Mechanism-informed diagnostic accuracy of blood biomarkers for sepsis-associated encephalopathy: a systematic review and Bayesian diagnostic network meta-analysis.

BACKGROUND: Sepsis-associated encephalopathy (SAE) is a common and devastating manifestation of acute brain dysfunction in sepsis, yet mechanism-informed blood biomarkers with clinically interpretable diagnostic accuracy remain uncertain. A growing range of candidates spanning innate immune activation, blood-brain barrier dysfunction, glial response, and neuronal injury has been reported, but their comparative diagnostic performance and biological hierarchy are unclear, partly due to heterogeneity in phenotyping and sampling timing. MAIN BODY: We performed a PRISMA/PRISMA-DTA-compliant systematic review and comparative diagnostic evidence synthesis, including a Bayesian diagnostic network meta-analysis, to prioritize blood-based biomarkers for SAE. PubMed, Web of Science, EMBASE, Cochrane Library, CNKI, VIP, and WFSD were searched from inception to October 2025. Studies were eligible if SAE case definitions and extractable (or reconstructable) 2 × 2 diagnostic data were available. To characterize heterogeneity, SAE reference standards were tiered (delirium-focused tools vs. broader encephalopathy definitions vs. unclear/mixed), and biomarker sampling was captured using an anchor-based context (anchor event and time-from-anchor). Comparative ranking in the Bayesian diagnostic network meta-analysis was based on the Advantage Index (S-value) derived from posterior distributions. Sixteen studies (n = 1,666) were included, covering biomarkers along an immune-BBB/glia-neuron axis. In pooled diagnostic test accuracy synthesis, blood-brain barrier/glial markers (e.g., GFAP, S100β) generally showed more balanced discrimination than classic neuronal injury markers in the included diagnostic settings. In Bayesian comparative ranking, upstream immune/alarmin candidates (TRAF6, S100A8) and day-3 S100β ranked highest; however, several top-ranked nodes were supported by sparse evidence (single-study nodes), and thresholds/platforms and sampling contexts were study-specific. Importantly, few included studies evaluated whether biomarkers add diagnostic information beyond established clinical predictors (e.g., illness severity scores and sedation- or metabolic-related confounders), limiting immediate clinical applicability. SHORT CONCLUSION: Blood biomarkers for SAE exhibit a biologically coherent immune-to-neuronal cascade, and their diagnostic performance appears sensitive to phenotyping and anchor-based sampling context. These findings are hypothesis-generating and support prospective head-to-head validation to determine whether mechanism-staged biomarker panels add incremental diagnostic value beyond established clinical predictors under standardized phenotyping and anchor-based sampling protocols. SYSTEMATIC REVIEW REGISTRATION: https://www.crd.york.ac.uk/PROSPERO/view/CRD42024531398, CRD42024531398.