Design and validation of white-light endoscopy model for predicting the nature of colorectal tumors
Objective:To analyze the clinical and endoscopic features of early colorectal cancer and construct a visualized and simple prediction model for assessing the nature of colorectal tumors based on white light endoscopy using a column chart to guide endoscopists in diagnosing early colorectal cancer and assist in selecting treatment methods.Methods:A retrospective analysis was conducted on the clinical, endoscopic, and pathological data of colorectal tumor patients who underwent endoscopic and surgical treatments at Beijing Shijitan Hospital, Capital Medical University, from January 2018 to December 2021. Emphasis was placed on the features observed under white light endoscopy. Independent risk factors for early colorectal cancer were determined through multifactor regression analysis, and a prediction model was constructed using a column chart. Six primary endoscopists with less than 6 months of endoscopic experience were randomly divided into a scoring group (S group) and a control group (C group) consisting of three individuals each. The two groups independently assessed the nature of colorectal tumors based on endoscopic images, and the differences in their assessments were compared to validate the model externally.Results:A total of 530 colorectal tumor patients were included in the analysis, including 287 cases of colorectal adenoma and 243 cases of early colorectal cancer. There was no statistically significant difference between the two groups in terms of gender, age, and lesion morphology (P＞0.05 for all). Multifactor regression analysis showed that lesion size (odds ratio [OR]=5.233, 95% confidence interval [CI]: 2.008-13.636, P=0.001), lesion location (left colon: OR=2.338, 95%CI: 1.329-4.111, P=0.003; rectum: OR=3.715, 95%CI: 1.692-8.160, P=0.001), villous features (OR=5.199, 95%CI: 3.057-8.842, P＜0.001), local depression (OR=5.162, 95%CI: 2.216-12.021, P＜0.001), and uneven surface (OR=5.583, 95%CI: 3.030-10.286, P＜0.001) were risk factors for early colorectal cancer. A column chart was constructed based on the identified risk factors to create a prediction model. A total of 110 endoscopic images of lesions were selected, including 50 cases of early colorectal cancer and 60 cases of colorectal adenoma. The assessment results in the S group ranged from 79.1% to 84.5% (average 81.8%), while in the C group, they ranged from 59.1% to 66.4% (average 63.03%). The S group performed significantly better than the control group in all cases (P＜0.05).Conclusion:The column chart prediction model generated based on multifactor logistic regression analysis of the white light endoscopic manifestations of early colorectal cancer demonstrates good predictive efficacy for the diagnosis of early colorectal cancer.
万方数据
