Predictive Power: Machine Learning in Colorectal Surgical Planning

A newly validated extreme gradient boosting (XGB) model improves preoperative prediction of omental metastasis in right-sided colon cancer and has immediate implications for surgical planning for patients at elevated risk.

Investigators compared six machine-learning algorithms and found the XGB algorithm delivered superior accuracy and discrimination. XGB achieved AUCs of 0.924 in the training set, 0.868 in the internal test set, and 0.766 on external validation; it also showed higher overall accuracy and greater net benefit on decision-curve analysis than the other five methods. These gains were consistent across cohorts and performance metrics and support the model’s role in preoperative risk stratification for omental metastasis.

The model relies on routine preoperative clinical variables — including tumor location, preoperative CEA and CA19-9, tumor grade and histology, and tumor size — all available in standard preoperative workflows without specialized assays or radiomic processing. That accessibility enhances feasibility for rapid clinical integration and local validation.

Accurate preoperative prediction can change the surgical plan by prompting a more targeted omentectomy, adjusted staging strategies, or preparation for extended resection and intraoperative assessment. Refined risk stratification enables surgical teams to tailor the extent of omental inspection and to recalibrate intraoperative decision thresholds when predicted risk is high. Model guidance complements, rather than replaces, surgeon judgment; false positives require cautious interpretation alongside intraoperative findings.