop people and help other people advance in their careers. I've had such an incredible fortune of mentors like yourself, Dr. Atul Butte and Kopecki, and many others who have helped me to develop in my own career and being able to pay it forward is a real gift. We're excited to be able to point as my successor, Josh glandore, the CIO at UC San Diego health, who I've worked with for many years, we are looking at other important roles, including the Chief Medical Officer role, which I hope to vacate at some point, so that other physicians can step into these roles as well. Developing people as part of this, I think that the impact of the type of work that we can help to lead at UC San Diego health and across the University of California Health is going to be really scalable. And as you mentioned, we already partner with other academic medical centers. What I think about a lot are the community health systems, right there 6000 hospitals and health systems in the United States, about 3000 hospital CEOs suggesting that hospital systems on average run about two hospitals, which is what UC San Diego has today. But only 800 of those 6000 are academic medical centers. So the other 5200 are community health systems. And the community health systems are the ones with the most margin pressure and the least ability to innovate. And so I think it's really our job to help reinvent how healthcare is delivered in partnership with vendors, who can scale these innovations, and ensure that they're delivered to all hospitals. A great example of this a decade ago, I was working with our partners at Epic on a product called Care Everywhere. That allowed us to exchange data with other hospitals. That was one of the first utilizers of the care everywhere product today. Every epic client, which accounts for over 2000 hospitals in the United States can exchange data, these type of innovations, we want our vendor partners to scale, but they often start with academic medical center innovations. And so that's one of the reasons that we're partnered with EHR vendors with Microsoft and others, on developing large language model innovations that we hope are going to impact patients across the country, and potentially even across the world, as we find ways that these workflows can be safely, equitably and ethically implemented, and demonstrate that they help patients. I'll tell you, Tom, one of the things I'm most excited about is how we can use large language models in the context of some of these amazing databases to deliver best practice care from patient data that may not yet have been researched, may not have been published. So I know you're familiar with the story of a patient with lupus that I helped to care for who we were able to make a patient care decision based off similar patients like her in the absence of peer reviewed published evidence. And I think that this type of innovation is going to come to the bedside, eventually to all of our clinicians across the country. As I think about the Center for Health Innovation and some of the work that we're doing there. We're partnered with the School of Engineering, where we've got faculty members like Dr. Kevin King, who have innovated a small sensor that sits under a bedpost, and this sensor is so sensitive, that not only can it measure sleep quality and tossing and turning, but when a patient is resting in their bed at home, it's sensitive enough to pick up what he calls a ballistic cardiogram at BCG. The ballistic cardiogram has a high correlation with an electrocardiogram. And so he can actually pick up cardiac arrhythmias from a sensor under bedpost. So we're working with him to scale that from research and small proof of concept and validation into actual use in our Accountable Care Organization across dozens or hundreds of patients who might be at high risk that allows him to partner with vendor companies to commercialize this and potentially deliver these types of innovations across the country in the world.
29:07
So Chris, you've given me now yet another example of showing that everything is going digital. Now your bedpost becomes another data generation device thanks to UC San Diego School of Engineering. Awesome. I'm gonna leave it there for us today. Thank you so much, Dr. Longhurst for joining us on the podcast as always, your insights, your vision, your drive and ambition to improve the quality of care for everyone. It came through so strongly in your statements. Thanks so much, Chris, for being with us today.
29:35
Thank you, Tom for having me on the podcast and for having this great podcast. Really appreciate it.