Natural language processing in healthcare: applications and value

Abstract

This thesis explores the application of natural language processing (NLP) in healthcare, emphasizing its development, challenges, and potential clinical impact. Part 1 examines the creation of NLP models across various healthcare contexts, highlighting technical and practical hurdles. Within these contexts, classification, topic modeling, and summarization techniques are applied to three types of healthcare data: clinical data, patient-generated data, and social media data. Part 2 evaluates the added value of NLP in clinical practice through two pilot studies. In one study, an NLP tool designed for analyzing patient experiences is evaluated in a specific department. In the other study, a system that records, transcribes, and summarizes clinical conversations is assessed in a mock setting. The general discussion describes the promise NLP holds for various applications in healthcare. Several challenges are mentioned, such as data quality and availability, bias, and limited clinical evaluations. Furthermore, the impact of large language models on the field is discussed. The thesis concludes with recommendations for hospitals and computer scientists, as well as an outlook on the future of NLP in healthcare.