Abstract Gastric cancer (GC) develops as a long-term process from the combined effects of multiple risk factors, during which genetic and epigenetic changes accumulate. The Correa cascade delineates the sequential transformation from normal gastric mucosa to GC, within which gastric intestinal metaplasia (GIM) represents a pivotal precancerous condition. Notably, incomplete GIM (IIM) carries a substantially increased risk of neoplastic progression compared with complete IM (CIM), with a pooled relative risk of 4.96 for gastric cancer; the sulfomucin-expressing type III subtype shows an even stronger association with cancer/dysplasia, with a pooled relative risk of 6.27 compared with type I/II IM. Endoscopy provides visualization of macroscopic mucosal alterations, whereas risk stratification at the histological and molecular levels requires the application of validated biomarkers. Although multiple molecular indicators have been reported in GIM, an integrated and clinically oriented synthesis of these markers is still lacking. This review summarizes recent advances in tissue- and serum-derived biomarkers that may help identify high-risk GIM. Tissue biomarkers reflect either the upstream molecular drivers of metaplastic transformation or the downstream effectors of epithelial reprogramming, such as aberrant oncogene activation, epithelial differentiation markers, transcription factors, trefoil factor family (TFF) members, and SPEM-associated proteins. Additionally, circulating biomarkers, including established serum indices and novel omics- and epigenetic-based signatures, offer non-invasive potential for early GC prediction. Incorporating candidate biomarkers into clinical workflows may enhance early detection and optimize surveillance strategies for patients with GIM. We also distinguishes phenotype-defining markers from candidate progression-risk markers in GIM. The next step is to test whether integrated tissue biomarker panels improve prediction of dysplasia or GC on top of OLGIM, EGGIM, and clinical risk factors in longitudinal cohorts.
Zhang et al. (2026) studied this question.