Agentic Patient 11: ""I stopped letting AI read my daughter's medical records" (Karlien Hollanders)

[00:00:00] Dear listeners, welcome to Faces of Digital Health with me Tjasa Zajc and a special series called Diogenetic Patient where we explore how patients use AI, what tools, what prompts they use, what are some of the guardrails that we should be mindful of if we reach out to AI for help when we are experiencing medical issues and at the same time how is AI impacting us.

[00:00:30] the relationship between patients and doctors. In today's episode, we are going to talk about the caregiver perspective in the case of having and taking care of a child with a rare condition.

[00:00:49] Karlien Hollanders, who is a pharmacist by background and lives in Belgium, became her daughter's medical record after her daughter was only two months old. Karlien Hollanders, who is a doctor of medical care, was diagnosed with a rare kidney condition.

[00:01:13] Karlien spent three years moving between Belgian specialists hospitals that could not exchange data with each other. So, 11 years later, she now has 408 documents in a national platform that runs on a federated approach.

[00:01:33] So, in order to get her full medical documentation from her daughter, she actually needs to download each particular PDF separately. We discussed how is she managing her daughter's care? What kind of a difference would it be if she could actually search through data more easily? What has she already tried in terms of new AI tools?

[00:02:02] And how does she look at the evolving European health data space and all the efforts that are going into the exchange of data, even across borders? The next day, we'll be right back in the future. Enjoy the show. And if you are interested in tips and ideas on how to use AI as a patient, go to facesofdigitalhealth.com slash agentic-patient,

[00:02:28] where you can find summaries of the key episodes with advice on how to approach AI. There's also a short video with six tips on where to start when you're starting off with AI as a patient and more. And if you haven't yet, subscribe to the podcast, leave a rating or a review wherever you listen to your podcasts,

[00:02:54] and check out our newsletter, which you can find at fodh.substack.com. That's fodh.substack.com. Now let's dive in the discussion with Carly.

[00:03:16] Carly, hi, and thank you so much for joining me on Faces of Digital Health and a special series called Diogenetic Patient, where I try to research how patients use AI very concretely, what prompts they use, what tools, what guardrails. We also cover discussions with researchers and clinicians about their experience that the use of AI by patients has on the discussions

[00:03:45] and the relationship between doctors and patients. And today I invited you to share a caregiver perspective, since you have a daughter that unfortunately had several medical issues already from very early on, which basically demanded for you to take care of a lot of her health data.

[00:04:15] So before we go deeper into that, can you maybe just create a very brief overview of the patient journey from your daughter and you as a caregiver? How did you go from, you know, just not worrying about health to now trying to figure out how to manage 400 PDFs from medical appointments?

[00:04:44] Yes. Thank you, Chesa, for having me. Yes. So she was my firstborn child and I was a hardworking mom and I had everything prepared to just stay home for three months. It is in Belgium and then start working again. But when my daughter was two months old, out of the blue, she did a complete renal failure. So her kidneys were not functioning at all. And so she almost died from one day on another.

[00:05:11] And then we had to go through several examinations to find out that she had a very rare disease called primary hyperoxalaria type 1. It's a very rare disease and it's actually a problem in the liver. And because of this problem, you get too high of levels of acetylic acid in the blood. And they form with calcium, they form stones. And then that broke down the kidneys to be very short about it.

[00:05:40] But it meant that she needed to start dialyze by the belly first as a baby. And then that wasn't enough. So also as a baby, she needed dialyze by blood, hemodialyze, which is very rare with babies. So I needed to travel to multiple hospitals to get very specialized care. Because she also needed a liver transplant to get rid of the cause of the problem, which was in the liver.

[00:06:08] And then later on, a kidney transplant. So having a baby with special care like that, liver transplant, kidney transplant, hemodialyze, made it necessary for me to go to multiple hospitals in Belgium with each of them having their own specialty. Which is okay for me because I understand that it's better to have specialized care in one hospital because you need special equipment, special teams around that problem.

[00:06:37] And also I was lucky to be mobile. I have a car. I was lucky to be able to live for three years with a lower income to travel around with my baby to those multiple hospitals. So I'm very blessed with that. But even though we have that special care in specialized hospitals, there was one big thing lacking. And that was the fact that those healthcare providers or hospitals could not work together.

[00:07:06] And so me being a pharmacist by background and also being able to speak both of our nation languages, which is Dutch and French, I was in the possibility to be her walking medical record next to her and to be able to each time to a new healthcare provider explain what she went through, make sure that things were going right.

[00:07:31] I often say I was like a mother lying standing above my little line, above my cup, to make sure that everyone did what they had to do and was aware of the path, things that she went through. And me being there was often necessary really to save her life. And so when she was better and so that lasted for three years to be sure that she would survive.

[00:08:01] So that was for three years long. I needed to travel around and make sure that she survived. And after when she survived, I made myself a patient advocate or a patient expert because I was aware of the fact that not every baby has a mom that is highly educated, that has a car, that can have medical background, can speak both languages, and all those things made that she would not have survived.

[00:08:29] And also we know a child with the same disease who sadly passed away when she was nine years old. So my daughter now is 11 years old. She's very good. Of course, she needs checkups every six weeks and she will need new kidney transplants in the future. But she's doing well. So I made it a little bit of my mission to make that happen. Yeah, yeah. I noticed that you used the word lucky for quite a few times.

[00:08:58] That you were lucky. Or yeah, that you were kind of lucky in this whole situation, which I thought was interesting. Yeah, I'm not at all lucky, of course. But yeah, you make the best out of it. So it's also what I was missing is being able to just be a mom. Because I was really more of that second healthcare provider in a team of healthcare providers around multiple hospitals, trying to connect one another.

[00:09:27] So I was more like a kind of coordinator. Yeah. Instead of just a mom. Yeah, I was more, yes, indeed a case manager. That's also something that you see now and then coming up, case managers, especially like, for example, cancer patients, which I think is good, but also makes me very sad because they always go towards cancer patients, diabetic patients, patients that have a, not a clear, but a known more or less pathway,

[00:09:56] where rare disease patients are often just as much as cancer patients in a hospital, like on a very regular basis, but are not seen as patients where they can do something like that for. But it's a good step. It's a good step that they see that it's necessary to have something like case managers. And if it works for those pathways, then it might in the future also exist for others. So, yeah.

[00:10:21] You are very interested in tech and how tech could improve interoperability, the understanding of medical data. So, knowing that the story of your daughter started 10 years ago, how do you think that your case could be managed differently if it was happening today,

[00:10:48] when we have so much of AI and tools that are available not just to clinicians but also patients and caregivers? Well, that's actually quite disappointing. So, in 10 years, I have not seen a lot of change in practice. We do, in Belgium, have a good way of sharing medical data, but it's always limited to medical data from secondary lines.

[00:11:16] So, from hospital data, primary care data from your house doctor or a diatist, a dermatologist, a psychologist, kinest, so physiotherapeut, those primary care healthcare providers, they don't share anything. It's not because they don't want to, but there's not really a way of sharing it on a platform. It does exist for all the data that comes from hospitals and lab results.

[00:11:44] And I did see an improvement there where some hospitals might not have been connected yet. They are better connected. While seven years ago, they still needed to burn a CD and send it by post, which really, when your daughter is in intensive care, makes you like say, what? How is that possible? So, they are better connected, but they only see data of hospitals and only data that is published. It's not really published.

[00:12:13] It's federated data that is connected by a meta hub, but it's only data that hospitals release, and it's often a release note when you leave the hospital or a referral letter, things like that. It will never be a complete operation when she was operated, a complete operation review. That will not be on this platform.

[00:12:37] But it is already good that these things are exchanged in a digital way, but it's often a dump of PDF files or HTML, like a website kind of record. But in case of my daughter, she's now 11 years old, and she has 408 documents there, which actually doesn't help.

[00:13:01] Hospitals that are on the platform, so you need to take into account also all the primary care stuff. So, how do you manage that? Do you just keep everything on paper printed? Do you also have a repository? There's online tools that could potentially enable that. As a patient and also as a caregiver, I have a mandate for the data of my daughter, so I can access that platform as well as a patient. So, not only my caregivers, but me as well.

[00:13:31] We have kind of the same filter technique. So, we can look for one doctor who made the document, a hospital and a date, and that's actually it. So, you cannot filter or look inside of the PDF files. So, there is not enough metadata, as we call it, linked to these published documents, which does not make it easy to go through those documents. So, I have access to that platform.

[00:13:58] And then, for all the data that comes from primary care, I just store it in my personal drawbox, and whenever I need it, I can find it. But actually, even there, there just isn't a lot of data from primary care. When you go to your house doctor, so your general doctor, you never get a report about what you've told that doctor. So, it will be in a system maybe, but in Belgium, we never get it afterwards.

[00:14:25] So, there might still be a lot of data around my daughter that I do not have as a patient myself, or as a caregiver. So, how do you use AI, for example? Because everything that you mentioned, leaving privacy aside, right, for the time being, could potentially be easily solvable with AI.

[00:14:53] You can create an offline app, just throw all the PDFs in, and then create an app that could potentially search easily through the data. So, I'm super curious, especially knowing that you're also quite tech-savvy, what you've tested so far, what you thought was interesting, what you thought you definitely wouldn't want to use. So, what's your experience with AI and all the documents that you have? Yes.

[00:15:18] So, those foreign documents are on that platform, but as I said, it's federated data. So, the data is still in the hospitals, and you just get like a key. In the MetaHub, you get a key, which can, like a repository, I think. You can ask like, what data do you have about the daughter of Carleen? And then it will show you that data, but it's still on a virtual platform. And I can download those PDFs, but I need to download them one by one.

[00:15:46] I cannot run AI on that platform because it's secured with my login. And also, it's from the government, so they will not allow AI tools to really go into that federated data system. So, I need to download them one by one, but if I have 408 PDF files, it's like three, four clicks to download one document. So, I had an IT friend look into that, and he has made me a little script in my Chrome browser

[00:16:14] so that I could like just download them all in two minutes' time. So, I've done that, and then I've saved them on my personal Dropbox. So, the good thing is I now have my data locally, although Dropbox is also in the cloud, but I could have it locally in a very easy way. The negative side of it is it's no longer up to date and live because I've taken it from the platform. If there are adjustments being done or new documents, it's just the moment of downloading.

[00:16:43] So, I would love it if in the future I could use AI directly on that platform or there would be a way of exchanging my personal data with AI tools. But nevertheless, now I have those documents on my computer, and so I wanted to play with AI. It's already, I think, four years ago, I think three, four years ago, that I started playing around with it. And so, I've used Liv, Liv, L-E-V-V.

[00:17:13] It was, it actually doesn't exist anymore. I think it was a startup in Oslo. By incident, through meeting multiple people, I got to know them. They had built, I think it's still in the App Store. You can download it in the App Store. You can upload your documents. And it was a very, very good tool. It actually made in a couple of minutes a summary of every document. So, all my documents had a summary in any language I wanted.

[00:17:42] It also extracted, like the doctor, the hospital, a summary of what was in that document. But then also, it organized all my documents by problem, more or less. So, it could filter my documents, and it organized them for these are files with eye problems, with bone problems, with liver, with kidney. These are operation files. These are blood examinations, exams.

[00:18:09] And so, suddenly, those 408 PDF files became linked to a problem or a part of your body. And it made it much more insightful. So that when I would see, for example, a new healthcare provider, which was, for example, a kidney doctor, then I could say, instead of, here are 408 documents, I could give him the 20 documents around kidney problems. So, that was already a first step.

[00:18:38] What I was lacking was more of a timeline, a timeline with, like, in those 11 years that she's existing now, what happened? What are major events? I know she's been operated 20, maybe 30 times. I really don't know the exact number, and nobody can tell me. Because in every hospital, we have pieces of the puzzle, but you never have the whole timeline.

[00:19:00] So, I also, for my mental health, it would help me because I've been in a kind of survival mode for three years to just make her survive. So, the only thing I did was eat, sleep, and be there as a medical coordinator. But I didn't know what was going on in the world. Everything went very fast, and I was living from the one life-saving moment to the other one.

[00:19:24] And so, it would be nice for me to have this timeline with key events and linking my documents to those key events so that I know this operation, what was the outcome, or why did we do it, what was before, what after. And so, I was looking for another tool, an AI tool, to be able to ask questions to my data. And so, the other thing that I have experimented with, also, I just looked it up, it still exists. It's humata.ai.

[00:19:53] It's a website, humata.ai. And you can, it's like a RAC principle. So, you upload your data, and you ask questions, and you can limit the temperature of inventing things. So, I could say, like, don't invent things, just look only in my data and answer my questions. And then I could ask, like, what is her average, I don't know, creatinine score, or how many times was she operated? Make me a timeline. It did help.

[00:20:23] I have some very cool things that I could do with it. For example, also make a summary about her medical history, but also make a summary for an orthopedic doctor. So, then suddenly there would be not a lot about her kidney and liver problems, but about her bones being broken and stuff like that. So, it did help me to prepare maybe like a summary for my healthcare provider.

[00:20:47] For me as a patient, to really give that timeline, it wasn't that accurate. Not accurate enough, because it's, it's, the AI will look especially in free text parts of the PDF files. And what is the problem? That's healthcare providers often copy-paste from their previous healthcare provider. So, if anyone ever made a good summary in words, in text, it often gets copy-paste.

[00:21:17] And so, the AI thinks it's more valuable to take into its own summary. And so, yeah, it's still not that good because it is lacking like raw data. It only can look in those release notes of hospitals. Yeah. How important is it, in your view, to have an overview of, you know, all the past things that have happened in history?

[00:21:46] I'm just thinking from the chronic patient perspective that sometimes you don't necessarily need to have every medical appointment because the history could be summarized in three sentences. And that's enough if you're not a very complicated chronic patient. So, obviously, every patient is different and rare diseases are especially complex.

[00:22:13] But I'm just wondering to which degree, on a broader level, do you think is the full picture with all the documents important and why? So, I'm just wondering if we are obsessing too much about all the documents or is there a reason that you would want to have all of them somehow connected? Yeah.

[00:22:43] I think the reason why I am obsessed with looking at the patient as a whole is because she has a lot of internal things. So, you have organ problems, kidney, liver, which automatically, in fact, influence a lot of other parts of the body. So, I have learned with my research afterwards when I was working as a consultant that there are special types of doctors who like to see a patient as a whole.

[00:23:13] And those are pediatric doctors, geriatric, I think you say in English, so for elderly people, and internal doctors, internists. So, those who look at organs or hormones, they are interested in having all the data of a patient because they need to make connections with one problem is connected to another part of the body. So, they need the whole picture of a patient.

[00:23:39] I've also talked, for example, to orthopedic doctors who do hip replacements. They are not at all interested in looking at every detail of a patient. It actually even makes it worse for them because they have way too much information and they just want to know, did you ever have bone problems maybe? Or they just want to know, did you ever already had a hip replacement? If so, where and when and what kind of brand was your hip?

[00:24:07] So, I think it's like very specific data that they are looking for. But for any of them, it's necessary to be able to ask specific questions to the past data. So, they all need those 408 PDF files. They just need other queries. They need other questions. So, in order to be able to make a kind of dashboard for each of them, you have the same data where you need to start from.

[00:24:35] You will just use other prompting to your AI or you will, if you don't want to use AI, you need to build other kind of dashboards which will respond to those types of healthcare providers or patients. Because as a patient, I also want other dashboards. I might be interested in what kind of procedures did I have and what was the cost of each and who paid me back what kind of money. I mean, it can be completely different.

[00:25:05] Or I want to know an overview of what I need to take as medication and when did they change my medication and why. Or how long has it been that I went to the dentist? I need to go once a year. When was the last time so I know if I need to reschedule? Vaccinations, tetanis for 10 years. When was the last time? Do I need to take another? I will have other questions than the healthcare providers.

[00:25:27] But I have noticed for myself that the main reason why I'm searching for the solution of sharing medical data is because I, as a patient or caregiver, have noticed that my healthcare providers don't know me. And that's something that I hear from every patient. It's that they are fed up with needing to explain time and time again what they went through, whole medical history. We often wait for weeks to see a specialist.

[00:25:55] And the specialists, when you finally enter that consultation room, they start with, okay, tell me, why are you here? And then, you know, oh, no, he didn't read anything before this consultation. And you only have 15 minutes. He's paid for 15 minutes. So you have 15 minutes to tell a story of 11 years and multiple operations.

[00:26:19] So you know that in the end, he will not be able to tell you anything because he was not prepared to have this conversation with you. And a lot of clinical patients bump into this problem of needing to retell their story time and time again. And if you're not medically trained, you might tell a different story that is not medically correct or you will leave out details that might be interesting. Yeah, yeah, yeah.

[00:26:47] That absolutely sounds quite frustrating. Indeed. And what's even more frustrating is that it's not the technology that is a problem. So now that I've been working in it for a couple of years now, I'm also working for three years as a consultant for our Belgian government. And everything is there. The technology is possible.

[00:27:12] So it's even more frustrating that the reason why it's not happening is because of people not wanting to work together, money involved, political things, not having something to gain. So it's all for the patient and for keeping citizens healthy. But there's no return on investment directly for everyone who is involved in this ecosystem.

[00:27:38] And so everyone wants it, but nobody really has neither the power or the money to make it happen. So it's very frustrating from a technical point of view, knowing that it is possible. But in the ecosystem, it does not happen. And sadly, I'm starting to become less naive and starting to lose my energy to keep fighting for this cause.

[00:28:05] So it sounds like you expect to have to keep the role of the caregiver as the integration layer between different specialists that you visit. Yes, indeed. Indeed. Now, still today, I mean, she will be operated the 9th of November every time now. It's for our teeth now. So it's a new kind of specialist. She's 11 years old.

[00:28:34] She gets her adult teeth. And again, it's me needing to tell them like, OK, yes, but she will be operated. She has kidney problems. So make sure this or that. Or they want to do a scan with contrast fluids, which is not good for kidneys. So they don't see the patient as a whole. They look at the teeth. And so they see a problem with the teeth. And they don't look at the medical background because they think it's a child. So it might be healthy.

[00:29:00] And so it's me needing to tell them, but watch out with that scan because she has kidney problems. She cannot have contrast fluid or unless you hydrate her very well before and afterwards. And it's also me telling, like, make sure because you will do holes in her mouth. It might be a reason of infection. She takes immunosuppressive medications. So this is a big risk for her. Things like that.

[00:29:27] I need to be continuously aware of what healthcare providers are deciding to do and if they are aware of her medical history and have looked at everything. So if you look at the last three years since the AI boom has started on the European market,

[00:29:49] I don't see yet like the proliferation of AI-based patient-facing apps as they do, as they exist in the U.S. where patients are also coming with their own scribes and their own apps to the doctor's office to record the chats and just save that and get different recommendations on those apps. But in your particular case, where do you see that maybe AI has impacted you the most?

[00:30:19] Where has it potentially disappointed you most? And also, what do you hope that will be possible because of AI? What a lot of patients, I think, use it for is to better understand their own medical letters so that if they have access to medical discharge letters, for example, to then ask, like, what does this mean? What does that mean?

[00:30:47] I often say it's a better Google way of reading your documents. Well, a couple of years ago when I was reading this, I was Googling some words to then understand my letter. Well, now I could upload it into ChatGPT, for example, and ask me to explain things or to question things. That's, of course, I think the most common used way of using AI as a patient

[00:31:15] is to better understand what has been written about you or also to second get, like, ask kind of a second opinion. Like, my doctor is telling me this. I think I also have these kind of symptoms. What do you think? I'm not really in favor of using AI for that reason without a healthcare provider. Me being myself a pharmacist by training, I can also chat with ChatGPT

[00:31:42] because I have my own medical background and it can give me ideas and make me, like, brainstorm about the data. But I can also see when the AI is going the wrong way, while a patient with no medical history might have wrong, like, even dangerous conclusions. So I especially would like to use it to go to help my healthcare providers

[00:32:07] get data out of her history so that they are prepared to have a consultation with me and they could query her data to prepare themselves for a consultation. But in order to test AI systems, if they are good in summarizing that medical data, I cannot longer do that because I don't want her real data to be fed to multiple AI systems. So I've used Lyft, I've used Humata, and I already feel guilty for that

[00:32:37] because I really uploaded 400 real PDF files with all her data in it, not anonymized at all. So, and she's now 11. If she would go find a job when she's 30 years old and she would ask, I don't know, ChatGPT, what do you know about her and her name? I guess that it could maybe answer her complete medical history because, as I've said, with the Oslo startup, they might not even exist anymore today.

[00:33:04] They probably did not have a good way of protecting their data, and I did fed it everything. So I have talked to a lot of big companies, like a lot of big companies, to give me a solution to anonymize my data in bulk. So I have 408 PDF files, give me a tool that I can run locally, that I do not need to upload them first into a cloud system

[00:33:30] because then I'm also providing my health data on a cloud-based platform, so that I could use a tool to first anonymize my data locally, and then I can play with AI on it. And I did not find a tool yet. So I'm stuck for, I think, almost one year and a half now, that I see a lot of big companies saying, like, oh, we have the perfect thing. You can organize your data. You can do this and that.

[00:33:58] And then when I ask them, like, can I first anonymize my data or can you, like, really say that my data is forever secure by giving it to you first? And then there's never a good answer. So I'm waiting for my possibilities to play with AI to first anonymize my data. And I'm also very scared because I think a lot of doctors are actually also using AI with non-anonymized data. From me, maybe, from others.

[00:34:27] And as a patient, you're not aware where your data is being sent to. Yeah, absolutely. It's one thing to have a national platform where, for example, all the discharge letters are stored, and then there's an audit trail of who accesses the data and when. And as a patient, at least in Slovenia, you can actually ask for an inquiry to get the list of people

[00:34:57] that looked at your medical record. And you can take action if any of the people were not supposed to look at your medical record. But once that data gets kind of thrown outside of any auditable trail, such as generalized models, then that's definitely a potential concern. Yes.

[00:35:26] And if it would be my own data, I would also not share it. So it's not because she's a child that I should now do that. So I'm very aware of her privacy, and I do not want to play around testing AI with her data. Given that you mentioned you live in Belgium, you have to speak several languages,

[00:35:54] you are also following how the European health data space is evolving together with all the other regulation that we have. So how do you feel as a patient living in Europe, which is very highly regulated? How does that impact how you feel about your data and where the technological development is going in Europe?

[00:36:25] Yes. With my daughter, I never had to go to another country. So I never experienced the need of carrying data around multiple countries. But of course, I do know a lot of people who live just on the border with another country, and they often need to go to one hospital in one country and then maybe the other one in another country. So it does happen a lot. And I did speak to a lot of people in other countries. And the problem that I have in Belgium,

[00:36:53] the fact of not being able to share your medical data between hospitals, it exists in any country. There's no other European country where you have a place where all your medical data is in one place and accessible, questionable, whatever. Often you see people presenting like, yes, we have that in our country. And then if you dive deep, it's only for, I don't know, private hospitals or only for people who are connected to this or that network.

[00:37:22] So it does not exist in any country. When we talk about what do you really need to know about a patient when you are abroad, for example? So if my daughter would be traveling to France and she gets in an emergency room, what do they need to know about her? It's maybe not those 408 PDF files. They just need to know what's really important. Like she had a kidney transplant, liver transplant, and her list of medication,

[00:37:51] maybe allergies, vaccinations, stuff like that. So that's kind of the patient summary in every country. But in Belgium, for example, the patient summary is still too much depending on manual inputs of doctors. And so that's not a good thing. That's why we have the AHDS coming up. And I think that's a good thing because this way we will have a structured way of getting a kind of patient summary

[00:38:17] with all the data that is really necessary to share in a structured way. Of course, it takes a lot of effort from all the countries while you have AI that could extract it. So it's kind of contradictory to say, yeah, let's build that, the couple of like, I don't know, 2030. So still four years to go while you already have AI that could fetch all that data

[00:38:46] from my 408 documents and like pre-fill all the structured EDS things. So I think we should use both of them, but the AI on the level of the software that the healthcare providers are using. So the healthcare providers now use software and they are not aware of, often doctors, for example, are not aware that they are putting in structured data. So if they use, for example,

[00:39:15] drop-down lists for a symptom or a problem or diagnose, they might not know the language that is behind it. For example, SNOMED CT code or something like that. So they are not aware that they are really putting in structured data because they often don't see the value of it afterwards. They are doing that and it just looks like administration, but it rarely comes back to them as an advantage. For example, a colleague of them could have done it. You see a patient for the first time

[00:39:44] and you get all that data automatically filled into your system. I think that will be a great advantage of the EDS that you are sharing, but also receiving. And if that could be done in a non-manual way so that if, for example, you are typing or you are dictating free text, I think there the AI should be used at the low level of healthcare providers so that the AI can then at the end say like, I've detected in your text these and these symptoms, these and these problems,

[00:40:13] and these and this medication. Do you want to write that away in a structured way? And then it can be shared to the AHDF. If it could go in this order, I think it will be very useful. And then it could even be more useful if that in itself could be used as a kind of metadata catalog so that the symptoms that are being detected in that consultation with free text are also used as a kind of filtering way

[00:40:43] of being able to go back to that original document in the future. So that in the future, if you want to know, did she ever have this symptom? Yes, that will be structured available. And I want to see the original document about this symptom. And then you can dive deep if you have the rights, of course, to look at that source data. So I think, I hope that this will be the way

[00:41:10] that the HDS will be working and developed. Yeah, yeah. I will check that. Because I'm also currently trying to better understand where we are with the implementation of AI and the requirements and just the progress that's being made. So stay tuned for that part. If I just return to AI for a bit,

[00:41:39] what's kind of the most useful advice that you've heard so far about using AI and as a patient or caregiver? And also what is potentially something that you would advise to others that are using it in that capacity? As a patient, other from just asking questions to chat GPT, I don't use AI as a patient on a regular basis.

[00:42:08] We do have a great example from ZAS hospitals. That's a group of hospitals in Antwerp in Belgium where they are using AI on their discharge letters to not only give the original discharge letter to the patient, but also a patient version. So, and it's already in production. So as a patient, you can download the original letter, but also a completely AI generated

[00:42:36] in patient language letter. And they also have made, I think three different granularities in it. So me as being also medically trained, I can use one that is made for patients, but still has a lot of technical terms which are explained in the document. But you could also say like, I'm a patient who hardly understands anything. So I want a very, very simple patient letter. And so generated automatically by AI,

[00:43:05] which I found very daring because we all know that it makes mistakes. And so it's the biggest hospital of Belgium who has really already launched it. So I think that's a great example of not keeping, not staying afraid to do something with the AI, but really still use it. Of course, it's a disclaimer two page before you see the letter. So it's still not there yet, but I think it's good that they experiment with it

[00:43:34] in boundaries that it's still safe. So another thing, but that's more behind the scenes when they merged multiple hospitals, they use AI to look into their old software systems to find problem lists. And they encoded it into FHIR. So in a structured way to then in their new software system, already begin with a patient record with structured problem lists in FHIR. Those are things that I like to see.

[00:44:04] It's often bottom up from the work ground where they are facing problems and they just experiment and they start with it. While on the government level, we are often still holding back a little bit afraid because of the mistake that AI can make. So you do see a movement of people on the ground. Like I see a lot more happening bottom up than top down and often top down also with the HDS, for example, you're putting out guidelines,

[00:44:34] but often you don't go deep enough to put it into the field. So you have regulation, which is not going deep enough to really say how they should implement it because they are also a little bit afraid and that's okay because giving too much information on how to implement it will restrict the software firms too much and then it's not useful or workable anymore. But giving not enough guidelines will also not make the bigger picture happen

[00:45:04] because it will be too much of freewheeling and so it's finding that right collaboration bottom up and top down to really make it work in practice so that it's easy to use no extra administration and you have a what's in it for me, for every player in the ecosystem. Yeah, yeah, yeah. Better clarity with direct links

[00:45:33] to the primary source. Yes. Especially when you have a complex condition, it very quickly happens that you forget. Maybe you know the name of the drug but you won't know the dosage or things like that when clinicians ask you. It's very useful to have that to the actual reliable point. Yeah, we have a structured medication scheme that is shared

[00:46:03] between healthcare providers but again, it's someone who needs to put it in manually and so it's never up to date and so I have my own Excel file with her medication scheme that I keep updated which is on the Dropbox and which I can share with anyone who needs it at an urgent moment. But again, it makes it necessary for me to be in the emergency room next to her to be able to open my phone, show the Excel file and share it so it's not the best way

[00:46:32] but the medication scheme on our platform is already outdated for more than five years because a pharmacist or a doctor needs to adapt it manually and that's really strange because in Belgium every box that you buy in a pharmacy is registered of course so it could already be useful to just give an overview of what she has bought for the last six months and that might already give an idea of her medication scheme

[00:47:03] and also our prescriptions are all electronic so I don't see technically why you cannot build an automated medication scheme it might not be 100% perfect of course but it will be better than the one that has been put in manually five years ago so if we take the current reality into consideration you know things are the way they are disconnected

[00:47:32] not ideal what based on 11 years of experience of managing data for your daughter and just kind of navigating different conditions what would your advice be on how to best prepare for medical visits what to prepare what to think about what kind of questions to think about and how can potentially AI help people do that

[00:48:02] hmm that's a good question because yeah I had that problem with our nanny if you can call it like that she has a problem on her shoulder already for years and she always tells I've been to so many specialists I've done so many things and still nobody can give me an answer of why I have this pain and how I should solve it and then I was thinking for her

[00:48:31] and I said like okay you're now going to a new specialist in hospital A and she went through so many people and now the question is how can this new specialist in hospital A know what she went through because she does not have that data herself and it will be scattered around hospitals but also private practices and so there is no one place where she has all her data so to prepare I told her like go back and look in your agenda

[00:49:01] when you went where and write an email to every of those doctors or like institutions to ask for your data because you have a legal right in Belgium to ask for your patient data in a digital way I think it's a European right to be able to ask your data as a patient in a digital way and I told her like ask it and give it to that new specialist so that he can find the missing links and do a good diagnosis

[00:49:31] of what you might have as a disease but also what he can still do as research to find for a solution because she already had echoes done things done so but it's a big advice to give to a patient I mean the patient needs to be able to send and then deliver it to someone else I mean you need to be half of an IT person

[00:50:00] to do that and then in the end that specialist will not be prepared because he's not paid to prepare that consultation so the first consultation will be here is all my data and then he will invite you for a second one she first needs to gather her data to be able to then let AI lose on it I also think that a lot of

[00:50:30] people think they have all their data but it's not the complete medical record and so it's dangerous to ask AI to make connections or to find a diagnosis because AI will only look in a part of your data that is available and so it will make maybe the right assumptions but based on not all of the data so the first step that I would say to a patient is gather all your data and store it yourself and then let AI

[00:51:00] lose on it and better than AI let the healthcare provider look at it but please make them look at you as a whole and that's a problem and that's where I am interested in all those personal vaults personal wallets that they are talking about because now I do it on a Dropbox it would be useful to have it in a vault that can also maybe structure it or maybe even

[00:51:30] create fire codes or link it to structured data or make it possible for me to share certain data but not all my data with other parties also if I want to use private apps that need my medical records that I could use them and give them access to parts of my medical data I think that is the

[00:52:10] for me a personal wallet sounds like I can carry my personal medical data with me and give it to someone as a healthcare provider or to something like an AI or an app to get insights into my data so at a lot of congresses you see all kind of flashy things and great technology being able to process your data but we are lacking the first step and that's good

[00:52:39] primary care data and secondary care but without having your whole medical records available the technology might be fantastic but you cannot do anything with it you need to source data first yeah yeah yeah it's a challenge that has been addressed and is known I remember years ago when blockchain boom was happening there was a lot of ideas on how

[00:53:09] maybe the only way to just have a full picture is if the patient has all the data either in an app or somewhere but somehow has an opportunity to integrate with different systems that he is in contact with so that data flows directly into that patient record that is complete but Carlyan thank you so much for

[00:53:39] sharing your story and your insights I guess the summary would be that it's still very important to be very careful when it comes to sharing your healthcare data as personal data especially thinking of the impact that sharing might have on the long run and we will

[00:54:08] definitely be in touch to see how the field develops as it is developing very fast so let's see where EHDS brings us and of course other regulation as well in a couple of years yes I'll try to stay positive and be there when you've been listening to Faces of Digital Health a proud member of the health podcast network if you enjoyed the show do leave a rating

[00:54:38] or review wherever you get your podcast subscribe to the show or follow us on LinkedIn additionally check out our newsletter you can find it