The German healthcare system is currently lagging behind digitalization. Doctors communicate their findings via doctor's letters, which are available as full texts in PDF format. Reading the reports, which are usually several pages long, is time-consuming and makes it difficult to compare them with previous findings. In addition, medical professionals have to manually type out important information in order to transfer it to practice management systems (PVS) or hospital information systems (HIS). This is not only cumbersome, but also prone to errors and worsens patient care. In addition, medical and pharmaceutical research can only use data from full texts with difficulty, which weakens Germany as a research location.

Analysing full-text documents and converting them into a standardized data format

As part of FHIR-Starter, researchers from the consortium are developing a software service that uses LLMs and natural language processing (NLP) to analyse full-text documents and convert them into standardized data formats. The FHIR medical data standard and the LOINC and SNOMED-CT coding systems are used for this purpose. Such uniform data standards hold great potential for the effective use of healthcare data and serve to support data exchange between software systems in the healthcare sector. The software service will offer open interfaces that enable service providers, healthcare software providers and secondary users of healthcare data to automatically incorporate the structured data into their systems. 

When developing the software service, there are two main challenges to consider: ensuring the reliability of the data as well as comprehensive data protection. But what exactly is meant by the reliability of the data? For AI models such as LLMs, truth is not a category. If an LLM does not have sufficient data, it invents information freely - it hallucinates. In the case of medical data, this can have fatal consequences. 

Special expertise is therefore required to ensure that the data structured using the software service is also reliable. Dr. Theresa Ahrens, head of the Digital Health Engineering department at Fraunhofer IESE, emphasizes: “If we want to use large language models in healthcare responsibly and thus exploit their potential, it is necessary to set up appropriate security mechanisms. With the development of the Uncertainty Wrapper, Fraunhofer IESE has already done important groundwork in this area, which we can build on in this project.” The Uncertainty Wrapper is a tool that quantifies, manages and reduces uncertainties in AI models. 

The second major challenge for FHIR starters is the responsible and GDPR-compliant processing of sensitive medical data. Standard LLMs on the market usually use servers abroad that do not meet the consortium's security requirements. Instead, the software service to be developed will therefore be based on open source LLMs that run on the users' own servers. This will create a self-contained and secure system that also complies with GDPR requirements.

Use cases for structured data in the healthcare sector 

The possible applications of the software service are many and varied. For Charité, for example, FHIR-Starter offers the opportunity to make medical data available to researchers in anonymized form in the future. Conversely, Insiders Technologies wants to focus on automating digitization processes in practices and hospitals with the help of the software service. The consortium is open to further cooperation beyond the existing project partners. 

According to Theresa Ahrens, FHIR-Starter also has the potential to take the use of electronic patient records (ePA) to a whole new level, independently of the project goals already defined: “The software service could allow doctors to display laboratory values over time or automatically create medication lists, for example. The structured data could be used to fully and meaningfully digitize the electronic patient record.”

About the Fraunhofer Institute for Experimental Software Engineering IESE

The Fraunhofer Institute for Experimental Software Engineering IESE in Kaiserslautern has been one of the leading research institutions in the area of efficient software and systems engineering for almost 30 years. With its applied research, the institute develops innovative digital solutions in the context of digital twins, virtual validation and simulation, digital ecosystems, cyber security and safety, among others. The focus of research and solutions is always on increasing efficiency while ensuring high quality, for example by means of AI support in software and systems engineering.  

Fraunhofer IESE provides support in overcoming challenges in diverse fields of application such as health, agriculture and food, mobility, production, smart city and defense. In well over 2,000 customer projects, the institute has transferred cutting-edge research into sustainable business practices and innovative products.  

Fraunhofer IESE is one of 76 institutes and facilities of the Fraunhofer-Gesellschaft. Together, they play a key role in shaping applied research in Europe and worldwide and contribute to Germany's international competitiveness.