Reproduce “Foresight -- Generative Pretrained Transformer (GPT) for Modelling of Patient Timelines using EHRs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zeljko Kraljevic, Dan Bean, Anthony Shek, Rebecca Bendayan, Harry Hemingway, Joshua Au Yeung, Alexander Deng, Alfie Baston and 4 more
Foresight, a Generative Pretrained Transformer developed by researchers at King's College London, models patient timelines by integrating both structured and unstructured Electronic Health Record (EHR) data through advanced natural language processing. The model demonstrates high clinical relevancy in forecasting a wide array of future biomedical events, with clinicians deeming 97% of top forecasted disorders relevant and achieving strong quantitative performance across diverse datasets.
Key insights
- Integrating unstructured free-text clinical notes with structured EHR data using advanced NLP significantly enriches patient representations, enabling more comprehensive and accurate predictive models.
- A generative, causal language modeling approach (GPT-2) allows for the simulation of entire multi-step patient timelines, moving beyond single-event predictions to dynamic trajectory forecasting.
- Clinician validation of forecasted concepts is crucial for ensuring the practical relevancy and utility of AI models in real-world clinical settings, bridging the gap between technical accuracy and clinical applicability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.