Machine learning model for recurrence-free survival in solitary resectable colorectal liver metastasis

Solitary colorectal liver metastasis (SCRLM) exhibits substantial heterogeneity in recurrence patterns after hepatic resection, yet individualized prediction tools for recurrence-free survival (RFS) are lacking.

In this multicenter retrospective study, 698 SCRLM patients undergoing hepatic resection were analyzed (training cohort: n=574; validation cohort: n=124). RFS was the primary endpoint. Three predictive models-random survival forest (RSF), Gradient Boosting Machine (GBM), and eXtreme Gradient Boosting (XGBoost)-were developed and compared. Model performance was assessed via concordance index (C-index), time-dependent area under the ROC curve (AUROC), and calibration plots.

The XGBoost model achieved the best performance, with AUROCs of 0.93 and 0.87 at 1 year, and 0.89 and 0.86 at 3 years, in the training and validation cohorts, respectively. Compared with the modified Clinical Score (m-CS), the model demonstrated significantly higher discrimination at 1, 2, and 3 years (all P < 0.001), and also identified a subgroup of patients more likely to benefit from postoperative chemotherapy. A user-friendly online tool was developed for clinical application: https://scrlm.shinyapps.io/scrlmapp/.

We developed and validated a machine learning-based model for SCRLM, enabling individualized recurrence risk prediction and guiding postoperative chemotherapy decisions. This approach may improve outcomes while reducing overtreatment.