Development and internal validation of an objective diagnostic model based on gastric filling ultrasonography for gastric stump carcinoma: a TRIPOD-compliant retrospective study

Background Gastric stump carcinoma (GSC) is a late complication of partial gastrectomy. Gastroscopy (GS) is invasive and poorly tolerated. Gastric filling ultrasonography (GFU) is non-invasive, yet its interpretation has been largely subjective. This study aimed to develop and internally validate an objective GFU-based diagnostic model for GSC. Methods In this single-center retrospective study, 93 patients who underwent partial gastrectomy and received both GS and GFU (March 2008–June 2024) were included (60 GSC, 33 benign; all pathologically confirmed). Ten GFU indicators were evaluated. Univariate logistic regression screened predictors; a final two-predictor model was fitted using Firth penalized logistic regression. Performance was assessed by AUC, bootstrap internal validation (1,000 resamples), cross-validation, calibration, decision curve analysis, and DeLong tests. Results Thickest lesion diameter and absence of submucosal echo interruption were retained. The Firth model was: logit(P) = −1.456 + 3.131 × (thickest diameter, cm)−2.287 × (submucosal echo interruption absent; absent=1, present=0). Apparent AUC was 0.917 (95% CI 0.862–0.973); optimism-corrected AUC was 0.886 (95% CI 0.822–0.946). Calibration was satisfactory (Hosmer–Lemeshow p = 0.694; bootstrap-corrected slope 0.87). At the cut-off of 0.655, sensitivity was 86.7% and specificity 81.8%. The model significantly outperformed subjective GFU ( Δ AUC = 0.075, p = 0.019) but not subjective GS ( Δ AUC = 0.024, p = 0.574). Conclusions An objective two-parameter GFU model showed good, internally validated discrimination for GSC and outperformed subjective GFU. It may serve as a complementary triage tool alongside GS in postgastrectomy surveillance. External validation is required before clinical implementation.