PhD defence T.M. (Tom) Seinen

Below is a brief summary of the dissertation: 

Although much information in electronic health records is now digital and standardized, a significant portion remains hidden in unstructured clinical texts. This presents opportunities for deeper medical insights but also challenges due to complex medical terminology, the large volume of data, and the lack of validated Natural Language Processing (NLP) techniques for Dutch clinical text.
This thesis introduces an innovative method using machine translations to create multilingual datasets, enabling the validation and improvement of concept extraction in languages such as Dutch. Furthermore, we explored how much richer information can be extracted from unstructured data compared to structured data.
Through a literature study and the use of various NLP techniques on a large general practitioner database, we demonstrate that valuable information from text can enhance clinical prediction models. We observed that combining structured and unstructured data often results in better models, depending on the context. We also found that the accuracy of clinical codes can be improved by leveraging information from text.
This thesis contributes to Dutch clinical NLP and points the way toward improved research and care through the integration of structured and unstructured medical information. Future developments in multilingual language models could lead to even more accurate and robust results worldwide.