Executive summary
For UK oncology multidisciplinary teams (MDTs), the daily challenge is managing a high volume of complex cases against a backdrop of ever-expanding genomic data, clinical trials, and evolving guidelines. NHS England has already published guidance encouraging more streamlined and effective MDTs. Artificial intelligence is emerging as a key enabler for this, but only if it is evidence-grounded, auditable, and safely integrated into the clinical workflow.
In 2024–2025, the market has matured significantly. Major evidence platforms like EBSCO (Dyna AI), Elsevier (ClinicalKey AI), and Wolters Kluwer (UpToDate Expert AI) have all shipped generative-AI assistants, promising faster retrieval with clear citations. Alongside these, oncology-specific knowledge bases like OncoKB and real-world data platforms like CancerLinQ offer deep, specialist support. For UK adoption, any tool must be assessed against the national guardrails: the DTAC, the NICE Early Value Assessment (EVA) pathway, and, for novel devices, the MHRA AI Airlock.
The MDT context: where AI evidence tools actually fit
The pressure on cancer MDTs is well-documented, stemming from case complexity, workload, and the need to incorporate genomic data and trial options. AI is not a single solution but a set of tools that can help at three high-value moments:
- Pre-MDT: Automatically compiling case packs with summaries, relevant guideline excerpts, and flags for genomic variants.
- In-meeting: Providing "cite-as-you-go" answers to complex questions, reducing the need for manual lookups and delays.
- Post-MDT: Assisting with the rapid documentation of outcomes and agreed-upon care plans.
Taxonomy of oncology AI evidence tools
A) General evidence assistants with Gen-AI front-ends
- Dyna AI (DynaMedex): Provides RAG-based (Retrieval-Augmented Generation) answers grounded in the curated DynaMedex and Micromedex content.
- ClinicalKey AI (Elsevier): A generative AI layer over Elsevier’s vast library of textbooks, journals, and drug information.
- UpToDate Expert AI (Wolters Kluwer): The new generative AI layer built exclusively on UpToDate's trusted editorial content, with an enterprise rollout planned from Q4 2025.
B) Oncology-specific knowledge bases / guideline navigators
- OncoKB (MSK): An FDA-recognised public database that provides detailed information and evidence levels for somatic alterations in cancer, designed for tumour boards and molecular MDTs.
- NCCN Guidelines Navigator: An interactive digital platform for delivering and navigating the NCCN guidelines (a comprehensive US-based resource).
C) Real-world evidence (RWE) and learning health networks
- CancerLinQ (ASCO/ConcertAI): A platform from the American Society of Clinical Oncology (ASCO) that provides insights and decision support by analysing real-world data from millions of oncology records.
D) Trial-matching AI (adjacent but MDT-relevant)
- AI-powered tools, such as Ancora.ai, are increasingly being used to automate the process of matching a patient's specific clinical and genomic profile to eligible clinical trials, a key function of the modern MDT.
Lessons from history: what not to repeat
The industry is rightly cautious, remembering the story of IBM Watson for Oncology. Its initial ambition outpaced its real-world validation, with studies highlighting that its recommendations were not always concordant with expert clinicians. This provides a crucial cautionary lesson: clinical utility, robust validation, and user trust are far more important than marketing claims of "intelligence" (STAT, SpringerLink).
What “MDT-friendly” should mean (a specification)
- Provenance-first outputs: All answers must have inline, clickable citations to the primary source (e.g., NICE, NCCN, a peer-reviewed paper) and be date-stamped.
- Genomics-aware reasoning: The tool should be able to understand and link variant data to evidence databases like OncoKB.
- Clinical trial awareness: It should be able to suggest relevant clinical trials, ideally from local portfolios and national registers.
- OS interoperability: The tool must be able to launch from within the main EPR (e.g., Epic, Cerner) using standards like SMART-on-FHIR, and ideally use CDS Hooks to provide proactive prompts.
- UK content pipes: It should natively ingest UK guidance, such as via the NICE syndication API.
- Safety UX: It must have clear uncertainty flags, an "abstain-when-unsure" behaviour, and explicit "seek specialist review" prompts.
UK governance & assurance (non-negotiables before rollout)
- DTAC: The Digital Technology Assessment Criteria is the baseline procurement standard for any tool used in the NHS.
- NICE Early Value Assessment (EVA): This is the key pathway for promising but early-stage AI, such as in cancer triage or radiotherapy planning, allowing for conditional use while more evidence is generated.
- MHRA AI Airlock: This regulatory sandbox is the correct route for novel AIaMD, allowing developers and regulators to test tools and refine rules in a supervised environment before scale.
- NHS AI Knowledge Repository: A central hub for publishing pilot protocols, case studies, and lessons learned to accelerate safe adoption across the system.
Comparative snapshot: what leading tools claim today
| Tool | Primary Corpus | Gen-AI? | Oncology-Specific Capability |
|---|---|---|---|
| Dyna AI | DynaMedex/Micromedex | Yes (RAG) | Strong for drug–cancer intersections |
| ClinicalKey AI | Elsevier (books/journals) | Yes (RAG) | Strong for deep literature/research |
| UpToDate Expert AI | UpToDate | Yes (RAG) | Deep, graded editorial pathways |
| OncoKB | Variant Data | No (Database) | Yes (FDA-recognised variant KB) |
| NCCN Navigator | NCCN Guidelines | No (Navigator) | Yes (US guideline engine) |
| CancerLinQ | Real-World Data | Yes (Analytics) | Yes (RWD decision support) |
Evidence standards & evaluation metrics
- Clinical appropriateness: Concordance with the final, expert-led MDT decision; impact on time-to-decision.
- Pathway impact: Increase in trial referral rates; reduction in time-to-treatment; rate of plan revisions based on new evidence.
- Safety: Rate of disagreements flagged and resolved; AI abstention rate; accuracy of citations.
- Operational: Clicks-to-answer; utilisation rate of in-workflow prompts (via CDS Hooks).
UK momentum in cancer AI
The UK is well-positioned to adopt these tools. The large-scale national breast screening AI trial and the new AIR-SP screening platform are building a robust evidence base. Furthermore, the successful rollout of primary care cancer risk stratification tools like C the Signs illustrates a proven pattern of governance and measurement that the MDT layer can now emulate.
Calls to action
- Cancer Alliances / ICSs: Run a 90-day MDT evidence-assistant pilot with one of these tools. Pre-register your metrics and publish your outcomes to the NHS AI Knowledge Repository.
- Vendors: Ship your products as "UK-ready"—with a completed DTAC pack, a NICE syndication plan, and readiness to integrate via CDS Hooks.
- Policy leads: Continue to align the EVA and AI Airlock pipelines for oncology AI to reduce duplicated assurance steps and speed up safe adoption.