Clinical Trials LLMs

Large Language Models in Clinical Trials


"Onco-LLM"

Multi Center Oncology AI Collaborative

Designing a Community-Driven, Oncology-Focused Large Language Clinical Trials Language Model and Retrieval Pipeline

Project Launch June 2023 - Completed October 2024

OncoLLM Initiative: Successful Completion


The Oncology Clinical Trials Large Language Model Collaborative Initiative (OncoLLM) was established to advance oncology clinical research by developing and validating a specialized large language model (LLM) framework through generative AI (Gen AI) integration. In collaboration with Triomics, the initiative fostered innovation and collaboration with the goal of evolving into an AI-driven learning model for clinical research. Designed to enhance efficiency, accuracy, and ethical integrity in oncology clinical trials, OncoLLM has reached a significant milestone with the successful completion of its first phase. During Phase 1, Triomics developed a robust and scalable AI-driven framework, demonstrating the transformative potential of LLMs in cancer research, clinical trial design, and data-driven decision-making. Building on this success, Triomics have now expanded OncoLLM into a comprehensive AI-powered oncology data platform think tank.


Transition to an Established AI Framework


With the completion of the OncoLLM Cancer Center Initiative, Triomics is excited to announce they have transformed this work into a fully established and actively utilized AI framework, integrating the best of OncoLLM with contributions from the broader Cancer Clinical Trials community. This transition ensures that OncoLLM continues to drive innovation, collaboration, and impact—supporting researchers, clinicians, and data scientists in enhancing trial design, patient recruitment, and precision oncology insights.  The Triomics collaboration has expanded to focus on leveraging advanced AI methodologies to extract high-quality, structured insights from unstructured clinical data. By integrating AI-driven data extraction, OncoLLM now bridges the gap between clinical trials, patient care, and evidence generation—accelerating data-driven advancements in cancer treatment and regulatory decision-making.


Scientific Validation


The OncoLLM model’s scientific validation was recently published in Nature Digital Medicine: [link] . This study demonstrates that OncoLLM can analyze complete patient charts with approximately 95% accuracy, identifying which portfolio trials at an institution are suitable matches for patients. Additionally, it provides assessments and explanations for trial criteria with about 90% accuracy.


In early 2024, OncoLLM was prospectively deployed at the Medical College of Wisconsin Cancer Center (MCW), where it maintained its high level of accuracy and led to:

• Proactive screening of all patient visits, with the AI system identifying 72% of all patients enrolled in Q3 and Q4 before their respective visits.

• Significant reduction (75%) in trial eligibility and chart review times.

• Increased interventional accruals:

• 39% increase in Q3 2024 over the same period in 2023.

• 27% increase in Q4 2024 over the corresponding period in 2023.


Triomics has been a valued technology validation partner, playing a key role in advancing this initiative. Institutions now have the opportunity to explore OncoLLM’s capabilities through custom sandbox access, enabling secure sharing of synthetic, de-identified, or identified longitudinal medical records for patient-trial screening and workflow evaluation. . or sandbox access, please contact Sebastien Rhodesby (sebastien@triomics.com).


Share by: