 So welcome here on Carl's West stage on the very first talk of the morning at 11 And I am very happy to introduce to you last Ronald He was previously at NIR specialist Kwonco and is now working for the health think tank HIA of the German Ministry of Health and he will give you a talk about a hackers guide to health care and how to improve life With health data Welcome last Thank you very much Good morning everybody. Thanks for making it out so early It's my first Congress and I'm a little overwhelmed with all the things that are happening especially at night So I really appreciate you all being here so early My name is Lars. I used to be a data scientist. I'm a data scientist by training and now I switched to Bureaucra bureaucracy To policy doing a policy stint my life kind of looks like this right now I work in the Federal Ministry of Health in a think tank that is advising policy sort of bringing in new ideas from the outside and Informing the rest of the world about what's going on inside the ministry and I lead the efforts on artificial intelligence Today, I'm here to talk to you about What to do with health care data Because I think there is a lot of talent missing in health care that technical people that hackers have And I think there's a lot of knowledge missing in the hacker community About what to do with this kind of data and so I'm trying to address five points today and sort of Get you acquainted with some of the stuff that is happening To get you started is anyone here a doctor. Can you raise your hand? Excellent, they're always the worst listeners This is a pathologist what pathologists do is not what you see in crime scenes What pathologists do typically is look for cancer? So whenever a doctor suspects cancer in a patient at some point in the journey They will cut a piece of tissue from the patient They will send it to pathologists the pathologists will harden it with special Chemicals will cut a thin slice of it and we'll look at it under a microscope like that one and they will basically look For cancer cells in that tissue The whole process seems very scientific, you know, it smells like chemicals. There's macroscopes involved It's fairly opaque even the doctor is sending in the original tissue It doesn't normally quite understand what's happening and in the end this pathologist will answer one question Is it cancer and if yes, how many and then sort of the Doctor that send the original tissue will work with that information and As data people we're pretty familiar with that kind of thing because it's a black box, right? It's input data comes in then Black box processing happens here in the form of this pathologist and then some output comes back and So I thought it's pretty natural as data people to ask. Hey, what is the accuracy of this black box algorithm in the form of this pathologist? turns out that's kind of a wild question in healthcare and I found the Results the the answer to the question fairly surprising and I want to share it with you This is a somewhat extreme example is very illustrative But it's in no way out of the ordinary It's a study done at a German University Hospital in Hamburg Where they compare the diagnosis that came in from the outside world Then they had an expert panel of several higher-ranking pathologists looking at the same tissue and Comparing what did the experts say versus? What did the original pathologist say? So we're basically comparing it's kind of a confusion matrix if you've heard that term before Where we're comparing the original prediction of a real pathologist With the closest we can get to ground truth, which is sort of a consensus opinion of experts And the results I thought were relatively shocking Most medical diagnosis fall on a spectrum from sort of oh good to oh god With this example, it's about prostate cancer If it's anything worse than oh good, so if it's sort of on the right of oh good Then your prostate will probably be removed to sort of keep the cancer in check And in the event that you have a prostate about half of us do You kind of want to keep it So you know having your prostate removed is not something that is a very pleasant experience You know it's it's a very Not nice surgery to have leave a lot of damage in the body so sort of Having a wrong diagnosis here is Important the first thing you see is that about two-thirds of all diagnoses the green bars in the middle are exactly correct Which means that one-third of diagnoses are actually not exactly correct And you can say hey, you know it doesn't really matter But what does matter is the case is from oh good Upwards because that's where you have a good surgery and not get surgery and what this data shows Is that from these five thousand four hundred patients? Out of those that actually by consensus opinion did not have cancer did not have prostate cancer or did not have an Immediate need to treat one-third of those patients had their prostate removed without the need to do that So it's a you know pretty drastic outcome for patients That You would really like to avoid now the question that I asked for myself is If this were the result of an algorithm of a machine learning algorithm or some other algorithm Would this be acceptable performance Maybe not but for humans as best we can do but I think being more transparent about How humans are just humans and maybe not particularly well suited to look at pictures all day I think would be a good thing to have more clarity in the healthcare system about what diagnoses actually mean and so this is my starting example This is called in the medical community. This is called intra inter observer variability So different observers look at it and their diagnoses vary depending on who looks at it Intra observer variability. So you show the same picture to the same doctor at a different time of the day also exists and it's not much better The example I showed you is Very illustrative but as I said this happens in a lot of medical fields Not only for prostate cancer, but for all kinds of cancer for all kinds of illnesses And again, it shouldn't be surprising doctors are not bad people doctors are just humans like us all and so mistakes happen And sort of it's one example of these questions that I think can be answered with data and that currently Maybe not explored enough another example would be who gets treatment when so who has access to doctors How long do they have to wait for appointments that kind of thing? How unhealthy are waiting rooms? So as people are waiting at their local doctor's office to be seen often for hours next to many sick people How unhealthy is that really? Does specialization into more and more specialized doctor's practices actually improve quality of care? Or does it maybe detract from it because people lose oversight? Can we predict the best course of treatment for a given individual to personalize medicine or? Does the probability that a given patient receives hip surgery so gets a new hip Does that depend on the reimbursement the doctor gets ie the price tag associated with that surgery? These are all questions that you would like to answer with data and the problem in healthcare is what I call the great irony of healthcare data Which is roughly this sort of the more legitimate use would be the harder It is for all those actors to get access to that data and you know the converse holds as a case in point in Germany in the year 2020 most patients still do not have some Official plan of their medication what meds are they taking this is important because you have these cross-effects inside effects of different Medication and when they're hospitalized you want to know what they are taking and when you prescribe a new piece of medication You want to compare that to what they're taking already? the Really active really engaged patients in Germany write these little papers by hand where they write down what has been Prescribed to them at some point and just carry that piece of paper with them And then if they go to a hospital and they get out of hospital again They would have to update that piece of paper. Maybe they do maybe they don't it's just not a good system So as a doctor who is working with patients every time I see new patients Have it really a hard time finding out what they're actually taking already and what I can prescribe to them Then we have this phenomenon of consumer electronics companies increasingly going into the health data space In a very smart way, and I think there's a lot of good things happening here but basically declaring their health care products as lifestyle products thereby circumventing a lot of those pesky Health data data privacy problems are saying this is not actually health data. This is you know lifestyle data And getting more access and you can sort of think you know why does Apple need to have health data At the same time, I think a lot of good things are happening here Patients are becoming more and more Empowered to make their own decisions. They're learning more and more and I think overall. It's a good thing The worrying part about this is that traditional health care providers might actually lose track of Or sort of lose the connection to the consumer electronics competition And then finally as we heard yesterday Health data does get lost and it's a terrible thing. This is probably the worst story I've ever heard about Singapore lost a whole database of HIV People with HIV and were started for it And you really don't want that to happen So what are the the learnings of a hacker community and what are the things that I want to talk about here? First of all, I want to say very clearly Keep hacking, you know, you don't want to lose this data and I think having people like the C2C community Being wide-hat hackers and making sure that people say on their toes is very very important for the safety and security of our health Care data because I don't think we will go back to a world where this is all on paper This will be digitalized and we will need to make sure that it's safe The second part I want to talk about is a little more subtle In data privacy under GDP are you have this idea of consent? So patients can donate their data patients can say you can use my data to do something and this Conversation currently especially in Germany, but all over Europe is very much driven by this idea between narrow consent and broad Consent where narrow consent is this idea that as a data subject I can say very specifically what can and cannot be done with my data And so I need to allow every single step of processing my data Whereas broad consent would say, you know, hey, I can say my data can be used for all kinds of cancer research and The issue that I see is that the idea of narrow consent is beautiful in theory and is a great legal idea But in reality leads to situations where for example a patient is being asked to sign a data usage form in The waiting room of a hospital by a study nurse I don't think that's the idea of freedom we had when we first said we want narrow consent because you have this power dynamic The patient wants treatment that is not a free choice what to do with the data And I think there's a trade-off here between narrow consent and being very specific what you can do with data and the user experience part of it of how you can give that consent and I think being more clear about the Chances we have with a broader idea of consent but having better user experience is something that would be extremely helpful in health data And in health care sort of similar to this idea of the cookie banners you get everywhere Where having one switch per tracking cookie at least for me doesn't really help me much I want to say cookies. Yes cookies. No on a very simple one-click solution and Then finally I want to give you some ideas of legitimate uses of things that you can do with health care data in maybe kind of an open data style Projects that you can work on to answer for example the questions that I started with So to get you started on something I thought I would tell you about a few data sets that I think are cool and Allow some hacking the first one are these two On the left you see a famous data set for pathology. So what we started with this is what? Human tissue when you sort of cut it and dry it and cut thin slices of it and look at a microscope and color it looks like is These pictures on the left you get them with a diagnosis and you can train computer vision algorithms to Hopefully improve on the quality of diagnosis that we currently have On the other side just to show you another example. These are pictures of birthmarks and The question is is there melanoma so that skin cancer or not both of these data sets are relatively well studied and great places to Start if you're into computer vision Then this is a bit of a German specific thing because health care is organized Nationally and so different countries will have different data sets But these are two data sets that I think are very interesting and probably underutilized the one is the so-called DRG browser where You're actually intended to download software which then allows you to see certain aggregate statistics about What billing codes are used for procedures are used so it tells you a lot about what health care is rendered in which numbers in Germany so the example here is actually birth and Under the billing code for birth what procedures were done so what specific procedures were happening you can download that software online Turns out that all the data that software displays is just in CSV files and install directory. You don't even need to install it But it's very interesting data to just get your fingers dirty on And the everything is a so-called quality Berichter is something that all hospitals have to publish and somehow I don't think this has gotten much attention in the open data community yet Tells you a lot of data about what types of procedures what types of diagnoses Hospitals are seeing and then some so I don't think this is necessarily super helpful for their Quality metrics that they're publishing there, but it tells you a lot about what kind of patients hospitals are seeing And if you pool this with other data like census data, I think there's a lot of interesting things you can do here this is another Last example to sort of get you thinking a little bit out of the ordinary on the left you can see that certain cities have published the Availability of their emergency rooms in different hospitals So basically for a given hospital in a row per time of day You can see how much room they have in their ER for specific diagnoses So you can say hospital a is getting flooded with gynecology cases, but they didn't like a for bone fractures And you can do all kinds of things like try and build predictive models on this Maybe somebody wants to archive the data somewhere I think there's a lot of interesting things that you could do with this type of data Of course originally it's intended to be used for ambulances to see which hospital to drive to And then finally on the right is a very well-studied example from the US Probably one of the richest data sets that we currently have in health care Mimic is a data set of patients in a hospital system on the east coast That contains almost everything that happens in a hospital so lab results doctors notes Medication all of these things that in Germany we don't really have access to are in this data set for research in a de-identified manner This data set is not public public, but extremely easy to get access to so another interesting case to dig into and see what's possible and There are some privacy concerns, but given that this data set is widely used in research already There's a lot of papers about it. I think this is a relatively innocent example to start with Now with all health care data there's a lot of issues and I want to talk about them briefly I See three main issues. The one thing is ground truth Is often lacking it's unclear what ground truth actually is and if you remember the example I started with you know, you have an original doctor's diagnosis and now you have a consensus diagnosis is that consensus necessarily better Why is it better if I think doctors agreed on it? You know, maybe they were wrong and the original doctor was right It's very hard to really find out what the right diagnosis would have been and I think a lot of research is lacking in that regard and this sort of carries over to a lot of different data points where It's extremely hard in health care to just trust the data that you have then second Semantics are surprisingly difficult in health care. What this means is You know somewhere in a hospital they store lab results for example, and there's just no standard whether they use milligrams per liter or Grams per cubic centimeter or there's all these different options and this one example sounds trivial But this is all over the place. How do you store a certain diagnosis? Is it the flu? Is it influenza? Is it influenza a? there's just many different ways to code this stuff and Having semantics, which is called sort of the mapping of what people use to what you can actually work with as a data person Is still I would say in its infancy And then finally and this leads me to my last two points There's a lot of sampling bias and resulting issues of representativeness And I think sort of deep down this comes from the fact that health care data is personal data And so, you know, you don't you can't just go and get access to all health care data for all patients so often you have these examples where Some data sets surfaces like the mimic data sets And then that's all you have and you work with that and that leads to a lot of representativeness problems in medical research, but I think in general in a lot of health care data projects. This is one example a Paper published in nature this year, which is very impressive. They predicted some really cool things But the only data they had was from the Veterans Association in the US. So a separate hospital system for US veterans And as you can predict their 94 percent male So Learning algorithms on data that is 94 percent male sort of Generalization error will be an issue and you would like these algorithms You would probably like these algorithms to work as well on women as on men At the same time, what else are you going to do? This is the only data that's available So you can't really blame them, but it's just an issue that is around in health care and that you need to take it take into account There's one thing that this also implies which is it's very hard to Certify that a certain medical device works as intended Because the regulatory bodies also don't have fully representative test data You basically going to have to believe the manufacturer the vendor of a medical device such as an algorithm that They used appropriate test data and that they use representative studies Now traditionally in pharmaceutical research, you run these randomized controlled trials I you have a status quo you have a new drug you sort of randomly give you a new drug to 50% of patients You use a status quo on the remaining half and then you compare the two groups Still the gold standards and evidence But extremely hard to do this in a representative way because you need to recruit patients to voluntarily take part in a trial It's expensive and sometimes it might not be possible because for example, you might be interested in how cancer drugs work in children We think that running medical trials and children is unethical So how are we ever going to get to a point where we can learn about the efficacy in children? And I think a really cool opportunity for evidence that we have with algorithmic data and algorithmic medical devices here is That we can collect test data as a regulatory body And so I think this is something that is currently being discussed internationally by the World Health Organization on the European level with the new commission and That we are doing a lot of work on which is this idea that Governments and regulatory bodies should collect test data in a representative manner So what would happen is on the top row? You sort of see the normal workflow here for deep learning, but it could really be anything There's public data. There's private data people train their algorithms on it And in the end they come up with a model on the far right and then the regulatory bodies Say on the European level would have a test data set that is actually secret That is sort of not to be shared with anyone that has a high quality standard And this can be achieved for example by mandating that you collect this data from hospitals all over Europe And you just say hospitals have to submit say one percent of the images that they have Into the secret body and now I have representative data that I'm not sharing and so I can use it as test data And I also can keep doing this over time so that I can account for population shift so for example in skin cancer you could say that in Germany the average patient in the next 10 years might have a Slightly darker skin tone than the average patient in the last 10 years because of migration patterns And so if you're working with algorithms you want to account for this population shift And you want to make sure that the algorithm that you certified Looking back on test data from the last 10 years will still work on the new patients of the next 10 years And so this will be a way to keep collecting this data and re-certifying algorithms Which finally it leads me to my last point before we open to questions I think what you have here is Just accuracy in the end But this doesn't really answer the question of what fairness means and I mentioned earlier that You know you want an algorithm that was trained on ninety four percent men to work as well on women But what does as well actually mean? You know is two percentage points difference good enough is four percent good enough sort of Not super clear from the outset what fairness means in these things And so I think this is a broader issue that we should discuss as a community Beyond health care, which is quantitative fairness, which is the idea that More and more decisions are made by algorithms instead of people Traditionally when you look at discrimination and fairness you try it to sort of Empathize with the actor So the judge would in the end the judge would say did he mean to discriminate or did he not mean to discriminate? With machines that doesn't really work anymore And so we need new ideas to describe what fairness looks like and what discrimination looks like and I think Approach is to quantify the fairness of a decision Exist the literature is relatively mature. This is one overview But somehow in Europe we don't really have that debate yet And I think it's long overdue that we start having this debate for health care and outside So my talk on a nutshell First of all keep hacking second of all think about what informed consent really means for data third of all do contribute legitimate uses The data I showed you the questions. I showed you our next are in the starting point Reach out if I can help with anything to connect you with data or ideas fourth demand evidence for medical devices and I think trying to Seize the opportunity we have to get better evidence is a really big chance and fifth promote quantitative fairness. Thank you very much Thank you last for your talk. So we actually do have a lot of time for questions We have two microphones in the hall and the signal angels. So please line up at the microphones and We start off with a question from the internet. So please signal angel wants to know if you You Yeah, I'm familiar with the literature Is not one of our main topics because we focus on digitalization So the checklist ideas pretty pretty convincing actually the idea was that in aviation Pilots were at some point required to fill out these checklists before takeoff entering flight and say, you know Yes, I did check this I did check this I did check this and sign it Which pilots hate it because you know makes you do things that you don't really want to do They were forced to do it and a dramatically reduced Incidences in an aviation and now I took a one that argues that the same thing should hold for doctors Because doctors also hate this idea that they're subjected to processes But checklist to Reduce a number of incidents as it happened The idea is relatively old and I think has been taken up in medical guidelines a lot But we don't specifically work on that because we focus on digital aspects Okay, microphone. Let's call it one How would you? legislate or decide what kind of Consent would be useful for medical data Currently it is I mean there is legislation around it It basically comes down all the way from GDPR down for German legislative bodies Currently your health data is deemed a special interest private data And so you need very specific narrow consent and One option that you have for health data in particular is that GDPR actually allows for exceptions to this idea of consent where you can say if a Important social interest stands against sort of your private interest and keeping your data personal You can say this data is available for research use so specifically for research This is something that Can be done, but I think You know you because it's so sensitive you want to give people the opportunity to maybe still opt-in But in a different way, maybe in a broader way or opt-out And I think that's why I'm talking about broad consent because I think narrow consent currently is too narrow And there would be ways to go broader and that would significantly Facilitate access to this data for health care Okay microphone to So you mentioned at the beginning this patient that had cat so it was diagnosed with cancer But he didn't had cancer was Operated but I don't think that was the fault of the doctor because this reduces to statistics you have the Precision record rate of like if you want only the patients that have truly cancer you have high precision but you miss a lot of patients that Have cancer so or if you want our patient with cancer you will have a lot of people that Don't have cancer so and the same problem you also have with machine learning. So you have the precision record trade-off and This reduces to actually two problems. You need better measuring devices so better devices that reduces this error the variance and of course more data so And self is not the problem of the Doctors statistics. So first of all, I agree with you. It's not the problem with doctors They're doing a fine job and doing the best job they can But sort of as data people if you look at this naively It's kind of weird that you have humans looking at pictures all day and are expected to diagnose a hundred percent correctly You know all day every day. They just look at pictures. That's not what humans are good at So it shouldn't be surprising that mistakes happen, but it's not their fault. It's just they're they're human The second part is yeah, you're right There's precision and recall, but what I'm arguing is that humans have poor AUC in their predictive qualities And we would want to have better performance In this type of algorithm Okay, please please on size questions. Okay discussion. Okay It's it's their follow-up question to that. Yeah, wait. Do you say then the threshold? Do you want more precision or do you want more recall? You want more AUC? Okay, is there a question from the internet that Sorry, could you repeat the second half of the question? You Yes, absolutely, I think first and foremost the medical practice often does not yet Sort of live in the year 2020. So I think a lot of gender bias exists implicitly in how treatment still is delivered So first of all, I think using existing healthcare data is helpful to show these biases and demand them to be reduced and then second of all, I Mean I can't predict the future But I could imagine that you find all types of understudied groups of patients that maybe historically because they didn't sign up to to be part of RCTs randomized control trials Were understudied and you can find gaps in the care that these people receive and women are one example But I think ethnic minorities would be another age groups would be another I would expect a lot of these type of insights to surface Okay, microphone one really short question. You mentioned this German quality assurance data set What does it contain except? besides admission rates does it contain outcomes or risk Adjustment etc You mean the the DIG data set? The DIG data set actually only contains aggregate data on the total proceed So the total episode so what was the total episode build ass? So there's a billing code and then there is procedures linked to that Other data sets exist some of them private some of them public access But I think if you want to have more detail on all the admission data you're moving into a world of currently private data or Protected research data because it's very hard to anonymize this data. And so there's always a risk of re-identifying it Okay, so mark for one again. Thanks. Thanks for your talk. I like the idea quite a lot with Building up representative data sets obviously But for me the question stays in my mind what you are achieving is a representative Data set but not necessarily also a high quality data set especially if you force this data collection up on Hospitals then you could end up with a really messy data set What do you thought about this? Very good point. I think it's not enough to just collect this data I think you would also need to actually invest in the quality of this data so you would basically need to hire an expert panel to Improve the data that you're collecting which means is a very expensive overall process You will not be able to do this for every type of diagnosis that you're interested in But only for the bigger fields that are becoming more and more mature, but one example. I think Memographies, so breast cancer screening Is relatively mature as a technology. There is several companies going on the market now I think there'll be one field where you're starting to collect International test data will be very feasible and probably worth it I'm looking at the signal angels there are questions from the internet No, okay, then mark from two please So looking at this from the perspective of the individual whose data is supposed to be these data sets in the end Is there any work being done or do you have ideas on how to so to say soften the blow if your data becomes public? this could be either unintentional if for example data leaks or Also, if I just agree to have my data be part of a public data set What kind of protections to have as an individual in that situation because I think that that Is obviously one big reason why you don't have access to sensitive data because it is sensitive and this is sensitive often for reasons that are Maybe something you could address with policy Yes, and no so first of all, I think what we're actually lacking is German verwendung's verbote so Penalized protection that even if you have certain data you cannot use it So, you know if certain types of data fall into your lap or if you happen to accidentally re-identify anonymized data You're obliged to delete delete that data Which does two things it makes the data less valuable on the markets So hacking for them becomes less interesting and second of all Those actors that are sort of in good faith will help protect that this data doesn't you know move through the world but I think They kind of fall short of really protecting you I would say one part is people tend to be a little too scared I personally think about what can be done with their healthcare data So I would be pretty careful with genetics data because we know there's a little information in there We don't know what data is in your genetics data. So yeah, maybe not And there are certain stigmatized data points like sexual health That you or me or psychological health that you maybe don't want out in the open, but besides that I think Yes, there are certain issues These issues we can address with legislation may be better than we are so Quantitative fairness can also go to health conditions and can say, you know, you cannot discriminate against certain health conditions as an insurance company That would take a lot of fears away that people currently have and then finally, I think the Idea is my idea is I'm willing to share my healthcare data if you're also sharing your healthcare data Because I think in a world where everybody's data is open a lot of the risks are already mitigated so I think being Conscious about how to do this And maybe not starting with sexual health data not starting with psychological health data Would be a way forward Okay, are you curing for the microphone microphone? Hi my question is basically that giving all the concerns that people have with Disclosure of data being detrimental to the personal privacy and issues Are we already at that point where we really need that? Considering health if there's a toxic substance a lot of like health issues are caused by needing to Analyze data on how toxic stuff would be in certain circumstances in the public sphere like Have we done all data disclosure that is not private that is not linked to an individual that is linked to male companies having a business model that generates health issues for the public and are we already Disclosing this data first before we harvest my data like how Does this count in to because in Germany if I'm recall right there was this issue of Coffee makers and there are the coffee brewing and they investigated that the cleaning in the morning of this let some Toxic substances lead in the first brews of the coffee, but they wouldn't disclose this data for saving the companies Not having a bad business model there So the question is is my private data the last resort here or is there a headway and wiggle room for improving our health without going harvesting my data first Why can't why does it have to be either or I think Asking for more aggregate data to be practically released is a good idea. I showed you some available aggregate data sets You can probably think about more But in the end I mean make no mistakes these aggregate data sets come from pooling individual people's data So you need to collect it at some point to then be able to do your analysis on it So I think even current ideas say, you know Maybe you cannot actually even as a researcher get access to one individual patients data But you can run queries on a lot of patients data and since you can't predict the queries that are interesting You need to collect individual data at some point Okay, Michael from to yeah Thank you for the talk, but you mentioned mammography as a good example for using Computerized detection and Any improvement so we had two waves of automatic mammography in the last 30 years the first one was a total mess cause the techniques were not safe enough good enough and They differed from side to side called they used analog techniques and things like that and the next wave was about 10 years ago where we used automated data sets and all the things that we had and It ended in a total mess too cause in 2015 when we did the first revision of this idea We saw that the people that used it either in the US or in the European Hospitals had the problem that there were too much false positive ideas. So we had too much too too much recalls and I think we lost a lot of trust and so how do we How can we avoid this and the next time we use it? I think what you're saying is exactly why I'm suggesting that we need Regulated test data to avoid making these mistakes on patients I think that I would personally think that technology currently is relatively mature And there are several companies going on the market again with these types of products and so precisely because In the past it didn't always work as an as intended you might be interested in ascertaining as a regulator that you have good test data and I think Reimprovement of the test data is a crucial point as you mentioned Okay microphone one, please Some important health data we have in Germany are the registers like the cancer register but they are organized on a federal level and we Had a lot of problems getting them running in a good way and now we have the new possibilities of the electronic patients file Coming up and also when we will have a change on our regulation with organ Transplant stuff. We are going to have some sort of Organ register like am I willing to donate or not? And are you thinking about is this register going to be? organized on a national level or on a federal level and do you think that the future of the electronic patient Yeah, the APA patient file will change the way the Registers like the cancer register are organized. Do you think that there will be some new interaction and how we use this data? Yeah Personally, I really hope so. So maybe for those that are not familiar Germany, I think like other countries has these Registers where we collect data on individual diagnosis To be used for research. So we say hey We have a lack of understanding of a certain cancer type or of organ transplants And so we collect data specifically for this purpose under a sort of a regulated exception And yeah, absolutely. I mean in a world where we have a national electronic medical record You would hope that this registry data can be included there Partially also because the electronic medical record could offer consent management where patients could be polled To give their consent for certain uses and having that on national infrastructure in a secure environment would actually be very desirable The second part of your question Who knows You know, who knows what's gonna happen with the national electronic medical records there They're not really available yet, and we don't know how people use them But in an ideal world, I think they could be used as a central information storage and sharing opportunity Not only with doctors, but also if the patient wants that with registries and with researchers Did I answer your question? Yeah, thank you just not the organ thing if you think that's going to be national or not But that's not my field of expertise. Unfortunately, I can't tell you that Okay, my form one again Hi, thank you for the talk. I'm I attended the presentation yesterday about the APA and so the patient file the case file and What was explained there is that they don't want to take the efforts for Actually having the option Opening the file for certain doctors just for like or certain parts of the own file for Doctors, it's more about all or nothing what you have to choose as a patient first when they want to implement it and Considering that private health companies already implement systems for data collecting with a Automated Solidarization for the the data which they send to research Centers so to say is there any effort by now from the German government or certain Institutions to Follow such an idea of Psydamize Psydamization to collect the data and keep the individual data with the doctors and if not, is there a Or where's the best place where to start for promoting such ideas? so the idea of pseudonym pseudonymizing and anonymizing health care data, of course is is widely spread and as well understood in the government I Think for the German patient records the national EMR That doesn't really well it could have worked but we decided to actually have central storage and guarantee privacy by encryption Which has certain advantages in particular I think Me personally I'm much more comfortable with my data being in sort of national infrastructure Then being in doctors offices Seeing the IT security of the typical doctors office in Germany So I think there's a lot to be said about having that in a national infrastructure type place the second part I Think the option so you what you currently can do is you can share all your data with one doctor You cannot choose what parts of your data in the current specification. I think the criticism has been heard and it will be possible to Share only selected parts of your data But again, I think what what doctors are saying on the other side of this discussion is What good is a subset of a patient's data? The patient might not know what parts of their previous diagnosis. I actually need for my work So I think the idea of withholding data from doctors is very unpopular with doctors and I think it might be much more of an opt out type situation where I don't want my Sexual health information shared unless it's explicitly needed, but everything else is okay Then it's an opt-in situation where it choose every single piece of data that I have in my patient record Okay, may I con size question? Yeah, okay, so Um When it doesn't really answer for me the question about the Psydomization because I think there's a lot interest of Yeah, gathering the data Psydomized to do research. So why couldn't be this a first step? Because with the new EPA coming If you have the access or if you have signed up for this electronic case File let me interrupt you for a second. Yeah, what you're saying with pseudonymization is totally right In the law that went into effect. I think two weeks ago the faugue digital effort organs because that's We actually installed a center to collect research data from health insurance companies And we have actually written into law a mechanism by which this data is pseudonymized So pseudonymization is in the law and it is being used Sort of from a data privacy standpoint that is just De-identifying that is not anonymizing because using external data. You can still re-identify the data and so it's sort of It's it's a hygiene factor, but it doesn't solve all these privacy issues that we have And so I think yes, we're doing it But at the same time you need consent and you maybe need in certain circumstances Effective anonymization techniques to really solve this bigger issue Okay, thanks. Okay. So this concludes our Q&A and thank last Very much for the talk and for the extensive Q&A ground of applause for him. Thank you