**Background:** Gastric cancer (GC) is the third most common cancer worldwide and in China. Intestinal metaplasia (IM) is a known precursor lesion that increases GC risk, but existing risk-stratification methods for gastric precancerous lesions have limited effectiveness. Prior evidence suggested that monoclonal gastric cancer 7 antigen (MG7-Ag), human telomerase reverse transcriptase (hTERT), and trefoil factor family 2 (TFF2) may be involved in gastric carcinogenesis. This study aimed to construct and validate a multimolecular prediction model incorporating these three biomarkers to identify high-risk IM patients and provide early-warning value for GC.
**Methods:** The study was approved by the Ethics Committee of Xijing Hospital, Air Force Medical University. In stage 1 (retrospective case-control), 17 patients with IM that progressed to GC (IM-GC group) were matched 1:1 by age, sex, and biopsy interval with three control groups: IM-NoGC (IM without progression), CG-NoGC (chronic gastritis without progression), and GC (gastric cancer tissue from the same IM-GC patients). The modeling cohort comprised the IM-GC and IM-NoGC groups (34 total). Immunohistochemical staining was scored semiquantitatively as the product of positive cell percentage (0–4) and staining intensity (0–3), yielding negative (0), weakly positive (1–4), moderately positive (5–8), or strongly positive (9–12). Logistic regression (inclusion criterion P < 0.1) identified predictors. Two models were built: Model 1 (MG7-Ag + hTERT + TFF2) and Model 2 (hTERT + TFF2). In stage 2 (prospective validation), 31 patients (8 IM-GC, 23 IM-NoGC) from October 2019 to December 2020 were enrolled. Discrimination was assessed by ROC AUC, calibration by Hosmer–Lemeshow test and calibration plots.
**Key Results:** In the retrospective cohort, MG7-Ag was overexpressed in the GC group versus others (P < 0.05), with no difference between IM-NoGC and CG-NoGC. hTERT was overexpressed in GC and IM-GC groups, with significant difference between IM-GC and IM-NoGC (P < 0.001). TFF2 showed low expression in GC and IM-GC groups, with significant difference between IM-GC and IM-NoGC (P = 0.001). Logistic regression identified hTERT and TFF2 as independent predictors. In the modeling cohort, both Model 1 and Model 2 achieved an AUC of 0.971. Model 1 had sensitivity 88.2% and specificity 100%; Model 2 had sensitivity 82.4% and specificity 100%. In the validation cohort, Model 1 achieved AUC 0.87, sensitivity 75.0%, specificity 87.0%; Model 2 achieved AUC 0.84, sensitivity 100%, specificity 65.2%. Calibration was good for both models (Model 1: chi-squared = 4.592 internal, 3.576 external, both P > 0.05; Model 2: chi-squared = 1.318 internal, 3.297 external, both P > 0.05). The annual GC progression rate in this study was approximately 8%, higher than the <3% reported in prior studies.
**Clinical Implications:** This study demonstrates that a three-biomarker panel (MG7-Ag, hTERT, TFF2) can effectively stratify IM patients by GC risk, with high discrimination and calibration in both internal and external validation. The model outperformed a two-marker model and previous risk-stratification approaches. If validated in larger, multicenter studies, this molecular prediction tool could guide surveillance intensity and early intervention for high-risk IM patients. Limitations include single-center design and small sample size.