#USA #Cambridge #AI Data Extraction #TriNetX #Clinical Notes

TriNetX Launches Human-Validated Data Extraction from Clinical Notes

TriNetX, the world's leading federated health data network, has announced the launch of its new feature that facilitates the extraction of validated data from clinical notes using natural language processing (NLP). This breakthrough, now available on the TriNetX LIVE platform and in downloadable datasets, aims to provide researchers with reliable data ready for analysis, extracted from over 800 million clinical notes and reports. The number is expected to surge to 1 billion by the end of 2026.

Clinical notes often contain valuable information regarding disease stage, functional status, and biomarker outcomes that enhance the structured record but pose challenges for reliable extraction. To address this, TriNetX combines artificial intelligence (AI) with profound clinical and informatics expertise, creating human-trained interaction channels based on an extensive collection of gold standard annotated notes. Unlike other approaches that stop at extraction, TriNetX rigorously evaluates each channel against human-annotated notes, assigning confidence scores to each fact and linking extracted data with structured electronic health record (EHR) data for the same anonymized patients. This enhances researchers' ability to expand cohorts and develop nuanced, clinically meaningful phenotypes.

Key Safeguards Enhancing Data Reliability

TriNetX's innovative method employs three critical safeguards to enhance the reliability of each fact extracted:

1. Human Oversight: Clinical experts meticulously review and correct AI labels, retraining models based on these corrections. Each module is required to achieve a precision score of 0.90 or higher when compared to human-annotated notes.

2. Clinical Corroboration: Extracted data is validated against expected clinical patterns, not solely relying on model metrics. For example, Eastern Cooperative Oncology Group (ECOG) survival curves showed lower survival rates for patients with a ECOG score of 2 or higher.

3. Continuous Monitoring: TriNetX actively tracks model performance as new healthcare organizations join the clinical notes program and new note types are introduced, refining models as variations emerge.

The initial results from the implementation of this feature demonstrate its promise, showing a significant increase in the coverage of smoking status from 22% to 67%. In a lung cancer cohort study, it accounted for six times more patients than structured data alone and revealed ECOG and Karnofsky functional status for 292,000 oncology patients without relying on structured fields.

August Calhoun, CEO of TriNetX, commented, "Targeting AI at a chart is easy. What distinguishes TriNetX is how we operate with scientific discipline and human expertise through a global federated network where data remain within the institutions that produced them. This means that clinical notes can serve as evidence that withstands scrutiny as AI becomes an integral part of daily research."

Dr. Matvey Palchuk, MD, MS, FAMIA, TriNetX's Medical Informatics Director, highlighted the inherent limitations of extracting detailed information from clinical notes accurately and meaningfully due to the complex nature of clinical language. "For this reason, validation is of utmost importance. Our goal is to extract facts that researchers can rely on to generate evidence."

Currently, the data extracted through NLP covers a wide range, including social history (tobacco and alcohol use), cardiology, pulmonology, general oncology, biomarkers, liver diseases, and more. These data are immediately available to members of the TriNetX network. Researchers looking for more information can visit trinetx.com/data/clinical-notes.

About TriNetX

TriNetX® serves as the global truth engine for better human health, driven by a secure, connected network of over 14,200 clinical sites across more than 20 countries. It provides reliable evidence that withstands the most rigorous scientific scrutiny in the world. By combining global data, unwavering integrity, and a unique fusion of AI and human intelligence, TriNetX accelerates discoveries and promotes better health outcomes worldwide. To learn more, visit trinetx.com or follow TriNetX on LinkedIn.

For media inquiries, contact:
Karen Tunks
Email: [email protected]