Serum guanylate-binding protein 5 as a novel biomarker for sepsis-associated encephalopathy: development and internal validation of a clinical prediction model in a multicenter retrospective case-control study

Background Sepsis-associated encephalopathy (SAE) affects 8–70% of critically ill sepsis patients and is associated with higher mortality, prolonged ICU stay and long-term cognitive decline. Its early diagnosis is difficult because no specific biomarker is available and bedside assessment is often confounded by sedation. Guanylate-binding protein 5 (GBP5) is an interferon- γ –inducible GTPase that promotes NLRP3 inflammasome assembly and has been implicated in sepsis-associated organ injury, but its clinical value in SAE is unknown. Methods In this multicenter, retrospective case-control study, adults (18–80 years) meeting Sepsis-3.0 criteria were enrolled at three tertiary hospitals between October 2024 and October 2025 and classified as SAE ( n = 145; GCS <15 not attributable to other causes) or non-SAE ( n = 186; GCS=15). Admission serum GBP5 and other laboratory biomarkers, including NSE, S100 β , IL-6, TNF-α, IL-10, CRP, and PCT, were measured. The cohort was randomly split into a training dataset ( n = 232) and internal validation dataset ( n = 99). Univariate comparison, multivariable logistic regression, ROC analysis and a nomogram were performed; performance was assessed by C-index, bootstrap-calibrated curves (1000 iterations) and decision curve analysis (DCA). Results Serum GBP5 was higher in SAE than in non-SAE patients (median 0.63 vs. 0.58 ng/mL, P < 0.001). Five variables were independently associated with SAE: GBP5 (OR 7.445, 95% CI 3.651–15.183), IL-6 (OR 4.472, 95% CI 2.192–9.121), APACHE II (OR 4.032, 95% CI 1.977–8.223), diabetes mellitus (OR 3.480, 95% CI 1.615–7.497) and SOFA (OR 2.364, 95% CI 1.134–4.927). GBP5 showed the highest single-marker AUC (0.813, 95% CI 0.757–0.861; cut-off >0.585 ng/mL, sensitivity 75.70%, specificity 86.40%), whereas NSE and S100 β were not retained as independent predictors. The nomogram achieved an AUC of 0.872 (95% CI 0.825–0.918) in the training and 0.864 (95% CI 0.791–0.937) in the internal validation dataset, with adequate calibration and net clinical benefit on DCA. Conclusion Admission serum GBP5 was independently associated with SAE and showed the best single-marker discrimination. A nomogram incorporating GBP5, IL-6, SOFA, APACHE II and diabetes mellitus performed well in both datasets. GBP5 may aid SAE risk stratification pending external validation.