Overview
Digital twin models—tools that simulate patient outcomes under different treatment scenarios—have the potential to transform cancer research and drug development. However, building reliable models requires large, high-quality datasets and a clear understanding of which modeling approaches perform best.
In this study, researchers used LLM-extracted data to train four models and understand their relative strengths and limitations. All models performed well, with strong agreement between predicted and observed outcomes and consistent results across patient subgroups.
Why this matters
These findings show that digital twin models can be successfully developed using real-world data and can accurately predict patients’ risk of death. This approach could help improve clinical trial design, provide deeper insights into treatment effectiveness, and ultimately accelerate the development of novel cancer therapies.