Splenic infarction after endoscopic glue injection for gastric varices: proportion, exploratory correlates, and clinical outcomes from a propensity score-matched analysis

Background and aim Splenic infarction after endoscopic cyanoacrylate injection for gastric varices (GV) is poorly characterized. We estimated its proportion among computed tomography (CT)-evaluable patients and explored clinical features and short-term outcomes. Methods We retrospectively screened 255 adult patients undergoing endoscopic glue injection for GV between January 2018 and February 2025. Eligible patients had CT within 30 days and no pre-existing infarction or alternative causes; imaging was clinically selected rather than protocol-mandated. Expanded 1:4 propensity score matching included clinical, laboratory, and index-session endoscopic features. Exploratory Firth regression addressed sparse events and complete separation. Results Of 183 CT-evaluable patients, 16 (8.7%) had radiologically confirmed splenic infarction. After matching, 16 cases and 64 controls were analyzed, although residual imbalance remained. All cases had ≥2 previous injection sessions versus 15.6% of matched controls, producing complete separation; the Firth estimate was therefore not considered a stable effect size. Exploratory models showed statistical signals for repeated injections, larger splenic diameter, large spontaneous portosystemic shunts, and greater index-session glue volume. These features may represent advanced portal hypertension and treatment complexity rather than independent risk factors. Most cases (93.8%) were managed conservatively. Thirty-day mortality and rebleeding did not differ, whereas hospital stay was longer (median, 14 vs. 10 days; p = 0.008). Conclusions Splenic infarction was identified in 8.7% of this selected CT-evaluable cohort. The observed features should be regarded as exploratory clinical markers, not causal or actionable predictors. Imaging selection, residual imbalance, 16 events, complete separation, and wide uncertainty preclude incidence estimation, precise effect-size interpretation, or individual risk prediction.