Even before COVID-19 overwhelmed hospitals, many healthcare workers were feeling engulfed by their workload, never having enough time to properly focus on their patients. Automating key tasks in the patient journey could help change this, so patients receive the best possible care.
Dr. Devin Singh, a pediatric emergency physician in Toronto, Canada, as well as the founder of Hero AI, a startup dedicated to empowering patients and healthcare providers with AI, came in to chat about how he was doing just this. Devin discusses applications of AI in healthcare, navigating the institutions that compose the healthcare system, his collaborative approach to bringing people from different fields together in order to cut the red tape surrounding the implementation of a brand new solution, and the future of healthcare.
Read on for the transcript!
Hi, I’m Monte Zweben, CEO of Splice Machine. You’re listening to ML Minutes, where cutting-edge thought leaders discuss Machine Learning in one minute or less.
This episode, our guest is Dr. Devin Singh. Devin is a pediatric emergency physician in Toronto, Canada, as well as the founder of hero AI, an innovative startup lab dedicated to empowering patients and healthcare providers with AI. His research focuses on using machine learning to solve some of healthcare's largest problems. Welcome, Devin.
Hey, Monte, thanks for having me. It's such a pleasure to be here. And I'm really excited to chat and, you know, connect with the audience, really looking forward to it.
Well, tell us about your journey. How did you get to where you are now?
Yeah, so like you said, I'm a pediatric emergency doctor working in Toronto. But the journey has been a bit of a exciting convoluted one. I did my undergraduate degree at Western University, and then worked for government for a few years as a business analyst there with the Ministry of Finance and revenue, actually, and then after that, jumped over to Australia to do med school at the University of Sydney, and then did my pediatric residency training and pediatric emergency medicine fellowship at sickkids in Toronto, and then I also did a, an AI fellowship at the University of Toronto as well with a real clinical focus on figuring out how to translate AI into healthcare practice and workflows. And then just because I love to learn too much so and punish my wife with school tuition fees, I did a Master's of computer science as well at the University of Toronto. So that's just wrapped up now. I'm really excited to like, have the oldest experience and education behind me looking forward to it.
Well, that's awesome. You're, you're a real slacker. You have your master's in computer science. Boy, I can't believe you've been so lazy.
It's all my wife, Monte, I gotta say she's gotten me through it for sure.
That's cool. Well, tell us more about your work with Hero AI. What's the problem you're trying to solve?
Yeah, that's a it's a great question. What I've realized through med school residency, and now throughout my master's of computer science is that we're sitting on this incredible technology, and infrastructure, we've got a ton of data being generated from EHR systems, we have a patient population that's highly motivated to engage and utilize technology, and a physician group and a nursing group and healthcare system as a whole. That is really overrun, putting COVID aside, you know, the burden that sits on health care systems, particularly in Canada, a publicly funded system is quite large. And so we've got this setup of technology data and motivated stakeholders, where now is really the time to figure out how to connect the dots together. And I really feel that machine learning and artificial intelligence can sit perfectly in that triangle, and allow for these types of technologies to be translated and create impact at scale in a way that's never been done before. And so that's what we're trying to solve is how do we translate that into the hands of patients and providers?
Thanks, Devin. And just for clarity, EHR, means electronic health record, the system that doctors use to store all their notes and all the information on all of their patience. Can you give me an example of how AI can be used in order to deal with that burden for the staff in this environment?
Yeah, totally. I mean, it's all about, you know, building capacity. And I think machine learning allows us to build capacity into healthcare workflows, in a really unique way because of its ability to facilitate aspects of automation. So let's take a simple example, kiddo comes in to an emergency department. Unfortunately, you know, ice skating, playing hockey broke his wrist common problem right now in the winter, right? When he or she comes into our emergency department, we assess pain, we treat that pain, but then we need to circle back and reassess pain. But you can imagine that there's a human element there. You know, the emergency department gets busy. We've, you know, not forgotten but have been pulled away because of different priorities to circle back to that kiddo. Right. And our ability then to adequately address someone's pain in an emergency department gets limited based on capacity issues. You can use basic machine learning algorithms, like NLP models, to understand when a patient is having pain to trigger an assessment for pain for that person through the interaction of a front end device like a mobile app, and then feed that back to providers. And that's just one very simple idea. have, you know the types of technologies we're building out?
Yeah, that's great. So first and foremost, we were talking about the staff a moment ago in terms of helping them deal with the capacity limitations. But what you just told a story about was really looking at patient outcomes and helping the patients in addition to helping the staff, is that a fair assessment?
Absolutely. All of this is very much rooted in this human element, right? It all comes down to we work in a healthcare system, that's very challenging one to get into to be able to have the privilege to work there, but to to even, you know, be functioning and providing this level of care to patients. And what are we motivated by is patient outcomes, patient care? That's really the heart of the type of innovation we're focusing on at hiroi. And we're trying to enable healthcare providers to do that for their patients.
I bet it's pretty cool for kids to to be talking to an AI and helping and having the AI help them feel better to.
Absolutely. And there's so much cool opportunity to think of like, how do you customize user interfaces based on the developmental age of a child? Right? So whether you're a five year old, and we're asking you about pain, or gathering symptoms, versus a 14 year old, there's a lot of nuances to how an interface needs to communicate despite the end goal being exactly the same. And despite the back end model being the same, the way it engages, needs to be done in an intelligent way. That's a really unique technical challenge. But it's super fun to work on.
Excellent. Well, let's transition to some more technical talk, what are some of the tools that you're using for your work?
Yeah, so I mean, really, as I alluded to, before, we've got these silos of opportunity, you've got a large EHR database that is then broken up amongst different institutions. And so one of the very simple objectives is getting all the right data into the right spot. And that involves, you know, partnering with community clinics partnering with patients in their homes, allowing them the power to share certain levels of wearable information or home data with the right providers at the right time. And then building a back end infrastructure that can ingest all of this synthesize this data and turn it into action. And so, you know, a lot of it is a I mean, splice machine is great at doing this, you know, a lot of it is pushing, pulling together data in the right space. But then our challenge with hero AI is building the machine learning models, to then integrate seamlessly into a clinical workflow, keeping the patient journey in mind, and making sure that we delivered the right alerts the right automation at the right time, never wanting to be obsessive about it, but also not wanting to miss the opportunities for efficiency.
Fantastic. So what we're talking about here is ingestion, getting the data from the disparate sources, and being able to consolidate it and make it useful. So moving to your journey in the implementation of AI into the clinical workflow, what's one specific challenge that you faced along the way you'd like to share with our listeners,
the first is actually just being able to translate solutions into clinical practice. And at a large institution, there are many, many barriers in place. And it's not because stakeholders are intentionally sort of road blocking you Everyone is very motivated to see these technologies translate has been my experience, like extraordinary enthusiasm. But the systems in place to allow for the approval and the movement forward of these types of new technologies weren't designed to vet machine learning. And so each step of the way, I find that not only are we translating new technology, but we're also helping an institution invent the process to vet the technology that then needs to be translated. And I found that that was really interesting and not anticipated right from the get go.
Okay, so summarizing, the challenge you just talked about, it really comes down to trust. And you not only had to innovate on the AI itself, and figuring out ways that you can improve patient outcomes, and assist the the medical staff with the burden that they have in the capacity that they're limited by, but now you have to take the whole organization and figure out a way where they can trust what you've done. And that sounds like a really interesting challenge.
Well, because what's interesting is when you these processes are put in place to mitigate risk. Right? And so then when you're going through this approval process, whether it's on privacy, data governance, you know, IP, all of these different things that get factored into when you're using large amounts of data, right? These pieces have all of these intense processes that are sort of historical. And when something doesn't fit, that equals risk to an institution, and the moment there's perceived risk, everything slows down, there's this perceived risk involved, when really the thing we're doing is actually, you know, even more robust and secure than a previous solution that was deployed. It's really interesting that way. And so it's about getting people over that risk, comfort, while also not recognizing our role, while also recognizing that risk is real. And we need to step cautiously together.
Can you give our listeners a specific way that you overcame that maybe for one of the deployments that you were successful with?
Yeah, so it's really around being one, understanding that this risk exists, this perception of risk exists, and how do you then work with someone collaboratively to overcome that, so understand, you're gonna run into a brick wall, and just when you know that there's a brick wall there, just just run right into it, when you when you want to, when you don't expect there to be these roadblocks. Things get, you get frustrated, right, and your timelines get blown. Everything, you know, everything goes off. But when you plan for those brick walls, it's okay. So I want like, anticipate that you're going to have to do that run right into them. And then be very collaborative with everyone around you, you know, legal privacy, other researchers, patients, physicians, just go into with that energy of understanding that this is what it's going to take and enjoy the process, right? People get really annoyed with these types of red tape roadblocks. But often these are opportunities to meet new people to collaborate, to chat to have meaningful conversation and human connection, and to even more intimately understand the problem. And so these pieces of red tape are actually really great opportunities to improve your solution. This is what I've been saying to myself, Monty.
You sound quite convincing. Oh, man. All right. Very good. Well, one more question. If you had to pick just one, just one idea, what do you see as being the most useful application of AI?
That's interesting. Because part of the struggle with me answering this is sort of in the immediate term, in sort of the five to 10 years, or in the long term, I'm gonna say, something that's sort of been on my mind with respect to seeing a surge of youth mental health issues arise, particularly related to the pandemic, it's been a real struggle. And there are a lot of challenges to addressing that. And so if I can make AI or machine learning Do one thing. And there was an ability for machine learning to solve in a high impact at scale way, Youth Mental Health to reduce suicide, self harm and those types of issues. I would likely pick that because this is a really difficult challenge that I don't see an immediate answer to you. But I live it every day with my job as a pediatric emergency doctor. And there are definitely a lot of personal and intellectual investment going into thinking through how I could help solve this problem for that community. So I would pick that.
Well, that's a very humbling answer. And it's clear that you've got this amazing combination of being a fantastic health care provider and technologist and it's been a pleasure. Thanks so much for joining us today. And I hope that we can speak again soon.
Absolutely. Thanks for having me. Happy to come back anytime!