Research Square · Oct 27, 2026
Routine clinical variables do not support pre-endoscopic triage for gastric intestinal metaplasia in H. pylori-seropositive safety-net patients
By Thrift A, Weaver N, Zarrin-Khameh N et al.
Purpose Gastric intestinal metaplasia (GIM) is a precursor to gastric adenocarcinoma and an actionable target for prevention. Pre-endoscopic risk scores may help prioritize endoscopy in heterogenous populations, but their additional value over H. pylori seropositivity alone remains uncertain. We evaluated whether routinely available clinical and demographic variables could identify GIM among H. pylori -seropositive adults in a safety-net setting. Methods We analyzed 273 asymptomatic H. pylori -seropositive patients (GIM prevalence 28.9%) recruited from primary care clinics in a safety-net health system. Random forest (RF) and Least Absolute Shrinkage and Selection Operator (LASSO) models were developed using 8 pre-endoscopic predictors spanning demographics, clinical, lifestyle, and medical history factors. Performance was assessed in a held-out test set using area under the receiver operating characteristic curve (AUROC), classification metrics, and decision curve analysis (DCA). Results Participants were predominantly Hispanic (92.3%), first-generation immigrants (90.5%), and low-income (72.9% with annual household income <$25,000). Age was the only candidate predictor associated with GIM (p = 0.033). RF and LASSO achieved held-out AUROCs of 0.558 (95% CI: 0.422–0.707) and 0.500 (95% CI: 0.371–0.643), respectively. Neither model provided clinically meaningful net benefit compared with endoscopic evaluation of all H. pylori -seropositive participants. Conclusion In this high-risk safety-net population, routinely available pre-endoscopic variables did not meaningfully discriminate GIM after H. pylori seropositivity was established. Findings suggest risk-score-based triage may be insufficient to determine which seropositive patients should undergo endoscopy. Systematic endoscopic evaluation may be more appropriate in similar high-risk settings, while future prediction models may require biologic markers that better capture residual GIM risk.