If artificial intelligence helps people in their everyday lives, why shouldn’t healthcare utilize this technology too? Listen to find out how AI is helping shape the future of healthcare, with our guest Jayson Marwaha, MD, MSc, a general surgery resident and postdoctoral fellow in surgical AI at Harvard Medical School.
Dr. Marwaha: AI is not meant to replace surgeons or replace any provider, really. It's meant to augment their decision making.
Christine: Welcome to another episode of MEDITECH Podcasts. We’re the leader in healthcare technology empowering you to be a more informed healthcare consumer and provider. Hear the latest from our friends and colleagues in the U.S., Canada, and abroad on topics we think you should know about.
Christine: Jason Marwaha, MD is a general surgery resident and a postdoctoral fellow in Biomedical Informatics at Harvard Medical School at Beth Israel Deaconess Medical Center. He is focusing on applications of informatics in surgery and conducting informatics research. His position is funded by the National Library of Medicine Biomedical Informatics Research Training Fellowship. Dr. Marwaha, it's a pleasure to meet you.
Dr. Marwaha: Thank you so much for having me.
Christine: I'm excited to talk with you about the role informatics plays in healthcare. Why is clinical informatics an exciting field to get into and for any healthcare leaders who are listening, why should healthcare systems hire a clinical informaticist?
Dr. Marwaha: First of all, Christine, thank you for having me on the podcast again. The incorporation of technology into routine clinical care over the next several years is inevitable. It's in the increasing role that it will play in how providers deliver care is inevitable. We need experts both on the technical side, as well as, on the clinical side. People who can sort of bridge the world between clinical expertise and knowing where technology can be useful and technical expertise knowing how the technology works. How it should be designed, implemented, deployed, evaluated, etc. Technology can be made more useful to physicians on a daily basis and those conversations can only really be started and should be led by informaticists.
Christine: Now, it has been interesting to see the evolution of that role and the importance of that role, especially during the last two years with the rollout of various, you know, telehealth programs, etc. It's really the clinical informatics that came to bear a lot of the brunt of some of that work and help with the processes and get that up and running. Pretty much at a very fast clip. I want to congratulate you on your fellowship. Tell us more about your work.
Dr. Marwaha: You know, I'm a surgical resident and so I'm also interested in informatics. So naturally, I think where my academic interest lies is in trying to figure out how to make this new technology I was just alluding to and new data science tools that I was just alluding to in the prior question useful to surgeons. I think that there's been a tremendous amount of interest in recent years in building and developing and validating new machine learning tools and data science tools. But the reality is that very little of it is actually used in clinical practice.
Why I think this broad research theme is very important in figuring out how to make data science useful to surgeons on a daily basis and how they conduct their practice every day is because of this phenomenon that I think we can all relate to, but I think is particularly noticeable in surgery because surgeons spend a lot of time not in the operating room, but also in the ICU and in the ED and lots of settings where you're bombarded was sort of with data all the time and that phenomenon is that as physicians, we confront an ever-increasing and already enormous amount of data about all of our patients every single day the second we walk into the hospital.
The challenge that presents is that it seems as though the amount of data we are confronted with far outpaces our human ability to process it all objectively. The reason why telephone numbers without the area code are only seven digits long is because that's the number of discrete data points that the human brain can kind of hold at once. You're confronted with many more than just seven data points about your patient. And ideally, you should be using all thousands of those data points in order to come to the best decision about how to take care of your patient. But an opportunity to use that data to its full potential presents itself in the form of data science and machine learning.
The computer's strength complements the human and that it is able to rationally learn from and process thousands of data points all at once and learn relationships between different predictors and different patient outcomes and apply those in a millisecond and those sorts of things. That is a huge opportunity to augment the human surgeon or the human physician or human provider to help them make the best possible decisions about patient care.
Christine: Expanding upon that, what is a chief problem in the surgical setting that you and all the work that the American College of Surgeons are collectively hoping to solve in your research?
Dr. Marwaha: The American College of Surgeons is the main sort of governing body for surgeons in the United States, but more specifically I think as it pertains to our discussion, I am a member of two committees on the American College of Surgeons. One is the American College of Surgeons Committee on Health IT and then the other is the American College Surgeons Committee on Emerging Surgical Technologies. And the sort of purpose or sort of north star of both is to really try to figure out how to unlock the value of new technologies that will soon play a role in healthcare on a routine basis in ways that make it valuable to surgeons and surgical patients.
It's funny how AI, as an example of one of many emerging technologies, and surgery touches our lives every day in numerous ways across numerous industries in simple things like Netflix, movie suggestions, directions into Google Maps and you know, it's just like the simplest things every day are powered by AI and similarly, API is run our lives. There are like when you check the weather on your phone. That's an API and it's kind of mind-blowing how none of that technology is really used at all by surgical providers or surgical patients on a daily basis and they really should be. I mean, I think it's time for healthcare to catch up. I think that, yeah, really trying to crack that code and bring these technologies that already deliver tremendous value to our lives in other arenas to surgeries is the ultimate goal.
Christine: Those were great examples of how AI is used in our everyday lives. Healthcare is a little newer to this journey. What are some of the other reasons why AI hasn't been more widely implemented in your field? And, then what could the surgical experience look like in the future with that AI?
Dr. Marwaha: There are so many tools, there are so many data science and AI and machine learning tools that exist that are used, again, you know, in many other industries and for many sort of remarkable sort of applications in our daily lives that have tremendous potential to be applied to surgery as well. And so, as an example, there's no reason why, in the future, we shouldn't have APIs that allow an ambulance to pick up a patient and you put in the patient's sort of basic information, name, whatever and you're able to very quickly retrieve a payload of medical history about the patient and what they're on and just all of their medical data. There's no reason why in the pre-hospital setting we should be so devoid of data and sort of make decisions about care and a void of evidence and a void of data. There's no reason why that shouldn't exist in the future. There's room in the entire clinical trajectory of the patient for informatics to play a huge role. I think there's room for informatics to really bring value at every stage of patient care.
What are the barriers? Briefly, in the same way that it brings value across the entire spectrum of patient care, barriers exist across the entire spectrum of development and deployment and evaluation of AI and other sort of informatics tools. An analysis that we've recently finished in our lab that's under review and hasn't been published yet is we sort of compared the past three years worth of published surgical AI tools and compared them against sort of established Best Practices for how AI and medicine should be built and found that the sort of median compliance with Best Practices is only 50%. Any given tool that has been built and published in a peer-reviewed journal and has been examined by experts in the field, really is only 50% compliant with Best Practices for how these things should be built and reported and evaluated diligently, and that presents a problem to its use. I mean, obviously if it's not built well in the first place then how was anyone going to have the confidence or the ability to to use it to deploy in their health system and use it.
This whole notion of performance degradation over time becomes an issue, right? So AI that was built and trained on patients a couple of years ago may have been great when it was run on patients a few years ago. There's a high likelihood that they experience some amount of performance deterioration because of data set shift. Patient care changes over time. Guidelines change over time. Provider behaviors change over time. And so something that was amazing two years ago might be totally obsolete now. And so even after you've successfully brought some piece of technology to the bedside, you have to kind of continue to monitor for whether it's still fulfilling the purposes it was intended to fulfill. Again, I think barriers sort of span the whole process of building to maintaining these types of new technologies.
Christine: This application of AI sounds very promising though. It certainly presents a transition for the way medicine has been practiced traditionally. How can we balance that surgeon expertise and intuition with the promise of machine innovations?
Dr. Marwaha: You know, there's this sort of common, I think, misconception that the purpose of AI and I think that this best conception has been to some degree debunked. At least, I mean, I think that people sort of are slowly starting to get the message that AI is not meant to replace surgeons or replace any provider really. It's meant to augment their decision-making for the very reasons I was stating at the very, at the beginning. AI is meant to make the human do what the human does best. It's not meant to replace the human altogether and I don't think it ever should, but the human strength is in capturing things that the computer is very complimentary and that it captures things that the human will never be able to capture. Right?
So any surgeon listening right now or any physician listening right now, will be able to relate to this. Sometimes you just step into a patient's room and the ICU and you look at them and you just your intuition tells you wow, that patient is sick. They need to go to the operating room or wow, that patient is sick. They need this medication or that medication and there's, it's hard to explain why you get that gut feeling. But that feeling, I think is very valuable. It presumably incorporates things like what the patient looks like. Are they sweaty? Are they gasping for air? What is their social situation like? Is their family and you know all these sort of soft things intangible, immeasurable things the human brain can measure but the computer cannot capture and so you need to be able to bring the two together because they do very complimentary things.
If you can train machine learning models not only on Lab values and imaging like people routinely do now to predict what will happen to patients after surgery. But also train it on that plus some objective measures of what the surgeon is thinking about the patient and what their assessment or what their impression is of how the patient is ultimately going to do which is I think a very valuable source of data that's quite hard to capture. If you can train models on both, then ideally you build a tool that predicts what will happen to patients and how to best care for patients in a tool that outperforms what either the human or the patient data alone could do. In fact, you have found that if you combine intuition data from the surgeon and physiologic data from the patient, you are able to build models that surpass the performance of either alone. Which I think is remarkable and I think should set a standard for how future machine learning models are built.
Christine: You're truly making a difference. I'm so happy to learn you have a platform to discuss important healthcare topics like these as a co-host on the AMIA podcast. Tell us about it.
Dr. Marwaha: Oh, yeah. So I am a part of the AMIA Clinical Informatics Fellows group, sort of official podcast and the AMIA Clinical Informatics Fellows is sort of a national group for clinical informatics fellows across the U.S. And so I arrange podcast episodes for a podcast called Go-LIVE. And yeah, you know as some examples, we've had Dr. Adam Wright from Vanderbilt who's a nationally recognized leader in clinical decision support talk about clinical decisions support and where it's headed. We've had other nationally renowned leaders and we've had a national renowned leader in telemedicine, Dr. Joseph Vidar who's the editor-in-chief of Nature Digital Medicine come on our show to talk about the future of virtual care and telemedicine and the role that it'll play in healthcare delivery in the future. And many others.
It's a very forward-looking podcast, which I think is particularly valuable both to myself but also to listeners because the future of healthcare will be largely driven bywork done in informatics. And so if you want to know what people in informatics are doing now and where they think the puck is now and also where they think the puck will be five, ten years from now. Then that's what the podcast is for. So I hope listeners of this podcast will check out the ASIF Go-LIVE podcast and enjoy it. Hopefully.
Christine: Absolutely, that sounds interesting. I'll plan on tuning in myself. It's been a pleasure having you today. We've always liked to close our podcast on a personal note. What's something we don't know about Jason Marwaha, MD?
Dr. Marwaha: I spent a lot of my time in school, finding activities that were meditative and would allow me to sort of get away from it all for an hour a day. The lab is a busy place and the hospital is a very busy place and you're surrounded by beeps and your pager is going off and so it's always nice to have sort of a mental retreat. I sort of always sought out activities that would provide that for me and I think one of the most satisfying and meditative activities that I do, I did a lot of also in college and medical school and onwards was rowing.
And so yeah, I rowed competitively in undergrad at Brown and I sort of continued recreationally in medical school and I just love it. I competitively rowed on a boat with seven other men. But I mostly, since then, have rowed sort of done sculling and that is the ultimate meditative experience. You're on a river alone. There's not a sound to be heard except for the splooshing of the water, which is, you know, very soothing and all the sounds of the city get sort of drowned out by the river. It's very peaceful.
Christine: Thank you very much for joining us today.
Dr. Marwaha: Thank you so much for having me. It's been a pleasure.
Christine: Thanks for tuning in, stay informed and subscribe to MEDITECH podcast and be sure to check out our resource page for links from this episode. We'll talk to you next time.