Oxford-built multi-agent assistant for cancer care to be piloted in collaboration with Microsoft

Through a strategic collaboration with Microsoft, the AI assistant has been integrated into Microsoft Teams and will be piloted at Oxford University Hospitals (OUH) NHS Foundation Trust, marking one of the earliest deployments of 'agentic AI' within a clinically realistic tumour board setting.

Multidisciplinary Tumour Board meetings (MDTs) are the gold standard for cancer treatment planning in the UK, convening radiologists, pathologists, surgeons, and oncologists to review diagnostic results and formulate treatment recommendations. However, rising caseloads are increasingly straining expert capacity.

A Cancer Research UK review identified that teams often have less than two minutes of discussion time per patient, and critical information gaps lead to postponements in 7% of cases. These constraints lead to treatment delays, missed research opportunities, and clinician burnout.

To address these challenges, Dr. Andrew Soltan, NIHR Academic Clinical Lecturer in the Department of Oncology, and Junior Research Fellow in Engineering at Jesus College, University of Oxford, has led the design and development of an AI assistant to support these crucial meetings. The tool, TrustedMDT, comprises three AI agents working in concert:

1. The Clinical Summarisation Agent - Analyses electronic health records, including radiology, pathology, and biomarker tests to produce concise, tumour-specific summaries

2. The Cancer Staging Agent - Applies international standards (AJCC/UICC) to determine disease stage

3. The Treatment Planning Agent - Drafts evidence-based recommendations aligned with professional guidelines.

'Because standard chatbots struggle with the high-stakes complexity of oncology, we developed a hierarchical multi-agent system,' says Dr. Soltan, Lead Investigator, and Specialty Registrar in Medical Oncology at OUH. 'In this architecture, each agent contains a dedicated team of sub-agents grounded in specific data with access to tools. This granular approach reduces the risk of ‘guessing’ because the system is required to reason through guidelines and explicitly cross-check its work against the patient's history.'

Through a collaboration with Microsoft, the Oxford team used the healthcare agent orchestrator to deploy their custom agents directly into Microsoft Teams.

'The technology must serve the workflow, not disrupt it,' adds Dr. Soltan. 'Using the Microsoft orchestrator, we embedded our agents directly into the Teams environment that is used in our existing MDTs. This positions the AI as a true ‘digital collaborator’, enabling colleagues to provide new information in real-time and probe the rationale of recommendations, all while the human remains the final decision-maker.'

A two-phase evaluation and pilot study has received approval to assess accuracy, usability, and technical performance at Oxford University Hospitals.

•    Phase I validates the tool using anonymous cancer cases, benchmarking AI outputs against expert decisions and physician preferences.

•    Phase II deploys the system in simulated MDTs with OUH clinicians to assess the user experience and how effectively the AI summarises information, supports discussion, and drafts treatment plans in a realistic clinical workflow.

The pilot study is supported by OUH Resident Doctors Dr Sajan Patel and Dr Jaya Sharma, who contribute clinical expertise and quality assurance, alongside DPhil candidate Edward Phillips, who contributed towards the development of the Treatment Planning Agent.

Dr. Ben Attwood, Chief Digital Officer for Oxford University Hospitals NHS Foundation Trust, said: 'At OUH, we are committed to exploring innovations that help clinicians prepare for and run cancer MDTs more effectively. Any adoption follows our established governance processes, including robust information governance and research approvals. We look forward to evaluating the TrustedMDT approach in collaboration with our clinical teams.'

David Ardman, Corporate Vice President, Cloud AI Platform, Microsoft Health and Life Sciences said: 'This multi-agent system created by Oxford represents a new frontier in healthcare AI by using multiple specialised agents to tackle different aspects of cancer care planning. By orchestrating these agents within Microsoft Teams, we enable clinicians to interact with AI in a dynamic, conversational way that fits seamlessly into their workflow. This approach reduces cognitive load while improving clinical decision support and patient outcomes.'

If validated, the tool could enhance communication between specialists, reduce treatment delays and expand clinical trial access. This study is a critical first step to demonstrate the potential impact of TrustedMDT and to generate evidence that will support technical refinement and guide future larger-scale trials. Further evaluation studies will be needed before the tool can be deemed safe enough to be used in clinical practice.

More information about the project can be found on the Department of Oncology's website.  

The CRUK report can be read here.