Persisting health data siloes

Although a public institution, the UK’s National Health Service – the fifth-largest employer globally – struggles with problems strikingly similar to those faced by multinational companies. Chief among them is data fragmentation.

 

While the NHS generates and stores staggering amounts of data – it has a reputation for storing some of the largest and most comprehensive health datasets in the world – its ability to exploit this asset effectively is constrained by entrenched data siloes.

 

Recognising this, successive UK governments have attempted to push primary and secondary care towards greater data integration, with the ambition of creating a single, shared source of health data accessible across the system to eliminate redundant data entries and inconsistent data records.

 

The latest milestone on this journey was the 2022 Health and Care Act. Prompted in part by the tragic consequences of poor co-ordination between health and social care institutions during Covid-19, the act marked a decisive shift in NHS governance: market competition was replaced as the organising principle with collaboration and integration.

 

The act also put integrated care at the heart of the NHS. Since 2022, England has been divided into 42 Integrated Care Systems (ICS), each overseen by a board responsible for organising healthcare services, including primary and secondary care, across a defined population.

 

Steps towards data integration

 

Institutional integration has also been reinforced by a mandate requiring health and care providers to comply with common information standards.

 

Having considered the scale of the job of unifying and integrating vast amounts of historic and real-time patient data to be analysed to reach better efficiencies, the UK government contracted the controversial US data analytics company Palantir in November 2023, to build and operate the NHS Federated Data Platform (FDP).

 

However, two years into the contract, it’s clear that adoption is still relatively slow.

 

In the middle of 2025, 75 NHS trusts used the platform actively – only one quarter of England’s 215 trusts.

Reasons for sluggish adoption are varied. First, creating the platform involves handling unprecedented volumes of sensitive patient data.

 

While the contract includes safeguards – Palantir is only a processor not a data controller, and data is strongly encrypted and must be retained in the UK – data privacy concerns remain a strong barrier to adoption, further amplified by the company’s involvement in the military and surveillance industries.

 

Second, healthcare professionals already using the FDP often complain about poor user experience, data integration challenges and  a centralised architecture that, rather than connecting existing systems, consolidates data.

 

Some of the feedback also suggests that legacy data systems used previously outperform the FDP in terms of usability, context and reliability.

 

A promising pilot to harness data integration

 

Running in parallel with national-level data integration initiatives are smaller, more targeted ones aiming to improve healthcare through better data flow. Cancer care, and particularly its treatment planning stage, is one of the most data-intensive and collaboration-dependent areas of medicine.

 

Comprehensive cancer treatment planning in the UK typically draws on four to five primary categories of data: electronic health records, tissue biopsy results, medical imaging such as CT, MRI, ultrasound or PET scans, and laboratory results.

 

Ideally, to reach a better trade-off between the therapeutic benefits of radio or chemotherapy and its toxicity in a more personalised treatment, patients’ genetic and molecular profiles must also be taken into account.

Cancer Multidisciplinary Teams (MDTs) that include surgeons, oncologists, radiologists and pathologists are tasked with synthesising this information to determine optimal treatment pathways.

 

In this high-stakes environment, not incorporating relevant data into decision-making, or bad communication between team members, can cause treatment delays or even fatalities.

 

Yet despite this enhanced level of risk, research shows that MDTs typically spend just three to four minutes discussing a patient.

 

This is precisely the kind of environment in which AI-driven decision support systems can add value.

They can, beyond any doubt, exceed human capabilities when it comes to summarising and analysing large datasets and identifying hidden patterns and sweet spots.

 

These are the super AI capabilities that TrustedMDT, an agentic copilot for tumour boards developed by Oxford University in collaboration with Oxford University Hospitals NHS Foundation Trust and Microsoft Research, aims to leverage.

 

Trusted MDT has three core agents: one for clinical summarisation that processes electronic health records into concise tumour summaries; a treatment planning agent drafting evidence-based recommendation based on guidelines; and a cancer staging agent that establishes the extent of the disease.

 

TrustedMDT only gained approval last December. The initial stage of the project focuses on validation using anonymous cancer cases and is followed by deployment in simulated MDT settings to evaluate how effectively the system supports clinical decision-making.

 

Even if successful, further evaluation will be required before the tool can be considered safe for live clinical use.

Although radically different in scale and focus, both FDT and TrustedMDT indicate that policymakers and health professionals are equally aware of the central role unimpeded data flow and the accessible, standardised data play in improving healthcare outcomes.

 

Enhancing the exchange and management of standardised, unified medical data is arguably the most compelling AI use case in healthcare – and probably the UK’s only chance to transform the NHS into a free healthcare system fit for the 21st century.