OR_Sempre Health_ Peter Bannister: this mp3 audio file was automatically transcribed by Sonix with the best speech-to-text algorithms. This transcript may contain errors.
Kyle Wildnauer-Haigney:
Hey everyone! Welcome back I’m your host Kyle Wildnauer-Haigney, and today I have the privilege of interviewing Professor Peter Bannister. Peter combines his role at Ada Health, where he leads collaborations with pharma, biotech, and consumer health partners to deliver AI-enabled care pathways for both rare and common diseases. With an honorary chair at the University of Birmingham Institute of Applied Health Research, affiliated with the BHP Centre for Regulatory Science and Innovation. Following academic research in medical and machine learning, he has gone on to develop, validate, and launch multiple products for diagnostics, surgery, and clinical trials. Peter also chairs the Institution of Engineering and Technology healthcare sector alongside several National Institute of Health Research funding panels and the HDR UK-Turing Welcome Health Data Science Ph.D. Leadership Program. That is a mouthful and I’m sure we’ll learn a lot about it with Peter. Finally, he is an inventor on multiple patents, has served on the Academy of Medical Sciences Council, where he was also selected as a future leader in innovation, enterprise, and research, and is co-founder of Migration Biotherapeutics, which integrates nanotechnology with cell biology to deliver new therapies for cancer. Obviously, as you can tell from his bio, he has accomplished a lot in his career and I’m very excited to welcome him to the show. Thank you, Peter, for joining.
Peter Bannister:
Pleasure to be with you today, Kyle. Thank you for inviting me.
Kyle Wildnauer-Haigney:
So maybe to kick things off, I’d love it if you could just share a little bit about yourself and what inspires your work in healthcare.
Peter Bannister:
Yeah, I’d love to. So I started off studying engineering and to be perfectly honest because as a kid I loved robots, I always find I’m a little sad when people look at Terminator 2 as an example of AI gone wrong, because actually for me, it was one of the many kind of reference points that inspired me to see what could be done with technology in a number of different ways. So I studied medical imaging, as you said, at university and got the opportunity to work embedded in a clinical neuroscience unit. So working with real patients, real data, and luckily enough to be peripherally involved, at least on some clinical studies that had real patient impact in the time I was working there. So I was very applied, and so I was very already very wedded to the application of technology in healthcare. But I think looking back, actually, what was almost more formative, or at least cemented my commitment to healthcare technology was, what I did shortly after that. I took what they sometimes call in the UK a busman’s holiday. So I took a step away from engineering in healthcare and I went to do engineering for jet engines, and what I did there was helping to develop machine learning, if you will, AI systems that could learn whether or not a jet engine was becoming sick. And the analogy there is that jet engines, like human beings, are very complicated systems, they’re very expensive systems, there are big consequences on many levels if they don’t work, even in a routine fashion. And hence there’s a lot of pressure on keeping them alive and healthy and working to their optimum potential and a bit like human beings. It may surprise you that jet engines are all very different to each other. You know, jet engines that fly in the Middle East have a very different characteristic to those that maybe fly in a tropical climate or in a Scandinavian climate. So you need to learn the individual and then you need to work out what’s normal. And at the point where it stops being normal, you need to know when to call in an expert. And that experience allowed me to apply what was very cool-sounding, cutting-edge AI technology but actually using it to try and scale up human experts. So working with Rolls-Royce, they have people who’ve been working for decades who can literally stand on a runway and hear an engine as it comes in and say, oh, okay, this part that’s maybe two inches long, it’s a little bit out of alignment, I know, I’m not exaggerating. So the challenge there was how can we take some of that human expertise and build it into boxes so that every single engine flying 24/7 can have the access to that same early warning insight? And it’s probably obvious to you from the way I’m describing it, that all of that can be transferred back into healthcare, and in fact, at the time I was working on this technology, my colleagues in the same research group were doing some of the things for patients in ICU, in intensive care. So even though it wasn’t in healthcare, it made me see how the real focus needs to be, and this is what drives me, is spending the time to understand what the problem is, what the opportunity is, and only then, proposing solutions. Because there’s always a temptation, technology is really cool. Everyone wants to work with the shiny new gadgets, and there’s always a risk that you rush in with what you think is the right technological solution because you like the technology, not because it’s necessarily a fit. So that, when I came back to healthcare after that sojourn into jet engines, I was really very focused around maximizing the amount of time I spent not building a solution, but actually understanding who needs the solution, what are their constraints, what are their expectations, and really importantly, to wrap that up, what evidence do you need to show them. So I think if I picked one word that drives me, it’s evidence. You need to prove to a doctor with different evidence that your solution works, then maybe the kind of evidence you need to show to a patient like you or me.
Kyle Wildnauer-Haigney:
Wow, I think that’s a very interesting background. I never have thought about the environmental and almost social determinants of health of jet engines and how it might apply to patients, fascinating. So maybe I’d love to hear, in your experience in what you’re doing at Ada Health today, kind of how are you positioned to support patients and then also, positioned to add value to the pharmaceutical value chain.
Peter Bannister:
So I think it’s worth looking back at where Ada started over ten years ago and Ada was founded by clinicians, despite the fact we have some amazing cutting-edge artificial intelligence technology, which is truly differentiating, I think what really sets us apart is the fact that we’re fundamentally built around strong established clinical evidence standards and clinical expertise. So what Ada set out to do was to provide an alternative to people like you or me, who often, when feeling unwell, will go and Google their symptoms. We know that even today, 80% plus of people will go and Google their symptoms, but what generally they’ll find when they search for their symptoms is the worst possible explanation for those symptoms, because to be fair to them, Google is not designed to be a diagnostic tool. So the whole motivation behind Ada’s technology, our symptom assessment platform, was to create a technology that could be easily put into the hands of a user like you or me because at Ada, we actually think of people before they come patients as users. The technology that could be used in the form of an app or through other common interfaces that would allow people to report the symptoms that they already feeling, and on that basis, follow through and have a question flow in exactly the same way a healthcare professional would seek to explore your symptoms, to come up with a set of suggested conditions that you may be suffering from. So to do that, Ada relies on two core components. One is the artificial intelligence component, the natural language processing, so sometimes referred to as a chatbot technology. So the way in which Ada asked questions using commonly understandable language takes into account your answers up to that point and uses that to ask intelligent next questions. So the question flow for each person using Ada will differ based on what they’ve said up to that point in the assessment. And what that results in is a very thorough what we call a pre-diagnosis, but an assessment of conditions which results in a ranked list to the user of what the most likely conditions you may be suffering from, and then what the severity is and what you can do about it. So for example, if it says, if it concludes that you’re most likely be suffering from something that demands emergency care, it will clearly indicate to you that you should go and seek emergency care, but we indicate the urgency as well as the appropriate next steps. So there’s the one component is the natural language processing, the human speaking interface, which is really important for engaging the user and making sure that you get the right information out of them at an accurate level to allow you to make that condition suggestion. The other component is what we call our medical knowledge engine. So that’s the underlying database, if you will, of all of the conditions that Ada is able to help surface. So today, Ada covers over 10,000 unique symptoms and links those to over three-and-a-half thousand unique conditions, which is able to accurately surface and suggest as a pre-diagnosis. And the way that we build that medical knowledge engine is, it’s not one of these black-box approaches that you sometimes hear about where people chuck data into an algorithm, and when the algorithm stops, they go, ah, that must be the right answer, let’s use that one. Which creates lots of questions about how do you evidence that was the right answer and how to the regulator, to the clinician, to the public, to the payers. What we actually do is we take a very rigorous approach using medical knowledge to, so we have a team of sixty MDs in Ada who are medical knowledge engineers and their job is to actually build that medical knowledge engine. So they continuously review peer-reviewed evidence, best practice from our clinical advisors, some of whom come via our engagements with our global life sciences partners. And they can, in a very agile manner in the software sense, update that engine, that medical knowledge base so that it can take into account the best peer-reviewed evidence knowledge about how to handle these symptoms and relate them to conditions. I think that really is what sets us apart. Alter that with the aim of helping users, patients, if you will, understand their symptoms and take the appropriate next step, know where to go to next.
Kyle Wildnauer-Haigney:
And so tell me about your business model and how the business gets in the hands of patients. Is it through partnerships? Large healthcare providers? Is it kind of in the payer approach? What’s kind of like the go-to-market and your other stakeholders that you engage with?
Peter Bannister:
So, so we have broadly two revenue streams, two channels. One is what we call consumer, which is the app, which anyone can download for free and use Ada to the fullest of its ability in terms of understanding your symptoms and suggesting conditions. And to date, there are over 12 million people globally who use that app, and we have over 25 million assessments. We are very proud of the fact that there are a quarter of a million 5-star app store reviews. And one of the things that I would say the real perk about working at Ada is actually we have a real-time feed which within the company which has all of those App Store reviews and look, they’re mostly 5-star, but what’s really good about them is the fact that you have, you hear stories from people saying, you know, for 17 years I’ve gone to doctors with these unexplained symptoms and Ada helped me understand my conditions and go and talk to the right expert, and now I’m actually on a course of treatment and so on and so forth. So it’s not replacing the doctor in any way, but it’s helping close the gap between uncertainty, but awareness of symptoms and actually knowing how to take action. So that’s the consumer side and available in 190 countries through the various app stores. The other part of it is, if you would be to be our enterprise offering, so we partner not only with life sciences companies, which is my responsibility in the business. So that’s pharma companies, consumer health companies, but we also partner with providers and payers. So a great example of that is Sutter Health in California. They use Ada’s technology, that symptom assessment platform is a digital front door for all of their patients. So if you’re at Sutter, what you’ll actually do is you go to the website. You’ll carry out the same kind of assessment you’re able to do in the app, but what we then can do is we can integrate that into the care that’s available to, in this case, Sutter patients or another provider’s patients. So you can actually then go directly from the end of an assessment to a telemedicine consultation or an online pharmacist or a find-a-physician service, or whatever range of even lower acuity solutions, depending on the severity and the most appropriate next steps. So we can then, and then we can go on and, in those contexts, then integrate all of that information, including the assessment report, which is designed, as a version it’s designed for the individual to understand, but also a version that’s designed for the healthcare professional to understand, including things like differential diagnoses that were ruled out during the assessment that could all then be passed into an electronic patient record system. So it becomes part of the longitudinal permanent patient record. So those patient-reported outcomes can be part of the ongoing care and support of the patient.
Kyle Wildnauer-Haigney:
You know, I have to say, I love this. I love this business model and approach. There’s two things that kind of jump out to me just hearing you talk about this. The first is, you know, I think we’ve all been there or have had friends, family members who have gone on WebMD, and it seems like everything that you Google your symptoms ends up in cancer or some terminal illness. And with Ada, it’s a more intelligent approach, right? It’s something that isn’t just a basic search criterion that is probably most likely designed for engagement entertainment because of ad revenue, but it’s actually designed to understand the clinical diagnosis. Is that is that fair to say?
Peter Bannister:
Absolutely, I would say intelligent is synonymous with clinical. It’s a clinical approach to doing that, but you’re absolutely right.
Kyle Wildnauer-Haigney:
And then the second thing too, I mean, I experienced this. I live down the street from Sutter Health, will have to go check them out. But so often when working with plans and providers, the onboarding process is extensive and you really just want to get to see the doctor. And it sounds like what Ada is doing is accelerating that doctor-patient relationship. So getting through all of the front-end process, triaging you to the correct provider, to the correct form of treatment, and then allowing the patient to actually interact with a human being to get that support and care that they need. While the front-end paperwork and triaging complex process is taken care of by this technology, is that fair to say as well?
Peter Bannister:
Yeah, I think that’s and that’s correct, and I think it’s as relevant when we work with life sciences, with pharma, or consumer health companies, as it is when we work with care providers of different kinds. Because in the life sciences sense, as I’m sure you and the listeners are aware, the role of pharma is to educate the healthcare professional about the indications and the inclusion criteria for best treatment for particular conditions. They are not in the business of engaging directly with patients when you talk about Rx. So the linchpin there, whether it’s a provider setting or a life sciences setting, is the same, it’s the healthcare professional, the physician. So whether you’re talking about is quite correctly, say, moving quickly through an integrated provider environment like Sutter to not over-alerting. Everyone ends up in A&E, that’s an economic problem and a bandwidth problem but also directly to people to the right service or whether you’re talking about making sure that someone with a set of symptoms that primary care physician maybe doesn’t understand, like a rare disease, has the opportunity to quickly move through to the right kind of prescribing physician. It’s exactly that use case. And I think what that brings with you is a challenge, which I think we’re very adept at solving, which is getting the right balance between trustable impartiality, but actually giving people the ability to take action. So we have an overall mission of achieving the improved health for a billion people with trusted medical guidance. And that may sound lofty, but it’s actually not as far off as it may sound on first hearing. And to do that, we need to actually look at a range of settings. So one of the ones that I didn’t mention when I was talking about our technology was actually Ada has a very explicit focus on low and middle-income countries as well. So I talked a few times, I know when I was describing the chatbot about talking in a natural language, but we not only talk in a natural, non-technical, non-clinical language to the user, but we also do that in multiple languages themselves. So Ada is available in currently 11 languages and that number is growing pretty rapidly that includes European languages, but it also includes Swahili and Arabic. And we’ve done a lot of work with NGOs, for example, in South Africa and Romania, making this technology available to help connect individuals who, for both awareness but also socioeconomic reasons, are not connected up to care that’s already available to them today. So I think that’s also a really, that is really fundamental to Ada’s DNA because something, a topic that I’m happy that people are much more conscious of certainly over the last couple of years is health inequality and specifically digital health inequality, going back to my kind of introduction about the difference between being, if you will, seduced by the technology versus being seduced by solving a real problem and seeing your evidence. There is, if I use, maybe misused the phrase, this is unconscious bias. If you jump in if you’re starting point is I’m going to build an app that improves people with this kind of condition, it doesn’t matter how much user empathy you do beyond that point, you’ve already excluded everyone who hasn’t got a smartphone and you maybe even find in some countries they don’t, they only have SMS and it may go further back, but we’re talking within the digital spectrum here. So doing that work proactively with partners has allowed us to understand how we can actively and consciously build technology, which doesn’t just make things better for people that have pretty good access to healthcare in a relative sense, but actually closes that gap between people who really have very poor access to healthcare, where maybe a few hundred miles, if not in a different country, there is a solution available to them for their condition.
Kyle Wildnauer-Haigney:
Yeah, that resonates with me in my experience. You know, I have extensive previous experience with diabetes population in the US and kind of in digital health wave 1.0, 2.0, wherever you want to call it, back in 2011, you see all these apps come out, but none of our patients had smartphones at the time. And so this app, even though it was potentially great, was not really an effective tool for the specific problem and for the patient. And so I really think what you just said about designing with the end-user in mind and really understanding their socioeconomic contexts is critical to developing a solution that actually moves the needle, and I think that’s just fantastic. One thing I’d love to get your sense on is how, what are some of the improved outcomes that you guys have tracked and reported on thus far? I mean, it sounds like you’re doing some incredibly important, impactful work. What are some of the studies that you really showcase when you’re talking about the impact of Ada to external parties?
Peter Bannister:
So there’s, as you probably gathered, there’s a lot of different ways that you can apply this symptom assessment platform technology. So the approach that we’ve taken has been quite use-case dependent. So a couple of great examples that I can think of are, I mean, in terms of health economic benefits, we did some work, but in fact, Stanford did some work with Sutter to look at the impact of Ada being used as a digital front door solution, and that was published a couple of months ago, and what they were able to show was that twofold benefit. Number one, that people got to the right level of care option much quicker and without Ada. So again, this is not the I got sent to A&E and then I realized, hang on, I just needed to go and see a physiotherapist so there was that efficiency saving, but what was really interesting was actually after that, people then spent a lot more time engaged with the other services that were available within that healthcare ecosystem that they would not have known about before. So a really tangible improvement in awareness of the different range of options, and as we all know, you know, you don’t have to go to an HDP all the time, there’s, one of the areas we work in actively, for example, with Bayer Consumer Health is what we call self-care, where the most appropriate, medically mandated course of treatment based on your symptoms, maybe go and buy something over the counter that’s not prescription based. So there’s some really interesting data around that that was conducted in a very rigorous way to look at the overall benefits where you take advantage of that very integrated Ada solution. I guess a couple of other great examples are in terms of the overall symptom assessment accuracy, we had a number of studies published, a key one in the British Medical Journal a couple of years ago comparing the performance of Ada as a symptom assessment tool to a number of other symptom assessment platforms, but crucially, benchmarking that against a group of very well qualified general practitioners, primary care physicians in the UK and the anecdotal summary was, for common conditions, good general practitioner is best. Ada is a very close second, and quite honestly, and you can look at the numbers they were they’re published in the journal, everything else is a long way back. But the best combination, particularly when you start to look at less common conditions, which is where we really add value, is general practitioner plus Ada. So the idea that you do an assessment, as I mentioned before, we want to make sure that when you come to your doctor with the results of your Ada assessment, that they don’t treat it in the same way as they will treat your Google search results. They go, okay, I get who Ada are, that they are built on clinical rigor, and I will trust the information that you already trust to factor that into my ongoing assessment, diagnosis, prescription, etc. as appropriate. So we, where you talk about rare diseases, we really add a lot of value because there are well, until a couple of weeks ago, there were 7000 rare diseases, but a week or two ago there was a paper published that said actually in the US they think globally there are now more than 10,000 rare diseases. And the statistics would tell you that even though they’re rare, they affect a lot of people either directly or indirectly. And the challenge is you can go to medical school for a very long time, and you’re never going to be able to, the best clinician in the world is not going to be able to diagnose. So there needs to be a system where we can effectively flag without over-concerning people that there’s a strong possibility of a rare disease and then show them how to move through to potentially a primary care physician who may have to refer them to a specialist or directly to a specialist, depending on the healthcare setting and the condition. So rare diseases are a very, very powerful use case for what we do, and linked to that have a number of papers. All of our papers are deliberately open-source. They’re available on our website under Ada.com/studies showing how we are able to, in many cases, cut out many years of uncertainty for people who ultimately are diagnosed for rare diseases, and not perform the diagnosis, but you can go straight from an Ada assessment in many cases to a specialist and have your diagnosis confirmed. Whereas unfortunately, we know for many people with rare diseases, despite the fact they’re very different and diffuse on average, you’re talking to 7 to 10 years, assuming it’s ever diagnosed. So that’s a very powerful use case with data to support it.
Kyle Wildnauer-Haigney:
Yeah, I mean, that’s incredible, the impact and the promise, right, of that man with machine combination and how it can really transform the care that’s delivered. I think there’s so much optimism and really data now that supports that. You know, maybe just kind of before we conclude, if you could just share with the audience kind of a closing thought and where listeners can collaborate with you.
Peter Bannister:
So, I may have to double back on this one a bit, but the, so what I would say is I’ve talked about rare diseases, but rare disease is in some ways the tip of the iceberg because there’s a whole category of misdiagnosed or overlooked conditions. So things like endometriosis, for example, or chronic cough. So just because they don’t have the tag of being rare, they nonetheless fulfill a lot of the same characteristics in that they are hard to diagnose and actually affect a large number of people. And I think for various reasons they don’t get as much attention and people working on solutions for them, as either more prevalent but well-known diseases, or on the other end of the spectrum, rare disease as well, although they’re not very common, the stories you hear from them are so heart-wrenching that you can’t help but sit up and go, we need to do something about this. But this middle ground of overlooked and misdiagnosed conditions is a really important one, and relate translates, as I mentioned earlier, to a lot of countries outside the West, lower-middle-income countries where actually they may not be overlooked or misdiagnosed in more affluent systems, but as you move away from richly resourced healthcare systems, they become, fall into that category. So I think we are actively collaborating with clinical partners, life sciences partners, healthcare provider partners who have an active interest in connecting people with care for those kind of conditions. And anyone listening to this who’s working in one of those areas, as well as some of the other use cases I’ve already talked about, we’re very enthusiastic about partnering and helping build more complete healthcare journeys so that we can move people whole steps from, hey, I now understand my condition a lot better because of Ada, what do I do next? Help me take those next steps and get me to that over-the-counter treatment or that healthcare professional. So much as we are about informing people better, what we don’t want people to do is come away from an Ada assessment, better informed, but then straight back on their web browser going, hey, I think I’ve got the disease now, what do I do about it? We are already completing the journeys for a lot of our users, but there’s a lot of work to be done to get to that 1 billion healthy, and to do that we need to partner with people who can provide those next steps and can help us understand those populations and how to access them and move them forwards as efficiently as possible.
Kyle Wildnauer-Haigney:
Well, Peter, thank you so much for joining today. Everyone listening out there, if you want to help support a billion patients and interactions to improve and to support the developing world, quite honestly, reach out to Peter and Ada Health. Thank you so much, Peter. It’s been a pleasure speaking with you.
Peter Bannister:
Likewise, Kyle. Thank you so much for the invitation. It’s been great to talk to you today.
Sonix has many features that you’d love including upload many different filetypes, advanced search, automated translation, automated subtitles, and easily transcribe your Zoom meetings. Try Sonix for free today.
