WEBVTT 1 "Britta Haenisch" (1929526528) 00:00:00.000 --> 00:00:20.000 Today we are looking forward to an exciting presentation by from our Refreq project, more consortium, followed by a round table discussion with different stakeholders from the regulatory, academic and also patient's perspective. 2 "Britta Haenisch" (1929526528) 00:00:20.000 --> 00:00:25.633 And now I would like to hand over to Claudia from Informat. 3 "Infarmed" (3012685056) 00:00:25.633 --> 00:00:58.250 Good morning, everyone, and on behalf of real prog consortium and informat. Welcome to this real prog workshop. Just a few housekeeping rules. Let me just share the slide. The workshop is being recorded as you noticed. During this workshop, please keep your microphone muted when you are not speaking, and if you have any questions or comments during the workshop, kindly use the chat to ask questions and provide. 4 "Infarmed" (3012685056) 00:00:58.250 --> 00:01:22.220 The the comments. As Rita already mentioned, this workshop is about lessons learned on exploring real data in relation with real forec project and this workshop focus on the description of the study population for ALS and breast cancer. We will start with the presentation with. 5 "Infarmed" (3012685056) 00:01:22.220 --> 00:01:42.220 1st result and challenges in all of implementation by Vera Erinstein and Mohammed Abulagin followed by a panel discussion. I will present briefly Vera and Mohammed. Vera is the scientific lead of work package one in Real foreign project. She's a professor for phase four studies. 6 "Infarmed" (3012685056) 00:01:42.220 --> 00:02:05.670 At Department of Clinical Epidemiology in Harrows University of Denmark. She has served a scientific coordinator of several large multinational multi database studies on safety of medicines and vaccines. Multi database study, including regulatory mandated and industry commission studies. She has experience working with major European databases including nordic registries. 7 "Infarmed" (3012685056) 00:02:05.670 --> 00:02:25.670 And databases in the Netherlands, Italy, Spain, United Kingdom, Angry France, Germany, United States, and most recently Portugal. She has served on the board of directors of the International society society for pharmacomiology and is currently a member of the steering group of the European Network of Centers for pharmacoupid. 8 "Infarmed" (3012685056) 00:02:25.670 --> 00:02:44.580 And pharmacovigilance and set. In that capacity she is continuously contributed to the development of methodological guidelines, most recently as the coordinator author of the guideline for conducting safety studies in pregnancy and breastfeeding using rerule data. 9 "Infarmed" (3012685056) 00:02:44.580 --> 00:03:04.580 Mohammed Abraghi is a researcher and PhD candidate at Sky. His work focuses on biomedical data science and artificial intelligence, particularly in the context of complex and neurogenerative diseases. Mohammed has worked with large scale data sets to. 10 "Infarmed" (3012685056) 00:03:04.580 --> 00:03:25.993 Such as UK and PPMI, and he's currently actively involved in the Real Foreig project. He has a background in pharmacy and a master degree in life science informatics from BIIT in Bond. I will now hand over to Vera to start the presentation. Thank you. 11 "Vera Ehrenstein" (7209472) 00:03:25.993 --> 00:03:33.452 Thank you very much to Claudia for the nice, introduction. Can you please confirm that you can see my slide in the presentation mode? 12 "Infarmed" (3012685056) 00:03:33.452 --> 00:03:36.331 Yes, we can see. Thank you very much. 13 "Vera Ehrenstein" (7209472) 00:03:36.331 --> 00:04:02.390 So good afternoon, everyone. Thank you very much for attending this workshop where we present the initial results of use case one in the real product project. I would like to gratefully acknowledge my colleagues of the work package two, associate Professorbet Harvag Puju and Statistician Bacos and also all the collaborating institutions and colleagues who have contributed with their hard work and expert. 14 "Vera Ehrenstein" (7209472) 00:04:02.390 --> 00:04:19.290 These into this presentation. So let's see if I can advance now yes she's taking one step back and presenting the real forward project as a whole. 15 "Vera Ehrenstein" (7209472) 00:04:19.290 --> 00:04:35.579 The realport aims to develop, optimize, and implement artificial intelligence methods for real world data analysis in regulatory decision making and health technology assessment along the product life cycle. 16 "Vera Ehrenstein" (7209472) 00:04:35.579 --> 00:04:50.639 This project is four years long and has started in 2023 and we just hit the finish stretch, we're about to complete this project at the end of this calendar year. 17 "Vera Ehrenstein" (7209472) 00:04:50.639 --> 00:05:08.489 This is a large scientific consortium consisting of ten partners from six European country and it is le led by Beefarm, whose leadership I also gratefully acknowledge. So. 18 "Vera Ehrenstein" (7209472) 00:05:08.489 --> 00:05:27.959 Just to give you a brief overview of the databases that contribute to the scientific results of Real for Reg, they are presented in this table, the data in Real for Reg come from four different countries from Denverk and Finland. 19 "Vera Ehrenstein" (7209472) 00:05:27.959 --> 00:05:44.699 Representing Nordic registry data from Germany representing claims data and from Portugal also representing a registry type data. All these data are accruing on individual level in the. 20 "Vera Ehrenstein" (7209472) 00:05:44.699 --> 00:06:00.239 Automated secondary national regional electronic health records database. And this is an example of the real world data that are increasingly used in both post authorization and pre authorization space. 21 "Vera Ehrenstein" (7209472) 00:06:00.239 --> 00:06:17.639 The, the overarching objectives of the Real Forect project is to develop new and optimized methods for analyzing real world data, including addressing the challenges and. 22 "Vera Ehrenstein" (7209472) 00:06:17.639 --> 00:06:33.539 Taking advantage of opportunities across different health care systems. Ultimately, we hope that this will support better decision making about medicines and other treatments and ultimately benefits patients and public health. 23 "Vera Ehrenstein" (7209472) 00:06:33.539 --> 00:06:49.319 And here you can see here some of the specific objectives of real for Reg is to enable the use of real world data in real world evidence to establish or better establish its value in the society. 24 "Vera Ehrenstein" (7209472) 00:06:49.319 --> 00:07:09.319 And to impact how real world data are used and analyzed, and this project aims to provide solutions to those objectives. There are four use cases scientific use. 25 "Vera Ehrenstein" (7209472) 00:07:09.319 --> 00:07:28.799 Use cases comprising real four egg. So the aim again is address the needs to demonstrate the potential of real world evidence in regulatory decision making and health technology assessment, and in that real four is employing these highly relevant use cases. 26 "Vera Ehrenstein" (7209472) 00:07:28.799 --> 00:07:47.249 Some on the disease side and some on the medication use side. So use case one and use case two are addressing the pre authorization and evaluation side of the product life cycle and use case. 27 "Vera Ehrenstein" (7209472) 00:07:47.249 --> 00:08:07.249 Three and four which are not focused today's presentation are addressing the post authorization side. So let's delve into the main topic of today, use case two. Let me talk a little bit about the background and the rationale of use case. 28 "Vera Ehrenstein" (7209472) 00:08:07.249 --> 00:08:26.009 This one. As some of you who are professionals in this area, they have probably seen the sentence many times in the papers saying that randomized controlled trials do remain the primary requirement for the initial authorization of treatments. 29 "Vera Ehrenstein" (7209472) 00:08:26.009 --> 00:08:43.439 Based on their efficacy and initial or short term safety. Real world evidence, and under that we understand evidence generated from real world data has been traditionally and already for a long time used. 30 "Vera Ehrenstein" (7209472) 00:08:43.439 --> 00:08:59.009 For assessing post authorization or long term safety and effectiveness and also for effectiveness monitoring of treatment. However, regulators, including the European regulator, the European Medicines agency. 31 "Vera Ehrenstein" (7209472) 00:08:59.009 --> 00:09:19.009 Have started to recognize the potential of real world data, so data accruing outside randomized trials to complement data from randomized controlled trials in regulatory decision making making, including not only post authorization but also pre authorization. 32 "Vera Ehrenstein" (7209472) 00:09:19.009 --> 00:09:35.219 And the value of real world data is especially potentially high in areas where randomized controlled trials have limited applications and that applies to rare diseases. 33 "Vera Ehrenstein" (7209472) 00:09:35.219 --> 00:09:53.009 Severe conditions, vulnerable patients subgroups and situations where there is lack of clinical equipboys. As traditionally these types of situations are not part of randomized clinical trial populations for a good reason. 34 "Vera Ehrenstein" (7209472) 00:09:53.009 --> 00:10:12.029 Because they, they exclusion of those increases internal validity of randomized clinical trial results. We have also those of us who work in this area have also been aware of some of the barriers. 35 "Vera Ehrenstein" (7209472) 00:10:12.029 --> 00:10:28.619 That stand in the way of use of real world evidence in the pre authorization and evaluation phase. Some of them are listed here, but one there is lack of awareness and knowledge among not only. 36 "Vera Ehrenstein" (7209472) 00:10:28.619 --> 00:10:47.129 The public, but also in some professionals about the available data sources. Number two, among those who know something about the data sources, there is some skepticism, sometimes justifies and concerned about concerns about data quality and validity. 37 "Vera Ehrenstein" (7209472) 00:10:47.129 --> 00:11:03.119 And those researchers who work every day with their hands with those data are really acutely aware about sometimes how often slow data access. 38 "Vera Ehrenstein" (7209472) 00:11:03.119 --> 00:11:19.769 4th studies even if the data are collected for regulatory and health technology assessment purposes. So, as I mentioned before, use case one is one of the use cases in workpage package one. 39 "Vera Ehrenstein" (7209472) 00:11:19.769 --> 00:11:35.519 And here, we, start by a very relatively simple analysis. We would like to conduct a basic epidemiologic study where we describe using real world. 40 "Vera Ehrenstein" (7209472) 00:11:35.519 --> 00:11:55.049 Data in several countries populations of patients with amiotrophic amiotrophic lateral sclerosis, ALS and breast cancer. And these types of studies are often conducted in our world to to understand the clinical course. 41 "Vera Ehrenstein" (7209472) 00:11:55.049 --> 00:12:11.999 And the prognosis of the diseases in the population. Why did we select ALS and breast cancer as kind of to showcase the use of real data in pre authorization stage? There are several reasons. Number one. 42 "Vera Ehrenstein" (7209472) 00:12:11.999 --> 00:12:30.989 These two diseases represent two diverse or I should say different pre authorization settings in which real world evidence use maybe especially relevant. Breast cancer is the most common and a very well started malignancy in women. 43 "Vera Ehrenstein" (7209472) 00:12:30.989 --> 00:12:48.929 In women breast cancer has well developed treatment protocols. It is also known that prognosis of breast cancer differs by patient characteristics. Specifically, lesson NO is known about breast cancer in men although it does occur. 44 "Vera Ehrenstein" (7209472) 00:12:48.929 --> 00:13:08.929 And because it is very rare, breast cancer is very rare in men. Men with breast cancer represent an understudied small subgroup with a less optimal prognosis than do women. ALS on the other hand, is a very rare. 45 "Vera Ehrenstein" (7209472) 00:13:08.929 --> 00:13:31.169 Motor neuron disease characterized by high fatality and absence of cure. Currently, pharmacher therapy with real user is the only approved option used for symptom relief among patients with ALS. So to summarize, because evidence from randomized controlled trial in rare. 46 "Vera Ehrenstein" (7209472) 00:13:31.169 --> 00:13:50.939 Subgroups of patients with breast cancer and also in ALS patients is difficult to obtain. It is difficult to obtain because the populations in those subgroups and those patient groups are small. These patient populations potentially may stand to benefit. 47 "Vera Ehrenstein" (7209472) 00:13:50.939 --> 00:14:10.939 From the use of real world data in the pre authorization stage. Here are the specific objectives of use case one. We start from the beginning, we start with having actual basic data access. 48 "Vera Ehrenstein" (7209472) 00:14:10.939 --> 00:14:26.249 Preprocessing including data management as an objective of this study. So we would like to learn about what it takes to access data in the. 49 "Vera Ehrenstein" (7209472) 00:14:26.249 --> 00:14:42.929 For participating countries, we would like to carry out the pro preprocessing tasks including quality control and as you will see later in this presentation, we would like to convert the. 50 "Vera Ehrenstein" (7209472) 00:14:42.929 --> 00:15:02.519 Diverse data from the different databases into a standardized common data model called OMOP for the analysis. Then we would like to provide a descriptive overview of breast cancer in ALS populations. 51 "Vera Ehrenstein" (7209472) 00:15:02.519 --> 00:15:17.939 Including three more bit characteristics, so pre disease characteristics, characteristics of the disease itself. We would like to look at treatments and finally we would like to look as per as progno at prognosis. Now I would like to stress that. 52 "Vera Ehrenstein" (7209472) 00:15:17.939 --> 00:15:35.939 We are not concerned here about the clinical or scientific conclusions from these analysis. What we would like to do, we would like to conduct a proof of concept study. We would like to compare results that we obtain against the published evidence. 53 "Vera Ehrenstein" (7209472) 00:15:35.939 --> 00:15:51.719 And to create the best, the best practice guidelines as a result of that. So when we obtain results on the same objectives from different herogeneous data sources. 54 "Vera Ehrenstein" (7209472) 00:15:51.719 --> 00:16:07.859 Would like to ponder examine on the sources of heterogenuity of the results across the country's specific patient populations and across different data types, across different coding practices, accessibility representativeness, completeness. 55 "Vera Ehrenstein" (7209472) 00:16:07.859 --> 00:16:24.209 Temporal variation and validity. So in terms of methods, we have conducted a cohort study and that this very simple graph represents. 56 "Vera Ehrenstein" (7209472) 00:16:24.209 --> 00:16:42.659 The study design. So for breast cancer, we identify in the routinely collected data in the databases incident is breast cancers and we say that the date on which breast cancer was diagnosed diagnosed was. 57 "Vera Ehrenstein" (7209472) 00:16:42.659 --> 00:17:01.019 Oh, the index date. We assess eligibility criteria five years before the index date and then we follow patients forward, and the end of the follow up shown here on the right depends dependent in this study. 58 "Vera Ehrenstein" (7209472) 00:17:01.019 --> 00:17:20.699 On the specific database dependent on the data refresh status at the time that the data were extracted. So this represents the study design for the breast cancer population, but it is pretty much the same for the ALS population. So we're doing basically two cohort studies here. 59 "Vera Ehrenstein" (7209472) 00:17:20.699 --> 00:17:40.109 So in terms of data access and management, as the 1st step, we always need to get ethical or other government governance, follow other governance procedures to be able to legally obtain and analyze the data. 60 "Vera Ehrenstein" (7209472) 00:17:40.109 --> 00:17:59.099 So we have filled out protocol based data applications in each country's data custodian. So we are not the owners of the data. We apply to data custodians to analyze these data on a pro on a protocol basis. Upon access, upon. 61 "Vera Ehrenstein" (7209472) 00:17:59.099 --> 00:18:19.099 Receiving access or receiving data from data custodians. We've been conducting local data checking to ensure that the data that we have requested have in fact been been delivered and those who work in this area know that more often than not, something actually maybe missing from what. 62 "Vera Ehrenstein" (7209472) 00:18:19.099 --> 00:18:42.199 You have requested, so it is very important to check that the delivered data or the access data are the data that have been requested. And then we do data management that consists of the routine cleaning and plausibility checks. And the next step would be of data access and management would be coding the data according to the protocol. 63 "Vera Ehrenstein" (7209472) 00:18:42.199 --> 00:18:59.369 According to operational definitions of study variables. And as you can imagine since we have four countries in this project, we, each data has its own flavor, meaning that the coding practices and some of the coding. 64 "Vera Ehrenstein" (7209472) 00:18:59.369 --> 00:19:19.369 Conventions and classifications differ across the different databases. So I'm going to pause here and I'm going to give the floor to Mohammed who will talk a little bit about this step of converting the native data to the format. 65 "Vera Ehrenstein" (7209472) 00:19:19.369 --> 00:19:25.989 And then I will take over again and talk about the results of use case one. 66 "Mohamed Aborageh" (788493568) 00:19:25.989 --> 00:19:36.993 Thank you very much, Vera. If for the current presentation, just a quick question, do I need to share the slides or can you maybe navigate them for me? 67 "Vera Ehrenstein" (7209472) 00:19:36.993 --> 00:19:39.972 It will be much better. I can I'd be happy to advance if you like. 68 "Mohamed Aborageh" (788493568) 00:19:39.972 --> 00:19:43.872 Thank you very much. 69 "Vera Ehrenstein" (7209472) 00:19:43.872 --> 00:19:49.165 I guess, right? Okay, here you are. Yep. 70 "Mohamed Aborageh" (788493568) 00:19:49.954 --> 00:20:01.289 Thank you very much. So, yeah, I will start by briefly setting the scene for how we actually bring a very heterogenous national data into a common data structure. 71 "Mohamed Aborageh" (788493568) 00:20:01.289 --> 00:20:20.159 And what you see here is the standard extract transform load workflow or the ETL, which is the backbone of any OMOP implementation. And we start with the national medical databases from different countries such as Denmark and Portugal and Germany. 72 "Mohamed Aborageh" (788493568) 00:20:20.159 --> 00:20:37.409 And these data sources differ substantially in the structure and the coding systems and at the level of clinical detail. The 1st step is extraction where we retrieve the relevant tables from the source systems and place them into a staging area. 73 "Mohamed Aborageh" (788493568) 00:20:37.409 --> 00:20:54.599 And at this stage, the data is still very close to their original country specific formats. The core of the process is the transformation step, and this is where most of the work happens basically. We clean the data, we normalize the formats such as the dates and the units. 74 "Mohamed Aborageh" (788493568) 00:20:54.599 --> 00:21:13.079 And crucially mapped local coding systems to standardized vocabularies that are used by the Omap common data model. And after that the transformed data are loaded into the data model which gives us a harmonized analysis ready representation across all countries. 75 "Mohamed Aborageh" (788493568) 00:21:13.079 --> 00:21:28.889 Once the data are in the common data model, we can use a shared analytical layer, the common tools, common definitions, and the reproducible methods while still respecting that the data originates from very different healthcare systems. 76 "Mohamed Aborageh" (788493568) 00:21:28.889 --> 00:21:47.159 I think we can move to the next slide. Initially, the ETL design was developed based on the metadata that was available to us up front. So the table descriptions, the variable lists, and the documentation that was provided before we had direct access to the data. 77 "Mohamed Aborageh" (788493568) 00:21:47.159 --> 00:22:03.179 However, once we obtain the data access to the real data sets, we quickly see that this initial design has to be adapted a little bit. And in practice we encounter different variable names, unexpected table structures, and in some cases information. 78 "Mohamed Aborageh" (788493568) 00:22:03.179 --> 00:22:21.059 That was documented but implemented differently in the actual databases. Another important adjustment was at the technical level where due to the size of some national data sets, we had to switch to an in process database approach rather than keeping everything in memory. 79 "Mohamed Aborageh" (788493568) 00:22:21.059 --> 00:22:41.059 Simply to make the ETL scalable and robust. The key point here is that the ETL development was not a one off task. It required iterative adaptation of scripts as we moved from documentation to real world data. And the main learning from this experience is that critical importance of early access to data and having accurate deep. 80 "Mohamed Aborageh" (788493568) 00:22:41.059 --> 00:22:58.109 Failed metadata available. Even the small discrepancies at this stage can have downstream effects on harmonization and analysis, and this is especially relevant in a multi partner setting like where aligning the expectations early can save substantial effort later on. 81 "Mohamed Aborageh" (788493568) 00:22:58.109 --> 00:23:16.829 Next slide, please. So building on the ETL process I just described, I now want to focus on one of its most critical components which is the target schema mapping. The central challenge here is that each country uses its own coding systems. 82 "Mohamed Aborageh" (788493568) 00:23:16.829 --> 00:23:36.829 Its own data structures and conventions. So on the left here, e.g., you see examples from Fillman Denmark, and Portugal, each with its own way of representing the same underlying clinical concepts. And then the target schema mapping is the step where we translate these country specific codes and structures into a standard. 83 "Mohamed Aborageh" (788493568) 00:23:36.829 --> 00:23:55.709 Vocabulary and schema as defined by the common data model. This mapping is essential because it allows us to align otherwise disparate data sources and to analyze them cohesively within a single framework rather than treating each country as a separate special case. 84 "Mohamed Aborageh" (788493568) 00:23:55.709 --> 00:24:15.449 In practice, this means that mapping local diagnosis, procedure and drug codes to OMOP standard concepts and assigning them to consistent tables such as, e.g., the person, the condition and the observation tables in the OMOP common data model and associated date fields. 85 "Mohamed Aborageh" (788493568) 00:24:15.449 --> 00:24:32.339 What's important to emphasize here is that this is not a purely mechanical lookup. Many concepts do not map to one to one concepts and decisions must be made about granularity, clinical meaning, and how much information can be preserved. 86 "Mohamed Aborageh" (788493568) 00:24:32.339 --> 00:24:48.329 Without introducing buys. And once this mapping is done, the OMOP schema gives us a shared analytical language. The same cohort definitions, outcome definitions, and analysis code that can then be applied across all the partners. Next slide, please. 87 "Mohamed Aborageh" (788493568) 00:24:48.329 --> 00:25:07.379 Yes, once we define the target schema, the next question is how can we actually map local medical codes to OMAP concepts in a consistent way? For this, we rely on Athena, which is a searchable vocabulary database that is maintained by the ODC community. 88 "Mohamed Aborageh" (788493568) 00:25:07.379 --> 00:25:24.959 And it serves as a central reference point that brings together many different clinical vocabularies and links them to the OMAP standard concepts. In practice, Athena helps us to identify a local code, whether it is a diagnosis or a procedure or a drug. 89 "Mohamed Aborageh" (788493568) 00:25:24.959 --> 00:25:43.379 And then find its corresponding concept together with additional contextual information. So the main three elements that are important to us here, 1st the domain ID, which tells us whether this concept is a condition, a drug, a procedure, observation or other clinical domain. 90 "Mohamed Aborageh" (788493568) 00:25:43.379 --> 00:26:03.379 And then 2nd is the concept ID, which are unique identifiers that allow us to represent the same medical concept consistently across all data sets and countries. And 3rd, the hierarchical relationships, which define the parent child links between the concepts and allows us to work at a different level. 91 "Mohamed Aborageh" (788493568) 00:26:03.379 --> 00:26:25.349 Levels of clinical granularity when needed. Together, these features make the vocabulary mapping more transparent and reproducible and they are essential for ensuring that the analysis performed on data are comparable across the partners. Next slide please. So to make this more concrete, this is an example of vocabulary mapping using athena. 92 "Mohamed Aborageh" (788493568) 00:26:25.349 --> 00:26:42.629 So here we have an example of concept which is gestational diabetes meditus. In hop up this condition is represented as a standard concept with a unique concept ID, e.g., the number that is in front of you here and it is derived from the Snowman vocabulary and classified as a disorder. 93 "Mohamed Aborageh" (788493568) 00:26:42.629 --> 00:27:02.629 By assigning this concept to the condition domain, we define where it belongs in the OMAP schema and how it will be treated in downstream analysis, and what is particularly important here are the hierarchical relationships which you can see visualized on the right. E.g., gestational diabetes has a parent concept such as diabetes, melitus and pregnancy. 94 "Mohamed Aborageh" (788493568) 00:27:02.629 --> 00:27:18.059 Which represents a broader, a broader clinical category. At the same time, it can have child concepts that capture more specific situations such as different diagnostic formulations or subtypes recorded in a local coding system. 95 "Mohamed Aborageh" (788493568) 00:27:18.059 --> 00:27:37.979 This hierarchy gives us analytical flexibility. If we want a very specific cohort, we can focus only on the exact gestational diabetes concept or if we want a broader cohort, then we can include the parent concept and automatically capture the related pregnancy associated diabetes conditions across all data sets. 96 "Mohamed Aborageh" (788493568) 00:27:37.979 --> 00:27:53.099 The key advantage is that these decisions are made once at the concept level and then applied consistently across all the countries regardless of how the condition was originally coded locally. You move to the next slide. 97 "Mohamed Aborageh" (788493568) 00:27:53.099 --> 00:28:09.779 Okay, while Athena is very powerful, there are some challenges that we encountered when using this in practice. So one category of challenges relates to the missing specific codes. 98 "Mohamed Aborageh" (788493568) 00:28:09.779 --> 00:28:29.779 Some country specific codes such as special reimbursement codes or social economic indicators simply do not exist in Athena because they are not part of international clinical vocabularies and a concrete example is shown on the right here where certain chemotherapy treatment regimens or country specific combinations are recorded as. 99 "Mohamed Aborageh" (788493568) 00:28:29.779 --> 00:28:50.069 Non standard. And in these cases we relied on specialized vocabularies such as the hemonk or the hematology and oncology vocabulary, but even then not all local formulations could be mapped cleanly to a single standard concept. 2nd and in some cases more fundamental challenge is missing entire vocabularies. 100 "Mohamed Aborageh" (788493568) 00:28:50.069 --> 00:29:05.369 E.g., extended national classifications like the Danish SKS classification system is not fully represented in Athena, and this means that some local quotes have NO direct equivalent in Omap and cannot be mapped one to one. 101 "Mohamed Aborageh" (788493568) 00:29:05.369 --> 00:29:21.809 The example at the bottom illustrates this, so highly specific damaged SKS code had to be simplified to a more general ICD ten concept and sub such as mapping the several subcodes of G 12.2 back to the base ICD ten category. 102 "Mohamed Aborageh" (788493568) 00:29:21.809 --> 00:29:41.809 As a result, the main learnings here that manual mappings were unavoidable. There will be some loss of specificity that is necessary and that mapping certain highly unique local codes was simply not feasible, and these trade offs are important to acknowledge because they directly affected the data granularity and must be considered. 103 "Mohamed Aborageh" (788493568) 00:29:41.809 --> 00:30:04.309 When interpreting downstream analysis. Next slide please. So beyond vocabulary vocabulary mapping, another major challenge we encounter relates to the missing information in the source data itself. So the common data model is intentionally very detailed and it assumes that a wide range of information is available and explicitly. 104 "Mohamed Aborageh" (788493568) 00:30:04.309 --> 00:30:22.739 Be recorded in the source system, but in reality this assumption does not always hold. In several data sets, we observe mixtures of structured data and free text. E.g., some information such as the number of dispensed drugs maybe embedded in text fields rather than stored in a structured format. 105 "Mohamed Aborageh" (788493568) 00:30:22.739 --> 00:30:42.739 And extracting this information often requires a manual review for custom passing which is feasible only for selected subsets of the data and not at a scale. We also encountered cases where key omap fields were not simply recorded consistently in the source data. A typical. 106 "Mohamed Aborageh" (788493568) 00:30:42.739 --> 00:31:04.949 Example e.g. is drug exposure information where variables were such as drug exposure end dates or days of supply were missing or incompletely captured and had to be derived. These gaps limit our availability or ability to represent treatment duration accurately and can directly affect the downstream analysis again. 107 "Mohamed Aborageh" (788493568) 00:31:04.949 --> 00:31:24.949 Such as exposure definitions or time at risk calculations. The main learning is that the importance of standardized input formats and the need for complete structured information especially for drug administration. And where this information is missing, assumptions or simplifications become unavoidable and these need to be clearly documented and con. 108 "Mohamed Aborageh" (788493568) 00:31:24.949 --> 00:31:45.809 Considered in interpreting results. Next slide please. Yes, it's also important to emphasize that introducing standard standards does not automatically remove the ambiguity. So a good example of this is the ADC classification system which many of you may already be familiar with. 109 "Mohamed Aborageh" (788493568) 00:31:45.809 --> 00:32:02.699 So ATC or anatomical therapeutic chemical classification organizes the drugs based on their anatomical therapeutical and chemical properties and assigns a structured code to each drug or group of drugs. As shown on the right, ATC is hierarchical with multiple levels. 110 "Mohamed Aborageh" (788493568) 00:32:02.699 --> 00:32:19.109 And it ranges from broad and atomical groups down to specific chemical substances. E.g., if we take metforman as an example, we can represent it at a different level. We're going to at a higher level, it belongs to the elementary tract. 111 "Mohamed Aborageh" (788493568) 00:32:19.109 --> 00:32:36.599 And metabolism group than to drugs that are used in diabetes and only at the most specific level does it represent performing as a chemical substance? The ambiguity rises because different analytical questions require different levels of this hierarchy. 112 "Mohamed Aborageh" (788493568) 00:32:36.599 --> 00:32:52.289 If you map drug exposures at a very broad ATC level, we may group together drugs with substantially different mechanisms of actions. On the other hand, if we map everything at the most specific level, we risk the fragmentation and the reduced statistical power. 113 "Mohamed Aborageh" (788493568) 00:32:52.289 --> 00:33:08.459 Especially in smaller data sets. So while ATC provides a standardized structure, the choice of mapping level is still a methodological decision and not a purely technical one. Next slide please. 114 "Mohamed Aborageh" (788493568) 00:33:08.459 --> 00:33:27.029 Another important example here is the ambiguity of standard vocabularies, e.g., that comes from RX norm, which we use extensively for drug mapping in omap. So RX Norm provides a standardized nomenclature for clinical drugs offering normalized names and unique identifiers. 115 "Mohamed Aborageh" (788493568) 00:33:27.029 --> 00:33:42.599 And linking linking medications across different drug classification systems. And if we compare it to ADC, it is much more granular and explicitly models relationships between ingredients, strengths, dose forms, and branded products. 116 "Mohamed Aborageh" (788493568) 00:33:42.599 --> 00:34:00.269 And you can see the diagram on the right, so it illustrates the, this complexity, a single active ingredient here, e.g., set regime can be connected to multiple clinical drug forms or multiple strengths or multiple brand names, and they're all linked through different relationship types. 117 "Mohamed Aborageh" (788493568) 00:34:00.269 --> 00:34:15.689 This switchness is extremely valuable, but it also introduces ambiguity at the mapping stage. E.g., if we map a local drug code, we have to decide whether we want to represent it at an ingredient level or a clinical drug level or the branded product level. 118 "Mohamed Aborageh" (788493568) 00:34:15.689 --> 00:34:35.689 And each of these choices would answer a different research question, including ingredient level mappings, e.g., support class based analysis while branded or those specific mappings are more appropriate for detailed pharmacological studies. Also in the nordic context, this is further complicated by country specific identifiers. 119 "Mohamed Aborageh" (788493568) 00:34:35.689 --> 00:34:55.289 Which as the VMR or the Nordic article number which was linked to the standard RX norm concepts by and why this linkage is extremely helpful, it still requires careful validation to ensure that the intended level of detail is preserved consistently across the data sets. 120 "Mohamed Aborageh" (788493568) 00:34:55.289 --> 00:35:13.409 And next slide. Yeah, so this slide brings together the previous two examples and highlights one of the most challenging situations that we encounter, which is the one to many mappings between standard vocabularies. So in this case the issue arises when a single ATC code. 121 "Mohamed Aborageh" (788493568) 00:35:13.409 --> 00:35:31.019 Represents a combination of products rather than a single active ingredient, and when we translate this into RX norm, which explicitly models ingredients that the single ATEC code naturally maps to multiple RNX RX norm concepts. 122 "Mohamed Aborageh" (788493568) 00:35:31.019 --> 00:35:51.019 And in this example and from a technical perspective, this is expected and correct. However, from an analytical perspective, it may introduce important decisions. So we treat, e.g., the question is, do we treat this exposure as a combination of therapy or do we split it into its individual component. 123 "Mohamed Aborageh" (788493568) 00:35:51.019 --> 00:36:09.629 Or do we define a higher level exposure group? And each of these choices would lead to a different cohort definition. What's crucial is that these are not errors in mapping but inherent ambiguities that arise when we bridge vocabulary designed for different purposes. Next slide please. 124 "Mohamed Aborageh" (788493568) 00:36:09.629 --> 00:36:27.959 Yes, this slide also highlights a slightly different but equally important issue. So this time between ATC codes and vnrs in principle, the VNR is expected to be a unique identifier for specific medicinal products. However, in practice we found some exceptions. 125 "Mohamed Aborageh" (788493568) 00:36:27.959 --> 00:36:45.359 So in some cases the same VNR is linked to multiple ATC codes, often because the product was approved for multiple indications or existed in slightly different formulations. And the diagram here illustrates the situation where we have two ATC codes that both point to the same VNR. 126 "Mohamed Aborageh" (788493568) 00:36:45.359 --> 00:37:04.679 And this challenges the assumption of a one to one mapping. When such mappings are translated into OMOP particularly via RX norm, this can lead to an inflation of drug exposure counts simply due to how concepts are represented. And this is simply a consequence of how different classification systems are designed. 127 "Mohamed Aborageh" (788493568) 00:37:04.679 --> 00:37:21.209 Again for different purposes. The key learning at this point was that relying solely on standard concepts can sometimes be misleading in these cases and instead manual verification of mappings becomes necessary or alternatively. 128 "Mohamed Aborageh" (788493568) 00:37:21.209 --> 00:37:41.209 The direct use of source concept identifiers such as ATC or BNR maybe more appropriate for certain analysis. Next slide please. Yeah, so so far I've shown different sources of of ambiguity in standard mappings and here we summarized the practical impact on the data itself. So. 129 "Mohamed Aborageh" (788493568) 00:37:41.209 --> 00:38:04.529 The core problem is what we refer to as data inflation, and in some cases a single health event in the native source data represented as one row can generate multiple rows once it is transformed into the format. And this typically happens when one source code maps to multiple standard concepts. So e.g., on the left here, you see a simplified representation of a native record. 130 "Mohamed Aborageh" (788493568) 00:38:04.529 --> 00:38:24.529 Where a reference ID, a vocabulary code, a date and a person ID together represents a single clinical event. And then after transformation, the same event appears as several entries in the OMAP condition occurrence table with its each with its own unique condition occurrence ID. And the challenge is that. 131 "Mohamed Aborageh" (788493568) 00:38:24.529 --> 00:38:48.060 That standard omap schema does not provide a built in mechanism to preserve the explicit linkage between these related roles. And as a result, it becomes a little bit difficult to reliably identify unique health events especially if analysis implicitly assume that each rock corresponds to an independent occurrence. If this is not handled carefully, it can lead to the inflated event counts. 132 "Mohamed Aborageh" (788493568) 00:38:48.060 --> 00:39:08.060 And this is why mapping decisions in and how we operationalize them in ETL pipelines is not it's not just a technical detail but we do have direct methodological consequences. Next slide please. So now I will show briefly how we address this issue. 133 "Mohamed Aborageh" (788493568) 00:39:08.060 --> 00:39:33.330 In practice, so the core idea of the solution is to separate the standardization from traceability and while Omap requires that each record has a unique occurrence ID, we introduced an additional identifier that preserves the link back to the original source event and completely for each native health event we generate a stable source level identifier such as a universal unique ID before transformation. 134 "Mohamed Aborageh" (788493568) 00:39:33.330 --> 00:39:52.740 And when this event maps to multiple OMAP pros, each of those rows receive its own condition occurrence ID as required by OMOC, but they all remain linked to the same source identifier. And this linkage is stored in a separate linking table which explicitly connects the source reference ID with the corresponding OMAP occurrence ID. 135 "Mohamed Aborageh" (788493568) 00:39:52.740 --> 00:40:10.790 This approach allowed us to accurately track which OMA pros originate from the same underlying health events during the analysis, and as a result, we can avoid double counting or reconstruct unique events when needed and maintain the transparency about how mappings were performed. 136 "Infarmed" (3012685056) 00:40:10.790 --> 00:40:16.710 We we need to explore it a little bit. Thank you. 137 "Mohamed Aborageh" (788493568) 00:40:16.710 --> 00:40:18.032 Sorry? 138 "Infarmed" (3012685056) 00:40:18.032 --> 00:40:20.704 We need to accelerate a little bit here. Yeah. 139 "Mohamed Aborageh" (788493568) 00:40:20.704 --> 00:40:30.690 Okay, so I have just one last thing is that the main learning is that the standard vocabularies and oma common data model is essential to harmonize the data across countries. 140 "Mohamed Aborageh" (788493568) 00:40:30.690 --> 00:40:50.340 And what matters ultimately is the ability to preserve the link between the standardized records and the original source events. And the last slide we just talk about the computational aspects, so this is the final slide. Depending on the population size and the observation period, the source data sets can become very large. 141 "Mohamed Aborageh" (788493568) 00:40:50.340 --> 00:41:08.940 And in secure data environments this is often compounded by hardware and software constraints. E.g., we encountered individual tables for around of around 65 gb, which is exceeds the available RAM in some environments. In addition, some secure platforms are not designed for multi user access. 142 "Mohamed Aborageh" (788493568) 00:41:08.940 --> 00:41:27.424 And so or modern machine learning workflows these constraints mean that the implementation choices are often derived not only by methodology but also by what is computationally feasible in practice and the main learnings here were the importance of incremental and database driven processing. 143 "Mohamed Aborageh" (788493568) 00:41:27.424 --> 00:41:41.612 Taking these constraints into account early is critical to ensure that the Omar pipelines remain scalable and yeah with that said, I am the presentation here and I pass back to Vera. Thank you very much. 144 "Vera Ehrenstein" (7209472) 00:41:41.612 --> 00:41:53.670 Thank you so much, the Joanna and Claudia, how much time would you like me to take for presenting use case one results? I know we're almost at the end. 145 "Infarmed" (3012685056) 00:41:53.670 --> 00:41:59.750 I think we can extend for more 07:10 minutes, ok? Okay. Thank you. 146 "Vera Ehrenstein" (7209472) 00:41:59.750 --> 00:42:16.080 So, I will continue presenting the use case one results and as much as you have just heard from Mohammed about etling of the native data to the format, most of the results I will present will be in the native format and I'll talk about it. 147 "Vera Ehrenstein" (7209472) 00:42:16.080 --> 00:42:32.880 About why. Also we're working on a publication for use case one results that would include data in both formats so stay tuned. So let's see. Okay, so. 148 "Vera Ehrenstein" (7209472) 00:42:32.880 --> 00:42:51.720 As with, etling data to a common data model format, there are surprises once you actually start looking at the data. So on the data access and analysis page we have planned to have a single analyst writing analytic code for the beautifully almost converted data. 149 "Vera Ehrenstein" (7209472) 00:42:51.720 --> 00:43:10.650 However, we have encountered I think in this project especially severe delays to data access because of limitations especially in Denmark to with respect to data access from external partners. 150 "Vera Ehrenstein" (7209472) 00:43:10.650 --> 00:43:29.070 So we had to change data custodian mid flight and to to enable us to actually conduct the project as planned, but as a, as a workaround to get some results ahead of the almopping of the native data. 151 "Vera Ehrenstein" (7209472) 00:43:29.070 --> 00:43:47.250 We actually went ahead and conducted all of the analy we planned analysis in the native, so three more format ahead of the analysis and the almost converted data. This had silver lining because of the challenges that Mohammed just mentioned. 152 "Vera Ehrenstein" (7209472) 00:43:47.250 --> 00:44:05.700 You don't really know when you apply the standard or more processes whether you're getting it right. So we thanks to thanks to the data delay and the challenges I should say we had the benefit of having the native data analysis native. 153 "Vera Ehrenstein" (7209472) 00:44:05.700 --> 00:44:24.840 Format data analysis results as a benchmark for the results that we would obtain in the OMOB after the OMOB conversion. But the analysis themselves were relatively simple, we were interested in obtaining descriptive statistics of the breast cancer and ALS populations, including incidence rates. 154 "Vera Ehrenstein" (7209472) 00:44:24.840 --> 00:44:40.590 Basic demographics, comorbidities treatments, and some of the markers of disease progression and survival. So let me show you some of the high level results as I mentioned, the publication is in preparation. Let's start with breast cancer. 155 "Vera Ehrenstein" (7209472) 00:44:40.590 --> 00:45:00.590 So we have along the way as we, as we obtain the results, always benchmark them also against the published evidence. Although as I mentioned, these results are not meant to be used in clinical practice. So the main message here in the graph on the left, you can see that breast cancer is primarily. 156 "Vera Ehrenstein" (7209472) 00:45:00.590 --> 00:45:17.220 I have female disease. You can almost not see the incidence rate in men, which is a very small blue rectangles. On the right hand side, you can see trends in incidence of the breast cancer over this study period. 157 "Vera Ehrenstein" (7209472) 00:45:17.220 --> 00:45:36.630 Data are available for this patient population from Denmark and from Finland. They're not available from Portugal because of the sum of the problems that were identified on the data access stage, so the the Portuguese team decided that would not be prudent to present these incidents rates data. 158 "Vera Ehrenstein" (7209472) 00:45:36.630 --> 00:45:57.565 From Portugal. Also, I need to say that we are still waiting for data access in Germany although this was planned as a four country analysis, we have data from three countries and that is also that also represents a real world challenge in data access for regulatory purposes. 159 "Vera Ehrenstein" (7209472) 00:45:57.565 --> 00:46:13.100 So here are some of the results in the demographic characteristics of breast cancer patients. I would like to highlight some of the differences in the types of data available from the three different countries, from which data were available. So for. 160 "Vera Ehrenstein" (7209472) 00:46:13.100 --> 00:46:35.340 For this specific data extraction, data unemployment, e.g., were available from Finland or not available from Denmark or from Portugal in Denmark data unemployment or other social economic indicators would technically be available, but it would not be possible to, have them accessed by 3rd party partners. 161 "Vera Ehrenstein" (7209472) 00:46:35.340 --> 00:46:55.140 With the with the ready script. So here are some of the comorbidities that we have analyzed. We looked at the prevalence of psychiatric disorders dementia and Alzheimer's specifically in cardiovascular diseases. 162 "Vera Ehrenstein" (7209472) 00:46:55.140 --> 00:47:10.650 You can see there is a lot of concordance with the exception perhaps of cardiovascular diseases in Thinland. We think that it has to do with completeness of recording maybe with some definitions. We're looking a little bit closely into that. 163 "Vera Ehrenstein" (7209472) 00:47:10.650 --> 00:47:27.990 What most of the results are presented for women because as you have seen this is a predominant group of these patients. When we look at treatments twelve months post diagnosis, you can see that there is some herogenuity in the prevalence. 164 "Vera Ehrenstein" (7209472) 00:47:27.990 --> 00:47:45.150 Of treatment modalities such as radiotherapy chemotherapy and total partial lymphoid and lymphoid indoctomy at anoctomy, but for we we see that for surgical procedures. 165 "Vera Ehrenstein" (7209472) 00:47:45.150 --> 00:48:02.895 Highlighted here, there is some relatively good concordance in the reporting of the, of the prevalence of these. And I think we think that this reflects generally a better recording of surgical procedures in the real world data. 166 "Vera Ehrenstein" (7209472) 00:48:02.895 --> 00:48:23.660 I want to look at the entire follow up, we have up to 20 years of follow up with some of these patients. We can see here some of the disease characteristics and I would like to bring attention to herogeneity of the way, e.g., the tumor characteristics are reported in the three. 167 "Vera Ehrenstein" (7209472) 00:48:23.660 --> 00:48:42.600 Countries, whose countries whose data are being analyzed. So e.g. Denmark and Portugal use the standard TNM staging diagnosis for breast cancer while in Finland and internal staging classification is used. 168 "Vera Ehrenstein" (7209472) 00:48:42.600 --> 00:49:02.600 And then you can see also that the distribution of the tumor size and the prevalence of metastas for metastasis is a diagnosis. They vary quite widely across the data sources and this probably represents for the represents for the most part different. 169 "Vera Ehrenstein" (7209472) 00:49:02.600 --> 00:49:21.300 Is in the data flow and completeness. When we look at signals of disease progression, we can see here that despite the heterogenity of the data sources, some things are recorded pretty similarly and probably equally well. 170 "Vera Ehrenstein" (7209472) 00:49:21.300 --> 00:49:36.330 We know that malignancy diagnosis usually come from cancer registries and there is not much measurement error or as we say misclassification in reporting deaf data, so we can see that here reflected in. 171 "Vera Ehrenstein" (7209472) 00:49:36.330 --> 00:49:53.700 The relative, relatively good concordance across the data sources in these, in these disease progression markers. Then you can see here when we look at overall survival of breast cancer in the three countries, the graphs look pretty similar. 172 "Vera Ehrenstein" (7209472) 00:49:53.700 --> 00:50:23.754 So very quickly we can see that men represent a much smaller group than women, men with breast cancer, and you can also see we observe in our real world data analysis that men are tend to be substantially higher at diagnosis than women with breast cancer, which has something to do also with potentially a poor prognosis in this subgroup. 173 "Vera Ehrenstein" (7209472) 00:50:23.754 --> 00:50:45.290 Okay, let me go quickly to the high level results for ALS. Here you can see incidence rates from all three countries for which we conducted the analysis Denmark, Binsland, and Portugal. On the left you can see that the data do reflect what we know from a literature slight proponderance of the male population in the als. 174 "Vera Ehrenstein" (7209472) 00:50:45.290 --> 00:51:09.671 As patients. You can see that the scale for incidence rate is much smaller than it was for breast cancer. On the right hand side, you can see trends over time. You can see them to be maybe more herogeneous than you you think because of the vergical scale being pretty stretched out, but there are some differences. 175 "Vera Ehrenstein" (7209472) 00:51:09.671 --> 00:51:38.060 Also you see here the populations of ALS are much smaller than the breast cancer populations, but you can see that there is some, there is good concordance in the demographics. You can see in all countries the median age and the social economic the median age is concordant and you can see that in the country where with the employment status available. 176 "Vera Ehrenstein" (7209472) 00:51:38.060 --> 00:51:46.891 Most of the majority of the of the patients are in the older age group and there are antiumers. 177 "Vera Ehrenstein" (7209472) 00:51:47.534 --> 00:52:20.753 With respect to clinical characteristics of ALS, I would like to call out general, the herogeneity of the data sources with respect to capturing things that are that are seen in this table. E.g., you can see a large herogeneity of other mild motor impairment. We also think that has to do with how data flows and what is actually recorded in the databases. 178 "Vera Ehrenstein" (7209472) 00:52:20.753 --> 00:52:45.540 When we look at treatment of the last twelve months post diagnosis, as I mentioned before, really use all is the only symptom relieved in treatment. This is also important to our plans for use case one use case two in this work package. Oh, 1st of all I would like to say that when we look at the things like intensive care and non invasive ventilation, you can see. 179 "Vera Ehrenstein" (7209472) 00:52:45.540 --> 00:53:01.380 Differences in how countries report these, as you can see for reliazole, you can see that really useful treatment is not reflected in the data in Portugal although we know that is used in Portugal. 180 "Vera Ehrenstein" (7209472) 00:53:01.380 --> 00:53:21.380 And then the prevalence of use is higher in Denmark than in Thinland, which might reflect differences in the populations but also probably reflects some of the differences in the data completeness. So when we look at prognosis, we have a very good concording. 181 "Vera Ehrenstein" (7209472) 00:53:21.380 --> 00:53:43.190 In the risks of respiratory infections and as we have seen in the breast cancer population, we have good concordance in the mortality data. And if you can remember in your mind's eye the way that overall survival graphs looked for breast cancer, you can see here that overall survival graphs looked much different. 182 "Vera Ehrenstein" (7209472) 00:53:43.190 --> 00:54:04.500 From the, for the ALS population the the drop in the 1st post diagnosis years is quite steep, and we can see that across all the databases that we have analyzed. As I mentioned before, most of the analysis are available in native data format in all countries. 183 "Vera Ehrenstein" (7209472) 00:54:04.500 --> 00:54:20.550 However, we are able to compare and we will compare in the publication results obtained from native and from this is just a teaser for you here. The 1st thing as an epidemiologist, you look at, you look at. 184 "Vera Ehrenstein" (7209472) 00:54:20.550 --> 00:54:36.810 With the size of the population similar from the different sets of analysis. See here you can see data from Denmark and Portugal, the countries for which both sets of analysis are currently available. And you can see here. 185 "Vera Ehrenstein" (7209472) 00:54:36.810 --> 00:54:53.070 That the, that the sizes of your population, the concordance is pretty good and so is actually the rest of the characteristics. So this seems to be reassuring for the future analysis. 186 "Vera Ehrenstein" (7209472) 00:54:53.070 --> 00:55:12.570 So to summarize the main findings and the lessons learned from this exercise, so as we have known, as we have learned again and again data access delays is an important barrier to timely generation of real world evidence, and correctness of data delivery cannot be automatically assumed and needs to be. 187 "Vera Ehrenstein" (7209472) 00:55:12.570 --> 00:55:32.570 Subject of standard quality control procedures for reporters, for researchers before they start using the data. The difference is the heterogenerity of the results across the country specific patient populations stems from various sources, it can stem from true differences, but also from artifacts of data flow. 188 "Vera Ehrenstein" (7209472) 00:55:32.570 --> 00:55:53.000 From, the completeness of recording from things like reimbursement practices and from the data providence. Are we talking about administrative data? Are we talking about claims data? That all matters. The herogenuity of the data, including data type validity and completeness should be used for assessing how fit for. 189 "Vera Ehrenstein" (7209472) 00:55:53.000 --> 00:56:14.280 Purpose the data are in relation to specific research and regulatory questions. And depending on the research question, the answer could be different for different data source. So in conclusion, we say that use of real world data and regulatory pre authorizations decisions requires expert knowledge. 190 "Vera Ehrenstein" (7209472) 00:56:14.280 --> 00:56:34.280 To assess its the impact of data limitations. Harmonization of data to the same structure does not remove underlying herogenuity of data quality flow and completeness, and I just want to stress that the results for breast cancer in alas are generated for illustration purposes only and should not be taken as indicative of disease epidemiology. 191 "Vera Ehrenstein" (7209472) 00:56:34.280 --> 00:56:43.875 However, they are broadly consistent with the published evidence. So thank you so much for your attention. I hope we have a nice discussion following this. 192 "Infarmed" (3012685056) 00:56:43.875 --> 00:56:51.126 Great, thank you, thank you very much Vera and Mohammed, and I will pass over to for the roundtable discussion. Thank you. 193 "Britta Haenisch" (1929526528) 00:56:51.126 --> 00:57:10.550 Yeah, thank you very much Claudia, and also thank you very much for my side to Vera Environment for this very exciting presentation. And now it's time for our round table discussion. 1st of all, I would like to introduce the discussions. We have Pierre and Himo. Pia is a issue. 194 "Britta Haenisch" (1929526528) 00:57:10.550 --> 00:57:30.550 Specialist at the finished medicine's agency and the development and information services. A primary responsibility is to lead the development of Finlands national Medicine's Data repository and its online services. PS over ten years of experience in HDA and worked in HDA team until 20. 195 "Britta Haenisch" (1929526528) 00:57:30.550 --> 00:57:49.590 22. She currently serves on the finished council for choices in health care Palco as a council member and vice chair. Palco issues national recommendations on whether hospital only medicines should be included on the publicly founded healthcare service basket in Finland. So very warm welcome to Pierre. 196 "Britta Haenisch" (1929526528) 00:57:49.590 --> 00:58:09.590 Then the next discussion this Lucia Artis. Lucia Artis Low is the health biologist and biologist. She currently works as a research and communication project technician at the Spanish platform of patient organizations. She is involved in the world of associations after being diagno. 197 "Britta Haenisch" (1929526528) 00:58:09.590 --> 00:58:27.750 Closed with congentle heart disease self operation success recovery, she realized that NO one knows the reliability of being sick better than the patients themselves and that their experience should be taken into account in all health related decisions. Also very warm welcome to the Sea. 198 "Britta Haenisch" (1929526528) 00:58:27.750 --> 00:58:45.630 The next discussion is Daniel Moralez. Daniel is a senior clinical epidemologist within the data analytics and methods task force at the EMA, also working closely with David U, and his background is in general practice and academia. 199 "Britta Haenisch" (1929526528) 00:58:45.630 --> 00:59:05.630 Well, welcome to you Daniel. And the last discussion is Ruben Fanake. Ruben is an associate professor and biostatistician at UMCU Tressed in the Netherlands. He previously served as a visiting scholar at Scanford University Center for Innovative study design. His research focuses on integrated. 200 "Britta Haenisch" (1929526528) 00:59:05.630 --> 00:59:35.450 Data into clinical trials and developing new endpoints that address both multi dimensionality and also patient preferences. Also, I warm welcome to you, Ruben. So now we have prepared some questions and I would like to ask the discussions to answer the respective question from your perspective, from your point of view, and please are taking NO longer than about 5 min. 201 "Britta Haenisch" (1929526528) 00:59:35.450 --> 01:00:00.560 That's for your answer. So, the 1st question, so we heard the presentation from Vera and and Mohammed and during the presentation there were some barriers also to reward evidence used in the pre authorization stage mentioned e.g. the lack of awareness about the data sources that are available or concerns regarding data quality or. 202 "Britta Haenisch" (1929526528) 01:00:00.560 --> 01:00:17.393 Also delays in data access. So in your opinion, what contribution can the stakeholder group you represent and belong to provide to overcome these obstacles? As 1st please Pierre. 203 "Piia Rannanheimo" (1370084096) 01:00:17.393 --> 01:00:22.748 Okay, thank you. So do you hear me well? 204 "Britta Haenisch" (1929526528) 01:00:22.748 --> 01:00:25.488 Yes, we do. Perfect. 205 "Piia Rannanheimo" (1370084096) 01:00:25.488 --> 01:00:51.550 So I have experience in one prior OMOP based project where we piloted the data from finished OMOP sensors in in three different use cases that concerned multiple myeloma SM SMA disease and and so on. And I I think we encountered the same issues that you see here like with with. 206 "Piia Rannanheimo" (1370084096) 01:00:51.550 --> 01:01:11.550 The, time, time delays and so on and so on. So, for me as someone who represents hates the A and decision making, I I think it's very important for us to describe the evidence need. 207 "Piia Rannanheimo" (1370084096) 01:01:11.550 --> 01:01:45.170 That we have to, you know, groups like this that run the analysis but also to the Omop centers and the national national institutes that sort of steer the the data management and the processes development nationally. So, so for me, I think it's important that we as HTA personnel. 208 "Piia Rannanheimo" (1370084096) 01:01:45.170 --> 01:02:05.170 Really describe our evidence needs thoroughly so that the systems can develop to to answer our needs. So for, e.g., here, I see that the press cancer and the. 209 "Piia Rannanheimo" (1370084096) 01:02:05.170 --> 01:02:25.170 Less examples are covered in in really high level and when we make decisions on new medicines, we always look at these specific indications that these products are intended for. So this is a really strict. 210 "Piia Rannanheimo" (1370084096) 01:02:25.170 --> 01:02:41.467 And also for the data and evidence that is produced that we would like to see sort of patient group specific results, I think that's my my 1st reflections. 211 "Britta Haenisch" (1929526528) 01:02:41.467 --> 01:02:47.428 Okay, thank you pierre. Then it's this year. 212 "Lucía Ortiz (POP)" (463020032) 01:02:47.428 --> 01:02:50.153 Hello. Can you hear me. 213 "Britta Haenisch" (1929526528) 01:02:50.153 --> 01:02:52.593 Yeah, we hear you well. Okay. 214 "Lucía Ortiz (POP)" (463020032) 01:02:52.593 --> 01:03:12.020 Eh, well thank you for inviting me and for having us patience here in the conversation. I won't be long answering this question. I think that from the patient organization side are mainly to increase trust, usability and speed of the real world evidence in prioritization, and how can we do this? 215 "Lucía Ortiz (POP)" (463020032) 01:03:12.020 --> 01:03:32.020 Well, we do this by improving awareness of data sources cause we patient organizations usually act as connectors between the participants, which are the patients and their regulatory bodies. And also we help them to translate technical documents into plain language so they. 216 "Lucía Ortiz (POP)" (463020032) 01:03:32.020 --> 01:03:35.430 But they can understand. 217 "Lucía Ortiz (POP)" (463020032) 01:03:35.430 --> 01:03:55.430 In every step, and we can also help them by focusing on virus and data that are actually relevant for them and we can help them define which endpoints or outcomes are meaningful eh in their real life. 218 "Lucía Ortiz (POP)" (463020032) 01:03:55.430 --> 01:04:03.594 To make some regulatory decisions. 219 "Britta Haenisch" (1929526528) 01:04:03.594 --> 01:04:09.033 Okay, thank you Lucia. So now it's it's Danielle. 220 "Morales Leaver Daniel" (540326912) 01:04:09.033 --> 01:04:35.810 Yeah, thank you very much for the invitation and it's great to see all the work being presented as well. I think in terms of the the elements in question one, I think in terms of greater awareness of data sources that are out there than from an EMA or European regulatory medicines network, we have the HMA EMA catalog of data sources, where data sources discri. 221 "Morales Leaver Daniel" (540326912) 01:04:35.810 --> 01:04:55.810 Descriptions of metadata can be added and eventually through the European health data space, there will be a a larger sort of data catalog anticipated covering other uses than than just medicine's regulation. In terms of the quality, I would. 222 "Morales Leaver Daniel" (540326912) 01:04:55.810 --> 01:05:15.810 Suggests that, you know, the, the DMA data quality framework was published at the end of 2024 and there's an extension to the data quality framework that's under review specifically for real world data because there are nuances specifically with real world data and we hope then that will improve both the, the, the data. 223 "Morales Leaver Daniel" (540326912) 01:05:15.810 --> 01:05:42.650 Metrics, their transparency and the reporting that can then hopefully help assessments as well. I think another aspect that the Open medicine's regulatory network can help is providing scientific advice. So, if there are uncertainties about its use and its relevance of real date of a specific question, then applicants can seek scientific advice before I conduct. 224 "Morales Leaver Daniel" (540326912) 01:05:42.650 --> 01:06:02.650 In the study or submitting, but equally data sets and data sources that are maturing, can also see qualification. And that is also a process that can be viewed. And, and I think delays in accessing, I think we've all faced that anybody who's tried to do projects with real data, and that's a. 225 "Morales Leaver Daniel" (540326912) 01:06:02.650 --> 01:06:22.650 Really difficult thing to address. So within the Darwin U network in some instances, but not all, we've been successful in trying to use blanket protocols that are tend to be for simpler study designs and and some IRB committees have accepted those. But we hope then something like the European health data. 226 "Morales Leaver Daniel" (540326912) 01:06:22.650 --> 01:06:37.871 Space will also facilitate that with standardized applications and, and maybe stricter timelines for for when maybe approvals or data can be provided. So we hope that that landscape will support as well. Thank you. 227 "Britta Haenisch" (1929526528) 01:06:37.871 --> 01:06:41.951 Thank you Daniel. So now it's Ruben. 228 "Ruben van Eijk" (1763132928) 01:06:41.951 --> 01:07:15.070 Yeah, thanks. So I I think I agree with what moses has already said. So yeah, from my perspective, so more the academic perspective, I think like the data access issue yeah that that's the the highest priority and and you saw it also in your project that that is causing like a main delay, so the project that we have ongoing, like where we try to use real world data, e.g., in clinical trials or where an industry partner tries to, require data from an academic institution or from one of the. 229 "Ruben van Eijk" (1763132928) 01:07:15.070 --> 01:07:35.070 The national registry, yeah. It it sometimes becomes almost infeasible just because like access to patient level data that that's very difficult and I I think that that's maybe, yeah, where the most effort is needed so how how can we, make this happen? And so I speak mostly from the ALS perspective, but what. 230 "Ruben van Eijk" (1763132928) 01:07:35.070 --> 01:07:50.460 What we have done there, e.g., is, is that we as a community have created like a common data set so where we already try to to use some something similar as like the, as the OPM system. 231 "Ruben van Eijk" (1763132928) 01:07:50.460 --> 01:08:10.460 Sorry or the OMAP system, where you try to standardize like all of the, the, the fields and and the outcomes and and try to, to merge that and make that open access to the community, so that's called proact at the moment. So that consists out of, of registries and consists out of trials and yeah, basically the whole. 232 "Ruben van Eijk" (1763132928) 01:08:10.460 --> 01:08:30.460 Our community is contributing, so both the academic institutes and like the investigator initiated trials, the registries are contributing, but also the industry are contributing as well and that that I think creates a very nice system industry itself is contributing but can also access these data then for their own development. So that, that is an. 233 "Ruben van Eijk" (1763132928) 01:08:30.460 --> 01:08:52.700 Important point and I think the, the detail of data that's often required for these projects is yeah, requires quite quite a lot of, of detail so if I reflect a bit on the results that were presented like like for ALS like very important prognostic datas, like quality of life or like actual disease status, yeah, these are usually not really. 234 "Ruben van Eijk" (1763132928) 01:08:52.700 --> 01:09:10.470 Captured in electronic health care records, but yeah, these are really the outcomes that that we're looking at in drug development and also in order to make prognosis. So I think that there are some effort needed as well, so how can we collect like these, you know, more difficult outcomes and. 235 "Ruben van Eijk" (1763132928) 01:09:10.470 --> 01:09:30.470 So, so one of the ways that we're setting up now in in Europe is to create like a, a a common registry so where each country tries to collect like a harmonized manner, a core outcome set, so where we collect like quality of life and functional measures of of patients, but also like the basic diagnostic information. So how. 236 "Ruben van Eijk" (1763132928) 01:09:30.470 --> 01:09:48.650 Long as the patient's ick and details about the diagnosis. So these data are not standard captured in electronic health care records, but yeah, so so that has to come really from the community to make such a registry I think and yeah, so I I will stop there for now. 237 "Britta Haenisch" (1929526528) 01:09:48.650 --> 01:10:09.230 Thank you, Ruben, so a lot of input to this question from the different perspectives concerning the data sources, trust and data sources, data access all these things are of course difficult to overcome the problems in most cases, but there are some initiatives, so that's very good. So I'm. 238 "Britta Haenisch" (1929526528) 01:10:09.230 --> 01:10:37.353 The other topic that we had in the in the presentation was about OMOP, and I would like to ask you, so what problems does Omap solve that the well designed local data model does not so concerning native data analysis, so what, what does what does OMOP solve and what are the advantages of, of OMOP or other common data models there? So we again start with Pia. 239 "Piia Rannanheimo" (1370084096) 01:10:37.353 --> 01:11:03.970 Thank you. So I I think the pace basic principle or aim to, to have like semantic intel probability across borders. That's the the key issue if that's accomplished and so to be able to combine data sets and so on. I think in federated analytics, the basic principle also to. 240 "Piia Rannanheimo" (1370084096) 01:11:03.970 --> 01:11:23.970 That the the data stays local and then the analysis can be run separately. That's also something that may speed things and create more trust to the system. One thing that was on my mind was that we don't. 241 "Piia Rannanheimo" (1370084096) 01:11:23.970 --> 01:11:58.280 Don't have many national disease specific registers in Finland, and I think that's one issue what the standardization could do that we could combine data from, e.g., university hospitals more easily. And when we do have quality registers, e.g., the, the level of detail in those registers is not uselessly so deep that we, we are on that like press. 242 "Piia Rannanheimo" (1370084096) 01:11:58.280 --> 01:12:12.331 In medicine states where we know all the details about the disease or the patient groups that we would like to have for fit for purpose this isn't relevant analysis. Thank you. 243 "Britta Haenisch" (1929526528) 01:12:12.331 --> 01:12:16.592 Thank you Pierre, then it's Lucia. 244 "Lucía Ortiz (POP)" (463020032) 01:12:16.592 --> 01:12:34.350 Thank you. Well I don't I agree with what Pia just said, but I don't think that Omog doesn't may compete with a good local model in how well it analyzes, but it solves the problem that the local model. 245 "Lucía Ortiz (POP)" (463020032) 01:12:34.350 --> 01:12:54.350 It has e.g. problems such as scalability and cooperability that happens across organizations and countries. And as an example of an advantage of the model, I would like to, if you don't mind, eh, talking about one of our tools. 246 "Lucía Ortiz (POP)" (463020032) 01:12:54.350 --> 01:13:14.350 That what is standardized by, well using the model and it's called grovy. Grovy is the 1st validated in scale that allows us to understand the impact of the disease in the psychological effective emotional. 247 "Lucía Ortiz (POP)" (463020032) 01:13:14.350 --> 01:13:34.350 And social aspects of the chronic patient's life. So in this way, thanks and because of using the model, we can use the results of this scale to highlight the invisible aspects of people with chronic diseases and use them in regulatory health. 248 "Britta Haenisch" (1929526528) 01:13:38.533 --> 01:13:41.732 Thank you, Lucia, so then it's Danielle. 249 "Morales Leaver Daniel" (540326912) 01:13:41.732 --> 01:14:14.890 Similar to other colleagues, I think one of the benefits of interoperability that allows once, once it's set up and there's obviously quite a high upfront cost in preparing and transforming data to a more common data model especially if it's for creating an ETL for the 1st time. But, but once established it certainly can facilitate or increase the speed of federated analytics that can be performed and also scalability. Also that there's governance bene benefits for cross border analyses. 250 "Morales Leaver Daniel" (540326912) 01:14:14.890 --> 01:14:31.687 And then the standardized analytics can reduce one aspect of heterogeneity which and maybe if looking running on your local data, there's some local assumptions required for protocol interpretation. Thank you. 251 "Britta Haenisch" (1929526528) 01:14:31.687 --> 01:14:35.692 Right? And robert. Yeah, I I agree. 252 "Piia Rannanheimo" (1370084096) 01:14:35.692 --> 01:14:36.751 You have everything that's said. 253 "Ruben van Eijk" (1763132928) 01:14:36.751 --> 01:14:56.160 So I I think like the, yeah, like the main use of of real world data especially in a rare diseases is like international exchange of data sets and there are like a common data model really, really helps. I see also risk of of using it because it seems to be very time consuming and, you know, like really to set it up and. 254 "Ruben van Eijk" (1763132928) 01:14:56.160 --> 01:15:16.160 So I, I was wondering if if like some aspects maybe you should prioritize like and others not so yeah maybe it's more like a hybrid form I can see. So like a full model might sometimes be like an overkill for for a project but it depends also I think on the ambitions. 255 "Britta Haenisch" (1929526528) 01:15:19.110 --> 01:15:39.470 Yes indeed. So there are of course advantages we have using Omar but also of course advantages using native data or embrace auto also heterogeneity of different data sources. So, we also presented some results of our use case one and. 256 "Britta Haenisch" (1929526528) 01:15:39.470 --> 01:16:02.569 And concerning ALS and breast cancer and our question would be, did you find an implausible finding for ALS or breast cancer or and then can you explain why it is implausible and speculate on the reasons for this? So this is also something that's maybe worth discussing here. 257 "Piia Rannanheimo" (1370084096) 01:16:02.569 --> 01:16:18.140 Well, I think my attention was on the question like was this really comprehensive for Finland? Was the data set comprehensive for Finland? Because I know that we have three almost data. 258 "Piia Rannanheimo" (1370084096) 01:16:18.140 --> 01:16:44.810 Data centers that are university hospitals. So I I think to assess what's implausible or not like more background information on the methods would have been needed and as as I said in my 1st response I I think like we always look at really indication specific sub. 259 "Piia Rannanheimo" (1370084096) 01:16:44.810 --> 01:17:07.636 Group from from subgroups these topics so so I I think for me reflecting to those current like decision maker and needs, this is this was on a like two broad level the the results. 260 "Britta Haenisch" (1929526528) 01:17:07.636 --> 01:17:11.053 Okay, thank you Pierre Lucia. 261 "Lucía Ortiz (POP)" (463020032) 01:17:11.053 --> 01:17:30.300 Yeah, well, thank you for the question. Here at Pop are our organization. We don't really discuss about specific disease because we represent patients with chronic conditions, so I don't really have to say much about ALS and breast cancer specifically. 262 "Lucía Ortiz (POP)" (463020032) 01:17:30.300 --> 01:17:53.284 But I think that those impossible results data mapping issues and spotting them early eh I think it could help to avoid wrong conclusions in regulatory or HDA decisions through the process. So thank you. 263 "Britta Haenisch" (1929526528) 01:17:53.284 --> 01:17:56.774 Thank you, Daniel? 264 "Morales Leaver Daniel" (540326912) 01:17:56.774 --> 01:18:29.560 I I think the impossible ones were slightly highlighted during the presentation and I think appropriately so recognizing the slightly a data capture aspect. E.g., I know the weather's much better in Portugal and that helps but in the, in the breast cancer analysis where nobody with a past history of psychiatric disorder, e.g.. So, I but I think that's sort of, I think the importance of understanding the data ahead of time characterizing the data and generating data metrics. But also importantly, I think it requires. 265 "Morales Leaver Daniel" (540326912) 01:18:29.560 --> 01:18:49.560 For for some of the things, an understanding of the clinical practice guidelines within that country as well, to really know. So I I don't think it could always be an assumption of data quality, I think they should go hand in hand and and that's sometimes a really challenging aspect especially with with federated analytics and the heterogeneity that we have. 266 "Morales Leaver Daniel" (540326912) 01:18:49.560 --> 01:18:53.193 Across Europe so. 267 "Britta Haenisch" (1929526528) 01:18:53.193 --> 01:18:56.111 Ruben. 268 "Ruben van Eijk" (1763132928) 01:18:56.111 --> 01:19:13.760 Yeah, so I was mostly reflecting on the LS data, so there, yeah, I noticed like a few discrepancies like within the care use e.g. releasual use or like some kind of medication use that that seems to be very different across countries or a little bit low. 269 "Ruben van Eijk" (1763132928) 01:19:13.760 --> 01:19:47.780 So, e.g., the 30 % really isol used like the main medication used for ALS seems seems off well we we know that like around 70 80 % of the people are using it. So, so, but yeah, it could could be like a true care practice difference across Europe, so it would be good somehow to position these data to like population based registries and what do we know from there? So how how well do those these results match with what is published already in the in the literature? Again, like what I said in the beginning. 270 "Ruben van Eijk" (1763132928) 01:19:47.780 --> 01:20:07.790 Is is that actually you're, you're interested in in much more detailed disease characteristics so it could be like like how how severely affected were these patients and how long were they sick and and this is this is very important to know clinically in order to know whether these data make actually sense. So for the oncology sort of breast. 271 "Ruben van Eijk" (1763132928) 01:20:07.790 --> 01:20:18.671 Like the the staging severity, so it would be good to have something similar for, for ALS so that that makes it easier to interpret the data in the end. 272 "Britta Haenisch" (1929526528) 01:20:18.671 --> 01:20:39.230 Indeed and also hetergenating results may also depend on recording and reimbursement practices, data prevalence so that all that you mentioned really yeah. Thank you. So, one of our last questions is so given the methodology use and also the results obtained so far in in real four. 273 "Britta Haenisch" (1929526528) 01:20:39.230 --> 01:20:54.780 How do you think a project like can be helpful in regulatory decision making or also for specific data evidence research research questions, and which impact will it have in your work? 274 "Piia Rannanheimo" (1370084096) 01:21:03.390 --> 01:21:36.620 Well, I'm speaking from from now from the HDA and national steering perspective so I I think real data is self evident that it's important in all states stages of medicines life cycle and for for us as HDA bodies we have a real task sort of to describe what are. 275 "Piia Rannanheimo" (1370084096) 01:21:36.620 --> 01:21:58.350 Evidence needs for projects like this, like every decision maker group, every steering group group has a little different evidence needs. So I think it's also relevant to really like specify which. 276 "Piia Rannanheimo" (1370084096) 01:21:58.350 --> 01:22:15.780 Sort of customer group are you aiming to serve with your data products and data services so that you can customize so regulators have a little, we have similar. 277 "Piia Rannanheimo" (1370084096) 01:22:15.780 --> 01:22:35.780 Data and evidence needs, but there are differences in in HDA perspective and and so on and so on. So I think that's crucial and then also we know that the sort of databases needs development and then also. 278 "Piia Rannanheimo" (1370084096) 01:22:35.780 --> 01:22:55.780 The processes where we sort of ask for data and and you produce evidence for us, they need development so perhaps these should be done in separate pro projects like developing the, the, the local data. 279 "Piia Rannanheimo" (1370084096) 01:22:55.780 --> 01:23:14.184 Data lakes and data warehouses and data management and then having these like international services for accessing data analyzing things jointly and so on and so on. 280 "Britta Haenisch" (1929526528) 01:23:14.184 --> 01:23:17.786 Yes, thank you Pierre. You see it? 281 "Lucía Ortiz (POP)" (463020032) 01:23:17.786 --> 01:23:34.440 Yeah, well, as patient organizations we think this project is useful eh, mostly because it can provide a stronger evidence for patient advocacy in order to making regulatory decisions. 282 "Lucía Ortiz (POP)" (463020032) 01:23:34.440 --> 01:24:04.652 It's more outcomes that are actually relevant for patients in the compensation. So, for us the impact of this project is really practical, because I think it can make real world evidence more reliable and explainable, so then it can be used in decision that affects patient access and quality of life. So thank you. 283 "Britta Haenisch" (1929526528) 01:24:04.652 --> 01:24:07.712 Thanks Lucia. Danielle? 284 "Morales Leaver Daniel" (540326912) 01:24:07.712 --> 01:24:34.390 These projects are very important. There's a temptation sometimes to go straight for the clinical information and evidence generation but understanding all the processes to get there are really important for trust and transparency. I think if we stick with the use cases of breast cancer and ALS, e.g., there's certainly great potential to establish a set of data metrics that could be. 285 "Morales Leaver Daniel" (540326912) 01:24:34.390 --> 01:24:54.390 Run against the different databases for for such reward evidence studies of you know breast cancer and ALS certainly to determine fitness for purpose, but also to identify data gaps that over time because databases have a life cycle may also be enriched through different means I. 286 "Morales Leaver Daniel" (540326912) 01:24:54.390 --> 01:25:14.390 Either through free text AI or something like that in future. So, I think that's a really good potential there. I think the other aspect is the challenging around OMOP. I have to remember that OMOP had a very traditional US evolution and we almost need to Europe fy it if that's such a word. Okay and. 287 "Morales Leaver Daniel" (540326912) 01:25:14.390 --> 01:25:34.390 That's kind of new ground. So I think all of those learnings is a temptation that everybody's doing this for the 1st time and this that might self introduce a bit of heterogeneity. So I think being transparent in how you devise that sort of mapping, try to get that out publicly and accept and adopt. 288 "Morales Leaver Daniel" (540326912) 01:25:34.390 --> 01:25:51.329 For certain areas, some things are very cancer specific, others are certain disease specific, even within cancer then. And then I think that can only help the landscape over time as well with the any recommendations for the learnings you have. 289 "Britta Haenisch" (1929526528) 01:25:51.329 --> 01:25:54.693 Indeed, thanks Danielle Ruben. 290 "Ruben van Eijk" (1763132928) 01:25:54.693 --> 01:26:27.140 Yeah, NO, I completely agree with what what Danielle was, was mentioning. I think I have very little to add. So I like I like to see this project also like as a as a benchmarking data set, so so how like from a epidemiological perspective, like how the disease is like the actual numbers on the disease and and and like how how prevalent is it, what are the differences between countries, so that could be very, very valuable, I think and and have a lot of impact. But yeah I think like the the exercise itself that's maybe the most valuable, and also to learn like what are the complicate. 291 "Ruben van Eijk" (1763132928) 01:26:27.140 --> 01:26:41.290 Locations there and maybe as a field, how can we adjust things so that that we can capture like missing aspects now and and and like how how would we like to progress? So I think it's an extremely important project. Thank you. 292 "Britta Haenisch" (1929526528) 01:26:41.290 --> 01:26:59.240 Yes, thank you Ruben. So as the workshop is near its end, now it's time to summarize so let me shortly sum up. So today we heard about the 1st results of use case one and the implications of the. 293 "Britta Haenisch" (1929526528) 01:26:59.240 --> 01:27:19.240 Findings. So the use of rework data and regulatory pre approval decisions requires of course expert knowledge especially when it comes to impact on the interpretation of also the data limitations and also when we have a look at the regulatory site of course evidence needs need to be adequately. 294 "Britta Haenisch" (1929526528) 01:27:19.240 --> 01:27:23.490 Addressed with reward data for generating reword evidence. 295 "Britta Haenisch" (1929526528) 01:27:23.490 --> 01:27:43.490 And it's also very important to be aware of the strengths and limitations of the data sources and the data quality and to conduct standard quality control procedures before the use of data. So that also the trust in the data sources can be increased also from the patient's side, e.g., so that. 296 "Britta Haenisch" (1929526528) 01:27:43.490 --> 01:28:03.490 Everyone is aware that other data sources are used also scientifically relevant and in the correct way. And sometimes it can also be time consuming to reach write a data source and to receive the variables you need as the data access can be quite at a bottleneck as we. 297 "Britta Haenisch" (1929526528) 01:28:03.490 --> 01:28:14.910 Or know from the different project and it can really be a kind of a barrier to timely reward evidence generation and that's a recurrent challenge across countries and a lot of data sources. 298 "Britta Haenisch" (1929526528) 01:28:14.910 --> 01:28:34.910 And of this this needs to be addressed properly with in more detail for the future. We also heard about the OMAP process, so the harmonization of the data to some structure does not remove of course the underlying heterogeneity of the data quality. 299 "Britta Haenisch" (1929526528) 01:28:34.910 --> 01:28:37.140 Of low or completeness. 300 "Britta Haenisch" (1929526528) 01:28:37.140 --> 01:28:57.140 And while common data models like Omap are very useful in interpreting results from different backgrounds and countries, it's still necessary to embrace also the heterogeneity of the native data, especially in view of different reimbursement practices in different countries, different recordings or the different data providence from claims data to registry based data and so on. 301 "Britta Haenisch" (1929526528) 01:28:57.140 --> 01:29:16.800 Of course, there are also some initiative also for better data access and also a catalog of different data sources especially the age, DS European health data we will large have laid large data catalog available and. 302 "Britta Haenisch" (1929526528) 01:29:16.800 --> 01:29:31.830 Yes, so the heterogeneity should also be used for assessing a given data source and its fitness for purpose in relation to the specific regulatory about also research question concerning real world data and real world evidence. 303 "Britta Haenisch" (1929526528) 01:29:31.830 --> 01:29:50.100 Concerning the regulatory side, it's also important to have a look at the scientific advice also because it can also help to get the suitable data sources involved especially when you think of industry related research or. 304 "Britta Haenisch" (1929526528) 01:29:50.100 --> 01:30:10.100 For market authorization processes. Okay, that's it for today. So now I would like to thank all the presenters roundtable discussions and all the workshop participants, of course, and I would also like to give a big thank you to the Informat team for organizing the work. 305 "Britta Haenisch" (1929526528) 01:30:10.100 --> 01:30:25.320 Workshop and a big thank you all to all refrect partners who contributed to the generation of the presented results today. So thank you very much again and have a nice afternoon. Goodbye.