CADDIE Version 1.0
Research type
Research Study
Full title
Multi-Centre, open-label, randomised, prospective trial to assess efficacy and safety of the CADDIE artificial intelligence system for improving endoscopic quality.
IRAS ID
251596
Contact name
Manish Chand
Contact email
Sponsor organisation
University College London
Clinicaltrials.gov Identifier
Duration of Study in the UK
3 years, 0 months, 1 days
Research summary
Research Summary:
Abnormal growths in the colon and rectum is referred to as polyps. Polyps can be harmless (such as hyperplastic polyps in the rectum) or harmful (for example adenoma polyps). Colonoscopy is considered the ‘gold standard’ for detecting polyps and cancer in the colon and rectum. Yet a major challenge in colonoscopy is the significant miss rates for these polyps. CADDIE is a multi-site study to evaluate the performance and safety of a novel investigational medical device called ‘CADDIE’ (‘Computer Aided Diagnosis and Detection for Intelligent Endoscopy’).The CADDIE device comprises a computer with an in-built software programme developed using artificial intelligence (AI) technology. CADDIE processes information contained in digital images in real-time during endoscopy procedures. It alerts the endoscopist to abnormalities by sounding a subtle alarm each time it detects a polyp in an image. This is accompanied by a coloured shape box, delineating the abnormal area, on the endoscopy screen. When the endoscopist presses on a foot pedal, it will reveal the CADDIE system’s suggestion of the type of polyp present (e.g. adenoma polyp).
The target population in the study is patients undergoing a scheduled colonoscopy as part of their routine clinical care. Consenting patients will be registered and randomised to the intervention (=routine colonoscopy with the CADDIE device), or to the control group (=routine colonoscopy). We will use the same endoscopist in both to minimise variation across the two study arms. To answer the main research question, we will compare the number of polyps detected by endoscopists when using the device, and when not using the device and whether the endoscopist diagnosis of the type of polyp present improves with using the CADDIE.
We will recruit 654 patients into the study and the study will last up to three years. There are no study-related follow up visits.
Summary of Results:
The main aim of the CADDIE Trial was to assess if CADDIE helped humans who perform colonoscopy ('endoscopists') to increase the number of polyps that are detected during colonoscopy, particularly pre-cancerous polyps called adenomas. CADDIE is a cloud-based artificial intelligence system which aims to help identify polyps visible in the endoscopy monitor as well as the type of polyp present. The study was conducted in nine NHS hospitals across the UK, with more than 25 endoscopists (the individuals performing the colonoscopy). The study included all the different professionals that perform colonoscopy, which include gastroenterologists, colorectal surgeons and clinical nurse endoscopists.We analysed more than 600 patients' data. We found that performing colonoscopy with the help of CADDIE increased the number of patients where at least one adenoma polyp or colorectal cancer was identified by 8.10%, mainly due to detecting more adenoma polyps. 33.30% of patients were found to have at least one adenoma polyp or colorectal cancer when the endoscopist used CADDIE, compared to 25.2% when the endoscopist did not use CADDIE.
After performing statistical analyses to account for variations between the two arms in the study in terms of age, sex and the reason for why the patients' had a colonoscopy performed, we identified that the odds of an adenoma polyp or colorectal cancer being detected in a patient increased by 46% when using CADDIE compared to without. In terms of the detection of any pre-cancerous polyp (either a sessile serrated lesion or adenoma) or colorectal cancer, the odds of detecting this in a patient increased by 47% when using CADDIE compared to without in the adjusted analysis. Importantly, we found that using CADDIE did not increase the length of time for a colonoscopy procedure. We found no evidence of additional safety concerns when endoscopists used the CADDIE system.
Regarding diagnosing the type of polyp present, for diminutive polyps (5mm or smaller), we found that the CADDIE system was more accurate in correctly diagnosing an adenoma compared to the humans' own diagnosis without the help of the CADDIE system. When analysing the polyps where the CADDIE system provided a diagnosis, the CADDIE device correctly diagnosed 87.5% of adenoma polyps compared to 70% for the humans' own diagnosis. There was no significant difference in the diagnosis of polyps that were not adenomas ('non-adenomas') between the CADDIE system and humans' own diagnosis without the help of the CADDIE system.
REC name
West Midlands - South Birmingham Research Ethics Committee
REC reference
20/WM/0219
Date of REC Opinion
5 Oct 2020
REC opinion
Further Information Favourable Opinion