Abstract
Outcome prediction in traumatic brain injury (TBI) guides treatment decisions. Biomarkers such as S100 calcium-binding protein B (S100B), glial fibrillary acidic protein (GFAP), neuron-specific enolase (NSE), and ubiquitin carboxy-terminal hydrolase L1 (UCH-L1) have shown prognostication value, but relative effectiveness is unknown. This network meta-analysis (NMA) explored the relative predictive value of these 4 biomarkers for mortality and functional outcomes after moderate/severe TBI.
Three databases (PubMed, EMBASE, and Cochrane Library) were searched using terms related to TBI, biomarkers, mortality, and functional outcomes (Glasgow Outcome Scale [GOS], GOS-Extended [GOSE]) until June 2024. Primary outcomes were mortality and functional outcomes (GOS/GOSE scores) at 3-12 months after injury. We included studies that provided biomarker levels, sensitivity/specificity values, and outcome measures in adults with moderate/severe TBI. RoB2 and ROBINS-I assessed the risk of bias. Data extraction followed Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. Random-effects models pooled standardized mean difference (SMD), sensitivity, and specificity. Bayesian NMA estimated odds ratios (ORs) and 95% CIs to evaluate indirect comparisons between the biomarkers.
Of 1,680 studies, we included 32 (n = 2,401). We found statistically significant effects for S100B (SMD = 1.35 [0.99-1.71],
= 83%), GFAP (SMD = 2.25 [1.40-3.11],
= 94%), NSE (SMD = 0.71 [0.12-1.29],
= 88%), and UCH-L1 (mean difference [MD] = 1.06 [0.72-1.40],
= 57%) in predicting mortality. For functional outcomes, S100B had SMD of 0.80 (0.58-1.01,
= 62%), GFAP had SMD of 1.03 (0.65-1.41,
= 89%), NSE had SMD of 0.73 (0.52-0.94,
= 41%), and UCH-L1 had MD of 0.86 (0.72-1.00,
= 0%). Earlier sampling periods (<12 hours) were associated with more consistent and reliable results across all biomarkers. NSE had the highest sensitivity for mortality (88%, 77%-93%), and UCH-L1 had the highest specificity (89%, 76%-95%). S100B had the highest sensitivity for unfavorable functional outcomes (74%, 55%-88%), and GFAP had the highest specificity (84%, 71%-91%). With low comparative heterogeneity, NSE had the highest rank probability for mortality prediction while UCH-L1 predicted unfavorable functional outcomes.
Our findings suggest that S100B, GFAP, NSE, and UCH-L1 have predictive value for mortality and functional outcomes in moderate/severe TBI. NSE was the most effective biomarker for predicting mortality. By contrast, UCH-L1 ranked highest for predicting unfavorable functional outcomes. Further research is needed to standardize protocols for measuring/providing data on biomarkers and to integrate them into predictive models for clinical use.