Introduction: Early detection of gastric cancer (GC) and risk stratification for metachronous gastric lesions (MGLs) remain unmet clinical needs, particularly in intermediate-risk populations where population-based endoscopic screening is impractical. Saliva has emerged as a promising non-invasive biofluid for cancer detection and monitoring.

Methods: Three complementary studies were conducted: (1) RNA sequencing (n = 54) and RT-qPCR validation (n=270) on formalin-fixed paraffin-embedded (FFPE) samples to establish a tissue-based gene expression signature; (2) a retrospective longitudinal analysis to assess its value in MGL risk prediction (n=168); and (3) a prospective study evaluating its translation into saliva using droplet digital PCR (ddPCR) in patients with gastric lesions and endoscopy-confirmed controls (n=197).

Results: Differential expression analysis identified 180 dysregulated genes in gastric lesions versus normal mucosa. A seven-gene signature was validated in independent datasets (TCGA-STAD/GTEx) restricted to stage I GC. Six genes demonstrated excellent diagnostic performance in FFPE tissue (AUC=0.96, 95% CI 0.94–0.99). In the longitudinal cohort, integration of the signature with clinical variables achieved moderate prediction of MGL risk (AUC=0.74, 95% CI 0.59–0.88). Translation into saliva revealed consistent dysregulation of one gene, whose downregulation remained independently associated with gastric lesion risk (aOR=0.54, 95% CI 0.29–0.91, P=0.035). A salivary-based model combining gene expression with demographics achieved AUC=0.78 (95% CI 0.69–0.88), sensitivity 0.82, specificity 0.67.