MatchMiner-AI: Open-Source Tool Bridges Gap Between Cancer Patients and Clinical Trials

Cancer clinical trials remain critical to advancing treatment options, yet they remain inaccessible to most patients—fewer than one in ten adults with cancer participate in therapeutic trials. Researchers have now released MatchMiner-AI, an openly available tool designed to help close this gap by automatically matching patients with trials suited to their medical profiles.

How It Works#

MatchMiner-AI combines several AI techniques to solve the matching problem. The system extracts information from unstructured clinical notes using open-source language models, summarizing each patient's medical history. In parallel, it processes trial eligibility criteria documents to identify the target populations each trial seeks. The platform then uses distilled embedding and re-ranking models—trained on synthetic electronic health record data rather than real patient information—to suggest relevant trial options and rank them by relevance.

The tool was developed in collaboration with practicing oncologists, ensuring clinical relevance at each step. Critically, the entire system relies on open-weight models and synthetic training data, making it transparent and privacy-protective.

Strong Performance Gains#

In rigorous testing, MatchMiner-AI significantly outperformed a baseline text-search approach. When evaluated on trial enrollment data, the system improved mean average precision (MAP) at 20 results from 0.44 to 0.95. On patients receiving standard-of-care therapies, performance jumped from 0.38 to 0.94—representing transformative improvements in identifying relevant matches.

When tested against a rules-based algorithm that relies on tumor genomic data, MatchMiner-AI retrieved trial options for every patient in a 50-patient test set. The existing genomic approach could only identify matches for 19 patients (38%) who had the required genetic profiling. Among those matches, 80% of MatchMiner-AI's suggestions were judged reasonable by an independent AI evaluator, compared to 53% from the traditional method.

Open Access and Transparency#

The researchers released the complete toolkit publicly: synthetic training data, model weights, inference tools, and demonstration interfaces. This open approach aims to democratize access to trial matching across healthcare institutions, removing barriers that have historically limited trial enrollment to patients at major research centers.

By combining open-source AI with privacy-preserving techniques, the work addresses a long-standing inequity in cancer care access.