New NLP-extracted clinical facts are confidence-scored, tested against human-annotated notes, and linked to structured EHR data for the same de-identified patients
CAMBRIDGE, Mass., Nov. 10, 2026 /PRNewswire/ — As artificial intelligence (AI) makes extracting information from clinical notes easier than ever, the harder question for researchers is whether those data can be trusted for real-world evidence. TriNetX®, the world’s leading federated health data network, today announced NLP-extracted Facts from Clinical Notes, now available in the TriNetX LIVE™ platform and in downloadable datasets. The capability gives researchers analysis-ready data points drawn from over 800 million clinical notes and reports, a base expected to reach about 1 billion by the end of 2026.
Clinical notes hold details such as disease stage, performance status, and biomarker results that complement the structured record, but are hard to extract reliably. TriNetX pairs AI with deep clinical and informatics expertise to build human-in-the-loop pipelines trained on a large corpus of annotated gold standard notes. Unlike approaches that stop at extraction, TriNetX tests every pipeline against human-annotated notes, attaches a confidence score to every fact, and links notes-derived data to structured electronic health record (EHR) data for the same de-identified patients, so researchers can expand cohorts and construct nuanced, clinically meaningful phenotypes.
Three safeguards support every fact:
Human oversight. Clinical experts review and correct what the AI tags, and models are retrained on those corrections. Each module must reach a precision of 0.90 or higher against human-annotated notes.
Clinical corroboration. Extracted facts are tested against expected clinical patterns, not just model metrics. For example, survival curves based on extracted Eastern Cooperative Oncology Group (ECOG) performance status showed worse survival for patients scoring 2 and higher.
Continuous monitoring. TriNetX monitors model performance as new healthcare organizations participate in the clinical notes program, and new note types arrive, refining models as variation appears.
Early results show notes-derived facts increased smoking status coverage from 22% to 67%, delivered 6x more patients than structured data alone in a lung cancer cohort study, and surfaced ECOG and Karnofsky performance status for 292,000 oncology patients with none in structured fields.
“Pointing AI at a chart is easy. What sets TriNetX apart is how we do it: with scientific discipline and human expertise across a global federated network where data stay within the institutions that generated them,” said August Calhoun, CEO, TriNetX. “That means clinical notes can inform evidence that stands up to scrutiny as AI becomes part of everyday research.”
“There’s no system that can extract every detail from clinical notes in a way that’s both accurate and meaningful. It’s a limitation inherent to the nature of clinical language,” said Matvey Palchuk, MD, MS, FAMIA, Chief Medical Informatics Officer, TriNetX. “That’s why validation matters. Our goal is to extract the facts that can be trusted for evidence generation.”
NLP-extracted Facts from Clinical Notes currently cover social history (smoking and alcohol status), cardiology, pulmonology, general oncology and biomarkers, liver disease, and more, and are available now to TriNetX network members. Learn more at trinetx.com/data/clinical-notes.
About TriNetX, LLC
TriNetX® is The Global Truth Engine for Better Human Health®. Powered by a secure, connected network across more than 14,200 clinical sites in 20+ countries, TriNetX delivers trustworthy evidence that stands up to the world’s most rigorous scientific scrutiny. By combining global data, relentless integrity, and a rare fusion of AI and human intelligence, TriNetX accelerates discovery and drives better health outcomes worldwide. Visit trinetx.com or follow TriNetX on LinkedIn to learn more.
Media Contact
TriNetX
Karen Tunks
Email: [email protected]
SOURCE TriNetX
