Medical Informatics Initiative: From building to using real health data

Contents

For ten years, the Medical Informatics Initiative (MII) has primarily created foundations

: standards, data integration centers, portals, and processes. At the anniversary symposium in Berlin, it became clear that the initiative is entering a crucial phase. The infrastructure is in place; now it needs to show that concrete applications for research and care can emerge from it. This was precisely the topic of the first technical session, where projects from cardiology, oncology, drug therapy safety, and AI presented their initial results.

The focus was no longer on building data spaces but on their practical use. How far the MII has come was

demonstrated by ACRIBiS, a project for “Improving cardiovascular risk identification through the synthesis of structured clinical documentation and phenotypes derived from biosignals.” The project directly addresses clinical documentation, standardizes routine data at the source, and combines it with biosignals such as EKGs. More than 3,800 patients have already been included in the cohort, and the aim is to reach over 5,000. Project coordinator Prof. Sven Zenker from the University Hospital Bonn said: “We still have a lot planned.” The project already uses standardized biosignal infrastructure, among other things, and modern AI models could be exchanged. According to Zenker, there is already international interest in reusing the dataset.

How large the gap between theoretical interoperability and lived healthcare practice can still be

became apparent in the “Integrate ATMP” (Advanced Therapy Medicinal Product) project. The Heidelberg-based project deals with advanced therapy medicinal products such as gene and cell therapies. According to Prof. Dr. Andreas Ziegler from the University Hospital Heidelberg, safety events have to be reported multiple times, sometimes still on paper. The project therefore attempts to merge routine data, registries, and patient communication into a common structure – with close integration to the MII infrastructure and the data integration centers. Patients are also to be more involved, for example via an app structure and a TI Messenger. At the same time, it became clear that such models can only function permanently if issues of financing and remuneration are also considered early on.

Dr. Christian Müller from the pharmaceutical company Bayer contributed the perspective of an industrial data user. He introduced himself as a “pharmacist, hunter, and collector of data at Bayer” and explained “Data-Insight,” a study that practically compares data from the Research Data Center Health (FDZ) and the Research Data Portal Health (FDPG). Müller explicitly stated that it is about “the Data Insight study, which has exactly this purpose, to compare the two registers.” The content concerns prevalences, incidences, medication changes, outcomes, and regional care patterns in ophthalmological indications – but also the question of how complex and reliable the use of the data actually is today.

Müller addressed points that are likely to be central to the next phase of the MII. He wanted to “put his finger on the wound” and described access from a user's perspective as complex. Although the portal is intended as a “one face to the customer,” in practice it seems as if “each center is trying to push its interests.” This is “not really comprehensible” and “not really customer-friendly either.” The criticism was not rejected by the MII but understood as an indication of an ongoing start-up phase. The term “ramp-up phase” was mentioned several times, during which procedures are tested and further developed under real usage conditions.

Data Integration Centers

Various universities also presented data integration centers, including Dr. Jan Erik Gewehr from the University Hospital Hamburg-Eppendorf (UKE) with a so-called “data hotel.” This is a secure processing environment where researchers can analyze pseudonymized data directly on-site. The data is provided in a Jupyter Notebook environment; Python, R, and Julia are available for analysis, among others. Each data space is created on a project-specific basis and receives its own pseudonymization. At the same time, Gewehr made it clear that the environment is primarily intended to help generate hypotheses. The data cannot simply be exported from the system.

The Data Integration Center (DIZ) in Jena also presented itself as a link between care, research, and industry. Efforts are being made there to implement interoperability and core datasets not only for research purposes but as closely as possible to healthcare processes. Danny Ammon explained that industrial components such as FHIR and terminology servers are now also being used for this purpose. ETL (Extract, Transform, Load) routes are outsourced to manufacturers where specifications are mature.

Heidelberg, in turn, reported on the practical setup problems of a DIZ. There are more than 250 source systems, some data on paper, many direct accesses to primary systems, and considerable effort in setting up ETL routes, data management, and quality assurance. Dr. Angela Merzweiler made it clear that each new project also brings new data flows and new requirements.

Personalized Oncology

With PM4Onco, it was also shown how strongly the MII is now influencing personalized oncology. The project links data from tumor documentation, molecular tumor boards, pathology, and patient questionnaires. The goal is to better evaluate therapy decisions and make them usable across locations. Project leader Melanie Börries emphasized: “For us, the patient is always at the center.” The ambition thus goes beyond pure data harmonization: the structures are to be built in such a way that they can effectively support clinical decisions. This also includes closer integration with cancer registry data.

“Interpolar” (Usual Care as a platform for IT-supported interventions) has a direct connection to clinical practice. The project is intended to help make medication-related problems in inpatient care and contraindications, interactions, and dosage errors visible early on. “So we want to make the blind spot of routine care measurable,” said Dr. Daniel Neumann from the University of Leipzig.

GeMTeX provides a building block for AI applications. The project is building a large clinical text corpus for AI-based language processing. So far, real, sufficiently processed clinical text data is largely lacking in Germany. Justin Hofenbitzer from TUM described this as a “big gap in Germany.” GeMTeX aims to close this gap with pseudonymized documents that should not allow any conclusions to be drawn about individuals, annotations, and later provision via the FDPG.

In several contributions and in the final discussion, it became clear that companies now perceive the MII infrastructure as a relevant access point to data. They wish to be involved earlier and more systematically. André Sander, Co-Managing Director of ID Berlin, wished “that we rely more on the solutions that already exist in industry.” He cited the Interopolar project as an example. For Sander, the impression was that “inpatient drug therapy in Germany is in a kind of blind flight.” He contradicted this with reference to existing systems. For example, “UKE started introducing a medication solution a year before the Medical Informatics Initiative.” Sander also referred to previous experiences with the MII. Mehmet Akyüz from TriNetX Europe – a platform that bundles real-world data from clinics and healthcare organizations to quickly identify patient cohorts for studies – added that cooperation between industry and healthcare sites could begin earlier from his perspective, ideally already at the project initiation stage.

Cooperation between Industry and Science

The Managing Director of the Technology and Methods Platform for Networked Medical Research, Sebastian C. Semler, also confirmed that closer cooperation with industry is not only desired in terms of content but often encounters hurdles in funding. It remains difficult to involve companies that are not fully eligible for funding as subcontractors in funding projects; in addition, there are “all the antitrust hurdles.” If society truly wants science and industry to work more closely together, “we need to refine these procedures again.”

On another panel at the symposium, a structural problem was also identified that goes beyond individual application projects: Not every scientific idea is developed further into healthcare. According to Dr. Susen Burock, from the Coordination Center for Clinical Studies (KKS-Netzwerke e.V.), young, dedicated researchers often come up with an idea that they want to test and document in a paper. When these individuals leave the university, further development often stalls. This is precisely one of the central challenges for the next phase of the MII: individual projects and publications must become robust, permanently usable solutions.

(mack)