Abschlussarbeit

Master Thesis

INTEGRATING MEDICATION LEAFLETS INTO A MOBILE APPLICATION USING FHIR AND AN LLM-BASED QUESTION-ANSWER PIPELINE

Kurzfassung / Abstract: In the European Union, it is a legal requirement for all medications to include a medication leaflet in the packaging, as it provides essential information about the medication. Challenges associated with paper-based leaflets have led to efforts to digitize and standardize medication leaflets. The standardization of healthcare related information is a central topic in the realm of digital healthcare, and a broad area of research explores different approaches to achieve this goal. Furthermore, the rapid growth in accessible healthcare data requires structuring and standardizing data to enable computer-based processing. This master’s thesis aims to automate the extraction, digitization, and standardization of information from medication leaflets for integration into a mobile medication management application. Therefore, an existing prototype of a mobile medication management application was further developed using Python for the back-end and Flutter with Dart for the front-end. OpenAI’s GPT-4o was used for the information extraction service. The extracted information was standardized using FHIR, particularly FHIR profiles. Finally, the FHIR resources were stored in a PostgreSQL database through the HAPI-FHIR server. The mobile application receives the FHIR resources by addressing the HAPI-FHIR server via RESTful API and HTTP requests. To assess the performance of the information extraction pipeline, the responses were systematically evaluated. The developed setup showcases an approach to information extraction by utilizing an LLM-based questionanswer pipeline. It illustrates the successful transformation of the extracted information into the FHIR standard through a template-based mapping approach. The evaluation revealed that delimited and straightforward information can be reliably extracted through the pipeline. However, recurring errors across different medication leaflets indicate that the question-answer pipeline requires further refinement, particularly for extracting complex and nested information. By addressing this challenge, this approach opens up possible new pathways in information management. This work demonstrates a way to combine the processes of information extraction and standardization and may lay the groundwork for further research which could also extend the approach to other areas of healthcare and broaden its applicability.

Zum Volltext: Download