Development and Internal Validation of a Clinical–Infectious Prediction Model Incorporating the Serum Pepsinogen I/II Ratio to Identify Histologically Confirmed Gastric Atrophy

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Background: Histologically confirmed gastric atrophy is a key precursor lesion in the intestinal pathway of gastric carcinogenesis. Serum pepsinogen I/II ratio (PGR) is a biologically plausible non-invasive marker of oxyntic gland loss, but its diagnostic performance varies across populations, assays, and clinical settings. Contemporary risk stratification increasingly favors multivariable prediction models over isolated biomarkers.Objective: To evaluate the association between serum PGR and histologically confirmed gastric atrophy and to quantify the incremental contribution of PGR to a prespecified clinical–infectious prediction model.Methods: This prospective observational study enrolled 120 consecutive adults undergoing diagnostic upper gastrointestinal endoscopy at a tertiary referral center between January 2024 and February 2026. Participants underwent fasting serum pepsinogen I and II measurement, polymerase-chain-reaction testing for Helicobacter pylori, assessment of Epstein–Barr virus (EBV) status, and histopathological examination of gastric biopsy specimens. Logistic regression was used to estimate associations with gastric atrophy. Discrimination was quantified using the area under the receiver operating characteristic curve (AUC), overall accuracy using the Brier score, and internal validity using bootstrap resampling.Results: Histologically confirmed gastric atrophy was present in 32 of 120 patients (26.7%). Median PGR was lower in patients with gastric atrophy than in those with intestinal metaplasia or epithelial dysplasia (4.87 vs 6.53 and 6.27, respectively; Kruskal–Wallis P=0.0076). After adjustment for age, sex, H. pylori status, and EBV status, PGR remained independently associated with gastric atrophy (adjusted odds ratio [OR] 0.771 per unit increase; 95% confidence interval [CI] 0.619–0.960; P=0.020). The apparent AUC increased from 0.644 for the clinical–infectious model to 0.732 after addition of PGR; the bootstrap optimism-corrected AUC was 0.669. The Brier score was 0.171. At a cohort-derived threshold of PGR ≤5.56, sensitivity was 81.3%, specificity 68.2%, positive predictive value 48.1%, and negative predictive value 90.9%.Conclusions: Lower serum PGR was independently associated with histologically confirmed gastric atrophy and improved apparent discrimination when added to a clinical–infectious model. However, the optimism-corrected performance was only moderate, and the threshold was derived in a small single-center cohort. PGR should therefore be considered an adjunctive triage biomarker rather than a replacement for protocol-based endoscopy and histopathology. External validation and clinical-impact evaluation are required before implementation.