Derivation and Validation of a Predictive Model for Advanced Colorectal Neoplasia Among Average-Risk Adults in China

As the burden of colorectal cancer (CRC) continues to rise and current screening methods have limitations in efficiency and accuracy, there is an urgent need particularly in China and other large populations to develop and validate precision risk prediction models based on individual risk factors with improved discrimination and generalizability to optimize the allocation of screening resources. Asymptomatic individuals aged 40-74 years were recruited from multiple hospitals in China, between January 2024 and May 2025. All participants completed a standardized questionnaire, physical measurements, and colonoscopy. A multivariable logistic regression model was derived to predict the risk of advanced colorectal neoplasia (AN), and a scoring system was derived from regression coefficients. Model performance was evaluated using discrimination, calibration, and risk stratification ability. Among 9617 participants, 673 AN cases were identified. The final model included six variables: age, gender, smoking, drinking, chronic appendicitis, and hypertension. The model demonstrated acceptable discriminatory ability (c-statistics: 0.656 internal validation) and good calibration. A scoring system (range: 0-37) classified individuals into low-, intermediate-, and high-risk groups. AN prevalence for each group was 3.46%, 6.25%, and 11.48%, respectively. The number needed to screen (NNS) improved significantly from 28.9 in the low-risk group to 8.7 in the high-risk group. The prediction model and scoring system developed enable effective risk stratification and are suitable for individualized assessment prior to colonoscopy in resource-limited settings, thereby improving screening efficiency.