There is a lot of noise around AI in healthcare. Opinion oscillates wildly about risk-benefits, but tangible opportunities exist to improve clinical outcomes and build a sustainable NHS for the future, all while stimulating growth in a domestic industry that struggles to compete with better-funded US counterparts.
Keir Starmer’s Government has announced steps to facilitate access to data for the world’s biggest AI companies in a secure and controlled way. As the CEO of a homegrown medtech company, I applaud the ambition, but before the doors are flung open, we need to remember that NHS data is a national asset that belongs to us all. It must be leveraged properly for the long term, not used as a one-off injection of cash for the Treasury.
Why not prioritise access to data for UK companies first? We are invested in the UK economy and the health of the NHS and have been doing the groundwork with regional partners for years. It would give us a competitive advantage that would open opportunities for commercialisation, growth, employment, and expansion overseas – all contributing to an increased tax base in UK tech.
NHS data is not North Sea oil – Palantir, Google, and the like do not need to develop local talent or supply chains to exploit healthcare data here. Paid access would represent much-needed income for the UK, but domestic growth and a seat at the table of AI superpowers do not necessarily follow.
To future-proof the NHS and avoid becoming the world’s biggest healthcare database for foreign firms, without the capability to grow ourselves, access to data is just the first half of the puzzle. The second is establishing routes to implementation and doing what we often struggle with in the UK – turning research into commercial success.
The right type of AI for the right problem
Critics argue that the Government should be focused on getting back to the basics of cancer services and that AI fails to address cancer as a systems problem. I believe that the right type of AI applied to the right problem can benefit everyone: the patient AND the system.
To achieve impact in a complex system like the NHS, we need to employ tools that can learn from data on a population scale. Machine learning does exactly that.
There is understandable nervousness around AI making healthcare decisions, but machine learning systems can, in fact, complement clinical intuition.
One example is “clinical decision support systems”, designed not to dictate a diagnosis or intervention, but to provide an extra level of information for decisions around a patient’s care.
Building trust
PinPoint’s Chief Scientist, Dr Richard Savage, was part of the SPIRIT-AI and CONSORT-AI initiative, establishing a benchmark for the quality of evidence in AI health interventions. Our collaboration with the University of Leeds, Leeds Teaching Hospital, and Mid Yorks Teaching Trusts has demonstrated how this can be translated to the real world – generating strong evidence whilst safeguarding personal data.
During development, we employed machine learning to analyse retrospective data from almost 400,000 patients – all data handled by us is anonymised at source. No personal data is transferred, nor could it be traced back to an individual. Our algorithms are also not ‘self-learning’, there is human oversight at every stage, using AI as a tool to refine and improve our software. Like any other medical product, there is a rigorous process of generating evidence, testing, and regulation.
How to support implementation in cancer services
When we think about deploying AI nationally, our minds naturally go to patient safety. What if cancers are missed? What if services are overwhelmed?
No system is perfect, and that includes NHS cancer services. Missed cancers are an unfortunate but accepted part of diagnostics with the technology and resources available today. Faecal immunochemical testing (FIT) is the recommended first test for bowel cancer when a patient presents with symptoms, but evidence shows that even this industry standard can miss between 5–17 per cent of colorectal cancers.
Machine learning has the potential to reduce these margins of error, but we need a tailored approach to deployment, leveraging scale without exposing patients to additional risk.
At present, new technologies are evaluated using small-scale pilot schemes. This is a good way to manage risk and measure the performance of a typical product, but the approach is problematic for data-based tools. By limiting scale, you simply cannot attain the level of statistical certainty needed to demonstrate impact.
The very thing that causes so many of the pressures on our healthcare system – the number of patients going through it – is also our opportunity. The scale is there, but if you don’t use it, you’re only exploring the tip of the iceberg – failing to achieve the step improvements that are possible as datasets increase in size and quality.
Recent government proposals are a step in the right direction, but a framework needs to be in place that standardises routes to implementation nationally.
Data governance should be standardised and IT systems connected so that checks and data sharing agreements in one region translate to another.
Financial incentives should be placed at the right link in the chain to encourage innovation, not reinforce the status quo. Payments for specific procedures, for example, may be well intended, but they disincentivise the adoption of alternatives that might reduce those costly, potentially invasive procedures.
Similarly, where budgets for supporting innovation exist, incentives should be in place to reward successful implementation, not simply count the total number of innovations entering the pipeline.
Standardised national guidelines and coherent incentives to support government priorities would remove the potential for conflicts of interest and simplify how SMEs interface with the NHS.
How PinPoint’s technology works
What PinPoint offers is ‘intelligent triage’. What does that mean? Our results show that if all patients referred for suspected upper gastrointestinal cancers – e.g. oesophageal or stomach – were given a PinPoint test and the 10 per cent at highest risk were prioritised for investigation, 57.4 per cent of all cancers would be included in that group. This is known as enrichment; in this case, enrichment of 5.74x. Put another way, whereas now, a consultant needs to see 17.9 people to identify 1 Upper GI cancer, in the PinPoint priority group, they only need to see 3.1 people.
For gynaecological cancers, that number goes from 22.2 to 5.6, and the same test is able to rule out the 20 per cent at lowest risk with 98.2 per cent accuracy. This is a huge advantage in the allocation of diagnostic resources and will keep improving with greater scale.
Cancer services: Back to the basics
The OECD estimates that, without action, the UK’s expenditure on cancer and cancer services will increase by 52 per cent per capita by 2050. We have a growing and aging population; demand on cancer services is increasing, but our capacity to deal with it is not. The need for innovation is clear.
Not all AI is the same, the term covers a range of different technologies suited to different tasks. I encourage people to look past the sensationalism and interrogate the technology on offer. At PinPoint, we believe that machine learning applied in the right way will save money, reduce backlogs and give NHS staff the one resource we can’t buy more of – time. AI can offer a route back to basics, delivered without additional burden or increasing budgets.
With the wealth of talent and innovation in the UK, we can be at the forefront not just of research but of adopting and scaling the right cancer services technologies to solve the 21st-century challenges of global healthcare.
Featured image via Shutterstock.
Giles Tully
DesignationCEO of PinPoint Data Science
