Survival Modelling Using Machine Learning and Immune-Nutritional Profiles in Advanced Gastric Cancer on Home...

Area of Science:

• Oncology
• Nutrition Science
• Biostatistics

Background:

• Patients with stage IV gastric cancer and chronic intestinal failure necessitate home parenteral nutrition (HPN).
• Assessing prognostic factors is crucial for managing these complex patients.
• Nutritional and immune-inflammatory biomarkers offer potential predictive value.

Purpose of the Study:

• To evaluate the prognostic significance of nutritional and immune-inflammatory biomarkers in stage IV gastric cancer patients receiving HPN.
• To develop an individualized survival prediction model using machine learning.
• To identify key predictors of overall survival.

Main Methods:

• Secondary analysis of 410 stage IV gastric adenocarcinoma patients initiating HPN (2015-2023).
• Assessment of Controlling Nutritional Status (CONUT) score and lymphocyte-to-monocyte ratio (LMR).
• Application of Cox proportional hazards models and a Random Survival Forest (RSF) machine learning model.

Main Results:

• Both CONUT score and LMR were independent predictors of overall survival.
• Higher CONUT scores correlated with increased mortality risk (HR=1.656), while higher LMR values were protective (HR=0.632).
• The RSF model showed high predictive accuracy (C-index: 0.985-0.986), with CONUT score being the primary prognostic factor.

Conclusions:

• Machine learning applied to immune-nutritional data provides a robust method for survival prediction in advanced gastric cancer patients on HPN.
• This approach aids in risk stratification and personalized clinical decision-making for nutritional support and treatment intensity.
• CONUT score and LMR integration into predictive models can optimize patient management.