An AI tool that identifies patients at risk of lung cancer from their clinical history could help GPs refer them for tests up to four months earlier, researchers say.
The algorithm, which can be used during a consultation, picks up clues in the free text of patient records.
Dutch researchers who developed the tool said using this method to pick up cases of lung cancer earlier had been shown to be more accurate than other risk prediction tools which relied on pre-defined factors such as whether the patient was a smoker or not.
To develop the AI tool, they analysed 525,526 patients, of whom 2,386 were diagnosed with lung cancer.
They used AI to look at both structured and free-text data and predict the diagnosis of lung cancer five months before diagnosis and four months before referral.
Reporting in the British Journal of General Practice, the team found it adequately referred 62% (1,480 of 2,386) of all patients with lung cancer around four months earlier than had been done by the GP.
But there is a cost to use of such an algorithm the researchers noted.
For every patient diagnosed four months earlier, 33 patients would need to undergo tests because of a presumed increased risk and about 40% of all patients with lung cancer would have been missed by the tool and referred four months later following usual care.
This ‘proof of concept’ analysis would need to be tested further in different countries and healthcare systems, the researchers said, but could also prove useful for other cancer types.
In England, the five-year survival for lung cancer is 16% and has improved only slightly during the past 40 years.
A paper earlier this month, also in the BJGP, found that GP practices who order more chest X-rays have fewer patients with cancers diagnosed at a late stage.
The approach should not be compared to screening – as is simply an algorithm that could be integrated into existing GP IT systems to support earlier diagnosis.
‘The algorithm can be fine-tuned to fit several settings, each with its own implications for clinicians and policymakers,’ they concluded.
‘For primary care practice, the current study has produced a model able to indicate a relevant risk of lung cancer several months earlier, with a potentially significant impact on prognosis.’
Meanwhile, an AI-powered system for detecting potential skin cancer has been approved by NICE today for use on the NHS for three years while further evidence on its effectiveness is collected.
