FHIR-Starter Project Aims to Structure Data in Electronic Health Records

Details

The project is a collaboration led by Fraunhofer IESE, involving Prof. Sylvia Thun’s research group at Charité Berlin and the AI company Insider Technologies. Currently, data within the Electronic Patient Record (ePA) primarily exists in non-structured PDF format, leading to difficulties in use, likened to a 'digital scroll.' To solve this, the project develops software utilizing LLMs and NLP, incorporating FHIR standards along with coding systems like LOINC and SNOMED-CT. This aims to enable doctors to automatically view longitudinal lab values or generate medication lists, achieving comprehensive and practical digitalization of the ePA. Furthermore, data will be anonymized for research use. Technically, ensuring data reliability and extensive privacy protection are key challenges—specifically preventing AI 'hallucinations.' Fraunhofer IESE is addressing this by developing an 'Uncertainty Wrapper' mechanism to quantify and manage uncertainty. The service aims to offer open interfaces for data transfer, with the goal of making the LLM as open source as possible. Funded over three years by the German Ministry for Economic Affairs, the project received 1.64 million Euros in funding. The initiative addresses the critical need to move beyond manual data entry and unstructured formats, which currently impede efficient care and research utilization across Germany.