Assessment of Zero-Shot Large Language Model (LLM) Assisted Clinical Trial Matching Processes: A Metastatic Cancer
Medrxiv : the Preprint Server for Health Sciences
|July 17, 2026
Summary
Large language models (LLMs) show promise in oncology clinical trial matching for patients with limited options. This study found LLM-assisted matching comparable to human review, highlighting potential benefits for cancer patients.
Area of Science:
- Oncology
- Medical Informatics
- Artificial Intelligence
Background:
- Clinical trials offer vital treatment options for oncology patients.
- Identifying suitable clinical trials is a significant challenge for patients and oncologists.
- Large language models (LLMs) are being explored to improve patient-trial matching efficiency.
Purpose of the Study:
- To evaluate the benefits and challenges of zero-shot LLM-assisted clinical trial matching.
- To analyze the performance of LLM-based matching against a human benchmark for a pancreatic cancer patient.
Main Methods:
- A zero-shot LLM was used for clinical trial matching.
- Results were compared to a manually curated "human benchmark" from ClinicalTrials.gov.
- Performance metrics (sensitivity, specificity, precision, accuracy) and qualitative content analysis of LLM reasoning were conducted.
Main Results:
- LLM-assisted matching achieved 81.1% sensitivity, 89.3% specificity, and 86.5% precision.
- Qualitative analysis indicated that 73% of errors could be mitigated through improved prompting and data access.
- The LLM's overall performance was comparable to human reviewers.
Conclusions:
- LLM-assisted clinical trial matching demonstrates potential for improving patient access to trials.
- This preliminary case study supports integrating LLMs into oncology care for patients with limited treatment alternatives.
