YOUR GO-TO INTELLIGENCE RESOURCE FOR THE SCHOLARLY PUBLISHING INDUSTRY

The European Society for Medical Oncology (ESMO) has released the ‘ESMO Guidance on the Use of Large Language Models in Clinical Practice (ELCAP)’, presented as the first structured framework to help patients, clinicians, and institutions integrate AI language tools safely in oncology. The guidance appears in ESMO’s peer-reviewed journal Annals of Oncology and aligns with a dedicated ESMO Congress 2025 session on ChatGPT and cancer care in Berlin.

ELCAP organizes recommendations by user type and translates high-level principles into 23 consensus statements for routine practice. Type 1 covers patient‑facing applications such as education and symptom support via chatbots, which should complement—not replace—clinical care, operate under supervised pathways with explicit escalation, and ensure robust data protection.

Type 2 addresses healthcare professional-facing tools for decision support, documentation, and translation, which require formal validation, transparent limitations, and explicit human accountability for clinical decisions.

Type 3 concerns background institutional systems linked with electronic health records for data extraction, automated summaries, and clinical-trial matching; these require pre-deployment testing, continuous monitoring for bias and performance drift, institutional governance, and re-validation when processes or data sources change. Clinicians should be aware of such systems operating in their environment, given dependencies on interoperability and privacy‑by‑design measures.

The document emphasizes that output reliability depends on input quality: gaps in documentation or partial patient queries can produce inaccurate or misleading responses, reinforcing the need for supervision and clear escalation routes.

ELCAP focuses on assistive LLMs that work under human oversight and support—rather than replace—clinical workflows and decision-making. It also acknowledges rapidly emerging autonomous “agentic” AI models that can initiate actions without direct prompts, noting these raise distinct safety, regulatory, and ethical challenges that will require future guidance. The guidance was developed between November 2024 and February 2025 by an international, multidisciplinary panel under ESMO’s Real World Data & Digital Health Task Force.