Inflammatory indices as independent predictors of in-hospital mortality in acute upper gastrointestinal bleeding

Abstract

Background: Acute upper gastrointestinal bleeding (UGIB) remains a high-risk emergency with non-negligible in-hospital mortality. Inflammatory indices derived from routine admission tests may provide additional bedside information reflecting inflammatory burden and physiologic reserve, but their prognostic value in UGIB is not well defined. Methods: In this retrospective single-center observational study, 502 consecutive adults admitted via the emergency department with UGIB (December 2022–2024) were analyzed. Demographic, clinical (including melena/hematemesis), comorbidity, bleeding etiology (variceal vs. non-variceal), endoscopy-related variables when available, transfusion status, and laboratory data were extracted from electronic records. Platelet-to-lymphocyte ratio (PLR), prognostic nutritional index (PNI), and neutrophil-to-albumin ratio (NAR) were calculated at admission. Predictors of in-hospital mortality were assessed using ROC analysis and logistic regression; two multivariable models were reported to address overlap between melena and hematemesis. Results: Overall in-hospital mortality was 9.4% (47/502). Non-survivors had lower admission hemoglobin, albumin, and platelet counts and higher INR. PLR, PNI, and NAR were associated with mortality in univariable analyses. In multivariable Model 1 (including melena), low PLR (OR = 4.734, 95% CI = 2.352–9.529; p < 0.001), low PNI (OR = 3.041, 95% CI = 1.488–6.216; p = 0.002), lower hemoglobin (OR = 0.896 per 1 g/dL increase, 95% CI = 0.809–0.992; p = 0.035), and melena (OR = 2.318, 95% CI = 1.131–4.747; p = 0.022) were independently associated with mortality. In Model 2 (melena replaced by hematemesis), low PLR (OR = 3.890, 95% CI = 1.894–7.989; p < 0.001), hematemesis (OR = 4.922, 95% CI = 2.385–10.157; p < 0.001), and lower hemoglobin (OR = 0.900 per 1 g/dL increase, 95% CI = 0.811–0.998; p = 0.046) remained significant, whereas PNI and NAR were not independently associated. ROC discrimination for PLR/PNI/NAR was modest (AUC 0.612/0.604/0.594, respectively), indicating limited stand-alone predictive performance. Conclusion: Admission PLR showed the most consistent independent association with in-hospital mortality across alternative multivariable models, while the independent contribution of PNI appeared model-dependent. Given the modest AUCs, these indices should be interpreted as adjunctive markers alongside clinical and endoscopic assessment rather than as stand-alone predictors. Prospective multicenter studies with standardized data capture are warranted to evaluate their incremental value within established UGIB risk stratification pathways.