A screening model for advanced colorectal neoplasia based on tumor markers and inflammatory indices: a retrospective study with an online risk calculator
Related Experiment Video
Updated: Jul 15, 2026
Competing-Risk Nomogram for Predicting Cancer-Specific Survival in Multiple Primary Colorectal Cancer Patients after Surgery
Published on: September 27, 2024
Chang Zhang1, Lei Jin1, Shihao Wu2
1Department of Gastroenterology, The Fourth Affiliated Hospital of Anhui Medical University, Hefei, China.
Objectives:
The goal is to design and validate a noninvasive screening model for advanced colorectal neoplasia(ACN) utilizing routine blood tumor markers and inflammatory indices, and to develop an online calculator for assessing individual risk.
Methods:
In this retrospective analysis, 1,290 patients who underwent colonoscopy and had full preoperative blood test results were included. Based on pathology, patients were categorized into a group with advanced colorectal neoplasms and another with non-advanced neoplasms. Patients were randomly assigned to a training set, making up 70%, and a test set, making up 30%.Candidate variables were initially screened using univariate logistic regression, and those with statistical significance were then included in a multivariate logistic regression model to determine independent predictors. To assess potential nonlinear relationships between continuous variables and the risk of advanced neoplasms, restricted cubic splines were employed. Prediction models were constructed using five machine learning algorithms and evaluated using the area under the receiver operating characteristic curve. The robustness of key predictors was further assessed through sensitivity and stratified analyses. SHAP was applied to interpret the final model, which was subsequently implemented as an online calculator.
Results:
Among the 1,290 patients, 210 were diagnosed with advanced colorectal neoplasms. CEA, SII, NLR, ALB, and PLR were identified as independent predictors, and their associations remained stable across sensitivity and stratified analyses. Among the models, XGBoost achieved the best performance, with an AUC of 0.956 (95% CI 0.936-0.976). SHAP analysis identified SII as the most influential predictor.
Conclusion:
A machine learning model based on key blood markers such as CEA, SII, and NLR can effectively support noninvasive screening for advanced colorectal neoplasms. This online calculator is a convenient and practical resource for evaluating risk on an individual basis in clinical practice.
