AI for robust decision support for multidisciplinary colorectal cancer team meetings

BEAGLE-MDT

Patient / Population Intervention / Exposition Comparaison Critère de jugement

In 15 synthetic colorectal cancer cases reviewed by five expert clinicians, the multi-agent AI platform BEAGLE®-MDT received high clinician ratings for case summarisation, management recommendations and trial matching.

N

15patients

Schéma

Blinded pilot evaluation of BEAGLE®-MDT, a multi-agent artificial intelligence platform for pre-MDT completeness checking, evidence synthesis, trial matching and structured summaries, on 15 synthetic colorectal cancer cases across disease sites and stages, with outputs rated independently by five expert clinicians and no comparator arm (local institutional registration with UHB).

Critère

Clinician ratings on a 5-point scale of patient summarisation, management recommendations and trial matching, including the proportion of ratings achieving the maximum score.

RésultatAcross clinicians, mean ratings for patient summarisation ranged from 4.50 to 5.00, with 73-100% of ratings achieving 5/5. Mean management ratings ranged from 4.07 to 4.87, with 40-87% achieving 5/5. Mean trial-matching ratings ranged from 4.33 to 4.93, with 67-93% achieving 5/5.

Ashraf S, Khan A, Beggs AD, et al. Presented at ESMO Congress 2026, abstract LBA18, 24 October. Not yet published.

Discussion et critique

This pilot shows that clinicians rated the platform's summaries, management recommendations and trial matches highly, which supports technical feasibility. It used only 15 synthetic cases and five raters, had no comparator such as standard MDT practice, and measured subjective ratings rather than patient outcomes, deferral rates or trial enrolment. Management ratings varied the most between clinicians, and the abstract does not report inter-rater agreement or any errors or unsafe recommendations. The study was funded by Histofy, which employs or is linked to several authors, and the authors themselves state that the findings do not establish clinical effectiveness and that prospective validation is needed.