Overview
Accurately identifying when prostate cancer worsens is critical for understanding how patients respond to treatment and how long therapies remain effective. In real-world data, prostate cancer progression can be documented in different ways, including clinician notes and changes in biomarkers like prostate-specific antigen (PSA), making it challenging to capture consistently at scale.
In this study, researchers enhanced a large language model (LLM)-based approach to identify real-world progression by incorporating both clinician-documented events and PSA-based indicators. Using data from over 370,000 prostate cancer patients in the Flatiron Health Research Database, they evaluated how often progression events were captured and how well these events aligned with downstream outcomes like treatment changes or death. The combined approach improved completeness of progression detection, with both clinician and PSA-derived signals contributing meaningful information related to outcomes.
Why this matters
This work demonstrates that combining multiple data signals can improve how prostate cancer progression is captured in large real-world datasets. By enabling more complete and accurate tracking of progression, this approach supports better measurement of outcomes like progression-free survival and can strengthen real-world evidence used in research and clinical decision-making.