Documented risk-based antithrombotic management and 30-day outcomes after endoscopic hemostasis for non-variceal upper gastrointestinal bleeding: a retrospective cohort study

Objective To evaluate whether documented adherence to a prespecified risk-based antithrombotic management framework was associated with 30-day rebleeding, thromboembolism, and net adverse clinical events (NACE) after therapeutic endoscopy for non-variceal upper gastrointestinal bleeding (NVUGIB). Methods This single-center retrospective cohort included 150 unique index admissions from January 2021 through December 2025. Documented strategy adherence was coded in 100 patients and non-adherence in 50. Rebleeding was evaluated only after successful initial hemostasis ( n = 141). Parsimonious Firth penalized logistic regression was used for the primary analyses. Thirty multiple imputations by chained equations with predictive mean matching and propensity-score overlap weighting were used as sensitivity analyses. Resumption timing was analyzed descriptively because temporal ordering did not support causal modeling. Results Thirty-day rebleeding occurred in 13/141 patients (9.2%), thromboembolism in 16/150 (10.7%), death in 6/150 (4.0%), and NACE in 29/150 (19.3%). Rebleeding was observed in 5/96 adherent patients (5.2%) and 8/45 non-adherent patients (17.8%; Fisher p = 0.026); NACE occurred in 12/100 (12.0%) and 17/50 (34.0%), respectively ( p = 0.002). Documented adherence was associated with lower rebleeding (adjusted odds ratio [aOR] 0.03, 95% CI 0.002–0.43) and NACE (aOR 0.04, 95% CI 0.007–0.20). The corresponding adjusted risk differences were −15.7 percentage points (95% CI −23.1 to −9.0) and −27.4 percentage points (95% CI −34.2 to −17.1). Multiple-imputation and overlap-weighted analyses were directionally concordant. Seven of 13 rebleeding events occurred on or before the recorded resumption date. Conclusion Documented adherence to a risk-based management framework was associated with fewer 30-day adverse events. Because exposure classification depended on clinical documentation, residual confounding and misclassification remain likely, and the large adjusted effects should not be interpreted causally. Individualized multidisciplinary assessment appears more important than any single predefined resumption threshold.