Integrated analysis of gastric microbiome and clinical features for the diagnosis of gastric neoplasms

Gastric cancer (GC) progression is closely associated with microbial dysbiosis. However, microbial alterations in the intermediate stage of gastric adenoma (GA) remain under-characterized. This study investigated stage-specific microbiome changes and evaluated predictive models integrating microbial and clinical features.

We analyzed 231 subjects (129 GA, 73 GC, 29 healthy controls [HC]) using 16 S ribosomal DNA sequencing of paired oral, non-lesional gastric mucosa (GM), and gastric lesion (GL) samples. Random Forest models were constructed using microbial genera and clinical features, validated via leave-one-out cross-validation.

Stepwise microbial dysbiosis was observed during GC progression. Alpha diversity was significantly lower in GC-GL than in HC-GM and GA-GL. Helicobacter pylori infection was associated with reduced alpha diversity in GL samples. Genera such as Streptococcus, Neisseria, and Haemophilus were enriched in GC-GL, while Phocaeicola, Faecalibacterium, and Bifidobacterium were depleted. A gastric neoplasm prediction model incorporating 33 microbial genera and six clinical variables (age, sex, BMI, hypertension, smoking, and atrophy) demonstrated superior predictive performance (AUC = 0.830) in distinguishing gastric neoplasms from HCs, compared to models using either microbial genera (AUC = 0.790) or clinical variables (AUC = 0.715) alone. The combined model also accurately distinguished between GA and GC (AUC = 0.807).

The gastric microbiome profile changes dynamically during GC progression, highlighting its potential role in tumor development. Integrating microbial signatures with clinical features enables robust classification of gastric neoplasms and may serve as a valuable adjunctive tool for diagnosis and risk stratification in endoscopic practice.