Combining biomarkers of BNIP3 L, S100B, NSE, and accessible measures to predict sepsis-associated encephalopathy: a prospective observational study

Background: Accurate identification of delirium in sepsis patients is crucial for guiding clinical diagnosis and treatment. However, there are no accurate biomarkers and indicators at present. We aimed to identify which combinations of cognitive impairment-related biomarkers and other easily accessible assessments best predict delirium in sepsis patients. Methods: One hundred and one sepsis patients were enrolled in a prospective study cohort. S100B, NSE, and BNIP3 L biomarkers were detected in plasma and cerebrospinal fluid and patients' optic nerve sheath diameter (ONSD). The optimal biomarkers identified by Logistic regression are combined with other factors such as ONSD to filter out the perfect model to predict delirium in sepsis patients through Logistic regression, Naive Bayes, decision tree, and neural network models. Main results: Among all biomarkers, compared with BNIP3 L (AUC = .706, 95% CI = .597-.815) and NSE (AUC = .711, 95% CI = .609-.813) in cerebrospinal fluid, plasma S100B (AUC = .729, 95% CI = .626-.832) had the best discrimination performance for delirium in sepsis patients. Logistic regression analysis showed that the combination of cerebrospinal fluid BNIP3 L with plasma S100B, ONSD, neutrophils, and age provided the best discrimination to cognitive impairment in sepsis patients (accuracy = .901, specificity = .923, sensitivity = .911), which was better than Naive Bayes, decision tree, and neural network models. Neutrophils, ONSD, and cerebrospinal fluid BNIP3 L were consistently the major contributors in a few models. Conclusions: The logistic regression showed that the combination model was strongly correlated with cognitive dysfunction in sepsis patients.