- Flatiron Health presents two new studies showing the potential of AI to advance oncology research across multiple tumor types
New York – (Business Wire) – Flatiron Health today announced new discoveries from its research.Using large-scale language models for scalable extraction of real-world progression events across multiple cancer types“Special Conference on American Cancer Research Association (AACR) in Cancer Research: Presented at Artificial Intelligence and Machine Learning held in Montreal, July 10th to 12th, 2025 at QC, Canada.
Leading by an interdisciplinary Flatiron team of clinicians, research scientists and machine learning engineers, the study demonstrates the power of a large-scale language model (LLMS) to accurately and efficiently extract real-world cancer progression events from unstructured electronic health record (EHR) data across 14 cancer types. The findings concluded that LLMS achieved F1 scores similar to the abstract and expert human, generating near-identical real-world progression-free survival estimates, highlighting the possibility that AI could expand high-quality clinical endpoint extraction for oncology research and care. This study was used in a recently published publication. Verifying the accuracy of LLM/ML extracted information and data (enabled) frameworks To assess the quality of the actual data LLM extracted, we compare performance directly by uniquely evaluating both LLM and expert human abstractions against overlapping expert human abstraction datasets.
LLM provided by scientists of the Optimized Flat Iron for this study was provided by Humanity– A leading AI safety and research company that builds the Claude family of models.
“AI and machine learning are fundamentally transforming the way oncology's real-world evidence is generated and used,” said Stephanie Reisinger, Senior Vice President and General Manager of Real-World Evidence at Flatiron Health. “This study illustrates how Flatiron leverages AI and multimodal data to unlock new insights from real-world data in oncology: accelerating clinical research, improving patient outcomes, and setting new standards for evidence generation in cancer care.”
“We look forward to sharing evidence that large-scale language models can approach them. In some cases, they even outperform human experts' performance in extracting important, clinically subtle cancer progression endpoints from EHR,” he says, practicing clinical data from Aaron B. Cohen, MD, Research Oncology Director, Flatiron Health, and clinical data from Bellevue Health in New York. “Scalable and high-quality extraction of critical and complex endpoints like progress opens new doors for new research, predictive modeling and more personalized patient care.”
In addition to this work, Flatiron Health also added:Design fairness: End-to-end bias assessment of LLM generated data“It further highlights the company's leadership in applying AI and real-world data to advance oncology research. Specifically, this study demonstrates the ability of an effective framework to assess both the quality and equity of LLM extracted data, highlighting the importance of ongoing bias assessments.
About Flatiron Health
Flatiron Health is a HealthTech company that uses data to expand the possibilities of Point of Care Solutions in oncology and to enhance smarter care for everyone with cancer. Through machine learning and AI, real-world evidence, and breakthroughs in clinical trials, we continue to transform patients' real-life experiences into knowledge, creating a more modern, connected ecosystem of oncology. Flatiron Health is an independent affiliate of Roche Group.
contact address
Media Contact:
Nina Tor
press@flatiron.com
