AI-assisted Colonoscopies Reduce Miss Rate by 50%
The most relevant cause of post-colonoscopy colorectal cancer (CRC) is the miss rate of colorectal neoplasia — the rate at which neoplastic lesions are not detected in a screening or surveillance colonoscopy. Some studies suggest that 52% to 57% of post-colonoscopy CRC cases are due to missed lesions at patients' colonoscopies. It's estimated that 25% of neoplastic lesions are missed following screening colonoscopy.
Mayo Clinic Gastroenterology and Hepatology, in collaboration with colleagues from around the world, found that using artificial intelligence (AI) in colorectal cancer screening produced a 50% reduction in the miss rate for colorectal neoplasia. Results of the study were published in the July 2022 edition of Gastroenterology.
"Colorectal cancer is almost entirely preventable with proper screening," says senior author Michael B. Wallace, M.D., division chair of Gastroenterology and Hepatology at Sheikh Shakhbout Medical City in Abu Dhabi, United Arab Emirates, and the Fred C. Andersen Professor at Mayo Clinic in Jacksonville, Florida. "The substantial decrease in miss rate using AI reassures health care providers on the decreased risk of perceptual errors."
A multicenter, multicountry study
This randomized control trial analyzed screenings of 230 patients, both men and women, with average risk for CRC. All participants underwent screening or surveillance colonoscopies at one of eight facilities in three countries: Italy, the United Kingdom and the U.S.
Each participant underwent two same-day, back-to-back colonoscopies. One screening was conducted with and one without AI. Patients were randomized into one of two arms. The test arm received a colonoscopy assisted by AI first, followed by a colonoscopy without AI. The control arm received a colonoscopy without AI first, followed by a colonoscopy assisted by AI.
Investigators conducting the screenings had a minimum of 1,000 completed colonoscopies and had an adenoma detection rate (ADR) between 20% and 40% or a polyp detection rate (PDR) between 30% and 70%. A maximum of three endoscopists per study site participated, and each investigator had a maximum of 90 randomized patients to screen.
Miss rate reduced using AI
Adenoma miss rate (AMR) reflected the number of histologically verified lesions detected at the second colonoscopy divided by the total number of lesions detected at both colonoscopies combined.
In the test arm, the arm with AI first, AMR was 15.5%. In the control arm, AMR was 32.4%. The polyp miss rate (PMR) was 16.9% and 31.1% for the test group and control group, respectively.
When compared with the control arm, the test arm reported lower AMR for nonpolypoid lesions (16.8% for test arm and 45.8% for control arm) and lesions less than or equal to 5 millimeters (15.9% for test arm and 35.8% for control arm). The test arm also had lower AMR for the proximal and distal colon than the control arm. False-negative rates were 6.8% in the test arm and 29.6% in the control arm.
"This computer-aided detection is highly accurate," says Dr. Wallace. "With decreased miss rates, we can ultimately increase our detection rates, detect colorectal cancer more accurately and save more lives."
The AI device used in this study had shown a substantial increase in ADR in two previous randomized control studies. This study indirectly validates that the ADR increase previously seen is specifically driven by reducing the miss rate risk. Larger studies are needed to assess the possible decrease of AMR for advanced adenomas.