TriNetX Launches Human-Validated Facts From Clinical Notes

Clinical notes hold some of the most useful facts in medicine, from disease stage to biomarker results, yet they are written in free text that computers struggle to read. AI can now pull facts out of notes quickly, but researchers want to know whether those facts can be trusted. In a press release dated November 10, 2026, TriNetX announced a product aimed squarely at that question.

This is a news summary of the company’s announcement, not medical advice.

What TriNetX launched

TriNetX, a federated health data network, announced NLP-extracted Facts from Clinical Notes. They are available in the TriNetX LIVE platform and in downloadable datasets. The company says the capability gives researchers analysis-ready data points drawn from more than 800 million clinical notes and reports, a base it expects to reach about 1 billion by the end of 2026.

The key claim is about trust. TriNetX says it builds human-in-the-loop pipelines trained on a large corpus of annotated gold-standard notes. Every pipeline is tested against human-annotated notes, every fact carries a confidence score, and notes-derived data is linked to structured electronic health record (EHR) data for the same de-identified patients. That linkage lets researchers expand cohorts and build more nuanced clinical phenotypes.

Three safeguards

The release lists three safeguards behind every extracted 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 checked against expected clinical patterns, not only model metrics. As an example, survival curves based on extracted ECOG performance status showed worse survival for patients scoring 2 and higher.
  • Continuous monitoring. TriNetX tracks model performance as new healthcare organizations join the clinical notes program and new note types arrive, refining models as language varies.

Early results

TriNetX reports three early results. Notes-derived facts raised smoking status coverage from 22 percent to 67 percent. In a lung cancer cohort study they delivered six times more patients than structured data alone. And they surfaced ECOG and Karnofsky performance status for 292,000 oncology patients who had none in structured fields.

These figures come from the company’s own release and have not been independently reviewed here. They do show why notes matter: performance status and smoking history are often recorded in narrative text rather than in coded fields, so structured-only studies can miss many patients.

Where it applies today

According to TriNetX, the extracted facts currently cover social history (smoking and alcohol status), cardiology, pulmonology, general oncology and biomarkers, and liver disease, with more areas planned. They are available now to TriNetX network members.

The company describes its network as spanning more than 14,200 clinical sites in more than 20 countries, with data staying within the institutions that generated it. CEO August Calhoun said that “pointing AI at a chart is easy,” and that the difference lies in scientific discipline and human expertise. Chief Medical Informatics Officer Matvey Palchuk added that no system can extract every detail from clinical language, which is why validation focuses on facts that can be trusted for evidence generation.

What it means

For data teams and health researchers, the announcement points to a shift in emphasis. Extraction is becoming a commodity; measured accuracy, confidence scores and links back to structured records are what make results usable in real-world evidence work. A published precision threshold of 0.90 gives buyers something concrete to question and compare.

It also shows a design pattern worth watching in health data generally: keep AI output tied to human-reviewed gold standards, monitor it as data sources change, and expose uncertainty rather than hide it. Federated approaches, where the data stays at each hospital, may help that pattern spread without moving sensitive records.

Bottom line

TriNetX is betting that researchers will pay for clinical-note data they can defend. If independent studies confirm the reported precision and coverage gains, validated, confidence-scored note facts could become a standard layer in real-world evidence platforms.

Sources