Latest / CareTalk: Healthcare. Unfiltered. / How AI is Transforming Prescription Efficiency

Transcript

How AI is Transforming Prescription Efficiency · CareTalk: Healthcare. Unfiltered.

• 0:00
  A simple Tylenol order takes up to 62 clicks in some electronic medical record systems.
• 0:06
  That's madness.
• 0:07
  And it's a key motivator for tech companies to combine artificial intelligence and voice recognition to simplify the process and reduce errors.
• 0:23
  Welcome to Care Talk, America's home for incisive debate about healthcare business and policy.
• 0:27
  I'm David Williams, President of Health Business Group.
• 0:30
  My guest today is Emmanuel Bilbeau, co-founder of Pozos, which is partnering with Microsoft's Nuance Division to allow clinicians to prescribe by voice and get point of care treatment recommendations personalized to each patient.
• 0:43
  Welcome to Care Talk, Emmanuel.
• 0:45
  Thank you, David.
• 0:45
  Very happy to be invited today and share with you great news about the partnership with Microsoft.
• 0:53
  Outstanding.
• 0:53
  Well, before we dig in, I'd like to encourage everyone to join the fast growing care talk community on LinkedIn, where you can dig deep into healthcare business and policy topics, access care talk content and interact with the hosts and our guests.
• 1:07
  And please be sure to leave us a rating on Apple or Spotify while you're at it.
• 1:12
  All right.
• 1:12
  I'll give people a second to do that.
• 1:13
  Now, Emmanuel, I want to start in a big picture and understand.
• 1:17
  What are some of the challenges in electronic prescribing?
• 1:20
  I've been around long enough to hear, you know, when they used to talk about the prescription pad and nobody could read the doctor's information.
• 1:27
  Now we got to electronic prescribing.
• 1:29
  Why are there problems?
• 1:32
  Yeah, for two reasons.
• 1:33
  Actually, prescribing is a very complex process for the medical reason, first the clinical reason, because prescribed a script for patients.
• 1:46
  is making clear that every drug is compatible with each other.
• 1:50
  It's like one by one, you have to check if it's compatible with the other drugs that the patient takes.
• 1:54
  You have to make sure that each drug is compatible with the patient condition, probably the patient renal function, patient disease, sorry, disease function.
• 2:07
  Then you have to adapt the dosage to the patient width, patient age.
• 2:12
  So it's a very complex...
• 2:14
  process from a clinical perspective.
• 2:17
  But as you said, when software were developed to help physicians prescribe electronically, at the end, it's even more time consuming because the very best EHR on the market is even like a crazy waste of time.
• 2:39
  You need at least 18 clicks to prescribe one medication.
• 2:44
  But you have to take into account that for the aging population, the average number of drugs in each script, it's seven.
• 2:54
  So seven times, you have to decide which drug and click on the drug name.
• 3:00
  Then which galenic form, click on the one you want to select.
• 3:04
  Click on the strength because there is different strengths and dosing for each drug.
• 3:10
  And then formulate.
• 3:12
  the dosage, which means like how many tablets you take in the morning, for lunch, for dinner, for how long.
• 3:19
  So imagine like a long prescription for your grandmother or grandfather.
• 3:25
  It could take more than 10 minutes.
• 3:27
  That's why we need to do something.
• 3:32
  And that's why we collaborate with Microsoft to help physicians save 10 minutes by patients.
• 3:38
  And we explain later why.
• 3:41
  The thing is, if I consider, sometimes you hear about, well, doctors don't like technology, but I think a lot of it has to do with, they do like technology.
• 3:49
  They're big adopters of iPhones.
• 3:50
  They use a lot of things electronically, but I think it's the difference between what they see every day in their life outside of work versus work.
• 3:58
  So you mentioned even if it's 18 clicks, you know, I go to Amazon and they're famous for one click.
• 4:03
  You know, I fill my shopping cart, you know, boom, boom, boom.
• 4:05
  Doesn't take me 10 minutes to spend a ton of money.
• 4:08
  And so I think some of it is just like sort of You know, the disconnect and I'm trying to understand.
• 4:12
  I still, I get it that it's complicated, but it could also be complicated to buy something from Amazon.
• 4:17
  So, I mean, and Tylenol is pretty basic.
• 4:20
  If you even have to prescribe it instead of just grabbing it off the shelf.
• 4:23
  I mean, how could it possibly take that many clicks?
• 4:26
  I mean, where, where does that really come from?
• 4:28
  You did explain it in a way, but I only counted up to about the possibility of like 20 clicks, but how could it possibly take that many?
• 4:35
  Yeah, that's, that's complex interfaces actually.
• 4:38
  I won't blame the software publishers for that because there are so many different kinds of prescriptions.
• 4:44
  I talked about tablets, but for some it's infusions.
• 4:48
  And you need so many options, so many possibilities to cover the whole possibilities of a script that interfaces are very complex.
• 4:59
  And that's why most of the medical software are moving to the voice because with this voice, you can shortcut this process of clicking everywhere on the software.
• 5:12
  And that's maybe the only solution.
• 5:15
  The other one is AI, trying to predict which drug you will prescribe for a dedicated patient and maybe generate options, suggestions for prescription, enabling physicians also to save time.
• 5:30
  Somehow with POSOS, we are combining both ways in one solution.
• 5:37
  So let's talk about AI first.
• 5:39
  One of the reasons that there's so much excitement about AI in healthcare is that there's a lot of really challenging problems.
• 5:46
  And if you look at all these problems of workforce, errors, quality, cost, we've been working on these things for decades.
• 5:55
  And you look at them to say, well, we've made some progress, but I think overall it's gotten worse rather than better in some ways.
• 6:03
  So you say maybe AI can be a solution.
• 6:05
  And it seems like that's possible.
• 6:07
  So if we look at AI first, and let's voice off to the side for a moment, what is the biggest opportunity with AI in prescribing?
• 6:17
  Or how does it address some of those main issues that you talked about before?
• 6:23
  I think AI is very interesting for us for one reason that somehow nobody really mentioned in the conversation around AI in health is that medicine is not a a perfect science somehow.
• 6:37
  And for exactly the same patient, you will see 10 physicians prescribing, like for instance, 10 scripts that are somehow different.
• 6:46
  And that means that there is no one solution or one great solution.
• 6:51
  There are many options.
• 6:52
  The same for diagnosis.
• 6:54
  For the same patient, you have like various diagnoses sometimes, and that's scary.
• 7:01
  But that's where AI is great because AI is great in processing the probabilities, providing the best solution.
• 7:09
  And that's also why not only for diagnosis, because we talked a lot about diagnosis for AI, but also prescription.
• 7:17
  There are great opportunities to suggest the most tailored prescription for a dedicated patient, taking into account all the patient history, comorbidities, current treatments, and making the best possible treatment for him or her.
• 7:37
  I talked before about the transition from paper -based prescribing to electronic prescribing.
• 7:43
  And while you had resolved certain issues, like you could read the prescription's handwriting because it was now filled in electronically, you started to have new problems that you didn't have before.
• 7:53
  For example, with a pull -down list, I might have to worry that the physician would write something I couldn't read and it looked like something else, but you didn't have the issue that you know, because of alphabetical order, they would pull down something, you know, one further down in the list that had nothing to do with what they were actually trying to do.
• 8:11
  Are there new problems like that with AI in prescribing?
• 8:15
  So maybe it's going to suggest something that nobody would suggest because it doesn't make any sense.
• 8:20
  Are there new kinds of problems that come up?
• 8:23
  And how do we, if we don't know that yet, how do we know and how do we become confident?
• 8:29
  The way you talk about AI like means you are thinking about all the generative AI models somehow.
• 8:36
  And those models are very effective when those using it can just verify quickly that it's right or it's wrong.
• 8:46
  If you can't verify quickly the accuracy of the results, then it shouldn't use it.
• 8:53
  So that's basically the same for prescription.
• 8:56
  If the physician is well trained, it will be able quickly identify that the options are relevant or not.
• 9:06
  So that's the way also we think about it.
• 9:09
  It's not like generating a prescription automatically and physicians just need to click and approve.
• 9:16
  But it's more about suggesting options that the physician will check and select.
• 9:24
  But options means right treatments, galactic forms, and dosage at the same time.
• 9:29
  more than just generate a prescription automatically.
• 9:33
  And that's something close to the maps that we use and that we transform into a GPS.
• 9:40
  Sometimes in Waze, for instance, you have two or three options and you select the one that makes more sense for you.
• 9:48
  That's the way we think about the best AI for medicine.
• 9:53
  Letting the physician always make the final decision.
• 9:56
  OK, so if I use your Waze example, I know that I want to go from point a where I am to point B and I know what point B is.
• 10:04
  Maybe it's a certain restaurant or store or a person's house in your, in your analogy here.
• 10:11
  Does the physician have to know the point B that they're going to?
• 10:15
  So for example, do you take a edit as a given that the diagnosis that they have is the correct diagnosis?
• 10:22
  And you're only saying given that diagnosis, here's these other ways to do it or Is it a broader decision support where you're saying, given all the characteristics about this patient, you're writing a prescription for Tylenol, okay, but maybe it should be something else because maybe you should be treating something else.
• 10:38
  How far back does it go?
• 10:39
  How does it tie it into the Waze model there?
• 10:43
  Yeah, we usually think about a patient with one pathology, one drug, but the reality is like a patient has several pathologies and several drugs in parallel.
• 10:54
  When you want to prescribe one additional drug for a new diagnosis you met, then you will think about your usual drug you prescribe for every patient because you trust this drug.
• 11:04
  But then you discover with the AI that the drug you are used to prescribe is not compatible with the other medication of the patient, so it's not compatible with its comorbidities.
• 11:15
  Then you have two options, for instance.
• 11:17
  I just give you an example.
• 11:20
  Choose another treatment for the new diagnosis or switch.
• 11:23
  a previous medication to another one.
• 11:26
  So that's all medication are compatible.
• 11:28
  This is a great example of two ways you have, and you will have to decide which one you want to select.
• 11:35
  That's the best analogy I can tell you for the ways.
• 11:39
  But physician has also his trust with one or several medications, just like you have a trust to one direction or another to go to your best restaurant.
• 11:50
  Sounds good.
• 11:50
  All right.
• 11:51
  So the danger of analogies is taking them too far.
• 11:54
  But I'd probably rather go to a restaurant where you live than where I live.
• 11:58
  So maybe that would be a good thing.
• 12:00
  I'm not going to do that together next time.
• 12:02
  Let's talk about voice.
• 12:04
  So you mentioned AI and voice as both being ways to get over some of these challenges of all the clicks.
• 12:10
  Separate from AI, what makes voice a good technology?
• 12:14
  Why have physicians adopted it?
• 12:17
  I would say that...
• 12:18
  that shortcut the complex interfaces where you have many options.
• 12:24
  And you've seen in the last conferences, even outside of a healthcare that people think about like switching from one software to just like something you will hold or keep in your pockets.
• 12:36
  That's somehow the same way like EHR publishers, EHRs being like electronic health records are thinking about the future.
• 12:45
  So voice is everywhere.
• 12:47
  Voice...
• 12:48
  you to control your software.
• 12:53
  But for the prescription side, because that's why I'm interested in, I'm a pharmacist, I may be like too much focused on medication, that will enable just to dictate the prescription, the way you think about it, and always have this assistance of an AI telling you that you are prescribing the right drugs, the right dosage, or there may be options to optimize a prescription or there is a huge risk in what you've said and you should absolutely correct it.
• 13:29
  So the way it works in terms of a user journey that you prescribe, you get the control based on your patient profile, you validate and you send it to your software or you correct and you send it to the software.
• 13:43
  But there is always this step where physicians can correct or validate the prescription, sorry for that, before he send it to the electronic health record.
• 13:57
  So if I think about, so you have this new collaboration with Microsoft and it's their Nuance division.
• 14:03
  Nuance started off as really a transcription company.
• 14:08
  They used to take in the dictation when it was analog and then they would type it up and they would provide it back.
• 14:15
  And then they evolved and then essentially acquired by Microsoft.
• 14:19
  And they do focus.
• 14:20
  I mean, they're integrated into, you know, they're integrated with this co -pilot kind of an approach.
• 14:25
  So they're using AI in the first place.
• 14:27
  So why, why did Nuance need to work with you?
• 14:31
  Why does Microsoft need to work with you as opposed to just including within Dragon, your functionality?
• 14:38
  The prescription part is probably the most complex step in in the use of your software.
• 14:49
  You will use your software indeed to write down the report of your patient physician's conversation.
• 14:58
  You will click in the software to select the diagnosis you made.
• 15:03
  But prescribing is a long process, as we said before, but also the most difficult step in terms of adapting and making the right choice.
• 15:13
  So here, adding puzzles to their portfolio somehow of solution, like partners, Microsoft is able to cover the whole patient journey in a physician office from the first discussion with the physicians, also the control of the software, we call Pilots, and at the end, the prescription, which is always the last step of a consultation.
• 15:38
  So it's a way for them to cover the whole patient journey.
• 15:42
  I think that's the first reason.
• 15:44
  And this very clinical part of the patient journey and physician interaction with the software is very hard for them to cover, just like Microsoft competitors, because you've seen that there are many, many competitors now in this copilot technology.
• 16:03
  And what you need also is a drug database to perform that.
• 16:08
  Because the way we manage this AI system for prescribing is to have a worldwide universal database covering all medication from all countries, covering all languages, making it possible to correct a prescription that anyone in the world will make.
• 16:29
  So that's the way also Microsoft think about the future, being able to scale every solution they partner with.
• 16:36
  And I think we are the best positioned in the worldwide to...
• 16:41
  enable that and that's why we decided to collaborate with us.
• 16:44
  It's more for the, it's not only for the use case we provide, but also the possibility to scale the solution everywhere.
• 16:51
  And I will tell you one funny story.
• 16:54
  I had last week a conversation with a Swiss EHR publisher asking if we can cover the three languages in Swiss, which are French, Italian, and Swiss German.
• 17:07
  And Swiss German is a very interesting language because it's written exactly like German, but they pronounce it in a totally different way.
• 17:15
  So we can manage the German parts in our software, but we are not able to cover the Swiss German way of pronouncing the prescription.
• 17:26
  So also from our side, Microsoft was the best partner to make it possible to use POSUS everywhere.
• 17:33
  that sounds good.
• 17:34
  Well, so Microsoft is obviously a global company.
• 17:37
  And you're talking about the ability to scale and what the needs are.
• 17:40
  And at the same time, healthcare markets and you know, the healthcare system is pretty different from country to country.
• 17:48
  Certainly the U S versus Europe broadly, but even France versus Switzerland or other countries in Europe and then Asia are different.
• 17:57
  What do you find, is the same, because it's very universal what you're doing, or are there certain things that need to change, not just based on like, you know, Swiss German, but.
• 18:06
  based on the actual structure of the healthcare system or the financing of it?
• 18:12
  Yeah, the clinical part is the same everywhere.
• 18:15
  You will decide to prescribe one drug or another.
• 18:18
  You will diagnose a pathology the same way in the US, in Europe, and in Asia.
• 18:24
  So the drug database we've developed, for instance, is not populated with text fields, but only international codings, which are the same everywhere in the world.
• 18:36
  So the clinical information we have is available and compatible with every country.
• 18:42
  The big difference is, of course, the patient journey in the system and also the price sensibility or reimbursement sensibility.
• 18:54
  And somehow, coming back to the ways analogy, in ways sometimes you also have different options based on the traffic.
• 19:06
  So here, I think for the US, for instance, you will have also to select which one drug or another based on the possibility for the patient to be reimbursed or the possibility to afford this prescription.
• 19:18
  So it's like an additional type of data that you have to process in your prescribing assistance, taking into account a patient's insurance coverage, for instance.
• 19:34
  At the end, the act of prescribing is really similar from one country to another, if you just take into account these economical points as well.
• 19:45
  You mentioned your background as a pharmacist, and one of the things pharmacists are typically concerned about is medication reconciliation, which is a big challenge.
• 19:54
  And I'm wondering if you could just describe from your perspective, you know, what are the issues with medication reconciliation and can AI be helpful there, whether or not Pozos is doing it?
• 20:05
  I'm not sure it's a very different process to the one we described before for prescription.
• 20:14
  The main issue actually is that physicians or pharmacists are partially blind because they are not fully aware of what the other physicians have prescribed or what the other pharmacist has given to the patients.
• 20:31
  decision they make in terms of prescription or giving medication to a patient could be wrong just because of his blindness.
• 20:41
  I'm not sure that the AI is the right solution for that.
• 20:44
  It's more about sharing information between all health system or physician, hospitals, and together with pharmacy.
• 20:54
  So once we will have a common electronic cash record shared with every clinicians, I think this reconciliation problem won't stay.
• 21:04
  So it's a question of sharing information to have the whole picture and then the right prescribing assistance to take into account this wider information that will be available.
• 21:21
  Emmanuel, my final question for you is about the vision that you have for the future.
• 21:25
  You've obviously just had this very exciting announcement of a collaboration with Microsoft, the potential to scale.
• 21:32
  Where do you see Pozos in five years?
• 21:37
  I announced this Microsoft collaboration, but I didn't talk about the collaboration we have with the EHR publishers worldwide.
• 21:45
  There are already 18 EHR publishers working with Pozos who embedded the solution into their interfaces.
• 21:56
  The vision we have in five years to have not only the only direct database that is really universal, as I said before, this capacity to prescribe with POSOS with a voice or still with your keyboard with a suggestion to automatically populate your prescription.
• 22:18
  But really, we want to be the central point, like enabling not only EHSR publishers, but also every medical app, every medical software to scale because using POSOS database, using our APIs is a great way to have the same software everywhere in the world.
• 22:37
  And that's the huge problem currently.
• 22:41
  Software are country -based because they are based on a country -specific drug database where software are basing their technical stack on a solution that can scale by itself.
• 22:54
  then they can have a real SaaS model and replicating their solution everywhere in the world.
• 23:01
  So we think about a more intentional point, more intentional software that will be available for every physician in the world.
• 23:09
  That's not only POSOS, it's POSOS together with a partnership, but we provide the great technical block for that.
• 23:15
  Well, great.
• 23:16
  That's it for yet another episode of Care Talk.
• 23:19
  We've been speaking today about AI and drug prescribing.
• 23:22
  My guest has been Emmanuel Bilbeau, co-founder of Pozo.
• 23:25
  Emmanuel, thank you for joining me.
• 23:27
  Thank you, David.
• 23:28
  So if you like what you heard or you didn't, please make sure to subscribe on your favorite channel.