Predicting the Prognosis of Radical Gastrectomy After Neoadjuvant Chemotherapy Based on Machine Learning Technology - A Multicenter Study in China

Authors
Qi-Chen He, Wen-Wu Qiu, Ji-Xun He, Jian-Wei Xie

Affiliations
Department of Gastric Surgery, Fujian Medical University Union Hospital, Fuzhou, China

Background
Neoadjuvant chemotherapy (NAC) can improve the prognosis of patients with locally advanced gastric cancer (LAGC), but there is currently a lack of precise models for predicting their prognosis.

Methods
This retrospective study included gastric cancer patient data from three centers in China. The internal validation cohort was randomly allocated to training and testing sets in a 6:4 ratio. COX regression was used to select feature variables, and six different machine learning algorithms were employed to build models. Internal validation was conducted using k-fold cross-validation technique.

Results
A total of 385 patients were included in this study, with 167 patients in the training set, 112 patients in the testing set, and 106 patients in the external validation set. Twelve variables were selected for the model through COX analysis, with the SVM model identified as the best predictive model after screening (AUC for training set: 0.96; AUC for testing set: 0.75; AUC for external validation set: 0.70). Patients were classified into high-risk group (HRG) and low-risk group (LRG) based on the optimal survival difference threshold. In the internal validation set, HRG patients had significantly lower survival than LRG patients, with average overall survival of 47.90 months (95% CI, 41.50 - 54.30) and 65.72 months (95% CI, 60.41 - 71.04) respectively (log-rank P = 0.001). Regarding recurrence types, the HRG group had higher rates of overall recurrence (HRG vs. LRG: 35.1% vs. 48.6%, P = 0.026) and peritoneal recurrence (HRG vs. LRG: 5.7% vs. 15.2%, P = 0.008).

Conclusion
The SVM model can help surgeons identify clinically significant key prognostic factors and facilitate postoperative monitoring and management of patients with neoadjuvant gastric cancer.