Related Experiment Video
Updated: Jun 20, 2026
08:50
Predictive Immune Modeling of Solid Tumors
Published on: February 25, 2020
A Multi-Model LLM Consensus Framework to Identify EHR-Predictable Eligibility Criteria in NSCLC Immunotherapy Trials
Abdul Muqeeth1, Yu Huang1,2, Jiang Bian1,2
1Department of Biomedical Engineering and Informatics, Luddy School of Informatics, Computing, and Engineering.
Summary
Improving oncology clinical trial enrollment requires simplifying eligibility criteria. Our framework converts complex criteria into standardized concepts, identifying those easily extracted from electronic health records (EHRs) for better patient matching.
Area of Science:
- Oncology
- Clinical Informatics
- Artificial Intelligence
Background:
- Clinical trial participation in oncology is hindered by complex eligibility criteria.
- These criteria limit patient accrual and access to novel cancer therapies.
Purpose of the Study:
- To develop a structured framework using large language models (LLMs) to standardize free-text eligibility criteria from Phase III PD-1/PD-L1 non-small cell lung cancer trials.
- To assess the predictability of these criteria from Electronic Health Record (EHR) data.
Main Methods:
- A multi-model LLM consensus pipeline was employed to convert free-text eligibility criteria into standardized umbrella concepts.
- Concepts were classified by their predictability using routine EHR data (structured and unstructured).
Main Results:
- Hundreds of eligibility traits were consolidated into a manageable set of clinical concepts.
- Approximately half of the concepts were deemed clinically important and inferable from EHR data.
- The remaining concepts were either low-value for prediction or required specialized testing.
Conclusions:
- The developed taxonomy prioritizes 'predictable targets' for Natural Language Processing (NLP) and machine learning models.
- This framework offers a blueprint for creating more computable eligibility criteria and EHR-driven prescreening tools.
- Enhancing EHR data utilization can streamline clinical trial recruitment in oncology.