Blog

Missed ESMO 2026? Our Scientific Highlights in Oncology

By Mariana Spínola, Sr Clinical Specialist at Biorce

At ESMO 2026 in Madrid, I spent most of my time in the AI sessions, from large language models to AI in the tumor board. AI literacy came up more than once, and ESMO's own guidance now lists it as a requirement for safe use (Wong et al., 2025).

I came back with one clear impression. Most of what we saw focused on the clinic and the patient. Less of it reached the protocol, where many of the problems those tools meet begin.


AI at ESMO 2026 Focused on the Clinic

The sessions we attended reflected how far AI has already entered practice:

  • Over 80% of physicians already use AI professionally, and around 40 million people use ChatGPT for health questions (Kather, 2026)

  • AI literacy now listed by ESMO among four requirements for safe use of large language models (Wong et al., 2025)

  • Eligibility screening cut from 4.5 minutes to one per lung cancer patient, with uncertain cases deferred to oncologists (INT Milan AI-ON-Lab, 2026)

  • An open trial identified for 92% of colorectal cases ready for tumor board discussion (Ashraf, 2026)

Each of these tools starts from a protocol that is already written. When its eligibility criteria or recruitment assumptions do not hold at real sites, the only fix is to change the protocol. In the sessions we attended, few speakers addressed that earlier step.

Why Oncology Protocol Amendments Matter

Every change to a protocol after approval leaves a record: an amendment. In oncology, that record is long, with 91.1% of protocols amended at least once, against 72.1% elsewhere (Botto et al., 2024).

Most of those changes, 82%, are unavoidable, driven largely by regulatory requests, changes in study strategy, and new data (Botto et al., 2024). The rest tend to trace back to drafting: eligibility criteria few sites can meet, untested recruitment assumptions, and late feasibility input.  Read more about the true cost of protocol amendments.

Our ESMO poster, AI-Native Protocol Generation and Avoidable Amendments in Oncology Trial Design, focused on that avoidable share. Across eight completed phase I–III protocols, independent experts classified each amendment change according to the Tufts framework (Getz et al., 2016). Where most changes were avoidable, an AI-generated first draft had already pre-empted 58% of them, with a range of 54.8–62.5% (Spínola et al., 2026).

In other words, much of what teams can prevent is visible before the first site opens. The pilot was small, and larger studies are now under way.


What ESMO suggests comes next

At the poster, a research nurse explained that an amendment stops recruitment at her site until the team is retrained and every enrolled patient signs a new consent form. AI can catch many of those changes while the protocol is still a draft.  

A regulatory affairs lead asked whether agencies would accept an AI-drafted protocol, given that each country has its own guidance. AI can check a draft against those requirements with a human in the loop, as the EMA and FDA guiding principles expect: an expert reviews and signs off every output (European Medicines Agency & U.S. Food and Drug Administration, 2026).

AI is already helping patients find trials. The next gains are in using it to design trials that patients can join.

References

Ashraf, S. (2026, October 23–26). AI preparation of colorectal multidisciplinary team meetings [Conference presentation]. ESMO Congress 2026, Madrid, Spain. [VERIFY]

Botto, E., Smith, Z., & Getz, K. (2024). New benchmarks on protocol amendment experience in oncology clinical trials. Therapeutic Innovation & Regulatory Science, 58(4), 645–654. https://doi.org/10.1007/s43441-024-00629-2

European Medicines Agency, & U.S. Food and Drug Administration. (2026). Guiding principles of good AI practice in drug development. [ADD URL]

Getz, K. A., Stergiopoulos, S., Short, M., Surgeon, L., Krauss, R., Pretorius, S., Desmond, J., & Dunn, D. (2016). The impact of protocol amendments on clinical trial performance and cost. Therapeutic Innovation & Regulatory Science, 50(4), 436–441. https://doi.org/10.1177/2168479016632271

Getz, K., Smith, Z., Botto, E., Murphy, E., & Dauchy, A. (2024). New benchmarks on protocol amendment practices, trends and their impact on clinical trial performance. Therapeutic Innovation & Regulatory Science, 58(3), 539–548. https://doi.org/10.1007/s43441-024-00622-9

INT Milan AI-ON-Lab. (2026, October 23–26). [Title of neuro-symbolic trial matching poster] [Poster presentation]. ESMO Congress 2026, Madrid, Spain. [VERIFY authors and title]

Kather, J. N. (2026, October 23–26). Generative AI, LLMs and agents in oncology [Conference presentation]. ESMO Congress 2026, Madrid, Spain. [VERIFY]

Spínola, M., Guidotti, L., & Bernardes, C. (2026, October 23–26). AI-native protocol generation and avoidable amendments in oncology trial design [Poster presentation]. ESMO Congress 2026, Madrid, Spain.

Wong, E. Y. T., Verlingue, L., Aldea, M., Franzoi, M. A., Umeton, R., Halabi, S., et al. (2025). ESMO guidance on the use of large language models in clinical practice (ELCAP). Annals of Oncology, 36(12), 1447–1457. https://doi.org/10.1016/j.annonc.2025.09.001

Get the inside track on clinical trial innovation Insights that pharma leaders actually read. No fluff.

Submit

Clinical Trials For All participating partner
AICPA SOC 2 certification

©2026 Biorce | All Rights Reserved

Partners with

Google
NVIDIA

Get the inside track on clinical trial innovation Insights that pharma leaders actually read. No fluff.

Submit

Clinical Trials For All participating partner
AICPA SOC 2 certification

©2026 Biorce | All Rights Reserved

Partners with

Google
NVIDIA

Get the inside track on clinical trial innovation Insights that pharma leaders actually read. No fluff.

Submit

Clinical Trials For All participating partner
AICPA SOC 2 certification

Partners with

©2026 Biorce | All Rights Reserved