A colonoscopy is a search. The doctor steers a camera through the bowel looking for adenomas, small growths that can turn into cancer, and removes them before they do. Whatever the eye misses stays in the patient.
Numerous trials had shown that AI could help: with AI tools for polyp detection, doctors found more adenomas. At the end of 2021, four endoscopy centres in Poland taking part in a trial called ACCEPT brought in AI tools for polyp detection. From then on, whether a given colonoscopy used the AI depended on the date of the examination.
That setup left a natural experiment behind. Researchers, including doctors at the Academy of Silesia, looked only at the colonoscopies done without AI, three months before the tools arrived and three months after, between September 2021 and March 2022. There were 1,443 of them, performed by 19 experienced endoscopists who had each done more than 2,000 colonoscopies.
Before the AI arrived, those doctors found at least one adenoma in 226 of 795 patients, 28.4%. In the months after, working without it, they found one in 145 of 648, 22.4%. That is six percentage points, a fifth of their previous rate. After adjusting for patients’ age and sex, exposure to AI was still linked to lower detection. With the AI switched on, the rate was 25.3%.
“To our knowledge this is the first study to suggest a negative impact of regular AI use on healthcare professionals’ ability to complete a patient-relevant task in medicine of any kind,” said one author, Marcin Romańczyk.
The authors were careful about what it shows. The study was observational, so factors other than the AI may have influenced the drop, and the doctors were all experienced, so the effect on less experienced endoscopists is unknown. The senior author, Yuichi Mori of the University of Oslo, raised a further question: earlier trials compared AI-assisted colonoscopy with doctors working without it, and some of those doctors may also have been affected by using AI the rest of the time.
Writing in the same journal, Omer Ahmad of University College London called it “the first real-world clinical evidence for the phenomenon of deskilling,” and warned against “the quiet erosion of fundamental skills required for high-quality endoscopy.”