Metabolomic Insights into the Predictive Landscape of chemoimmunotherapy in Gastric Cancer: Towards Precision Medicine with Machine Learning

Abstract: Objective Chemoimmunotherapy has emerged as a promising therapeutic strategy for patients with gastric cancer (GC). However, significant heterogeneity among individual patients and the development of treatment resistance severely limits the proportion of patients who derive true clinical benefit. Consequently, there is an urgent need for accurate tools to predict therapeutic response and guide clinical decision-making. Design A comprehensive metabolomic analysis was conducted on 369 plasma samples obtained from 108 patients with gastric cancer undergoing chemoimmunotherapy. Machine learning techniques were used to create two models for predicting and monitoring clinical efficacy. Results The machine learning-based prediction model, utilizing baseline metabolite expression, achieved a high predictive performance with an area under the curve (AUC) of 0.857 (95% CI: 0.784–0.929), an accuracy of 0.787, and a specificity of 0.837. Furthermore, the clinical efficacy monitoring model demonstrated robust performance, yielding an AUC of 0.861 (95% CI: 0.666–1), an accuracy of 0.84, and a specificity of 0.812. Regarding pathological outcomes, dynamic changes in Indolelactic acid and Undecanedioic acid, which exhibited significant negative correlations with Becker scores, indicating improved tumor regression. Conversely, Deoxycholic acid and Adrenaline showed significant positive correlations with Becker scores, suggesting an association with poor pathological response. Conclusion The results of this study reveal the metabolic status of GC patients after chemoimmunotherapy, creating models capable of independently predicting and monitoring clinical efficacy, providing strong evidence for the advancement of precision medicine in the field of GC.