Language modelling for infectious disease detection in general practice
Language modelling for infectious disease detection in general practice
This thesis by Theodoor Homburg introduces natural language processing (NLP), an AI technique for understanding human language, as a new instrument for infectious disease detection in general practice (GP) care. At every consultation, the GP records a free text note describing what the patient reports, what the GP examines, and what is agreed upon. These notes hold a wealth of information that has remained largely untapped, and Homburg applies NLP to unlock it.
The COVID-19 pandemic was the catalyst for this work: some questions could be answered with existing research methods; others, such as who had COVID-19 before testing became available, required new methodological approaches.
The first part describes the pandemic's impact on GP care: accessibility and continuity came under pressure, government measures affected care utilisation more than regional infection rates, and the decline in regular GP care had adverse consequences for vulnerable patient groups.
The second part describes the development of AI methods applied to GP consultation notes: from manual text analysis to automated NLP models, culminating in ERNIE, an AI-framework that flags emerging infectious disease outbreaks early, even when the disease is not yet known.
For the first time, language models have been shown to analyse GP consultation notes at scale and with high accuracy. This opens up new possibilities for GP care: improved public health surveillance, with faster and more accurate recognition of infectious disease outbreaks, and better insight into clinical signals invisible in diagnostic codes. The data have always spoken; this thesis teaches us how to understand them.