Abstract
BACKGROUND AND AIMS: Colonoscopy is an effective screening tool, but it is resource-intensive and carries procedural risks. We aimed to develop and validate a biomarker-based risk prediction model (RPM) to improve triage of average-risk patients for colonoscopy by better predicting high-risk adenomas (HRAs). METHODS: We assessed blood-based biomarkers of glucose metabolism, liver enzymes, lipids, ferritin, and carcinoembryonic antigen (CEA) in 980 average-risk individuals undergoing screening colonoscopy. Logistic regression was used to predict HRAs, defined as adenomas ≥10 mm, villous histology, high-grade dysplasia, or ≥3 adenomas. Biomarkers were modeled as continuous variables; variable selection was performed using logistic regression and least absolute shrinkage and selection operator. Model performance was evaluated with area under the curve, reclassification indices, calibration statistics, bootstrap validation, and decision curve analysis. RESULTS: Glucose and CEA were significantly associated with HRAs, while no biomarkers were linked to high-risk sessile serrated lesions. Nine RPMs were developed; 3 were selected for validation. A model including all univariately associated biomarkers (CEA, C-peptide, gamma-glutamyl transferase, glucose, hemoglobin A1c, triglycerides) achieved an optimism-adjusted area under the curve of 0.66 (95% confidence interval, 0.60-0.71). Reclassification indices showed that adding biomarkers improved stratification, with the all-biomarkers model yielding the highest net reclassification improvement. CONCLUSION: Biomarker-enhanced RPMs modestly improved prediction of HRAs compared with models using only clinical and demographic variables. These findings support the potential of biomarkers to optimize colonoscopy resource allocation, but additional markers and refinement are needed to establish clinical utility.