What question did this study set out to answer?

This study aims to evaluate how well an AI system aligns with expert multidisciplinary team (MDT) decisions in treating breast cancer.

May 24, 2026Open Access

682P Implementation of a geriatric oncology care pathway for older patients with breast cancer in a regional public hospital in Portugal

Key Points

This study aims to evaluate how well an AI system aligns with expert multidisciplinary team (MDT) decisions in treating breast cancer.
Retrospective analysis of 200 patients with early breast cancer discussed in MDTs across two hospitals.
Data collected included medical records and was processed through an AI system for treatment recommendations.
Central expert panel reviewed both AI-generated and MDT recommendations for concordance.
The AI system achieved 95% concordance with expert consensus in treatment decisions.
Inter-rater reliability of 0.93 indicates strong agreement between AI and expert recommendations.
Mean generation time for treatment recommendations was 27 seconds per MDT record.

Abstract

Background: Explicit concordance between Artificial Intelligence (AI) and Breast Cancer MDT decisions are underreported in current literature.Aim: To evaluate concordance rates, performance metrics and efficiency between an oncologistdesigned AI system and routine breast MDTs, with prospective central expert validation.Methods: Retrospective study of 200 consecutive patients with early breast cancer discussed at routine MDTs at two international sites (Mater Hospital, Sydney, Australia and Saifee Hospital, Mumbai, India) in 2025.Inclusion criteria required histologically confirmed breast cancer, complete medical records (pathology, radiology, ECOG performance status, molecular profile), and documented MDT recommendations.Anonymized clinical data (age, TNM staging, biomarkers, comorbidities) were extracted from electronic medical records and input into Gemini models.The AI system generated treatment recommendations blinded to historical MDT decisions.Both were subsequently reviewed by a central expert panel. Results:The AI system achieved 95% perfect concordance with expert consensus; in 5% of cases, AI recommended an acceptable but less-preferred alternative.Subsequent adjustment of the AI tool integrating this clinical feedback allowed for dynamic refinement.Mean generation time was 27 seconds per MDT record.In terms of performance metrics: inter-rater reliability demonstrated Fleiss' 0.93, with an integrated decision-making score of 96/100.Model consistency was 98%; Monte Carlo simulation yielded 97% confidence in predicted treatment pathways. Conclusions:Our study demonstrates exceptional concordance between AI-generated and expert MDT treatment recommendations in breast cancer, with rapid processing times.These findings provide multiple potential opportunities for clinical workflow integration with AI, in resource constrained settings, or those with limited access to subspecialised oncology opinion.A prospective, international multi-institutional study is underway, implementing AI-augmented tools in real-time breast MDTs to further validate clinical utility and impact on decision-making efficiency.

Bookmark

View Full Paper