OBJECTIVE: To develop and validate a machine learning framework combined with a nomogram for predicting recurrence after radical gastrectomy in patients with vascular and neural invasion.

METHOD: Machine learning models, including Random Survival Forests, Decision Survival Tree, Extreme Gradient Boosting, and a nomogram, were developed and assessed for their ability to predict recurrence-free survival in patients who underwent radical gastrectomy for non-metastatic gastric cancer with "double invasion".

RESULTS: A total of 559 patients were included in the study, and the machine-learning models demonstrated higher c-index values than the nomogram. The Random Survival Forests model had the highest c-index of 0.791, followed by Extreme Gradient Boosting (0.788) and Decision Survival Tree (0.728). Our refined nomogram harnessed the power of the Random Survival Forests algorithm to weave together the critical influence of nine variables: patient gender, age, the tally of positive lymph nodes, the surgical approach to gastrectomy, the tumor's positional characteristics, and the molecular biomarker expression profiles, including CD56 and FHIT, along with ki67 levels and the tumor's maximum diameter. All models showed good calibration with low integrated Brier scores (< 0.1), although there was calibration drift over time, particularly in the traditional nomogram model. DCA showed an incremental net benefit from all machine learning models compared with conventional models currently used in practice.

CONCLUSION: Random Survival Forests have surpassed traditional machine learning and nomograms in predictive accuracy, yet nomograms remain vital for identifying high-risk patients and guiding postoperative care. Combining nomograms with advanced machine learning in a hybrid model enhances patient care, provides critical insights, and supports informed clinical decisions for gastric cancer cases with "double invasion".