Systemic inflammatory biomarkers (NLR, SII, PNI and FPR) combined with CEA for predicting advanced colorectal neoplasms: development and temporal validation of a machine learning model.

BACKGROUND: Chronic systemic inflammation is closely associated with the initiation and development of colorectal tumors. Biomarkers reflecting inflammatory status may help detect advanced colorectal neoplasms (ACRN), which include advanced adenoma and colorectal cancer. Nevertheless, the diagnostic ability of combined inflammatory and nutritional indicators for ACRN has not been fully clarified. This study evaluated whether several systemic inflammation markers-NLR, SII, PNI, and FPR-together with CEA, can improve the identification of ACRN. METHODS: A retrospective analysis was conducted in individuals who received colonoscopy from December 2016 to December 2024. Eligible subjects were randomly assigned to a training set and an internal testing set. Patients enrolled between January 2025 and January 2026 were used as a temporal validation cohort. Correlation analysis together with restricted cubic spline (RCS) modeling was applied to assess the associations between inflammatory markers and ACRN risk. Machine learning approaches were then used to construct prediction models based on NLR, SII, PNI, FPR, and CEA. Model performance was assessed by discrimination ability, calibration, and clinical usefulness. An online calculator was also established to support individualized risk estimation. RESULTS: A total of 1330 individuals were analyzed, including 252(18.95%) cases diagnosed with ACRN. Higher values of NLR, SII, FPR, and CEA were linked to a greater probability of ACRN, while PNI showed a negative relationship. These indicators also displayed clear changes across different stages of disease. Among the evaluated machine learning approaches, the XGBoost model showed the strongest predictive ability. The AUCs reached 0.960 in the training set, 0.944 in the testing set, and 0.908 in the temporal validation cohort. In addition, a web-based calculator was built to support personalized risk assessment. CONCLUSION: Systemic inflammatory and nutritional biomarkers combined with CEA demonstrated robust predictive performance for identifying advanced colorectal neoplasms and may provide a convenient tool for early risk stratification in colorectal cancer screening.