ESMO basic requirements for AI-based biomarkers in oncology (EBAI)

M Aldea; M Salto-Tellez; A Marra; R Umeton; A Stenzinger; M Koopman; A Prelaj; K L Kehl; S Gilbert; M-E Leßmann; J Lipkova; L Provenzano; F Meric-Bernstam; S Halabi; J Wu; A Pellat; K P M Suijkerbuijk; B Besse; B Ryll; C Marchió; M Crispin-Ortuzar; R Fehrmann; J Vibert; D Ferber; C Pauli; A Valachis; Federica Corso; T J Brinker; J Mateo; N Harbeck; E C Winkler; F Lopez-Rios; R Perez-Lopez; G Pentheroudakis; S Delaloge; C Benedikt Westphalen; J N Kather

doi:10.1016/j.annonc.2025.11.009

ESMO basic requirements for AI-based biomarkers in oncology (EBAI)

M Aldea
, M Salto-Tellez
, A Marra
, R Umeton
, A Stenzinger
, M Koopman
, A Prelaj
, K L Kehl
, S Gilbert
, M-E Leßmann
, J Lipkova
, L Provenzano
, F Meric-Bernstam
, S Halabi
, J Wu
, A Pellat
, K P M Suijkerbuijk
, B Besse
, B Ryll
, C Marchió

M Crispin-Ortuzar, R Fehrmann, J Vibert, D Ferber, C Pauli, A Valachis, Federica Corso, T J Brinker, J Mateo, N Harbeck, E C Winkler, F Lopez-Rios, R Perez-Lopez, G Pentheroudakis, S Delaloge, C Benedikt Westphalen, J N Kather

Research output: Contribution to journal › Article › peer-review

41 Citations (Scopus)

64 Downloads (Pure)

Abstract

Background
Artificial intelligence (AI) is expected to introduce an increasing number of biomarkers in oncology. To bridge the gap between oncology and computer science, it is timely to define recommendations for AI-based biomarkers suitable for routine clinical use. Here, we propose the ESMO (European Society for Medical Oncology) Basic Requirements for AI-based Biomarkers In Oncology (EBAI).

Design
The EBAI framework was developed using a modified Delphi methodology, involving a multidisciplinary panel of 37 experts who participated in four structured consensus rounds.

Results
AI-based biomarkers were classified as ‘class A’ (AI quantification of established biomarkers), ‘class B’ (indirect measure of known biomarkers using AI-based alternative methods, to be deployed as pre-screening tests), and ‘class C’ (novel AI-derived biomarkers, with C1 for prognosis and C2 for prediction of treatment effect). The EBAI framework addresses AI biomarkers for clinical use. Ground truth, performance, and generalisability were considered essential; fairness was recommended. Minimal validation requirements indicate that class A requires concordance studies, class B analytical validation, class C1 high-quality retrospective real-world or clinical trial data, and class C2 additionally requires clinical validation in prospective clinical trials for the prediction of response to a new treatment. All biomarker studies should report multiple evaluation and calibration metrics, with a clearly defined primary objective. Generalisability should be demonstrated across all intended use settings, including variability in data acquisition, post-processing, and population characteristics. Biomarkers must not be applied to other cancer types or modalities without supporting evidence.

Conclusions
EBAI defines criteria for AI-based biomarker adoption in routine use, providing a common language for physicians, AI developers, and researchers.

Original language	English
Pages (from-to)	414-430
Number of pages	17
Journal	Annals of Oncology
Volume	37
Issue number	3
Early online date	12 Feb 2026
DOIs	https://doi.org/10.1016/j.annonc.2025.11.009
Publication status	Published - Mar 2026

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 3 Good Health and Well-being

Keywords

biomarker
EBAI
cancer
validation
scale
artificial intelligence

Access to Document

10.1016/j.annonc.2025.11.009Licence: CC BY-NC-ND

ESMO basic requirements for AI-based biomarkers in oncology (EBAI)
Copyright 2025 The Authors. This is an open access article published under a Creative Commons Attribution-NonCommercial-NoDerivs License (https://creativecommons.org/licenses/by-nc-nd/4.0/), which permits distribution and reproduction for non-commercial purposes, provided the author and source are cited.
Final published version, 1.25 MBLicence: CC BY-NC-ND

Cite this

Aldea, M., Salto-Tellez, M., Marra, A., Umeton, R., Stenzinger, A., Koopman, M., Prelaj, A., Kehl, K. L., Gilbert, S., Leßmann, M.-E., Lipkova, J., Provenzano, L., Meric-Bernstam, F., Halabi, S., Wu, J., Pellat, A., Suijkerbuijk, K. P. M., Besse, B., Ryll, B., ... Kather, J. N. (2026). ESMO basic requirements for AI-based biomarkers in oncology (EBAI). Annals of Oncology, 37(3), 414-430. https://doi.org/10.1016/j.annonc.2025.11.009

@article{da157f241647470f9f1c656a932d3fdf,

title = "ESMO basic requirements for AI-based biomarkers in oncology (EBAI)",

abstract = "Background Artificial intelligence (AI) is expected to introduce an increasing number of biomarkers in oncology. To bridge the gap between oncology and computer science, it is timely to define recommendations for AI-based biomarkers suitable for routine clinical use. Here, we propose the ESMO (European Society for Medical Oncology) Basic Requirements for AI-based Biomarkers In Oncology (EBAI). Design The EBAI framework was developed using a modified Delphi methodology, involving a multidisciplinary panel of 37 experts who participated in four structured consensus rounds. Results AI-based biomarkers were classified as {\textquoteleft}class A{\textquoteright} (AI quantification of established biomarkers), {\textquoteleft}class B{\textquoteright} (indirect measure of known biomarkers using AI-based alternative methods, to be deployed as pre-screening tests), and {\textquoteleft}class C{\textquoteright} (novel AI-derived biomarkers, with C1 for prognosis and C2 for prediction of treatment effect). The EBAI framework addresses AI biomarkers for clinical use. Ground truth, performance, and generalisability were considered essential; fairness was recommended. Minimal validation requirements indicate that class A requires concordance studies, class B analytical validation, class C1 high-quality retrospective real-world or clinical trial data, and class C2 additionally requires clinical validation in prospective clinical trials for the prediction of response to a new treatment. All biomarker studies should report multiple evaluation and calibration metrics, with a clearly defined primary objective. Generalisability should be demonstrated across all intended use settings, including variability in data acquisition, post-processing, and population characteristics. Biomarkers must not be applied to other cancer types or modalities without supporting evidence. Conclusions EBAI defines criteria for AI-based biomarker adoption in routine use, providing a common language for physicians, AI developers, and researchers. ",

keywords = "biomarker, EBAI, cancer, validation, scale, artificial intelligence",

author = "M Aldea and M Salto-Tellez and A Marra and R Umeton and A Stenzinger and M Koopman and A Prelaj and Kehl, \{K L\} and S Gilbert and M-E Le{\ss}mann and J Lipkova and L Provenzano and F Meric-Bernstam and S Halabi and J Wu and A Pellat and Suijkerbuijk, \{K P M\} and B Besse and B Ryll and C Marchi{\'o} and M Crispin-Ortuzar and R Fehrmann and J Vibert and D Ferber and C Pauli and A Valachis and Federica Corso and Brinker, \{T J\} and J Mateo and N Harbeck and Winkler, \{E C\} and F Lopez-Rios and R Perez-Lopez and G Pentheroudakis and S Delaloge and \{Benedikt Westphalen\}, C and Kather, \{J N\}",

year = "2026",

month = mar,

doi = "10.1016/j.annonc.2025.11.009",

language = "English",

volume = "37",

pages = "414--430",

journal = "Annals of Oncology",

issn = "0923-7534",

publisher = "Elsevier Ltd",

number = "3",

}

Aldea, M, Salto-Tellez, M, Marra, A, Umeton, R, Stenzinger, A, Koopman, M, Prelaj, A, Kehl, KL, Gilbert, S, Leßmann, M-E, Lipkova, J, Provenzano, L, Meric-Bernstam, F, Halabi, S, Wu, J, Pellat, A, Suijkerbuijk, KPM, Besse, B, Ryll, B, Marchió, C, Crispin-Ortuzar, M, Fehrmann, R, Vibert, J, Ferber, D, Pauli, C, Valachis, A, Corso, F, Brinker, TJ, Mateo, J, Harbeck, N, Winkler, EC, Lopez-Rios, F, Perez-Lopez, R, Pentheroudakis, G, Delaloge, S, Benedikt Westphalen, C & Kather, JN 2026, 'ESMO basic requirements for AI-based biomarkers in oncology (EBAI)', Annals of Oncology, vol. 37, no. 3, pp. 414-430. https://doi.org/10.1016/j.annonc.2025.11.009

TY - JOUR

T1 - ESMO basic requirements for AI-based biomarkers in oncology (EBAI)

AU - Aldea, M

AU - Salto-Tellez, M

AU - Marra, A

AU - Umeton, R

AU - Stenzinger, A

AU - Koopman, M

AU - Prelaj, A

AU - Kehl, K L

AU - Gilbert, S

AU - Leßmann, M-E

AU - Lipkova, J

AU - Provenzano, L

AU - Meric-Bernstam, F

AU - Halabi, S

AU - Wu, J

AU - Pellat, A

AU - Suijkerbuijk, K P M

AU - Besse, B

AU - Ryll, B

AU - Marchió, C

AU - Crispin-Ortuzar, M

AU - Fehrmann, R

AU - Vibert, J

AU - Ferber, D

AU - Pauli, C

AU - Valachis, A

AU - Corso, Federica

AU - Brinker, T J

AU - Mateo, J

AU - Harbeck, N

AU - Winkler, E C

AU - Lopez-Rios, F

AU - Perez-Lopez, R

AU - Pentheroudakis, G

AU - Delaloge, S

AU - Benedikt Westphalen, C

AU - Kather, J N

PY - 2026/3

Y1 - 2026/3

N2 - Background Artificial intelligence (AI) is expected to introduce an increasing number of biomarkers in oncology. To bridge the gap between oncology and computer science, it is timely to define recommendations for AI-based biomarkers suitable for routine clinical use. Here, we propose the ESMO (European Society for Medical Oncology) Basic Requirements for AI-based Biomarkers In Oncology (EBAI). Design The EBAI framework was developed using a modified Delphi methodology, involving a multidisciplinary panel of 37 experts who participated in four structured consensus rounds. Results AI-based biomarkers were classified as ‘class A’ (AI quantification of established biomarkers), ‘class B’ (indirect measure of known biomarkers using AI-based alternative methods, to be deployed as pre-screening tests), and ‘class C’ (novel AI-derived biomarkers, with C1 for prognosis and C2 for prediction of treatment effect). The EBAI framework addresses AI biomarkers for clinical use. Ground truth, performance, and generalisability were considered essential; fairness was recommended. Minimal validation requirements indicate that class A requires concordance studies, class B analytical validation, class C1 high-quality retrospective real-world or clinical trial data, and class C2 additionally requires clinical validation in prospective clinical trials for the prediction of response to a new treatment. All biomarker studies should report multiple evaluation and calibration metrics, with a clearly defined primary objective. Generalisability should be demonstrated across all intended use settings, including variability in data acquisition, post-processing, and population characteristics. Biomarkers must not be applied to other cancer types or modalities without supporting evidence. Conclusions EBAI defines criteria for AI-based biomarker adoption in routine use, providing a common language for physicians, AI developers, and researchers.

AB - Background Artificial intelligence (AI) is expected to introduce an increasing number of biomarkers in oncology. To bridge the gap between oncology and computer science, it is timely to define recommendations for AI-based biomarkers suitable for routine clinical use. Here, we propose the ESMO (European Society for Medical Oncology) Basic Requirements for AI-based Biomarkers In Oncology (EBAI). Design The EBAI framework was developed using a modified Delphi methodology, involving a multidisciplinary panel of 37 experts who participated in four structured consensus rounds. Results AI-based biomarkers were classified as ‘class A’ (AI quantification of established biomarkers), ‘class B’ (indirect measure of known biomarkers using AI-based alternative methods, to be deployed as pre-screening tests), and ‘class C’ (novel AI-derived biomarkers, with C1 for prognosis and C2 for prediction of treatment effect). The EBAI framework addresses AI biomarkers for clinical use. Ground truth, performance, and generalisability were considered essential; fairness was recommended. Minimal validation requirements indicate that class A requires concordance studies, class B analytical validation, class C1 high-quality retrospective real-world or clinical trial data, and class C2 additionally requires clinical validation in prospective clinical trials for the prediction of response to a new treatment. All biomarker studies should report multiple evaluation and calibration metrics, with a clearly defined primary objective. Generalisability should be demonstrated across all intended use settings, including variability in data acquisition, post-processing, and population characteristics. Biomarkers must not be applied to other cancer types or modalities without supporting evidence. Conclusions EBAI defines criteria for AI-based biomarker adoption in routine use, providing a common language for physicians, AI developers, and researchers.

KW - biomarker

KW - EBAI

KW - cancer

KW - validation

KW - scale

KW - artificial intelligence

U2 - 10.1016/j.annonc.2025.11.009

DO - 10.1016/j.annonc.2025.11.009

M3 - Article

C2 - 41260261

SN - 0923-7534

VL - 37

SP - 414

EP - 430

JO - Annals of Oncology

JF - Annals of Oncology

IS - 3

ER -

ESMO basic requirements for AI-based biomarkers in oncology (EBAI)

Abstract

UN SDGs

Keywords

Access to Document

Fingerprint

Cite this