GPs use AI to boost cancer detection rates in England by 8%

Artificial intelligence that scans GP records to find hidden patterns has helped doctors detect significantly more cancer cases.

The rate of cancer detection rose from 58.7% to 66.0% at GP practices using the “C the Signs” AI tool. This analyses a patient’s medical record to pull together their past medical history, test results, prescriptions and treatments, as well as other personal characteristics that might indicate cancer risk, such as their postcode, age and family history.

It also prompts GPs to ask patients about any new symptoms, and if the tool detects patterns in the data that indicate a higher risk of a particular type of cancer, then it recommends which tests or clinical pathway the patient should be referred to.

C the Signs is used in about 1,400 practices in England – about 15% – and was tested in 35 practices in the east of England in May 2021, covering a population of 420,000 patients.

The results, published in the Journal of Clinical Oncology, show that the cancer detection rate rose from 58.7% to 66.0% by 31 March 2022, while those practices not using the system remained at a similar rate.

Bea Bakshi, a GP who created the system with her colleague, Miles Payling, said: “It could be a scan, an ultrasound, or they could need to be seen by a specialist at a clinic.”

Patients are tracked through the C the Signs system to remind doctors to check test results and referrals elsewhere. “Our system has detected over 50 different types of cancers,” Bakshi said. “The key thing is that it’s not only an earlier diagnosis, but a faster diagnosis.”

Bakshi and her colleagues also sought to validate the tool by assessing 118,677 patients in a previous study, which found that 7,295 were diagnosed with cancer and 7,056 were successfully identified by the algorithm.

Where the tool came to the conclusion that it was unlikely a patient did have cancer, only 239 out of 8,453 went on to have a confirmed cancer diagnosis within six months (about 2.8%). Bakshi developed the tool after meeting a patient in hospital who had been given a late diagnosis of pancreatic cancer and died three weeks later.

“It stayed with me as a problem area,” she said. “Why are patients with cancer being diagnosed so late?”

The UK has three cancer screening programmes, for bowel, breast and cervical cancers, but there are 200 different types of cancer which may be asymptomatic or create symptoms easily confused with other conditions.

“Two-thirds of deaths are in the non-screenable cancers and the ones that we aren’t screening for,” Bakshi said. “Patients visit GPs between three and five times before they are recognised as being at risk of cancer. GPs detect an average of eight cases of cancer a year.”

GPs use guidelines from the National Institute for Health and Care Excellence to make decisions about when to make a cancer referral.

“They’re quite comprehensive guidelines, but no GP can remember them all,” said Peter Holloway, a GP who chairs the Primary Care Group for the East of England Cancer Alliances and a co-author on the study.

“We know a lot of cancers present with vague symptoms and some are difficult to define and don’t necessarily correlate with our guidelines.”

Holloway saw a patient in his early 60s who had diarrhoea and some lower abdominal pain. “Very common symptoms, and not things that would lead you to do a referral for suspected cancer,” he said. But the C the Signs tool recommended a faecal test.

“The test was positive, he was referred and he turned out to have colorectal cancer, which was diagnosed early and treated successfully,” Holloway said. “He’s fine, but if we’d followed the rigid guidelines, he may not have got referred for several months.”

Related: TechScape: Can AI really help fix a healthcare system in crisis?

The NHS England Long Term Plan for Cancer aims to have 75% of all cancers diagnosed by stage one or stage two by 2028. The NHS is also conducting research into whether the Galleri blood test, which attempts to detect DNA from more than 50 different types of tumours, is effective. The trial began in September 2021, and 140,000 people have been tested.

Holloway said that decision support systems such as C the Signs were an important part of cancer detection, alongside improving patient awareness of different types of possible symptoms of cancer, and getting better access to diagnostic technology such as CT and MRI scanners.

Prof Peter Johnson, National Clinical Director for Cancer at NHS England, said: “Despite increased demand and pressure on services, record numbers of people are being checked and treated for cancer, and we are now diagnosing a higher proportion of cancers at an early stage, increasing people’s chances of survival.

“We know we have a lot more to do to help people with cancer get the care they need, and using the most up-to-date technology is an important part of our work to reduce waits and find cancers earlier, such as ‘teledermatology’ to diagnose skin cancers, or community lung trucks, and home testing for bowel cancer.”