Artificial Intelligence can detect signs of lung cancer up to four months more quickly than family doctors, a study by researchers at Amsterdam’s UMC teaching hospital has shown.

The researchers developed an AI-based model which analysed doctors’ notes of 525,526 patients at practices in Groningen, Amsterdam and Utrecht between 1989 and 2021.

Of the patients whose medical notes were analysed, 2,386 developed lung cancer but in 75% of the cases the disease was already in a late stage at the time of the referral, making their chances of survival less likely.

The researchers wondered if the long-term data in the family doctors’ records might contain hidden information that could be used for early detection, which can significantly improve patients’ prognoses.

This proved to be the case, said professor of medical informatics Ameen Abu-Hanna said. Based on the doctors’ notes, the algorithm picked up signals up to four months earlier, not only by detecting terms such as “smoker” or “coughing up blood” but also in more general information about the person’s health.

More research will have to show which data is picked upon by the algorithm, Abu-Hanna said.

The model also has disadvantages, the researchers said. Some 33 people needed  to be screened to find one early case and the testing would put more pressure on an already stretched health system, they said.

Last week, the public health council RVS warned that the increasing availability of health tests is turning people into patients unnecessarily.

The government should shift its focus from testing to prevention, the council said in its recommendations.

However, even a diagnosis which discovers the cancer four weeks before the referral is made would improve a patient’s chances, they said. Some 80% of people diagnosed with stage 3 or 4 lung cancer die within a year.

There is currently no mass screening for lung cancer. In 2023, the Antoni van Leeuwenhoek cancer hospital started a study into the effectiveness of screening, involving a trial of 400,000 heavy smokers. The results of the study are expected in 2026.