Aug. 19, 2023
Patients diagnosed with inflammatory bowel disease (IBD) are at increased risk of colorectal cancer. To help detect precancerous or cancerous polyps at an early stage in these patients, current guidelines recommend beginning surveillance colonoscopies 8 to 10 years after IBD diagnosis and every 1 to 2 years thereafter.
Background mucosal inflammation in patients with IBD can make colorectal lesions, especially flat ones, harder to identify. When evaluating colorectal polyps endoscopically, gastroenterologists use a variety of scoring systems to describe the morphology of a lesion or to predict its histopathology. The Surveillance for Colorectal Endoscopic Neoplasia Detection and Management in Inflammatory Bowel Disease Patients International Consensus (SCENIC) statement, published in Gastrointestinal Endoscopy in 2015, currently recommends that endoscopists characterize polyps using a modified Paris classification system.
According to Nayantara Coelho-Prabhu, M.B.B.S., a gastroenterologist at Mayo Clinic in Rochester, Minnesota, understanding how accurately endoscopists grade the morphology and predict the histology of colorectal lesions in patients with IBD using currently available classification systems is important.
"Use of the modified Paris classification system for evaluating IBD lesions has not yet been validated. The results of this assessment directly impact real-time treatment decisions and patient outcomes," explains Dr. Coelho-Prabhu. "The structure of the polyp can dictate the type of therapy performed, specifically the type of resection technique optimal to ensure complete removal. Some structural features may suggest a higher likelihood of precancerous changes within the polyp known as dysplasia. Conversely, certain features may indicate a lower risk of malignancy, allowing for less invasive treatment options."
To analyze these issues, Dr. Coelho-Prabhu and colleagues conducted a study analyzing interobserver agreement (IOA) among endoscopists using the modified Paris classification system to evaluate visible colorectal lesions and the accuracy of endoscopist pathology predictions in patients with IBD. The results of this study were published in Gastrointestinal Endoscopy in 2023, with Dr. Coelho-Prabhu serving as corresponding author.
"The key finding of our study suggests an unsatisfactory level of agreement between endoscopists when using currently available rating systems to assess colorectal polyps in patients with IBD, for morphology classification as well as for histology prediction."
Methods
In this study, 10 senior and four trainee endoscopists from five tertiary care centers graded 100 deidentified endoscopic still images and 30 videos of visible colorectal lesions found in patients with IBD. Endoscopists were asked to review each image and provide a single response for each of the following: standard Paris classification, modified Paris classification, lesion border assessment and histopathology prediction. Researchers then measured the agreement among these responses using Light's kappa coefficient, and they assessed the consensus among the ratings according to strict majority. Participating endoscopists had access to a teaching guide containing the Paris and modified Paris classifications during their reviews.
Results
Overall, the study data demonstrated a very low IOA for Paris and modified Paris classifications and low accuracy and IOA for lesion histopathology prediction.
- The Light's kappa coefficient for all study endpoints (using both image and video data sets) was from 0.32 to 0.49. In a subgroup analysis comparing data from endoscopist trainees and data from senior endoscopists, the Light's kappa coefficient was less than 0.6, with a slightly higher IOA among trainees.
- Paris classification: The overall IOA was 0.41 for images and 0.42 for videos.
- Modified Paris classification: The overall IOA was 0.42 for images and 0.41 for videos.
- Lesions with the lowest IOAs or no consensus were mostly classified as Is, IIa and mixed using the Paris classification system, and as sessile and superficial elevated using the modified Paris classification system. The difference between these categories is whether the lesions are elevated less than or greater than 2.5 mm. The data show that this is clearly very difficult to ascertain. The more important aspect is whether this difference is clinically relevant in terms of outcomes, and there are no data to support any differences in outcomes.
- Lesion border assessment: The overall IOA for border prediction was higher in videos (0.49) than in images (0.32). However, it was still poor. This is relevant because only lesions with well-demarcated borders are candidates for endoscopic resection, while poorly defined lesions should be removed with surgical resection.
- Histopathology prediction: The overall IOA for histology in still images was 0.39. When compared with ground-truth pathology, the overall accuracy of histopathology prediction was 59% for senior endoscopists and 57% for trainees. The overall IOA for histology in videos was 0.37. When compared with ground-truth pathology, the overall accuracy of pathology prediction was 52%. This evidence strongly supports the need for development of artificial intelligence (AI)-powered polyp detection and characterization tools in this patient population.
- The association between the Paris classification and the predicted histopathology was strong (P < 0.001) in both still and video images.
"The key finding of our study suggests an unsatisfactory level of agreement between endoscopists when using currently available rating systems to assess colorectal polyps in patients with IBD, for morphology classification as well as for histology prediction," says Dr. Coelho-Prabhu. "Further research and innovative strategies are needed to develop more-accurate lesion characterization systems and to avoid missing precancerous lesions or leaving these lesions unbiopsied and unresected in patients with IBD."
Dr. Coelho-Prabhu explains that future research focused on colorectal polyp morphology and histopathology prediction will help clinicians refine surveillance protocols and help determine the optimal intervals for follow-up colonoscopies to detect and identify high-risk lesions.
"Additional research to help us better understand how a lesion's features may impact its dysplastic potential is also necessary to create better endohistologic scoring systems," says Dr. Coelho-Prabhu. "An ideal scoring system should be easy for practitioners in the community to use and have high interobserver agreement and reproducibility."
Dr. Coelho-Prabhu and co-authors also emphasize that having an AI tool to detect and characterize colorectal lesions in patients with IBD may assist in tackling all of the above challenges.
"Our research team also recently developed a novel AI tool for computer-aided detection of colorectal lesions in IBD," explains Dr. Coelho-Prabhu. "Automated characterization of these lesions is an ongoing research effort. Understanding the morphological changes associated with IBD-related lesions can help us develop new models powered by AI to assist clinicians in accurate lesion recognition, interpretation, treatment and surveillance. The results of our efforts to develop this AI-based tool were published in iGIE in 2023."
For more information
American Society for Gastrointestinal Endoscopy and American Gastroenterological Association. SCENIC international consensus statement on surveillance and management of dysplasia in inflammatory bowel disease. Gastrointestinal Endoscopy. 2015;81.489.
Guerrero Vinsard D, et al. Interobserver agreement of the modified Paris classification and histology prediction of colorectal lesions in patients with inflammatory bowel disease. Gastrointestinal Endoscopy. 2023;97:790.
Guerrero Vinsard, D, et al. Development of an artificial intelligence tool for detecting colorectal lesions in inflammatory bowel disease. iGIE. 2023;2:91.
Refer a patient to Mayo Clinic.