A machine learning-based predictive model for lymph node metastasis in elderly patients with early gastric cancer and prognosis.

Lymph node metastasis (LNM) plays a key role in the treatment of patients with early gastric cancer (EGC), especially elderly patients (over 65 years old). The accurate preoperative prediction of LNM is still a challenge, and studies of elderly patients are lacking. The purpose of this study was to develop and validate an effective model based on machine learning to predict the risk of preoperative LNM in elderly patients with EGC. The data of 1578 patients from 2010 to 2020 were extracted from the Surveillance Epidemiology and End Results (SEER) database. Five different machine learning algorithms were applied to construct the models. The external validation data were obtained from 127 elderly patients in our hospital. The area under the receiver operating characteristic curve (AUC), accuracy, sensitivity and specificity of the models were compared and evaluated. A Cox regression model was used to determine the prognostic factors for overall survival (OS) in elderly patients with EGC. Of the 1578 patients included in this study, 165 (10.46%) had LNM. Multivariate logistic regression analysis revealed that differentiation grade, T stage, tumour size were independent predictors of LNM. The performance of the random forest (RF) model was the best, with an AUC of 0.771 in the training cohort, 0.733 in the internal validation cohort and 0.796 in the external test. Multivariate Cox regression analysis confirmed that sex, age, primary site, tumour size, T stage, radiotherapy and RNE were independent prognostic factors for OS in elderly patients with EGC. Based on RF model, clinicians can develop more reasonable treatment plans to improve the prognosis of patients and reduce the burden on the medical system and patients.