TriNetX adds human-validated data extraction from clinical notes
TriNetX has launched a clinical note data extraction capability that uses natural language processing, human review and clinical checks to produce research-ready data. The company says the feature is available through TriNetX LIVE and downloadable datasets.
- TriNetX has launched NLP-based extraction from clinical notes.
- The system uses human annotations, confidence scoring and clinical corroboration.
- Data has been extracted from more than 800 million notes and reports.
- Smoking-status coverage rose from 22% to 67%, according to TriNetX.
- The feature is available in TriNetX LIVE and downloadable datasets.
Turning unstructured notes into research data
The system has extracted data from more than 800 million clinical notes and reports, a figure TriNetX expects to reach 1 billion by the end of 2026. The notes can contain information such as disease stage, functional status and biomarker results that may complement structured electronic health record data.
TriNetX says the extracted information is linked to structured EHR data for the same de-identified patients. This allows researchers to expand cohorts and create more detailed clinical phenotypes instead of relying only on fields already formatted in medical records.
The data currently covers areas including social history, tobacco and alcohol use, cardiology, pulmonology, general oncology, biomarkers and liver diseases. It is available to members of the TriNetX network.
Three checks support each extracted fact
TriNetX says its approach combines artificial intelligence with clinical and informatics expertise. Its extraction pipelines are trained on a large collection of human-annotated clinical notes, and each pipeline is tested against those annotations.
The company describes three safeguards. Clinical experts review and correct AI labels, with the models retrained using those corrections. Each module must reach an accuracy of 0.90 or higher against human-annotated notes.
The extracted data is also checked against expected clinical patterns, rather than model metrics alone. TriNetX gives survival curves based on Eastern Cooperative Oncology Group functional status as an example, with lower survival among patients whose score was 2 or higher.
Finally, the company monitors model performance as more healthcare organizations join the clinical notes program and as new note types are added. It says the models are refined to account for variation across those sources.
Reported gains in coverage
According to TriNetX, note-derived data increased coverage of smoking status from 22% to 67%. In a lung cancer cohort study, it covered six times more patients than structured data alone.
The system also identified ECOG and Karnofsky functional status for 292,000 oncology patients who had no corresponding information in structured fields. These results illustrate the company’s stated aim of making information in clinical narratives available for real-world evidence research.
Why the validation matters
TriNetX says clinical language cannot be extracted completely and accurately in every case, making validation central to the usefulness of the data. Its chief executive, August Calhoun, said the company’s approach combines scientific discipline and human expertise within a federated global network, where data remains at the institutions that generated it.
The company says the capability is designed for researchers who need evidence that can withstand scrutiny while using information that is often absent from structured records. TriNetX operates a network connecting more than 14,200 clinical sites in more than 20 countries, according to its company description.
