Performance of Ulcer Features in Predicting Malignancy Among Gastric Ulcers Diagnosed on Upper Endoscopy.

Objective: Even though the prevalence of malignancy within gastric ulcers is low, surveillance endoscopy is routinely performed after gastric ulcer diagnosis resulting in unnecessary costs and risks. Endoscopic appearance may be used to identify ulcers with malignant features and guide decisions regarding the need for surveillance endoscopy. Our aim was to assess the predictive value of several endoscopic ulcer features with the risk of prevalent malignancy in patients diagnosed with gastric ulcers.

Methods: Patients with gastric ulcers were identified using endoscopic reporting software in 2 hospitals in Houston, TX, from February 2019 to July 2021. Malignant and benign gastric ulcers were defined using ulcer biopsy histopathology, and ulcers that had healed on surveillance endoscopy were also classified as benign ulcers. Potential endoscopy-related predictors of malignant ulcers included: Forrest classification, location, size, elevated border, irregular border, and background gastric atrophy.

Results: We identified 338 patients with gastric ulcers, and 150 (44%) had at least one surveillance endoscopy. Malignant ulcers were found in 41 patients (12%). The strongest predictors of malignancy were irregular border [area under receiver operating characteristic (AUROC): 0.89, 95% CI: 0.80-0.97], gastric atrophy on histopathology (AUROC: 0.87, 95% CI: 0.78-0.96), and elevated border (AUROC: 0.84, 95% CI: 0.73-0.95). A multivariate model including corpus/cardia location, irregular border, elevated border, and gastric atrophy on histopathology had the best discrimination for predicting malignant ulcers (AUROC: 0.96, 95% CI: 0.93-0.98) with low false negatives (0.4%).

Conclusions: A model combining corpus/cardia location, irregular border, elevated border, and gastric atrophy on histopathology best-predicted malignancy in gastric ulcers and may identify patients with the most benefit from surveillance endoscopy.