- il y a 2 semaines
The Intelligent Ally: How is AI Reshaping Medical Practice?
Catégorie
🤖
TechnologieTranscription
00:03Sous-titrage Société Radio-Canada
00:30I'm sure over the next few days there's going to be a huge amount of discussion about artificial intelligence and
00:36how it's being applied.
00:38One area where we're already seeing multiple applications across the industry is healthcare.
00:46So today we have a group with us here to really discuss how AI is being deployed in healthcare and
00:54how it's reshaping medical practice.
00:57So I'm delighted to welcome you all. If you don't mind, please starting by introducing yourself. Jean-Claude, maybe you
01:05could start if that's okay.
01:06Sure. Good morning, everyone. I'm Jean-Claude Sagbini. I'm President and Chief Technology Officer for Lumeris.
01:13A bit of background on Lumeris. We're about 1,000 employees fully dedicated to transforming primary care.
01:24We've been at it for the past 15 years.
01:28We partner with health systems, bringing in technology and know-how to transform primary care.
01:36Our market is in the U.S. And our flagship product is Tom. Tom is two things.
01:44It's an agentic AI platform, end-to-end from data to patient interactions and physician interactions.
01:54Tom is also a primary care team member that supports primary care teams on that journey.
02:00So excited to be here. This should be a fun conversation.
02:04Great. Thank you. Alex.
02:06Bonjour. Good morning. Hi, my name is Alex. I'm the President of Tencent Healthcare.
02:11And Tencent is a very, very large company. So we're the healthcare arm of this.
02:15And we provide, one, consumer-facing healthcare services through our WeChat platform.
02:22The second part is we provide AI and cloud solutions to hospitals, pharmacies, life sciences researchers, and pharmaceutical companies.
02:31I think third, we actually have a dedicated AI for life sciences lab, where we actually do some of the
02:37frontier genomic research, drug discovery research using AI.
02:41But I think fourth, and probably our most important arm, is that we have a lab that focuses on reducing
02:47health inequality.
02:49And we call it AI for inclusive health lab, where we actually combine all of the technologies and solutions that
02:54I just talked about,
02:56specifically addressing the issue of narrowing the health inequality between populations.
03:02Great. Thank you, Alex. Clara.
03:04Good morning, everyone. So I'm Clara Léonard. So Product Medical Director at Doctolib.
03:10And Doctolib, we're basically developing solutions for healthcare professionals on one side and for patients on the other side.
03:18With basically two missions, everything that can actually improve the daily lives of healthcare professionals.
03:24So helping provide them better care, help them reduce the workload and improve the comfort at work, increase their activity.
03:32And on the patient side, improve access to care. Maybe some of you have been user of Doctolib.
03:37And also manage the relationship with their professionals.
03:41Today we're being used by 400,000 healthcare professionals in Europe and 80 million patients.
03:47And now it's about what we can bring with AI to all of these users.
03:51Great. Thank you. And Chez.
03:53Good morning. My name is Chez Partovi. I'm the Chief Innovation Officer at Royal Philips.
03:59A physician by training. I practiced for about 12 years and joined Philips about five years ago.
04:04And just in case you don't know Philips, Philips is a health tech, med tech company.
04:09I know that you may know it for its other things, but we divested of all the other technologies that
04:15we used to do, like TVs and so on.
04:17We licensed the brand. We don't make light bulbs.
04:18Philips is purely a health tech company now, an MRCT, angio-ultrasound, and so on.
04:24So it's a pleasure to be here. Look forward to this terrific discussion on transforming healthcare.
04:29Awesome. So great to have you all here.
04:30So I think just to kind of get us warmed up, there's so much hype really now about artificial intelligence.
04:38I think probably like 90% of the conversations that people are going to be having here at VivaTick will
04:42involve it in some way.
04:44Clearly, AI is being used in healthcare and all kinds of like very, very clear use cases at the moment.
04:50So I'd just like to just level set as a first question and ask each of you, like, where are
04:57you seeing the most impactful gains from AI today?
05:00Is it diagnostics? Is it workflows? Or is it patient engagements?
05:04And maybe we could start with you, please, Jean-Claude.
05:08Like, where are you seeing the maximum gains at the moment?
05:11Yeah, look, the maximum gain is where we have the largest problems.
05:17And the largest problems are access.
05:20We're having conversations on both in the United States, where our market is, but also globally.
05:29Most humans on the planet do not have access either to proper care.
05:34Some of them don't have access to any care at all.
05:37So the biggest impact we're seeing are applications that are reaching the patients, engaging with the patients,
05:45helping them navigate their care, and then either providing them, you know, with care that is not being provided to
05:53them
05:53or plugging the gaps in care that they have today.
05:57Okay, great.
05:58Alex, what about you?
05:59Where are you seeing the most impactful gains?
06:00I think in addition to what Jean-Claude has mentioned, I think in the area where AI,
06:05we have seen from a productivity and quality improvement for physician care.
06:10Now, at the end of the day, I think the physicians do have to be responsible for their diagnosis.
06:15But I think now they are a lot more confident in what the latest generation of AI can actually help
06:20them with.
06:21And I think it's also the hype has died down, and now they are a bit more realistic of what
06:26AI can and cannot do.
06:28And I think that's a big difference compared to, let's say, five years ago.
06:31And you mentioned that you are doing research with life sciences.
06:35Are we seeing real gains there in terms of, like, fundamental scientific research as well?
06:39Are you excited about that?
06:41Well, I think absolutely.
06:42I think last year's Nobel Prize speaks for itself.
06:45And I think we are only barely scratching the surface from a point of view of having very, very personalized
06:51science and care.
06:53Yeah.
06:53From that point of view, and I think just like genomics, I think just the size and breadth of the
06:58data is way too much for traditional bioinformatics to actually analyze.
07:03And I think this is where we have yet to see, actually, the leap that we are seeing in the
07:07general AI outside and the hype on the part of life sciences.
07:12I think the golden year is actually the next five to ten years to come.
07:15That's exciting.
07:16And Clara, from your perspective, are the maximum gains happening in kind of workflows, patient engagements?
07:22Well, I think for sure, number one objective is that healthcare professionals spend zero time on admin or financial tasks.
07:30So, objective one.
07:32And then objective two is, of course, to transform the way care is being delivered.
07:37There was a quote, I think, from the French Academy of Medicine last year that said that it would actually
07:44be a lost opportunity
07:45for patients if we did not leverage AI for care delivery.
07:49So, I think it's a bit of a moral obligation for all of us to invest in AI to make
07:54better clinical decisions
07:55and transform the way care is being delivered.
07:59Because if you think about it, doctors today, they're a bit data archaeologists, right?
08:05So, they need to dig into the data, into their file.
08:08They need to mobilize what is the right medical knowledge, knowing that medical knowledge is doubling every five years.
08:14So, being able to actually mobilize all this data and take the good clinical decision,
08:20this is something where AI can actually bring a lot of value, in particular now,
08:26because I think there have been, like, AI has been around in healthcare for a long time,
08:29in particular in diagnostics or imagery.
08:31But with all the progresses that have been done recently, in particular around language,
08:37so natural language processing, large language model, data and knowledge today in healthcare
08:42is being transmitted via language, oral language or written language.
08:46And I think all of these progresses now open a new chapter for innovation in AI
08:50that will then really support professionals to deliver actual better clinical decisions.
08:56Great, thanks.
08:57Shez, where are you seeing the maximum gains?
08:59You know, if you want to look at that, simply look at the specialty in medicine
09:03that has been engaged in AI the longest.
09:05It's probably radiology.
09:07Yeah.
09:07Radiology has been at the forefront of many things.
09:09Yeah.
09:10And because of the advent of Pax many years ago and the digitization of radiology,
09:13they've also been leaders.
09:14Yes.
09:14In advanced applications of AI.
09:18And in that space, you can clearly see how AI has gone past the hype
09:22and is delivering better care for more people.
09:24And as an example, in an advanced visualization workstation that we have at Philips,
09:30when physicians are looking at CTs of the chest looking for lung cancer,
09:35there's a clear demonstration that using AI you can detect the anomalies 26% faster
09:41and you can also identify up to 29% lesions that wouldn't even be seen by, if you will,
09:49looking at it without AI.
09:50So you can see that there is actual bonafide value in using AI
09:55and the actual applications are there here today.
09:57It's not simply hype.
09:59And radiology is a great place to look.
10:01For sure, there's other places as well.
10:03Now it's permeating everywhere.
10:04But in radiology, you can see that for sure it's having positive impact
10:07for delivering better care for more people.
10:09Yeah.
10:10Yeah, no, absolutely.
10:10And there's that great example at Moorfields Eye Hospital in London
10:14where I think now they don't actually involve human beings in any of the X-ray sort of analysis.
10:20It's all done by work that was done by Google DeepMind and a professor called Pierce Keane.
10:25So we talked a little bit about access earlier on,
10:28and maybe Alex, I can come to you to discuss this.
10:33Tencent has an integrated healthcare and AI.
10:37It's pretty much everyday platforms.
10:40like WeChat, which is obviously like the everything app in China.
10:44What do you think you've learned from that?
10:47What are the insights from that, particularly around sort of like building trust with patients
10:52and also delivering healthcare to sort of like underserved markets and underserved areas?
10:58So I think I myself am very lucky, right, to be working within the WeChat platform because I think trust
11:08is gained not just because you have a cool technology that solves a problem that they want.
11:13It's actually built up through time, and I think a lot of our users have a lot of trust in
11:19WeChat on how we manage the data, on what they can do on it.
11:23And I think that's number one, that it's the continual interaction that makes them believe that you are actually doing
11:30the right thing for them.
11:31But I think secondly is, in order for us to deliver care to, let's say, 1.4 billion people equitably,
11:40we actually have to build a lot of the unsexy infrastructure.
11:44Now, let me give you two examples.
11:46One infrastructure that we have to build is China runs on the national health insurance.
11:51Now, if all of the transactions online, whether it's actually teleconsult or a prescription, if that payment system is not
12:01integrated and you have to self-pay and then get reimbursed, that is a barrier.
12:05It's not a barrier of access, it's a barrier to payment and reimbursement.
12:07So what we have built up is that it's basically online real-time health insurance reimbursement with the national insurance
12:14system so that you can get the same online health services and pay for it at the same time.
12:20I think that's one.
12:20And the other one is healthcare can be very fragmented for users in China because we don't have a very
12:28– the primary healthcare system is still developing.
12:31And so what ends up being patients are being their own GP, finding different specialist doctors for their own problem.
12:38And then with it comes with a basically data silo.
12:42And what we are trying to do is actually also build one healthcare ID that institutions across China can start
12:49using and using that ID to actually make sure that the health record and everything can be matched along the
12:55way.
12:56I mean, those are, I think, very, very unsexy non-AI, but I think those are actually critical digital infrastructure
13:02that one has built in order for AI and other digital healthcare services to thrive.
13:08Okay, great.
13:09Sean-Claude, I'd love to bring you in at this point.
13:12The idea of, like, agentic AI seems to be the kind of – the area that we've settled on is,
13:17like, the kind of, like, the real area where we're seeing a lot of conversations in a lot of organizations.
13:23And you advocate for this in order to address primary care shortages.
13:28Can you talk a little bit about how that model works and really what makes it scalable in the way
13:35that we, you know, Alex and the way that Tencent is working with WeChat, which is, you know, a vast
13:41platform with a lot of scale?
13:43Yeah.
13:44You know, what you were saying, Alex, was highly resonating with me because we also see agentic AI being deployed
13:53but with a platform approach.
13:55You need to start with the unsexy basics of data.
14:00And we actually put lots of AI R&D in making sense of the data.
14:06Data comes from a variety of forms, electronic health records, insurance data, social determinants of health data, etc.
14:13And starts with a platform to make sense of the data.
14:17Then it goes into creating insights from that data.
14:21Up until now, physicians, doctors were always complaining because the only insights they were getting is what they were doing
14:28wrong.
14:29It's like, hey, I could determine something you're doing wrong.
14:31Do something about it.
14:33That was not scalable.
14:34But nonetheless, we have to extract these insights.
14:37And then there's an agentic set of services that are now, as of the past two or three years, able
14:44to take action.
14:45And this is the exciting part.
14:46So you have to invest in the infrastructure, invest in the insights.
14:51But now we're able to develop and layer that can take the action, take the action on behalf of the
14:57physician, engage with the patient, engage with the physician, the entire care team to deliver the outcome that's needed.
15:05And Clara, Dr. Lib has recently launched an AI tool or an AI product.
15:13You've been pioneering, obviously, in digital health infrastructure in Europe.
15:18How are you embedding AI into your tools for doctors and also for patients?
15:25And maybe I'll come back to the product that are existing because we so on the health care professional side,
15:34we have a wide range of solutions that we offer to to health care professional ranging from productivity tools.
15:42Right. To manage the practice to patient relationship management as well.
15:47So online booking, video consultation, messaging, preventive care as well.
15:53Care cooperation as well.
15:55So we're the largest care cooperation network with instant messaging, referral, being able to exchange about a patient case on
16:03the platform and clinical and financial software.
16:06So a full actually operating system basically for health care professional and on the patient side products to access care.
16:13So online booking, I think this is what most of you might know us for.
16:16So online booking, search online, manage the relationship with a health care professional and manage their own health or the
16:24one of their relatives.
16:25Now, I will basically we will invest a lot in AI and use it in two ways.
16:30One, we will integrate it in all of those different products to make them more intuitive and more contextualized.
16:38So think of a clinical software, so electronic health record for a health care professional.
16:43There is no real reason that it's the same interface for a gynecologist and for a physiotherapist because just the
16:50workflow of a consultation is different.
16:52So adapting it by specialty makes a lot of sense.
16:55But you can go even one step further if you're thinking of a gynecologist, a patient that comes for a
17:00pregnancy follow up or a patient that comes for a contraception consultation.
17:04Same thing.
17:05You would not want to actually surface the same data in order for the practitioner to run the consultation.
17:11And as the consultation go, then you might want to actually trigger some additional data points to the health care
17:18professional.
17:18So how to actually adapt those tools to surface the right information to the practitioner at the moment where he
17:25needs the most.
17:26So that's one example on how to use AI to actually make those existing solutions just more intuitive and personalized
17:33and adapted to a context.
17:35And on the other hand, we're developing full, I would say, AI native product.
17:41So actually, we did not launch one, we launched five in the past 12 months.
17:45So we launched one consultation assistant, which is able to automatically document the consultation by voice, right?
17:52So just by understanding the consultation with the patient, transcribing and summarizing that in the in the electronic health record
18:01of the professional.
18:02We launched a patient data codification tool.
18:06As we were saying, a lot of the data is in free text today, so they cannot really act upon
18:11it.
18:11So we developed a solution to actually recognize medical concept and structure all of this information in the doctor software.
18:19We launched a phone assistant.
18:21So it's basically an assistant that's able to understand patient requests that are calling the practice and filter and generate,
18:30I would say, a asynchronous response.
18:32We launched a request for the secretariat.
18:34We launched a billing assistant, even to also suggest some billing codes to the professionals.
18:39So wide varieties of tools.
18:42And we're now, I think, moving to the phase where we are working on more clinical AI.
18:47So how can we support the practitioner in the clinical interview?
18:51What are the good questions to ask based on what the patient has just said?
18:56What are the potential risks that the patient has?
18:59So a bit of a prevention assistant and also a care plan generation assistant.
19:05So based on everything that has been said in the consultation, what does the state of the art medical knowledge
19:11basically recommends for this given patient in terms of exams, in terms of prescription.
19:16So actually go into the clinical recommendation space.
19:19So that's something we're working on.
19:21Okay, great.
19:22Thank you.
19:22Can I share this, please?
19:23So it's interesting.
19:24So anyone in the audience here trying to make AI for healthcare?
19:27Okay.
19:28So I have a sound bite for you.
19:31Trust is traction.
19:33Trust is traction.
19:36Philips did a survey of, what, 16,000 patients and 2,000 clinicians to understand what led to AI adoption.
19:45And 79% of physicians and nurses said that they're very optimistic, but what they worry about is bias in
19:54the data and liabilities of using it.
19:57So trusting that the data is the data pipelines gives actually represents the patients that are taken care of and
20:04understanding the liability.
20:05So trust in that.
20:07Then you look at the patients and it's interesting 59% of them were only bullish on AI.
20:15So 20% trust gap between clinicians and patients.
20:20And then you ask the patients, well, what would make you trust using AI?
20:24And they said, look, I don't mind using it for scheduling.
20:28That's fine.
20:29Check in.
20:30That's fine.
20:31But if you're going to care for me, I'm not sure I trust it.
20:35I'm not sure I trust its privacy.
20:38I'm not sure I trust what it's going to do.
20:40And then when you ask them, what would make you trust it?
20:43Yeah.
20:44They say, add the clinician in the loop and their trust went up to 86%.
20:48So you have basically, you want to build AI in healthcare.
20:52You begin with the question of how do I earn the trust of the clinician?
20:57How do I earn the trust of the patient?
20:59And your AI will fly off the shelf.
21:03I'd like to ask you also, Chez, about access.
21:06You know, Alex was talking about sort of like ensuring that there's sort of like greater ability for people,
21:12particularly in rural areas, to have access to healthcare.
21:15How are Philip's thinking about that?
21:16Yeah, there's a great example.
21:19I have often talked about this if you haven't heard this example.
21:23It is my favorite example of improving access in underserved communities with AI.
21:29So in the first trimester of pregnancy, a woman that's pregnant should have at least one ultrasound
21:35to check for the health of the baby.
21:37It's the World Health Organization's recommendation in the first 24 weeks.
21:42And of course, getting that sonogram, getting that ultrasound requires not only the equipment,
21:47but a sonographer who needs to have training, know what to look for, do the right kind of scan.
21:52And many parts of the world that in fact, even in developed nations, there are maternity deserts where there is
21:58not availability.
21:59So, Philips has a handheld ultrasound called Lumify and working with the Gates Foundation,
22:04we worked on an algorithm and built an algorithm where, with about 15-20 minutes of training, an extender
22:11simply swipes the belly of the pregnant woman three times down, three times across.
22:18No training needed beyond that.
22:20And the AI algorithm then, there are about five measurements that determine how a baby is doing.
22:25It's like, it's called the length of the baby, the heart rate, the position, where the placenta is, five parameters.
22:31And the algorithm automatically determines those five parameters.
22:34It doesn't even give an image.
22:35It just goes three sweeps down, three sweeps across, you get five numbers.
22:41With 15-20 minutes of training anywhere in the world.
22:45And then you know, mother can go home, baby's fine, numbers aren't right, you should have a follow-up.
22:51Okay.
22:52And so that's an example of AI improving access to care in underserved communities.
22:56And it's probably the best example I know of this, bridging the divide using AI.
23:01And it's a beautiful example.
23:02It's actually being used in different parts of the world.
23:04And Jean-Claude, I'd love to come back to you.
23:06Just building on what Chez was saying about trust is traction, you have noted the unpredictability of AI systems in
23:13the past.
23:14How do you think we can create the governance frameworks that balance innovation, excuse me, balance innovation, but also mitigate
23:22risk?
23:23Yeah.
23:25So governance is important, but I want to go just for a second related to that, Chez, to your point
23:30on trust.
23:34When we, with Tom, our product, Tom is a, we made Tom be a primary care team member.
23:40And we gave doctors control over what services to let Tom do.
23:47So just like you have a team as a doctor and you say, I'm going to let you take blood,
23:52but not do a consult.
23:53Same thing with Tom, you start with, I'm going to let you do preventive services, scheduling reminders.
24:01I will now, now I trust you more.
24:03I will let you do chronic condition management, check-ins, maybe only for hypertension initially, maybe goes for CHF, maybe
24:10expand for COPD, full control with the physician.
24:14So to the point of governance, it's both governance, but also what deployment model do you do to allow your
24:24customers, the health system to build their own governance in their own organization, right?
24:29So, so then if you put yourself in their mindset, their governance is, we are going to start deploying early
24:37services.
24:39We are going to report to our governance body at a health system.
24:42What are the results?
24:44Those are low risk capabilities.
24:46And then from there on, they will get internal approvals to allow Tom, in our case, to expand the capabilities
24:52that Tom can do.
24:55That's our, that's our, that's our approach.
24:58I'd love to get others, the thoughts of others, and maybe Alex and Clara, we can move on to you.
25:03Like in terms of governance, how are you thinking about it?
25:05Trust and governance, those kinds of challenges.
25:08Maybe Alex first.
25:10I think we have sort of two sets of governance questions.
25:14One is specifically around the AI.
25:17Yeah.
25:17And the other one I think is so healthcare specific.
25:19I think the data and the scope of the regulation, because right now there's already a very, very well run,
25:27whether it's the FDA or the NMPA in China, EMA, right?
25:30On how to actually manage AI as a medical device.
25:33Now, but I have to say that I think that has to be evolved because I think that, that constraints
25:39a lot of,
25:40I will not say bad limitation, but I sometimes think a necessary limitation to the real life situation of how
25:48AI can actually be deployed.
25:51For example, for what Jean-Claude was talking about, it's really about, people use the word agentic.
25:57Agentic to me, the meaning is that it's basically closed loop, fully functional.
26:02They can execute.
26:02But I think we are at a point right now in healthcare, we're not right there yet.
26:07Because I think it's not just about the trust in the actual AI.
26:11I think it's a societal trust in what is the role of the healthcare system and the healthcare system plus
26:17AI.
26:18I think people are, we are still struggling.
26:20I think all of us are facing the same sort of headwind, right?
26:23We want to do more and more and more.
26:24But yet there's actually a lot of resistance coming from the health system because of the very issue that we
26:31discussed.
26:32So I think the idea is for the governance is how can we scope it so that with essentially low
26:39risk, medium risk,
26:40then the high risk, which is then about automatic autonomous diagnosis based on radiology, right?
26:49Is this what we wanted to go to?
26:51Because from a health source, human resources planning and health system building perspective,
26:57I think all of us in the industry know that that has to be the way to go.
27:01But I just don't think that right now there's enough of a conversation, right?
27:06Of what that will look like and how we should be able to govern and regulate and manage that when
27:11the day comes.
27:12Clara, any thoughts on that governance question?
27:16No, I'm aligned with what has been said.
27:18And I think on the regulation points, I think evaluating in Europe, yes, there is some stringent regulation.
27:27We see that more as an opportunity, also as an opportunity to actually build the best AI,
27:33because it also forces you to ask yourself the question on how to actually create good quality AI that actually
27:40meets the standards.
27:41So either privacy or quality standards that is required in healthcare.
27:45Yeah.
27:46Because at the end of the day, this is what does build trust with healthcare professionals.
27:50So we see that more as a framework in which we operate that give us guideline into what to look
27:56for and how to be even better into the AI we develop.
27:59So I think, yes, it can be cumbersome at times, but we see it more as an opportunity to leverage
28:05and align with you.
28:07I think at the end of the day, we would need to think of health system plus AI together.
28:11What are the benefits of that?
28:13And potentially, what are the reimbursement pathways for AI solutions as well?
28:18So right now, we're still quite far away from that. And I think we should start working on it sooner
28:23than later.
28:24Shez, Agentic AI, Alex described it as closed loop and can execute, if that's correct, Alex.
28:30How are you thinking about it at Philips?
28:32You know, so at Philips, the way we look at it is you can think of three plus one.
28:39Three plus one. So we have three A's and one more A.
28:44We look at it as are we automating something, completely automating.
28:49And in fact, Claire, I love what you said. Physicians and nurses should not be doing any administrative tasks.
28:54We believe those should be fully automated.
28:56And to the extent that they can be full closed loop agentic where the agent does the work.
29:01That's amazing. And we have some of those examples at Philips.
29:04The second A is augmentation.
29:07That's the human in the loop where you're really augmenting the ability of a human to do work with,
29:12I guess technically you may not call it agentic unless the agentic really is providing the augment and then the
29:17decision is the clinicians.
29:19The other third A that we focus on a lot is agility.
29:23AI gives the ability for an organization to respond to new data in an agile way faster than humans can,
29:31because it can work all the time, 24 seven, every second, microsecond.
29:35And so the ability to become agile is something that AI can bring to our organization, literally becoming frictionless and
29:42slipstream.
29:43The plus one is because we also believe it's important to watch for adverse events.
29:50So it's not about releasing automation, releasing augmentation, firing up agility and then saying adios, good luck.
29:58But it's about watching for adverse events.
29:59So we believe that the practice of medicine itself will fundamentally change where there are people also watching agents do
30:06work as watching over adverse events potential.
30:09And so I'm sure Tom, in a way, there are clinicians watching over.
30:13So augmentation, sorry, automation, augmentation, agility and watch for adverse events are the four A's.
30:20I think it was three plus one because three are positive and one is a watch out is the way
30:24in which we look at it.
30:25OK, thank you. Jean-Claude, coming back to you, we're seeing sort of like a real trend in developed healthcare
30:32systems across the world to push patient care away out of hospitals and into the home.
30:38Do you see any kind of like promising AI applications that will enable that care continue?
30:49Yeah, the medium that we and I'd love to hear what others see, but the medium that we have found
30:57in a fascinating way, the easiest medium has been voice and text.
31:03You know, as we start getting into sort of new modalities and apps and whatnot, it started getting a bit
31:08tougher.
31:08But but especially with the new advances in AI voice has started to become very natural.
31:16We are, you know, after AI engages with patients, we do talk to them in person and get direct feedback,
31:23not just, you know, so we do quantitative, but also qualitative.
31:27And and there's a big convergence on voice to create a personal connection with AI.
31:33I personally started creating connections with AI.
31:37You know, I can now, you know, I don't like certain personalities of AI.
31:41I like others.
31:41I so so the medium is voice, but also the other trend that we're seeing is is the impact of
31:49wearables.
31:49Yeah.
31:50So up until this new space of agentic AI, wearables had been, you know, pushing streams of data.
31:58And, you know, the idea is that physicians are, you know, need to look at all the stuff that's happening
32:03with their patients and do something about it.
32:05That was not feasible.
32:06Therefore, the promise never happened.
32:07Now there's an opportunity where we started to see all of that data from wearables show up somewhere where there
32:14are agents that are looking over that data, analyzing it and truly surfacing the exceptions or the trends, etc.
32:22So I would say voice and this new revival of wearables and being able to use truly use wearables data
32:30at scale.
32:30Okay.
32:31Maybe just to compliment on what you were saying, because maybe to quote, I remember my master thesis back in
32:38the day where we had actually studied what are the drivers of adoption of innovation in healthcare.
32:45Trust was one of them for sure.
32:47And ironically, the clinical value was not the top driver.
32:53So the fact that it actually brings a clinical impact was not the top driver.
32:57The top driver was the comfort it brings to the practitioner.
33:02And voice, I think, goes exactly in that direction.
33:05And interestingly enough, so they were providing the example of looking at the speed of adoption of anesthesia versus washing
33:14hands in hospitals.
33:16I would say clinically speaking, washing hands in hospitals saves a lot more lives than anesthesia per se.
33:25But when it comes to comfort, just operating on a patient asleep is much more comfortable than having it awake.
33:35So it turns out anesthesia was just adopted in a matter of couple of years, whereas it took decades to
33:40actually spread the habit of washing hands in hospitals.
33:43I think just when it comes to now AI and taking that into consideration, that does mean that we need
33:49to take this comfort piece really into account when developing innovation.
33:52It's not just, hey, it brings clinical impact.
33:55It's really how does that really smoothen your daily life and really fits in your workflow in order to be
34:02adopted.
34:03And Alex, in terms of that driving innovation piece, do you see a big movement if you're working primarily through
34:12WeChat or WeChat is obviously a significant platform at scale for you.
34:15Do you see both sides of that equation, so both clinician and patient, sort of like both moving in the
34:24same direction when it comes to increased use of the home as the place where treatment occurs?
34:30And I think where we see most of the movement or adoption or the demand driver, I think it's actually
34:39what Zhang was talking about.
34:40I think it's actually in the area of self-care and prevention.
34:43And I think this is where we actually want to nudge because from a health system perspective, whether we're looking
34:49at Europe, US, China, any other countries in the world, we are looking at a full-blown doctor, nursing and
34:55healthcare professional shortage.
34:56And so one of the things that we are doing is actually embedding sort of wearables, right?
35:02And actually, I think this is where AI become more of a coach to do a lot more of the
35:07primary prevention.
35:08A lot of the chronic disease care where one would argue that if you have to wait in line to
35:15see a doctor or versus having a trustful AI to actually help you with that,
35:18I think the barrier for them to do weight loss management, all of that upstream, less clinical, what we're seeing
35:26that the adoption is actually far greater by the patient and consumer.
35:30But I think with that, there's also been a missing link.
35:34It's great that we are empowering a lot of patients using AI, using wearables to do a lot more self
35:39-care, to do prevention, or to do significant prevention of chronic disease management.
35:45But I think then is how can we actually get some of those summary data and progress back to the
35:51clinician so that that is, again, integrated into their long-term care.
35:57And so the way I see that is I think because of the supply, demand and balance, I think we
36:03see patients a lot more willing, maybe it's different elsewhere, to try new tools because I think they're short of
36:09options.
36:10Whereas, I think the doctor, I think the adoption is coming, but I think the lack of adoption sometimes actually
36:17defines systemic issues, like reimbursement, like the regulation, like who is liable for the use of AI.
36:25So there's a lot of the governance question that comes back to sort of inhibiting or slowing down the adoption
36:31of AI and agentic AI within the healthcare system.
36:36Shez, any thoughts of this in terms of like the movement to the home?
36:39You know, actually, I'd love to switch a little bit because we talked a lot about the home, but has
36:45anybody here really thought, what does the hospital room of the future look like?
36:49I mean, today you walk into a hospital room, all this equipment, all these wires, patients tied to the bed,
36:55can't go anywhere.
36:57Is that what it's going to look like?
36:58I mean, the nurse comes in, takes a pressure a few 15-20 minutes, wakes them up.
37:04Now, the reality is that the data that comes off of your ventilator, of the IV pole, of all the
37:11digital readout, all that data is going to be fed to agentic models that are deciding things on the fly
37:18in real time in a hospital setting.
37:20It's equivalent to the concept of the home because you're talking about wearables, but in the hospital, we'll have wearables
37:24too.
37:25We will not be tied to the bed with wires, and it will not just be the ICU.
37:30It will be every part of the hospital, and there's agents that are monitoring the wellness of the patient, what's
37:35going on, and those agents are actually talking to the patient about how they're doing in the hospital.
37:39They just had a blood test that came back from the lab.
37:42The patient will have a conversation with the agent, well, what did it show?
37:45What's next?
37:46Why am I scheduled for an MR tomorrow morning?
37:49The hospital room of the future will be data converted to these conversations, whether it's with clinicians, whether it's with
37:57nurses, whether it's with patients, whether it's with the families who are not even at the hospital.
38:01Today, Philips liquidates one in two beds in the US, all the data that comes from anything in the room
38:08is going through Philips capsule.
38:10We foresee a time when that data is conversationally talking to nurses, talking to doctors, talking to patients and family
38:19members, and the hospital room of the future will be silent, wireless, and not like anything you've seen today.
38:28One question I have to ask you, you did a piece of research recently that suggested that clinicians lose 45
38:35minutes per shift just to administration.
38:38Is that going to be the future?
38:40So this is what comes to your terrific comment on physicians and nurses should not do anything administrative.
38:46In that same survey I spoke about where we uncovered the trust gap, when you talk to clinicians, they said
38:52that three out of every four said they spend time, unnecessary time, looking for data.
38:59And then what's worse is half of them said that they lose four hours a week simply looking for data
39:06to act on for patient care.
39:09And that for absolute sure should not be occurring.
39:12And in fact, really, this is a part of the agility opportunities for AIs to make that data imminently, immediately
39:18available at the point of care.
39:20So the right action can be taken for sure that should not be the future.
39:23Actually, one last comment on that survey, which is in some ways depressing and a call to action to us.
39:28So physicians were asked, look, compared to five years ago, there's been tech innovations.
39:34So tell me, compared to five years ago, are you spending the same amount of time, less time or more
39:41time on administrative work?
39:43Doctors and nurses.
39:4445% said, compared to five years ago, I spent the same amount of time on administrivia, administrative trivial things.
39:53What was depressing is 35% said, I spend more time now than five years ago on administrative tasks.
40:02So only about 20% said that they spend less time.
40:04So we as an industry have a little bit of work ahead of us.
40:09OK, well, let's talk about what's ahead of you.
40:11We've got a little bit of time.
40:12So I think maybe just to sort of like maybe end the session, I'd love to get each of you
40:17to give us a sort of some insights on how you think AI is going to shape health care in
40:24the next five to 10 years.
40:25What's going to be different if we were at Viva Tech in 2035?
40:30What would health care look like?
40:31Sorry, Jean-Claude, I'm going to unfair, but I'm going to come to you first.
40:35Give us a sense of like how you think things are going to be different in the next five to
40:3810 years.
40:39Look, I will humbly start by saying we don't know.
40:43Yes, I think that's fair.
40:45So we and others are in the process of creating that future.
40:51The only thing I know for sure is it will not look like what we are imagining it, but we
40:56are going down a path where we are creating it.
41:00I will then go back just a bit to why we're at it, why I do what I do, why
41:06we do what we do is in the United States, a hundred million Americans don't have access to primary care.
41:15And the ones that have access have access to suboptimal primary care.
41:21There are 500 million primary care hours available in the country.
41:26We need 2 billion hours on top of the 500 to do primary care for everybody.
41:32So if personally I'm successful at what we're doing, we would bridge that gap and provide primary care to everybody
41:43in the country.
41:44It will only be done through this concept of primary care as a service with agentic AI, with everything that
41:51we talked about here.
41:52It's a combination of all of these capabilities that will bring that, but our vision is that primary care becomes
41:59available to everybody who needs it at the proper levels.
42:03Okay. Thank you, Alex. Your vision.
42:06My vision is, I think this is one thing that I mentioned briefly is about genomics.
42:13I think genomics is another language that we haven't fully grasped with and the potential of this.
42:19I think actually my vision is in five years time, we can then using much cheaper genomic analytics to then
42:27upfront screen and diagnose cancer at much earlier stage, in stage one, stage two, than ever before.
42:35Now, that comes with a lot of healthcare burden and what to do with it.
42:38I think this is an important question to be discussed.
42:43But I think a lot of the healthcare costs right now is wasted in very late stage diagnosis, end of
42:50life care.
42:50If we can then start reshaping the healthcare system so that more time can spend on primary care, upstream care,
42:57focusing on health rather than diagnosing and treating the diseases, right?
43:02I think this is where I think AI has enormous potential.
43:04Great. Thank you. Clara.
43:07I compliment in saying I think healthcare will be a lot more personalized.
43:13But personalized not only to your medical condition, but to everything which is around you.
43:19And also what are your own preferences?
43:22So what is your medical literacy?
43:24So I don't think someone who is very knowledgeable about his disease will be treated in the same way that
43:29someone is not.
43:30What are your preferred channel to actually being cared for?
43:34Some people actually like to read content.
43:36Some people like to we change the messaging.
43:38Some people like to go see a doctor face to face.
43:40Some people like to chat with an AI.
43:42So I would say to that as well.
43:45And also a lot more preventative.
43:48So a lot more preventive care by being able to identify risk factors that today are honestly factored in the
43:56traditional care pathway, I would say.
43:58Yeah.
43:59And being able to, of course, looking in genomics as well.
44:03But sometimes we don't even need to go that far.
44:06I was talking to a cardiologist that told me, you know, I need to know where the patient is living.
44:12Because if the patient is living in a very loud environment, there is a higher risk of re-infarctus.
44:19So like onset of heart attack.
44:20And that's something today, it's not present in the patient files, right?
44:24So if you can actually take that as one of the data points to start doing prevention at this level,
44:30taking into consideration all the different aspects of the patient, not only just the static patient file, then we can
44:37go much further.
44:37Thank you.
44:39Sure.
44:39After a great three thought leaders, I'm supposed to then come something original, everything.
44:43Look, maybe I'll add plus one to everything you said.
44:46And then I'll take it to the last step, which is for sure precision therapy.
44:52Yes, you'll have precision medicine in the sense of personalize the way the personal coaches personalize and the way the
44:57clinicians interact with the agents.
44:59And without exception, that will go all the way to therapies with personalized vaccines, personalized drug therapies, personalization all the
45:05way.
45:06So the value chain will be end to end personalized, not just in the upstream coaching prevention, but in the
45:11downstream, the actual therapies.
45:12And it will be a beautiful future.
45:14I love that.
45:15It's going to be a beautiful future.
45:17Thank you all.
45:18We will be back here in 2035 to see if you are right.
45:22But thank you for sharing your insights today.
45:24It's been a really great conversation.
45:26Thank you.
45:27Thank you very much.
45:28Thank you.
Commentaires