New multicentre study shows how AI-drafted clinical notes can also capture structured cancer data as part of routine documentation.
Madrid, ESMO Congress 2026. Gosta Labs and clinicians from four European cancer centres presented real-world results at ESMO Congress 2026 in Madrid on Sunday, 25 October, showing that AI-assisted documentation in oncology can do more than save time on note-writing. It can also produce structured clinical data that is otherwise slow and costly to collect.
Oncology teams are under growing pressure to record structured endpoints such as performance status, tumour staging and adverse events. These data points drive treatment decisions, trial eligibility and real-world evidence, but they usually require separate manual abstraction on top of the clinical note.
The poster, Reducing the structured data burden in oncology (216P), analyses 15 months of deployment data, from July 2025 to September 2026, from Gosta's oncology AI assistant in routine care at Tumor Zentrum Aargau in Switzerland, Lapland Central Hospital and Turku University Hospital in Finland, and Centre Oscar Lambret in France. Across nearly 4000 oncology consultations, Gosta drafted the clinical note from the consultation and, from the same note, extracted ECOG performance status, TNM stage and CTCAE adverse event grading. Clinicians reviewed the note and every extracted value before transfer to the electronic medical record.
Key results:
- Under two minutes to finalise a note: The median edit time was 1 min 54 s per edited note, for consultations lasting a median of 16.5 minutes. About a third of notes (34%) were transferred to the medical record without any edits.
- Structured data at scale: The workflow produced 3,341 ECOG and 2,743 TNM extractions and identified 718 adverse effects for CTCAE grading, all as a by-product of routine documentation.
- High clinician acceptance: Clinicians kept 99% of TNM values and 90% of ECOG values unchanged. For CTCAE, 89% of adverse effects were left unchanged, and the remaining edits were split evenly between removed symptoms and grade changes.
- Improving over time: At Tumor Zentrum Aargau, ECOG concordance rose from 75% in the first quarter of use to 83–93% in later quarters.
- Works across settings: The approach fits very different consultation styles across three countries, with median encounters ranging from 11 to 30 minutes by country.
Concordance in this study reflects clinician acceptance of the AI's values rather than accuracy against a reference standard. The results support the feasibility of generating real-world oncology data directly from everyday clinical work, and the partners' next step is to validate the extracted data against an independent reference standard.
“Treatment decisions in oncology rest largely on clinical trial evidence, yet trials include only a small share of the patients oncologists see every day. When structured data is produced as part of routine documentation, every consultation adds to the picture. At Gosta Labs, our aim is to help cancer centres understand how treatments work for the patients they actually treat,” said Henri Viertolahti, Co-Founder and CPO, Gosta Labs.
“In a busy clinic, ECOG, TNM and toxicity grades are often recorded inconsistently, or only as free text. Now they’re captured as structured data during the consultation itself. That gives us reliable data for tumour boards, monitoring outcomes of treatments and trial screening, without adding extra work for our clinicians”, commented Dr. Razvan Popescu, Medical Director at Tumor Zentrum Aargau and lead author of the study.
Interested in the full poster?
If you'd like a copy of the poster or want to discuss the results with our team, reach out at henri@gostalabs.com