Episode 25: 'THE GODFATHER' with Dr. Anthony Chang

Welcome back to Season 3 of the Not Mini Adults Podcast.  To kick off Season 3 we have a very special guest and spend our time talking about the importance of data and AI and what it will mean for our children's wellbeing now and in the future.
 
 This week we are talking to Dr. Anthony Chang.  Dr. Chang is the Chief Intelligence and Innovation Officer as well as Medical Director of the Heart Failure Program at Children’s Hospital of Orange County.  Dr Chang also founded the Medical Intelligence and Innovation Institute (MI3) at CHOC Children’s. 

Dr. Chang holds an MD from Georgetown, an MPH from UCLA, an MBA from the University of Miami, holds a certification from MIT on AI, and has an MS in Biomedical Data Science from Stanford.

Anthony has been called “Dr. AI” by the Chicago Tribune having published numerous peer reviewed papers in ML and AI related to medicine.  He is passionate about how AI can transform healthcare and works tirelessly in advancing the use of AI.  

Dr Chang is also one of the original founders of the International Society for Paediatric Innovation and to us is the Godfather of Paediatric Innovation.

Visit our shop here to purchase a copy of the Thinking of Oscar Cookbook - Made with Love or Face Coverings. THANK YOU!

Thinking of Oscar website and contact details can be found here.

Follow us on Twitter here or Instagram here.

Theme Music - ‘Mountain’
–
copyright Lisa Fitzgibbon 2000
Written & performed by Lisa Fitzgibbon,
Violin Jane Griffiths

Podcast artwork thanks to The Podcast Design Experts

Transcript

Anthony: 0:00

Medicine up to now has been like a string ensemble, but mainly string instruments. Now we have the availability of trumpets and clarinets and drums. It doesn't mean that the violinist needs to drop their violin and learn these new instruments. Just learn how to play with those other musicians so that we can have a much bigger range of music that we can play together. So if you equate that to, you know, the very complex chronic disease patients, it's like expecting a string ensemble to play Beethoven's Ninth Symphony, he just can't do it enough justice. And you need other instruments. Actually, Beethoven's Ninth has singers as well, right. So you need as many types of instruments as you can to really bring that Symphony to life.

David: 0:53

Welcome back to the not mini adults podcast pioneers for children's health care and wellbeing. My name is David Cole. And I'm joined by my wife, Hannah. And together we are the co founders of UK children's charity, Thinking of Oscar, it's been quite a year for us, as I'm sure it has for many of you. And we're sorry, it's taken us a while to come back with our third season of the podcast. We're so grateful to everybody who has downloaded, shared and enjoyed the pod. And we're really delighted that so many medical students have chosen to listen and learn more about some of the amazing people that we have been lucky enough to talk to. We have what we hope you will agree is a wonderful set of conversations coming up over the coming weeks. And once again, we are humbled to be able to share these conversations with everyone. We started this podcast in 2020 with three aims, firstly, to continue to educate ourselves, so that when we're asked to support a project, we're able to be in the best position to do so. Secondly, we strongly believe that you can change the world by sharing your story, and therefore, we wish to share as many stories as possible. And finally, we hope that those listening will be inspired to want to help to do more in children's health. Before we get going with this week's episode, we want to say a quick thank you to our comms team who have worked throughout the start of 2021, helping to spread the word of the podcast and share the stories that we've been lucky enough to discuss. So thank you to Susie, Annie, and Carrie. And now without further ado, let's discuss our guest this week. You heard a clip from him earlier and those that know him will know that we are delighted to welcome Dr. Anthony Chang to the not many add ons podcast to kick off season three. Dr. Chang is the chief intelligence and Innovation Officer as well as medical director of the heart failure programme at Children's Hospital of Orange County, and he founded the medical intelligence and Innovation Institute, MI3 at choc children's also, Dr. Chang holds an MD from Georgetown, an MPH from UCLA, an MBA from the University of Miami, and also holds a certificate from MIT on AI. And finally, he has an MS in biomedical data science from Stanford. Anthony has been called Dr. Ai by the Chicago Tribune, having published numerous peer reviewed papers in machine learning and artificial intelligence related to medicine. He is passionate about how AI could transform healthcare and works tirelessly in advancing the use of artificial intelligence. Dr. Chang is also one of the original founders of the International Society for paediatric innovation. And to me, is the godfather of paediatric innovation. Anthony is most importantly, a father of two young girls, who he also discusses during our conversation, we can think of no one better to kick off our third season of the not mini adults podcast, and we hope you enjoy our conversation as much as we did. Anthony, Hi, thank you so much for joining us on not mini adults podcast.

Anthony: 3:43

Thank you very much, David and Anna, and I just want to publicly thank you for everything that you do to help children around the world. It's phenomenal work. So thank you

David: 3:52

Thank you. So I mean, a lot of the stuff that we try and do is actually been inspired by the work that you have done with I spy, and we'll talk a little bit about that. But you know, you and I met a few years ago at the peace 2014 conference, which was a massive eye opener for us in terms of understanding just what the world is looking for, what the world is doing and the importance of bringing innovation into into paediatric care. But to me, the thing that I always talk about, and I used to do use you anecdotally, Anthony, is just to talk about how you've gone from your clinical studies or you know, your clinical work, but brought with it data science. And I think that's always been a fascinating thing. So if you could talk maybe just to begin with just a little bit about your background and how you've brought the two together. I think it's a wonderful example for everyone.

Anthony: 4:40

First of all, I've always been a math, computer, chess nerd in early school years. And that led to several mentors that really emphasise mathematics to me, one was a pathology professor at Harvard Medical School. Another one was a heart surgeon. from my fellowship training. They introduced me to complex math and chaos theory and I was already very enamoured with biostatistics, which is not common for clinician to actually like statistics. And I always did my own statistics for my papers. So I've always loved math. And I think I told you the story that when IBM Watson beat the human contestant on Jeopardy, was a really good wake up call for me to go back to school and learn the current data science, which looks nothing like statistics that I learned before. But it was helpful to have that background going in. And after three and a half, long years of education really gain a really valuable insight into not just applying data science to healthcare, but also I think, change me fundamentally, as a clinician, which I didn't expect, neither did my programme director expect that on my exit interview, I told him that I didn't expect being changed as a clinician, because as a cardiologist, and intensivist, as you can imagine, I was always on system one thinking just reacting and fast thinking. And three and a half years at Stanford taught me that there's a lot of data science available now you can pump on the brakes occasionally, if you can to make a an even better decision than you previously thought. So it's been wonderful for me in terms of thinking of ways to deploy data science and AI to clinical medicine. But also, I think it fundamentally changed me as a clinician.

Hannah: 6:31

So when David talks about Watson, I spent 18 years of my career in IBM as well. So I'm familiar with the story. But he's always talked after that the early times of Jeopardy, and kind of, which was exciting for so many people it was, it was just a really brilliant illustration of, you know, maybe where tech could be going. But then, yeah, I've heard him time after time talking to colleagues and customers about augmenting decision making. And it was really easy in the earlier years, I think more so than now to, you know, for the press to pick up on AI replacing our jobs. And there's been a firm kind of putting on the brakes of actually, it's not about necessarily about taking jobs, but as much as anything about improving our decision making abilities. When you said it changed you as a clinician, is that what you're suggesting that he was giving you opportunity to reflect? Is that what you meant? Or have I made that up?

Anthony: 7:26

Yes, no, you're spot on. I think most clinician, especially seasoned clinicians, that are used to making rapid fire decisions are always working on you know, Daniel Kahneman system one thinking they're fast thinking, they're reacting part of the brain. In situations, I don't have to make such a rapid decision, I've learned to pump on the brakes, and use more system to when the situation allows that. But wouldn't it be wonderful if even in system one thinking situations we are armed with system two already built in? I've spent most of my career in a cardiac intensive care unit for children, and would be so wonderful to have clinicians that work in that setting, to have the availability and to leverage data science during the system one fast decision process, so we have the best of both worlds, essentially. So yeah, I think what what you got from me is absolutely correct, that I think we must try to get this technology and methodology into clinical medicine to make the best decisions that we can, every time, not just when we are sort of bringing our a game and we're we're not tired. Lots of times clinicians are very fatigued, or mired in a situation that they're just not thinking creatively any longer. So we need to relieve the burden as much as we can. So that the best part of the human clinicians, which is the complex problem solving, the creative problem solving can be really, really unleashed as best as we can.

David: 9:00

I think you and I've, again, discussed this previously. And I know that this is also a personal kind of quest for you. But the thing from for me about AI is if I think about when Oscar was in hospital is that his doctors were ringing around to colleagues on the other side of the world or waiting for them to wake up. So you've got that kind of lag time trying to see whether or not they had seen a child with his condition or any correlations in there and the democratisation of data has always been that kind of that driver for me in terms of resetting my professional career moving into healthcare and thinking about AI and being able to actually put the right right data into the hands of the right people at the right time. And that's kind of my driving force and you know, a lot of what we're trying to do from from thinking of Oscar perspective, and I know that it's very similar for you, but why is it so important for paediatrics in particular, and what have you seen and how do we think that we can start down that journey and, and really make a difference?

Anthony: 9:58

Well, it is it Even more essential for paediatrics because we have varying sizes for patients. So many, many more permutations of the same problem. We have rare diseases, that are often undiagnosed or diagnosed not in a timely fashion. And we have so many more permutations basically, of even basic diseases. So we really need democratisation of data, information and knowledge. So we had to first work on sharing data. However, we can safely, however safely we need to get it done. And I think we're getting closer to have mechanisms that will enable us to do that, every institution has some hesitation to share their data. But, you know, there's sort of more current ways of thinking about that. And one is, obviously increase the security aspects. But also, there are mechanisms now, such as federated learning where you actually push the models of AI to the local data, keep that local data where it is, and then push the model parameters back centrally to a meta model. So I think that's another way we can get around the institutions not being perfectly willing to share their own institutional data. And then I think once we solve the the enigma of data sharing, then I think the information and the knowledge will come very quickly, because the AI methodologies as you know, David, is actually kind of ahead of schedule. So we're just not leveraging that technology enough. And it's mainly because humans with the data repositories are still trying to figure it out how to share this data safely, in such a way that we can take advantage of the amount of data. So you're absolutely right, we need to work on the data sharing the data governance of all the institutions, but in a way, that's good, because we should be doing that anyway. So the AI is, I think, a tremendous Northstar. For us to try to reach to really improve healthcare, the best reason I can think of is to gain insights, without having to go to sort of the top children's hospitals by anyone's definition, I have always objected to that kind of ranking, and sort of putting certain children's hospitals in the elite category. I think that should be democratised. I think ranking is great for sports, but terrible for children's diseases

Hannah: 12:35

That is so interesting, the comment that you just made about democratisation and applying it to the hospitals and children's access to care, because in the two series of podcasts we've done so far, when we talked to people about what would you like to fix, then the topic that has come top of the list every single time is about the inequality of access to great care, and you've just cycled back around it again. But the question I had for you was, what was always had always worried me was that what we were describing, leveraging our AI in order to improve outcomes. And, and you talked about, you know, unleashing the real value of great clinicians and the human side of that expertise. It just felt like it was, you know, a real dream and a vision rather than a reality. But what you're describing feels more touchable. So when do you think we are, you know, is realistic to imagine that there are going to be different outcomes as a result of more widespread availability of this type of technology and shared insight.

Anthony: 13:37

On July 15 2028, I'm just kidding. I think it's a, it's going to be a lifelong journey, thinking that machine and deep learning as good as they are, are already showing a little wear and tear in terms of what it's capable of doing. These are great statistical tools. But what we really need in a really big way, is all the clinicians to get involved provide knowledge, insights, creativity into these sort of knowledge machines. So that we can really, really move AI forward into what I call the cognition or the smart AI era. And I think we'll see in our lifetime, I think we'll see in a decade or two, but it's going to take a lot of work. It's almost like machine and deep learning is sort of the easy part of AI is as simple as scaling complicated they can be, it's still relatively straightforward. Essentially, these are labelers, right? They label different things in different situations. But hopefully with the advent of technologies like the Transformers that are around now for natural language processing, and cognitive architectural elements that can be built in to various aspects of deep learning, and some consideration for changing the way we share data, the way we look at databases. I think, in this decade, we can see some, I think, really, really big steps forward. And looking at all of this as a really smart symphony of tools. I just gave a talk just the hour before to Ireland, with clinician saying that we all need to learn how to code and programme and I say, Well, no, just think that medicine up to now has been like a string ensemble, but mainly string instruments. Now we have the availability of trumpets and clarinets and drums, it doesn't mean that the violinist needs to drop their violin and learn these new instruments, just learn how to play with those other musicians so that we can have a much bigger range of music that we can play together. So if you equate that to, you know, the very complex chronic disease patients, it's like expecting a string ensemble to play Beethoven's Ninth Symphony, you just can't do it enough justice. And you need other instruments. Actually, Beethoven's Ninth has a singers as well, right. So you need as many types of instruments as you can to really bring that Symphony to life. So I think the violin is gonna stay in their seats, you don't have to run out and learn how to programme and code. Matter of fact, in some ways, it's actually better that you don't, because then you can actually not be distracted by a new domain, but work with a domain experts, like data scientists to actually work, you know, very cohesively, and harmoniously together to bring projects to life.

David: 16:35

I absolutely love your analogy. And it does make me think that the opportunity that we have now is actually to capture that data, if we think back maybe even as 5-10 years ago, and in some countries, the UK included, we're still not capturing the data digitally. So, you know, we talk about artificial intelligence, and machine learning and all the rest of it. But actually, unless the data is there to be utilised, these systems can't see it at all. So for our children, for example, their clinical data, as well as many other many other aspects of their life is going to be stored in a digital format there, you know, what, what we call the digital exhaust is going to be, you know, far superior to us. So therefore, hopefully, there will be that opportunity, but you're still seeing medical institutions not utilising that and not being able to capture that data. And, you know, a lot of the conversations I have in my professional life is all about signals, you know, being able to spot signals, hopefully, before they cause any adverse, you know, events. And that's going to be so important for us moving forward. The next question, Anthony, is for anybody listening to this, and we get quite a few students, as well as clinicians. What would your advice be in terms of you know, where to start? What should they be looking to do? Either just on an everyday basis? And I guess you've touched on that a little bit. But one of the fortunate things, I think, that we have at the minute is that we have a tremendous amount of great ideas, because we're seeing a bit more of, you know, that kind of democratisation of data and the computer speeds necessary in order to share that data. But what would your advice be to you know, paediatricians that are looking to actually make a difference?

Anthony: 18:09

Well, I think it's a very personal journey. And this new era, I don't expect everyone to just kind of become a passionate advocate for artificial intelligence. I think, as smart and as good a doctor as you can be, because that's very valuable, especially in this coming decade, in artificial intelligence, when we're looking for, you know, sort of these cognitive elements to build into the AI machinery, it's going to really require very, very good doctors with insights and wisdom that we can basically find a way to programme into these machines that you know, want to build for decision making basically. On the one hand, we need better curation and better organisation of data. On the other hand, we also have to work in parallel ways that we can do, accomplish things, you know, without hundreds of 1000s of patients of the same disease category. So it's always good to have both of those strategies in place, because I think those two strategies are very synergistic. But I think for anyone that's listening, I think, just follow your passion. There's always a way you can have your passion and medicine and or data science be used very wisely, in particularly this coming decade. This coming decade is going to be I think, significantly more challenging, because I think the low hanging fruit was medical image interpretation with deep learning. And I think, the next phase of taking on decision making. So although the real holy grail of clinical medicine is going to be here and i think that challenges to solve those issues, and I think within a decade or at most two decades, we won't really be calling anything in medicine, artificial intelligence anymore, because it's all embedded that just how we practice. You know, this new paradigm of clinical medicine, which is, I think, just hugely exciting. I'm a little bit envious of younger people these days. But certainly I think the next 25 years will probably be the most exciting 25 years in medicine ever. I see so many signals and trends heading in that direction. I think I'll make medicine fun again, and make medicine great again. And also I think I'm predicting as opposed to my AI guru friends. I'm predicting there'll be more doctors and more radiologists than ever before and not less.

Hannah: 20:31

And apart from the evolutions that you described from a technical side. I was also mulling over, sort of how do you get there. And what I'm thinking about was scale, because AI works and assisted decision making works, because you have very large sets of data, where algorithms can be trained, and in order that you can derive insight from these very, very large data sets. But I'm thinking about the scale of the data. And given that we don't have the sharing on a global scale, that it's going to be most useful yet. Is this an example where you can start small, because you can't start that small? Because you're not going to have sufficient volumes of data in order to derive insight? So for example, are there things that you can do at an institutional level or I know that I spy is one of the networks that you're extremely involved in, for example? And so then, are there ways of collaborating across existing formal and informal networks where somehow you get some volumes that become valid?

Anthony: 21:31

I think all depends on the disease. But I think certain diseases are so rare. And we need as much data as we can for those few patients, that you absolutely probably going to have to data share as much as possible. So we have to think of solutions with those types of patients, and then their patients with diseases, they're not so rare. And then you can, I think he's still needed to collect data that maybe not absolutely essential that you have, you know, 100 000's of those patients records to learn from, but it's always good to have more good quality data. But on the other hand, as I said before, we have to work in parallel strategies that will enable us to do so in an AI type of research. Without that many patients and one of my areas of interest is how can you work with little data. And part of the I think art of working with little data is going to be something that I've mentioned before, which is bring in cognitive elements to that data set. So you're not entirely dependent on deep learning and massive data sets, you can actually have cognitive shortcuts that bypass lots of data. And I think the clinicians will play a major role in that this decade in terms of providing clinical insights and a knowledge that will really not require treatment or diagnosis, with a lot of parameters and a lot of data. So AI is going to get even more sophisticated. It's not we're gonna head from the statistical era of machine and deep learning to a cognitive era. As you know, I'm a big fan of IBM Watson and what it was trying to accomplish, we just need to continue that philosophy of the synergy between machine and human intelligence is going to be far better than either machine or humans alone.

David: 23:29

Absolutely. During this this journey, you've started up AI med, so people can find it. And I'll put the link in the show notes around it. But that has enabled you to look at lots and lots of different not just paediatrics but lots of different walks within the medical field and where AI can fit in. But my question is during that time, or from your own experience, can you talk about learning from where we as in humans have failed, you know, in the last few years, which has actually meant that we can accelerate? And we can learn from that

Anthony: 24:00

Great question. As I reflect back, I think maybe one of the major takeaways is not to be daunted by these level of sophistication and the level of difficulty of artificial intelligence. I think it's easily distracting for a lot of people to realise that they can't code, they don't understand IT. So they're kind of out of the loop, when in fact, you should be in the middle of a loop because as a clinician, especially a seasoned clinician, that is the most valuable resource, I think in AI in medicine. Obviously, data scientists are not far behind. And not to be daunted by the level of knowledge that you need for AI but just take it in, learn a little bit at a time, learn how this relates to clinical projects and problems that need to be solved. And I think what we're missing and this leads to the second takeaway from the last decade, which is I think, where there's lots of great data science and healthcare. But unfortunately, not nearly enough clinical relevance that we need, I always say the perfect model can have zero impact. And we don't want that, obviously, we want maybe good and great models with, you know, moderate to great patient impact. Because the perfect model, which I think a lot of data scientists try to do, is almost sometimes clinically irrelevant. And 0.99, 0.98 on the area under the curve doesn't really impact on patient care, what we really need to focus is on how we impact on patient care, in terms of delivering better outcomes. It's almost like you need two scores, just like you know, skating competitions, you need a score for technical and you need a score for the art. I think we should have two scores for AI projects, you need a technical score for the data science, and perhaps you need a clinical relevance score. And then combined is I think, what we should do. As a matter of fact I'm editor of a journal, maybe I shouldn't insist on that for my own journal submissions, that we score them on technical quality as well as clinical relevance.

David: 26:08

I think that's a brilliant point. And actually. I can't believe that, given the conversations that you and I've had in the past, that it's taken us this long to talk about the patient to a degree. And I think always, certainly I know that you always think this way. And certainly we do. But it's all about the patient. It's all about bringing the right information, in order to make the correct decision for the patient, you know, rather than just playing with new shiny tech,

Anthony: 26:31

But you may have noticed that peach 2040, as well as AM meetings typically starts with a patient and story. That's sort of the signature of my meetings. Because I want everyone to myself included, by the way to remember that we are talking about patients and families and outcomes and life and death. We're not talking about support vector machines, and deep learning. We are talking about all of these things. And as you may remember, I became a patient myself a year and a half ago. And, you know, pulmonary edoema is a string of two words that you can use data science to find patients. But I realised that whenever we find pulmonary edoema, in a chart, that patient suffered greatly because of that term, you know, so it really adds a emotional and human weight to those terms. They're not just kind of search terms for data science projects, people paid greatly for that data and those vocabulary words that really, really are part of any project, but people pay the price for those.

David: 27:36

We've spoken a little bit more, you mentioned that there was initiative that you started Pedes 2014. So looking kind of, you know, 20-25 years out, but just briefly, what do you think medicine is gonna look like by the end of this decade? So 2030? What do you what do you think we will be,

Anthony: 27:52

I think we will start to see bigger dividends from data science, we will start to see true elements of precision medicine. We will see more institutions willing to share data, and work towards solving common problems and work on common goals, more and more. I think in a way, it took a pandemic for us to realise that, you know, the humans really need to work together much closer than we've been, especially for children. Thankfully, this pandemic has not affected great numbers of children. But if it did, this aspect of sharing data and working together will be even more of a mandate from all of us. So I do see very positive trends for the future and very optimistic about the future. And you know, what's encouraging, at least here in the US more people are interested in going to medicine at all levels than ever before. It's almost like this pandemic has been a clarion call for everyone to get involved in health care. So I do see great promise for healthcare and medicine for the next 10 to 20 years.

David: 29:01

Yeah, and Fingers crossed. You know, a lot more people want to go into paediatrics as well, as a result of that. And interestingly, I was hosting a panel for the Royal College of paediatrics in the UK and Ian Hennessey here, I think you know, it all the hay actually started to think about what were the implications of being if COVID had been reversed. So actually, it was children that had been infected, not adults. How would we have coped with that? And how would we have done it? And it's, obviously it's a horrific thought. But I think asking those questions and turning those scenarios on their head and trying to be prepared for it is obviously key.

Anthony: 29:36

I don't think we should have like wargames equivalent for a pandemic that affects mainly children. I think that would be an amazing idea and project and a wake up call for all of us now, what if instead of seeing older people dying on ventilators, that only children are in those ICU's? And we actually a few of us actually talked about that possibility. You know, what if we ran out of ventilators for children, it was a horrible shock for us to even have that for adults. But I can imagine for children, it will be a very, very terrible world to live live through to see that ever. So that means we need to be smarter, smarter than the viruses, which is not easy. The viruses are actually on the AI side of really, really good complex adaptive system. And they're pretty. There's the certain intelligence to that system. So we need to be smarter.

David: 30:26

Anthony thank you so much for joining us. We ask one final question, if you don't mind, which is if you could change anything in paediatrics, what would it be?

Anthony: 30:34

I would get rid of anything that puts institutional hubris and agenda ahead of the children. So we'll get rid of ranking systems, I would make sure that every hospital raises their game, and we democratise high quality care as much as possible and it may be against hospital marketing strategies. But I think we ought to put institutional agendas aside and put children first and let's rank diseases rather than hospitals. I think that will be a great change in philosophy and strategy for us and children are the future. So we absolutely need to have a very different mindset for children.

David: 31:14

Totally agree. And I think on that note, your girls are probably getting restless. It's about quality nine at night for us, and we can hear the pitter patter of tiny feet upstairs, we better go and check that our children are in bed. But thank you so much, as always a pleasure to speak with you. You know, thank you for everything that you're doing and really fingers crossed that we can we can meet up in the not too distant future as soon as we're able to travel again. Yes, thank you so much David and Hannah. Thank you. Thank you so much to Dr. Chang for joining us on this week's not mini adults podcast. We we had such a wonderful time and we really hope that you enjoyed the conversation to next week feels like a doubleheader because we are delighted to say that we have one of Dr. Chang's great friends, Timothy Chu, who will be joining us to talk about his project around bringing a paediatric health cloud to hospitals all over the world. We really hope that you can join us then please do subscribe to the podcast. And if you're enjoying it, please do leave us a review as well. We hope you'll join us again next week.