Prediction model to prioritize colorectal cancer screening with visual colonoscopy

ObjectivesVisual colonoscopy is a standard method for colorectal cancer screening but carries unnecessary operative risks. This study aimed to develop a clinical prediction model to identify patients at high risk of colorectal polyps or benign conditions detectable via colonoscopy.SettingData were routinely collected during mass screenings: December 2022 at Kumpawapi Hospital, January 2023 at Nhonghan Hospital, and April 2024 at Wangsammo Hospital. All participants with positive fecal immunochemical tests were included.MethodsA retrospective delayed-type cross-sectional study was conducted. Predictors included male sex, age, family history of colorectal cancer, prior colonoscopy, smoking, alcohol use, diabetes, clinical symptoms (e.g. altered bowel habits, weight loss, decreased stool caliber), and hematocrit level. Polyps were biopsied and histologically examined. A clinical prediction model was derived using multinomial logistic regression, selecting predictors based on clinical relevance and face validity. Patients were classified into three risk groups (normal, benign, malignant). Model absolute accuracy was assessed by comparing predicted versus actual classes. Polytomous discrimination index (PDI) was evaluated.ResultsAmong 1071 patients undergoing colonoscopy, 66 (20.3%) had benign polyps, 148 (45.4%) had non-advanced adenomas, 103 (31.6%) had advanced adenomas, and 9 (2.7%) had colorectal cancer. Final predictors were male sex, age (year), family history of colorectal cancer, alcohol use, and abdominal pain. The model showed an absolute accuracy of 0.484 (95% CI, 0.454-0.513) and a PDI of 0.426 (95% CI, 0.396-0.456).ConclusionsThe model showed fair discrimination in identifying high-risk patients. This prediction rule may support shared decision-making for elective colonoscopy and help prioritize patients for urgent screening.