In October 2023, the NHS announced a £21 million investment in AI tools for lung cancer diagnosis. It is the latest expansion of AI in the health services. But amid the fanfare around artificial intelligence, we need to stop and consider: is it worth the money?

We analysed this latest funding to consider the benefits it could provide and whether they are likely to outweigh the cost of investment.

The cost of cancer

The new technology announced by the NHS is being rolled out across 64 trusts in England this winter. Using AI to analyse X-ray and CT scans might help clinicians diagnose lung cancer faster and more accurately.

Limiting the impact of lung cancer has obvious advantages for people and their families. But there are also wider economic benefits.

We know from a previous Frontier analysis that reducing our exposure to preventable cancer risk factors – like tobacco and UV radiation – could create billions of pounds in gains for the NHS.

The new tools will not be able to prevent cancer cases, but how much of the economic costs could they cut through better diagnosis?

How might AI help?

If we cannot prevent a case of cancer, there are two routes to lessening its impact. The first route is faster diagnosis: reducing the time it takes to arrive at a diagnosis once someone presents with symptoms.

The second route is earlier diagnosis: correctly identifying cancers when they are in their earlier stages and are more likely to be treatable.

The AI technology announced by the NHS could help with both routes in different ways.

The tools have the potential to support a faster diagnosis by making clinical review more efficient. This would reduce the diagnosis waiting time for patients – there is currently a two-week target, which is only met in around 75% of cases. It is hoped the technology could raise this figure back to the +90% levels seen before the COVID-19 pandemic.

The technology could also support earlier diagnosis of lung cancer. This is because it could improve accuracy and reduce the number of false negative results, ensuring cancer cases are identified at an earlier stage.

The cost of false negatives

To analyse the NHS’s investment, let’s use ‘earlier diagnosis’ as an illustrative example.

The main benefit of the tool is a reduction in false negatives. Correctly identifying more lung cancer cases at earlier stages would reduce the number of deaths (and the associated economic costs) and lower treatment costs.

Our first step is to determine the current cost of false negative diagnoses.

There are 54,000 lung cancer patients in the UK, costing the NHS around £1 billion a year. A definitive rate of false negative diagnoses is hard to find. But, the results of one major study showed that 17.5% of lung cancer cases that progressed beyond stage 1 had initially been missed or misdiagnosed.

That would mean false negative lung cancer diagnoses create an excess cost to the NHS of £16.7 million per year and result in 858 avoidable deaths.

Calculating the economic gains

How much impact could the new AI tools have on these figures if they were to be rolled out across all NHS trusts?

If the technology could reduce the rate of false negatives by just 10% in its first year, that would save 191 patients from progressing to later-stage cancer.

This would potentially save 85 lives, with a benefit to the UK economy of £40.9 million. The NHS would also save £1.67 million in treatment costs.

Assuming the AI benefits last a further five years, the economic gains would total around £235 million.

What’s more, this does not include the benefits of improved quality of life for patients, the positive impact on their families and the combined impact of faster and earlier diagnoses.

And if the new technology frees up more time for radiologists to perform other tasks, this would create further gains for the NHS.

Costs vs benefits

Our calculations show that the new technology needs to have a relatively small impact to justify the £21 million investment. The AI could potentially pay its way within the first year.

But, we do not yet know how effective AI will be in reducing false negative diagnoses. If our assumption of 10% is too high and the technology only manages to reduce the rate by 3%, for example, the investment becomes more challenging to justify.

For more information contact Alex Charlwood on alex@arclex.com