A Framework for Curating High-Quality Lung Cancer Real-World Data Using Natural Language Processing
Year: 2024
Type: dissertation
Abstract: Real-world data (RWD) within unstructured text is essential to determine treatment effectiveness in diverse populations. This thesis aims to scale the process of obtaining RWD using natural language processing (NLP) to extract unstructured data from medical text within lung cancer patients' electronic health records (EHR). We utilized either a pre-trained chat model or fine-tuned a language model to organize textual data into a spreadsheet format from EHR. Subsequently, we evaluated its performance against manually curated data. We showed that we can efficiently and accurately obtain lung cancer biomarkers and progression data with NLP, using a combination of different text reports, and this data can be used to compare treatments in an RWD scenario. Our results highlight the variability of medical text data and the need for a fit-for-purpose solution to generate RWD.
Source: TSpace (University of Toronto)
Author: Ajaj, Rami
Institution: University of Toronto
Language: English
Subfield: Health Informatics
Field: Medicine
Domain: Health Sciences
Open Access status: green
Funder: University of Toronto
OpenAlex ID: W7225768709
