Health AI Forum, Part 3: AI Meets the Healing Arts

Episode Topic: AI Meets the Healing Arts 

Join Dr. Michelle Hermiston, an Iowa farm kid turned global oncology leader, as she uses AI to bridge the gap between childhood leukemia survival rates in Vietnam and much higher survival rates in high-income countries.

Featured Speakers:

• Michelle Hermiston, Dean of the College of Health Sciences, VinUniversity in Hanoi, Vietnam 

Read this episode's recap over on the University of Notre Dame's open online learning community platform, ThinkND: https://go.nd.edu/197946. 

This podcast is a part of the ThinkND Series titled Health AI Forum.

Thanks for listening! The ThinkND Podcast is brought to you by ThinkND, the University of Notre Dame's online learning community. We connect you with videos, podcasts, articles, courses, and other resources to inspire minds and spark conversations on topics that matter to you — everything from faith and politics, to science, technology, and your career.

• Learn more about ThinkND and register for upcoming live events at think.nd.edu.
• Join our LinkedIn community for updates, episode clips, and more.

Transcript

Welcome & Introduction

1 0:06

So I am so pleased to announce, uh, and certainly to introduce, uh, Dr. Michelle Hermiston. She's the Dean of College of Health Sciences at Vin University in Hanoi, Vietnam, where she is building cutting edge programs to train the next generation of health leaders. Before that she spent over 20 years at us, I'm sorry, UCSF pioneering advanced in pediatric oncology, immunotherapy and global cancer care across Southeast Asia. She now advises international organizations on education and cancer research worldwide. Please join me in giving a warm welcome to Dr.

Speaker 2 0:46

Michelle Hermiston. Thank you for

Vietnam Needs Assessment

Data Gaps, Privacy, and Regulatory Roadblocks

Why Vietnamese Kids Had Different Outcomes

Speaker 3 0:55

the kind introduction and the opportunity to be here. this meeting has been fantastic and was definitely worth the long flight. I, have not taken the typical academic route. Um, I actually, uh, grew up in the Midwest. Um, I'm first generation college grad. Grew up on a farm in Iowa. spent eight years in St. Louis and then moved to, there we go, San Francisco for, my residency and pediatric hematology oncology fellowship. And then had the good fortune of being able to stay for, um, almost a quarter of a century, which seems like a long time. Um, and then, about 11 months ago, moved to Vietnam to become the first Vice Dean of Health Research, and then became Dean about a month ago. and when I was in Vietnam, I had the pleasure of meeting net Natasha. and in 20 minutes I somehow ended up getting invited to give a keynote talk here. And I, I don't do ai. I, I, I, I appreciate the use of it and in my current job where I'm, um, expected to accomplish a lot in a short amount of time, um, it's a very necessary tool, but I certainly am not an expert. but this is the slide that, um, made him say, I need you to come give a, a keynote talk. And so, I, have had the privilege in my career of spending time being a bench based physician scientist, moving into clinical trials, ultimately getting to develop and lead our, CAR T-cell, which is cutting edge immunotherapy program at UCSF and being involved at a national level in the children's oncology group and getting to design a lot of clinical trials. During all of that, I, met a resident who asked would I give a talk in Vietnam on HLH, which is a rare, very rare disease, but much more prevalent in children of Southeast Asian descent. And I said, sure. I was also the fellowship director and I was always looking for fellows. And so I was like, tell me about your project and what you're interested in. And he's like, I wanna improve cancer outcomes in Vietnam. He had grown up in Vietnam. He had done a rotation when he was a. Medical student at Harvard, and was very traumatized by what he saw there. Overcrowded hospitals, a very stressed, um, system. And it's like, sure. And so I went in 2015 and then did a big pivot in my career, and became much more involved in global cancer. And I think as I was sitting and listening to the opening remarks, for this meeting, I was like. Wow. My upbringing as a Catholic kid actually I think influenced a lot of that in that passion for equity, and, um, justice. Um, and I think, you know, something I did, I never really appreciated, I think until that moment. I think I'm the only pediatric oncologist in the room. Um, so I'm gonna give you a crash course of why I think I have the coolest job in the world. leukemia is the most common cancer of childhood. Universally fatal if untreated. And to me, this is the poster child for why science matters. Um, due to clinical trials and intensive therapy, we've taken this and we as, um, nine out of 10 children are actually on a national clinical trial in the United States, and that is what has pushed these survival curves up so quickly. But in the 1960s, almost uniformly fatal disease and because of cooperative group trials ran at a national level, we've been able to push the survival curves up to around 90%, which is really fantastic. And I think it highlights the importance of team science. I think the thing that's frightening to me right now as an outsider looking at the US is that funding for this is going away. And so. and you know, we don't cure a hundred percent of kids, so the job isn't done. I think even though we can cure most kids, when you're a doctor on the front line, 90, I can't fix 90% of a kid or 10% of a kid. It's all or none for that child. And I really don't like going to funerals. And so, you know, I was very interested in can we, what do we know about how do we cure this group of kids at this level? And. One of the answers to that was, I'm gonna go back for a second. As you notice, the survival curves got closer and closer and closer together. And while we used to make lots of impact, we weren't making much change in our last several clinical trials. And I think that reflects the fact that. At some point more chemotherapy is just more chemotherapy and more toxicity, but it doesn't actually improve your cure rates and you need something new and different. And the game changer are, was immunotherapy. And so CAR T cells take our normal healthy T cells out of the body. We engineer them in the laboratory to specifically recognize the leukemia cells. We expand them in the lab and we re-infuse them into the patient. And that has been a game changer in my field. Um, and from, um, also adult patients who have lymphoid malignancies. Um, this is, Emily Whitehead. She was the first child to receive these about a decade ago. She had ly relapsed leukemia. The outcomes for that population of kids was less than 10%, you know, and she's a long-term survivor. Has graduated from college, is super cool. And this is the exciting part. So for the population that these cells were tested in and approved in, in the United States, we only cured about 10% of those children. Mostly'cause we couldn't get them into remission because they were resistant to all of the chemotherapies we had. Car T cells still don't cure a hundred percent of these kids, but we get 80% of them into remission. Half of them are still, you know, cancer free after, usually after a transplant, not without even, you know, another transplant at 12 months. And so that's very exciting. But what I started to become bothered by is that, you know, I had the privilege of doing this, really exciting, caring for these, kids with really exciting therapy. We cure 90% of kids with nothing fancy drugs that are on the World Health Organization Essential Medicine list. And it's not just because we know how to do this well in the United States, if you look at any cooperative group around the world in, uh, high income countries, so in US and Europe, DFIC is in Boston, Moss Bore is in Singapore. St. Jude's, we all do really well and we do really well with drugs that should be available to any child anywhere they live. And so what is the state of childhood cancer in low and middle income countries? We don't have perfect data. it's, and so that makes it a little hard, but all of the modeling would suggest that about 80% of childhood cancer burden is in low and middle income countries. And you could say name your favorite disease. 80% of the world lives in low and middle income countries that don't have the same access to healthcare that we do. We've been talking about all these great things we can do. Over the past several days. But I think thinking about how do we take this technology, how do we use AI and actually get it to patients that live in most of the world, is an interesting and important question. And that's what I'm gonna talk through. How do we've approached that in our project, and now with the advent of ai, how can we potentially accelerate that? There's an increased prevalence of childhood cancer in low and middle income countries because the age population is younger. and there's also an increased incidence, and this is because we've actually done a really good job globally at decreasing under five mortality in most places around the world. Kids are no longer dying of malnutrition or infection. And so that means they live long enough to get childhood cancer, um, which you would predict based on what we know. And so this is the modeling from Carlos Rodriguez Galindo, whereas in the red line as under five mortality drops, cancer goes up significantly. And in Vietnam where I've been, um, working, um, and spending time in the last 10 years thinking about, how do we improve outcomes under five mortality has plummeted, which is fantastic, but every year the hospital gets more and more and more crowded. And so how do we deal with that? And so, to me, this is what it is. The driver for a lot of the work that I do, zip code is the number one predictor of outcome in kids with cancer. And, to me that's a disparity that we need to think about how to address. So even though the medications are theoretically available, the outcomes are somewhere between, depending on your zip code five and about 60% globally. And so how do we fix that and can we use these new emerging technologies to do it faster and better around the world? So, I give this talk often. To, pharmaceutical companies, um, because I do clinical trials, I have a lot of colleagues in that space and then they hear about what I do in Vietnam and then they become interested and then like, can you give a talk? And it's been interesting'cause I've been doing this for about 10 years. the shift, in people who are starting to become interested, I'm not sure that it's really for ethical or moral reasons, but. Even if it's for financial reasons, it's exciting that pharma is starting to become interested in partnering in these areas. So I argue cancer doesn't discriminate can happen to anybody and children. your ha bad habits or those things, environmental exposures play very low risk. It really does seem to be mostly bad luck for most kids. Kids don't choose their zip code to which they're born, and so then the question is. Do we have an obligation to all children, to improve therapy? And if we think we do, what would it take to make, childhood cancers in, improve in these settings? pharma has become interested because we've ran out of ni treatment naive patients in the United States, especially for like CAR T cells. And there's an appreciation that. We need to have more genetic ancestry, diversity in our studies. And so going to other places, has suddenly become interesting. Um, and one of the first things I did when I, um, moved to Vietnam was to actually, establish a clinical trial unit and to establish that infrastructure. So how do we improve outcomes? And this is where I think it gets challenging when we think about using ai. Improving outcomes in cancer is complicated. It's not, you know, there's multiple places where you need to think about diagnostics, treatment. You know, do you have the space? Is it safe? How do we get people who are remote? talking about rural medicine I thought was so interesting'cause it's the same problems that we face. I think, you know, both in California and Vietnam are long, skinny countries fairly similar and a lot of the logistic problems are identical in both places. You know, clinical investigation, you know, training people, complex procedures, psychosocial support, NGOs, government policy, all of those things are important and have to be addressed. And so how do we do that? The typical academic approach, is, you know, you work on your little part of the elephant, um, and it's very siloed. It can lead to lots of papers, but doesn't necessarily produce solutions. And so in our global cancer program at UCSF, we took, very mindfully an intentional, approach of multidisciplinary team science that was really focused on solving the problem. so our team had physician, had nurses, had pharmacists, had some, public health scientist and, child life specialist, and really took a holistic approach at trying to, address the problem of infrastructure. oh, interesting. the approach that we use, stems from our educational approach at UCSF in terms of putting. The problem, you are interested in the center of the puzzle and appreciating that to solve that problem, you need multiple scientific lenses. And I think one of the things we need to think about as we think about new technologies and ai, and accelerated approaches is that you need to think about all these things. You know, COVID is a nice example. No one of these areas of, science actually cured a single person. You really need that collective integrated approach to address these things. And so, I found it satisfying last night when, Dale, and the Telestra, group was talking about how do they approach. You really need a strategic plan. and that was how we started in Vietnam. And I think as we're thinking about how to address these very large complex systems, pro, pro, problems. globally a strategic co-design approach is really important. And so, one I, this was a team of people, this was back in 2016 and we did a needs assessment and we took a national approach. One of the problems if you only partner with one hospital is that you improve things there, but that can create inadvertent disparities because you're lifting things up there and not in other places. And so we partnered. Ultimately with all 14 hospitals that care for children with cancer, and blood diseases in the entire country. and in that needs assessment, we saw a number of things. And I think as we think about these technologies and how to apply them in these settings, you have to actually go and see what the setting looks like. So this is Children's Hospital number two, which is the, largest and busiest hospital in the country. Um, they see about 10,000 kids a day on the ward. Um, they had 183 inpatients that day. They only have 120 beds. And in the country of Vietnam, there are only two hospitals that have a one child per bed policy. Think about that one child, you know. I take, you know, at UCSF, we have only single rooms and they're double HEPA filtered and they're huge. And there's a bed for the parent, you know, and in this setting, you know, open Windows immunocompromised kids for the physicians neutrophil count under 500. and the infrastructure is such that, you know, the international standards for, nursing ratio for PT Mon is four to one. Their nursing ratio during the day is about. One nurse per about 15 to 20 kids, and at night it's one to 40. So when we start thinking about how do we implement these things to make it safer, you have to think about the context that that is happening in. And that's healthcare systems in, you know, much of the world is very different from what we are privileged to have. So in our assessment, what we saw was a large number of children with cancer, lots of opportunity for impact. An overwhelmed staff with limited infrastructure. They were highly motivated and dedicated physicians and nurses, and I think that's been the most amazing part of this entire project. There was no mechanism for training, so a peds hem onc to become, to be able to write chemotherapy. They did an adult rotation, adults, and got an adult, a certificate. Adults get completely different cancers than kids. Their metabolism is different, so. No, pediatric specific training and also no mentorship. what I think we had as an opportunity was a very inclusive international network with proven models and expert, um, expertise. There was no research being done and the opportunity for developing, collaborative group with in Vietnam and a cancer registry and the idea that cancer is hopeless terminal illness. Um. Awareness support from local community and government is a way to address that. And so, because we were actually multiple institutions, we, developed our own group that we, um, that uh, has members from, actually multiple continents at this point for Southeast Asia pediatric hematology oncology. And our goal was really to improve cure and outcomes of children with cancer and blood diseases. And our mission was to develop a cutting edge, competent, compassionate workforce that would be retained because turnover was a huge issue. and to think across the spectrum of who does it take to actually cure these children, that you can't focus on only one group, but you really need to take a holistic approach. And I think as you think about using AI too. Develop new approaches. You have to think about the whole spectrum of what it takes from a patient journey perspective. And then, um, because I'm a big believer in those survival curves and that research drives improvement and outcomes, engaging in collaborative research, qua continuous quality improvement and advocacy were, key things that we wanted to work with our partners in developing. We initially focused on three areas, training and curriculum building, national consortium and unified approach, and data, QI and research. we came up with a very detailed plan. I won't go into the details, and I think importantly we linked it to tracking the outcomes and having ways to measure were we successful. The one thing I will say a little bit about, because I think it gets to what I have taken away from this meeting as one of the core things. We need to think about collectively as a community in terms of how do we use new technologies to enhance, um, and address these problems. And that's training and curriculum building. So if you only remember one thing about this partnership is key and there should never be anything about me without me. And so you have to have your partners at the table. And I've seen multiple times where people design these beautiful things. That are completely irrelevant in the context of what they're working. And so we've been very, purposeful in having our partners there at every step of the way and building trust so that we can actually accomplish things together. And I think from a global health perspective, if you're successful, you are no longer needed on this kind of like a, you know, losing your job in many contexts can be a good thing. Um, we partnered with the University of Medicine and Pharmacy and they gave us several constraints. They wanted us to develop an education program that was patterned after the American Board of Pediatrics and graduate medical education requirements globally. They wanted us to teach in English, which is good'cause I don't speak very good. I can order a beer, but that's about it in Vietnamese. but teaching in English, because it was the language of international meetings and wanting to get their, trainees on an international state and then fitting it to the constraints of the com co, um, country. So in the US training is three years there. We had to do it fit to a two year feedback. Our one constraint was a succession plan to ensure stability and sustainability and to, identify who would be taking over before we started. And so we designed this as a train the trainer. we, with a plan of when it would be self-sufficient and went to the Ministry of Health, um, which from a P kid perspective, you know, in Iowa there was kind of a wild experience. Um, the site going to NIH study section except you're in the room while they critique you. but they did ultimately approve the program and we started in 20, 19. and then the pandemic happened. Um, but we were able to go through that. One of the things that I think I'm most proud of from this project is that we really thought carefully about how to do this. And when you develop something new, you don't have to follow the rules. and so we really thought about how do you train the 21st century physician? And I think this has relevance as we think about how to use these tools that are emerging, to, better enhance, education of our, of our workforce. So, you know, from the perspective of how do you train people to treat and address health problems that have not yet emerged and how do we train physicians to use tools, technologies, and treatments we haven't even invented. And I think two relevant examples, COVID didn't exist right when I was a medical school student. And you know, we have had to deal with that problem. I ran a CAR T cell program that didn't exist when I was a medical student either. And I think in the context of the current era, and this is the challenge, this was about a year after COVID 982 results. If you searched PubMed, you know, less than a year later, that had gone up to 165,000 references. And that's the challenge is there's so much data. How do we teach people to find what is relevant? And that's the beauty of, I think, of AI and many of these. becoming capacities of being able to use, these tools to help us be better physicians and to learn and know more. The problem is the, uh, quality of the data and is it applicable, and I'll talk about that in a second. We also decided that we would take a paradigm shift. And the way that education, I think historically has been done for physicians and certainly in Vietnam, is that you focus on understanding and memorizing. You know, so I study, I learn all this stuff. I take the test, I pass it. I forget. Most of what I've learned and a real focus on memorization and recall an alternative approach is to think about the fact that almost everything I need to know as a physician is in the palm of my hand. And how do I teach people? To access that information. And in particular, I think that's important if English is your second language. Um, and so everything we did was really focused on applying, analyzing, evaluating, and creating data, and information and new processes. We used something that we had, I led the inquiry curriculum at UCSF, and we applied this approach as the central framework for our training program. Identify a good question. Find the relevant literature, evaluate that data, synthesize, apply it to your patients, recognize gaps in design solutions, and then reiterate that pro um, process. And we have multiple components to the curriculum. the fun foundational knowledge, but importantly also clinical skills, developing meaningful, research skills. formal training and quality improvement and, a large portion on professional development and leadership. I think one thing that happens globally is when you train people, they immediately will be thrust into leadership positions because there's a vacuum in that space. And so preparing people for that. And then there's the logistic, um, consideration. Um, how do you train people if there's nobody in the country that's actually been formally trained? And so our approach was to co-teach with international and local faculty, preset a goal of developing additional academically focused graduates that would then lead the program. Um, and we had it all nicely, uh, planned out. Um, it was dangerous to be a friend of mine'cause then you were gonna be teaching and everybody did this for free. Everything was donated pro bono. But we had started with 54 faculty on four continents and truly the content experts of areas of P team on globally, they were, plan was that they would teach in one for one week in country in teams of two to three. did a virtual orient faculty development. And at UCSF, I never had that many people show up for an hour and a half on a Sunday night. But, people were really excited about this. and you know, and we now have a, we're up to, I think, 78 faculty with a long waiting list, of people that want to be engaged in this. And so I think that's one of the really refreshing and exciting things is just that people want to give their time. and, I also learned a lot about Zoom and ended up getting tasked with leading our virtual learning environment at UCSF because I had been doing it in this context. but we are, this was our first class of co, cohort of graduates and we're now in year, um, our seventh cohort. Um, and this is going forward very nicely and I think having a ripple effect because our trainees are now training graduates, our, our graduates, and training the next generation of pediatric hematologists, oncologists, as well as treating, provincial physicians. and we've also used this to develop infrastructure in the region. transplant is an emerging modality there, and so we have a monthly virtual meeting, um, for transplant at a national level now has expanded to, I think nine countries in the region. and importantly having people who actually know how to do this in a setting that's, similar to Dr. Julia Palm. Is a transplant expert from Chile and she's been instrumental in helping build this infrastructure within Vietnam and the region. So that's sort of how we approach that. how are we thinking about it now? And, I've tried to, cultivate friends who have expertise in AI because I think that's one of the really interesting and exciting things about this is it allows us to accelerate generating equitable outcomes. And we've heard lots of different things at this meeting, so I wasn't going to go into those in great detail. But we really can think about how to integrate, every part of the patient journey, um, using these tools to enhance and potentially, accelerate improving outcomes. So from diagnosis. Medical imaging, pathology and risk stratification. All of those projects are ongoing actually at VIN University and several institutions in Vietnam. Thinking about treatment from surgical, robotics, pharmacogenomics, precision medicine, drug repurposing, and drug discovery, all can, uh, I think be ac um, accelerated, um, using, um, uh, the algorithms that all the computer scientists are developing. In terms of operations, we heard earlier today about authorization, refills, appointments, remote monitoring. Um, we're now doing a project where we're, trying to develop, use, um, essentially, avatars to help with patient navigation.'cause that doesn't exist. And when you have a ratio of one nurse to 20 patients, there's no time to do that. And so can we, use virtual approaches to help assist with that? Proactive health management, I think is one of the most interesting things from the perspective of AI and the information that is there in terms of being able to collect information from people. Personalized, um, medication reminders and symptom tracking. And one of the things I learned from one of my mentors is you can't cure cancer if they don't take their chemotherapy. And so something as simple as having a, um, a remote monitor in the cap of the to know, did they take it off and sending a text, and you can do that in a resource limited setting. Everybody has a phone there. and so I think those are things that are exciting and helpful. And then I think the thing we don't, I haven't, I haven't heard much talk about at this meeting, but I think is really interesting and important is pa how these new technologies can also empower patients. sometimes in a good way, sometimes, in ways that aren't as helpful, but it ha does allow for enhanced health literacy with, um, interactive educational content. Patient navigation via chatbot and avatars and connection to peer support groups. I think the part that is hard, and I think it's one of the values, that I have learned from being in country and on the ground, is you have to do that in a culturally uh, competent way. And so I was super excited last night at the poster, session, Angelica Gar, uh, Garcia Martinez had this is doing this beautiful work. Look at an at empowering caregivers through ai, a culturally tailored chatbot for nutrition support in pediatric cancer. To me, this is super cool because there's no time to do teaching in in these settings, and so can we automate some of this and how do you do it in a way that is culturally appropriate? So I practiced medicine in San Francisco for 25 years, about 25. About 40% of our patients are of Asian genetic, ancestry. About 40% are Hispanic genetic ancestry. Food is central to both of those cultures in terms of, how parents care for their kids and those things. How those two cultures approach it completely different. And so coming up with ways that are culturally sensitive and can be adapted, I think is really important. And so this excited me because I think a lot of these technologies don't take into account, necessarily the, the local context. I think when we think about AI and healthcare, there's also a lot of challenges and things we have to be wary of. first and foremost, and we've been hearing about this for the last two and a half days, is the limit is data. what you put in is what you get out and how we train these, different algorithms really depends on the data that you put in. much of that data does not include people, in resource varied settings. There's also the privacy concerns and the ethical issues that arise with the use of sensitive health data. We have lots of rules about that in the United States. Those rules don't exist in other places, but we still, I think, have a moral responsibility to protect people's identity. and, personal data. There's lack of regulatory frameworks that hamper widespread adapt adoption, and the regulations are variable in, and have lots of complexity. The thing we thought would be easiest to do in Vietnam was just to set up a cancer registry that has ended up being almost impossible. Because they cha we were partnered with St. Jude. They have a beautiful platform for doing that. We were, and we got a grant in collaboration with them to, to implement that. and then they passed a law in both China and Vietnam that with all of the Facebook things that you can't put data into the cloud, it's really hard to do a, registry without that. And so then we. The work around was let's get REDCap and all of the hospitals that's free, and have this, you know, reprogram this to have it, at a local level. So there are ways to work around these things, but the regulatory context, can make things challenging, and a lot of delays. And it's helpful to think about that upfront. integrating AI into clinical workflows. People have asked me how is it, um, what is the uptake in Vietnam? It's so much faster to hit there than here. Um, I think partly because they don't have all the regulation, partly because they're so overwhelmed and overcrowded that anything that can make their job easier is something that people will grab onto very quickly. Um, and so you, you hear a lot more in that context than you do in ours, which is interesting. and then I think at the end of the day is you have to ask, does it save or increase time spent on non-patient facing tasks? So I've lived through three different, four different, four different medical EMRs, um, electronic medical records and I can tell you that documenting before we had any of those things was so much faster.'cause I can talk really fast and I dictate well, you know, and while I appreciate, the electronic medical record and. Preserving data. I also know that, you know, we make dot phrases and we pace forward and all of those things, and it takes a lot more time. And as the, you know, the law is that the attending has to write the note. So all of those things can, while, while intentions can take more time and in a setting where they're so busy, thinking about those things carefully is critical. I also think it's important to remember that one size does not fit all. And this was another poster I really enjoyed last night from, Zi Tech. Um, and she is looking at Swahili, in these line, and how do you process that in this, you know, into these different, um, data sets and algorithms. And one of the challenges is that. There's a gazillion different dialects of that. And so how do you, code that and, translate that into electronic information? And so I think thinking about those things, if we want to make our data more inclusive, is gonna be important. I think the other challenge is garbage in, garbage out. so this, I led the inquiry curriculum at UCSF for 10 years before moving to Vietnam, and we showed this as the second slide. On the first day of medical school. Half of what you learn in medical school be shown to be dead, wrong, or out of date within five years of your graduation. The trouble is that nobody can tell you which half. So the most important thing is to learn. to learn is how to learn on your own. That statement should actually terrify all of you who are doing AI and data. You know, we do this big data, we throw it into the system, um, and that's how we train, you know, and this iterative process. But if half of that is wrong in five years, and we're using data from the last 15 years in healthcare, that should make alarms for all of us. and so I think thinking about this iterative process and how quickly we have to update things, and that it's a very dynamic situation in terms of when we think about healthcare is critical. I also think we have to think about, internal validity. And what that is is essentially, what makes it true is the data good when you run an algorithm. And if you do it on a, I was writing a paper, I have expertise in a rare disease called histiocytosis, and there were several of us from around the world who were gonna write a pos, a review on, you know, kind of current statement. So we asked AI to write it first for us, actually did a really good job. But what was fascinating was that it did lots of quotes of several of us. And all of the references were completely wrong. And so if you run, you know, if you, I use chat GPT all the time, but you're like, if you run a question just to check it in an area of your content expertise, and then you look at the reference, you know, it's usually a white paper or something that's been cited, you know, a lot, not necessarily the good data and the part that I think is missing in this context. Is we also need to teach the bots are, is the data good? Is there internal validity? Is it scientifically solid? and so we teach this, um, particularly in Vietnam and we do, a series of journal clubs to help them think about these things and to really learn this skillset. And it's so interest. They're student led, faculty facilitated, and it's so interesting. To watch them learn how to really critically analyze data. But one of my worries is that in this, modern era where it's so easy to type in a question and get an answer back, losing that critical eye and skepticism, around internal validity equally important. And this is the third slide that we show, um, both our fellows in Vietnam and our medical students is really thinking about external validity. Does it apply to the patient population or system in front of us? And you know, I haven't heard any comment on this the entire meeting, but it's actually really important. So does the data come from people that look like the patient in front of you? And is there a way that we can actually teach these systems to test and look for external validity? This is an example of why this is important. So these are two different studies. This one was done in France. This one was done in the United States. It took kids that were treated in France on French protocols or in the us, on US protocols at children's oncology group sites, and looked at their outcomes based on genetic ancestry and in two different continents, two different, you know, therapeutic regimens. If you look at the kids who were Vietnamese, they did substantially worse than kids who were non Vietnamese. we now have some hints as to why that is, but I think it raises important questions about biology, unique social determinants of health. we know that Vietnamese children have a much higher rate of a polymorphism, so a change in one of their genes that metabolizes one of the key drugs that we use. If you are homozygous for that mutation, that polymorphism, you need only 10% of that drug to get the same levels as somebody who lacks that mutation. and so these kids die of toxicity because they're profoundly, neutropenic in the toxicity of the drugs. So those things are really important to think about in terms of how do we apply the answer from chat, GPT. To the patient in front of me are, you know, we're patients that look like mine in there. And if you have it, if you ask in your, questioning to whichever, chapa, algorithm you like best, you know, restrict to Vietnamese patients, you get nothing back. You know,'cause it, that doesn't exist in the literature and that holds for most children in low, um, middle income countries. And so thinking about how do we improve the data is important. And then I think the final thing that I'd like to comment on from the, the perspective of a pediatric oncologist is how do we retain the humanism in the era of ai? And there are some simple truths, caring for children. This is a slide that we, share when we are teaching communication skills. But I think it's important to think about from a humanism, compo, perspective. Caring for children with I is hard. Despite our best science, we are not able to cure every child. And people, often say, you know, oh, it must be so hard to be a pediatric oncologist. I'm like, no, I mean cure most kids. That's great. And for the kids that we can't cure because my field is so evidence-based, it's really because their disease is smarter than our science. And that makes it easy to pivot and to walk with a family. to maximize the time that they have with the child. and to do your best to make the memories that they have, meaningful. Every child and family's journey is different, but I think, you know, the important thing is that we can always care. and so how do we retain that in the, art part of medicine, in the era of ai? And I loved what you said about the human touch'cause I think that's what it is. That's what excites me about the era we are in is that my, I so hope that all of you computer engineers can enable us to have more time to actually be with our patients and do the fun part of being a doctor instead of the less fun. this is a, Chris. Adrian is, um, was actually the day he started Fellowship. Um, he was one of my PT Mont Fellows at UCSF. He was also named one of the top 20. authors under the age of 40 by the New Yorker, and he wrote this very interesting article called The Question. It was an essay that was in the New England Journal of Medicine that comes from the cocktail party question. So people ask you at a party, what do you do? And you say, A lot of times I'll just say I'm a pediatrician. Sometimes I'll say I'm a pediatric oncologist, and people go, oh, that's so hard. That must be so set. And then they always have to go to the bathroom and you end up standing there by yourself. You don't know how to talk to you, but I, I actually think I have a great job. but what Chris picked up from interviewing all of the faculty in our division was that multiple people told him, talking is the most important procedure of our profession. And I would say that that holds for much of medicine how we talk to people. Our words can be medicine and. I don't think that the computer's ever gonna be able to completely, replace the, the touch and the words that we can use with people. My hope is that they'll give us more time to do that. So in terms of thinking about all of these different things that AI can do, what I really hope is that we'll bring back time so that we can actually do the job that, is the most fun part of our job. So in terms of takeaways. I think we're at the tip of the iceberg. one of my faculty's like, oh no, we need to do a big survey to think of where, you know, to see where people think AI will take us in five years. I'm like, it's changing so fast, we don't have a crystal ball. but I think thinking about what we should worry about and how we can utilize this to be able to take better care of patients is important. The challenges are, significant. I don't think they're insurmountable. I think we have to have a strategic plan with all stakeholders at the table, and that takes into account the local context. We need to support generation of reliable and unbiased data, and that's the part that I hear missing in this. We're taking all this data and we're putting it in, but the data we really need still has to be done in many ways, the old fashioned way. Um, so we understand con these different contexts. I think a perfect example was a talk I heard from a colleague in India a couple of months ago, and they took Western protocols, applied them, to kids in India and saw huge toxicity rates because they have so much antimicrobial resistance in that context. They modified the protocols to their context, their survival rates are better than ours. And so I think that there will be come a point where we will see much more bidirectional flow of information. It's hard to, um, deescalate therapy in the United States from an ethical perspective. If I can cure 90% of kids, should we step back on therapy? It's hard to sell that. I think we can learn a lot from resource varied settings where it's safer to do that, and we're gonna find out we don't need to give as much toxicity and exposure to these nasty drugs as we are currently doing. and then I think we need to keep the ethical aspects at the front. uh, and I think we have to be bold and, you know, to think big, to be a catalyst and to think really end to end in this process. I have had the privilege of working with an amazing, uh, number of people from literally all over the globe, and this is really a very collective, effort that, um, I've had the privilege to be involved with. I wanted to say one thing about why,'cause I've gotten this question a million times. Why did I move to Vietnam? I, I did have a really good gig at UCSF, but I'm really passionate about wanting to improve access and equity in healthcare. And in Vietnam, I think we have an opportunity to be a leader in the region and beyond and to develop a roadmap for both education and healthcare for other emerging countries. And you know, when you look at those survival curves, emerging countries, I don't feel have should have to go through the five decades to get to 90% survival. Can they leapfrog this and can we, learn from that collectively at a global level. and then at the end of the day, it still comes down to the kids. This was a child who was on that doorstep, um, at the National Children's Hospital in, I have permission to use this picture, in, um, Vietnam about three months ago. And that was him two weeks ago. and we were able to get him, you know, compassionate use because of these pharma relationships to a drug.'cause he was. Blasting through, all of the traditional chemotherapy, it makes a huge difference. And so I think having the opportunity to increase access to these things is really important. So I'll stop there and say thank you. Um, I am hiring so anybody who wants a real adventure overseas, come talk to me, and I'd be happy to answer any questions.

Q&A

Speaker 2 44:55

Yeah. A question. So, thank you so much.

Speaker 4 45:01

So my question is, so we were, so you've mentioned a couple of times, you know, needing to adapt to the, the context you're in, right? So, I'm curious, you know, how when you went there, went to Vietnam, how you, you know, what approach you took to understanding the context you were in. You know, given that it's probably very different from ours, you know that, that it is very different from ours. You know, there's probably a lot of nuances to under understand, so I, yeah. Just curious what, what approach you took to help you.

Speaker 3 45:29

we were lucky in that we had somebody who was an ethnographer on our team, and took copious notes and then we fed them into deduce and decoded them. and then we, I, I've had, I often took students with me who would play that role of just collecting the observations. Um, so that was part of it, you know, just being observant and curious. I think the second part was, meeting with the right people, so getting stakeholders around the table. I think the third is that you have to be really patient trust. Um, Vietnam is a really interesting place, you know, and the United States didn't necessarily do great things over there for a period of time. And so there's always that question of how you'll be perceived, is a people that live very much in the here and now, but trust takes time. And so showing up and coming back, following through those things, then you start to learn. we supported, a national consortium, based on the Children's oncology group. And I think the fact that we were very clear and transparent upfront that this is a bi-directional partnership, that we are here to support you, but being very clear about, you know, they want to do research, they wanna be on an international stage, that's really important culturally for them. I don't need to be the last author or the first author. This should be about you and how can we work together to get your story out there? And so I think those are the things, but it's very iterative. You can't fly in once for two days and think you understand their context. and then I think the other thing that's really important is that when you're designing things, you need to have partners at the table. and so, I think a lot of global health, you know, there's this whole concept of co colonialization. And global health and people coming in, taking data, taking samples, taking it out versus coming in saying, wow, this is a really interesting problem. How can we work on this together? Why don't you guys seek, you know, we'll do this in parallel in our two settings, and then we can troubleshoot, and those kind of things. So it's mindfully wanting to build capacity. and then you partner with them to actually do the studies and collect the data so you can actually feed it into the system and that becomes part of your database and you can start to pick out is it similar or different, and why is that? And thinking about it from a molecular level all the way to a systems level.

Speaker 2 47:51

Thank you. The other.

Speaker 5 48:08

So thinking about ai, I know some of the solutions I've worked on in the oncology space has been more on, um, diagnostics, imaging. So more of using the readily available data. Given that you've seen the full spectrum of a patient's journey from, you know, the. The onset of symptoms through, through the end of care. What would you see as sort of your, your first step of an area that you would like to solve for? where would I like your

Speaker 3 48:34

help in solving for to? I think that, you know, so where are the, the pain points? Like what are the things, so. That's a mo that's a really complex problem and I think there are multiple things you do can do and they don't have to be done necessarily sequentially. So when I partner with people or when I have students who wanna work on a project, I usually start with, what are you interested in?'cause you're more likely to follow through if it's something you're passionate about. I think when I hear you say diagnostics, that's a huge issue in that context. you know, learning how to do that, the fact that you can now look at specimens together or code them and it, and all of those things over zoom and teach that way is a game changer. the system is very, very overwhelmed there. So if we can come up with, you know, that's where I think we can make a really big difference right away is accurate diagnosis using AI assisted methodologies. same thing true with, you know, if you have, you know, the Bach, my hospital's the largest public hospital there, they have 19,000 outpatient visits a day. Can, I'm like, I can't imagine that get me like, that's uber Mayo. and um, so if we can use AI assisted methods to do the first preview on all of the mammograms, that would be huge. And then you can, you know, follow that up. So I think the diagnostic space is really important. I think the second place that to me feels like, um, we can make huge impact. Is the operational part. So at the National Children's Hospital, they usually have 200 or so chemotherapy visits a day. They don't know until families show up in the morning who's coming. Um, like, you know, I, like, I write my chemo like a week in advance. And if you forget, you know, you're getting paged by the pharmacist at, you know, 10 o'clock at night in the us and so. Coming up with systems that allow them to track patients to accurately forecast how much drug do they need. And those things, I think is a game changer and it's not that hard to do. And then I think the third area is really around how do we use AI and these technologies to get time. People, if you are a, you know, one nurse per 40 patients, like we wanted to implement, there's something called, pediatric early warning, SY SY system pews that's based on vital signs and those kinds of things. you have to have somebody actually do the bio signs to know that they're act, you know, and if you have 40 kids that you're responsible for, that's not gonna happen. you know, so we actually are, you know, like educating parents on how to do that. But, you know, can we automate some of these things to buy time for people so that they can give better care? I think those are the, the low hanging places to really start. but there's, um, there's a lot to do, but it's, I think the thing that's also exciting in this setting, Is that you, you're starting at a much lower place or with a lot more complicated complication. You know, things that you have to fix. So you see progress really quickly and it's super satisfying and, and fun. And and I think, you know, people in these settings are doing, you know, I don't think might have an easy job. They have a really hard job and they do it because they're so dedicated to these patients and these families. And so it. they make things happen in ways that don't happen in the United States. That's really fun and, and satisfying. So yeah, if you have any ideas, I can find a place to test them and implement them and people to partner around that.

Speaker 6 52:12

Thank you. So, um, this is just so heartwarming. I, it's a wonderful talk for the end of the day. Um. I was just wondering, you know, especially when I think of partnerships. Mm-hmm. And you've outlined a lot of things where, partnerships could be really, really helpful to you and what can be done and, and amazing opportunity there. Are there things that frankly, we in the US should be learning from what you've been doing in Vietnam so that your partnerships not only are feeling good about what they're giving, but they actually can see it's gonna help them scale something better in the US too.

Speaker 3 52:56

Yeah. I think, yeah, you're bringing up exactly like the centerpiece of our global cancer program at UCSF is all around bi-directional partnerships, and that's why I have a waiting list of faculty because I think what people realize is that you gain much more than you give in this context, and you learn a ton of what is helpful here. So, I went there because I have expertise in this disease called HLH. When I started working with the doctors there, and we did, I mentored somebody to do a research project and you start to learn from that. You're like, oh my God, I'm completely overtreating these kids. I, you know, how you do it is so much better. Let's deescalate care and you know, and now that's kind of becoming how we do this in many places in the United States. Um, and so I think that we learn both directions. and that's the value. And there's, you know, if 80% of the world's population with name your favorite diseases are in those settings, there's a ton we can learn. I think that will be applicable in our context. and so at least for childhood cancer, we're seeing a lot of that. And I think there's an appreciation that, We, you know, if we cure 90% of children, we're overtreating a whole lot of kids. And can we learn in these contexts of other ways to do this ways that are safer or better? and in, resource varied settings where you have much higher rates of antimicrobial resistance. They have to modify the therapy. In order to not have kids die of toxic deaths. And when you see that those modifications, which are usually less intensive, give you the same cure rate that informs what we do here. And so I think that there's a lot of ways that this can go back and forth. know, learning how to give good care and get good cure rates with, you know, fewer personnel. There's a, there's a lesson to be learned there. So, I mean, I'm, I think I'm always learning things there. It's really interesting. Great. Thank.