Serial measurements of neuron specific enolase, glial fibrillary acidic protein and neurofilament light for mortality prediction in patients with severe acute brain injury in the ICU

Introduction

Prognostication in patients with severe acute brain injury (SABI) admitted to the Intensive Care Unit (ICU) can be challenging. In some cases, decisions regarding the continuation or withdrawal of life sustaining therapy are made at an early stage of hospitalization [1]. There are several pitfalls in this process [2]. An optimistic prediction of outcome may lead to prolonged hospitalization and potentially survival with severe disability, possibly against the patient’s wishes. A pessimistic prediction may lead to withdrawal of life-sustaining therapy, possibly denying the patient a chance for good recovery [3], 4]. A fast and reliable tool to predict long-term neurological outcome after sustaining SABI is needed. Such a tool likely consists of a combination of several diagnostic tests, leading to a multimodal approach, similar to the guidelines used for prognostication in patients who remain in coma after cardiac arrest [5]. Unfortunately, such guidelines have not yet been fully developed for ICU patients with SABI due to other causes, e.g. traumatic brain injury (TBI), subarachnoid hemorrhage (SAH) or stroke. Prognostication in patients after cardiac arrest primarily focuses on reliable prediction of a poor outcome. For other types of SABI in the ICU, it should be considered to shift the focus on identifying patients with a chance of a good outcome as this could help prevent erroneous withdrawal of life-sustaining therapy.

Biomarkers released into the blood upon cellular injury of the brain could be a valuable addition to a multimodal diagnostic approach as they have shown promising results in assessing brain injury [6]. Several serum biomarkers have been investigated. The BIO-AX-TBI study reported neurofilament light (NFL) concentrations approximately 10 times higher in patients with moderate-severe TBI compared to trauma patients without TBI in the first few days. Furthermore, NFL concentrations were associated with functional outcome at 12 months [7]. Richter et al. studied the incremental value of biomarkers determined at admission in TBI patients when added to the established prognostic models IMPACT (International Mission for Prognosis and Analysis of Clinical Trials in TBI) and CRASH (Corticosteroid Randomization after Significant Head Injury) [8]. They found that several biomarkers added predictive value to the models: glial fibrillary acidic protein (GFAP), NFL, S100β and ubiquitin c-terminal hydrolase L1 (UCH-L1). The timing of biomarker assessment plays a critical role in their utility, as each biomarker follows a distinct temporal trajectory. Neuron specific enolase (NSE) typically peaks between 48 and 72 h and is currently recommended by international guidelines to identify patients with a poor outcome after cardiac arrest [9]. In TBI, the highest serum concentrations of GFAP can be detected during the first days, whereas NFL concentrations typically demonstrate a slow increase over time, peaking at 1–2 weeks after TBI [10], 11].

Current biomarker research focuses mostly on specific causes of brain injury, such as TBI, cardiac arrest, or sepsis-associated encephalopathy, but these studies often also include patients with mild or moderate brain injury. Patients with SABI admitted to the ICU represent a distinct subgroup where more accurate prognostic tools could be especially valuable to guide clinical decision-making. Although the underlying etiologies may vary, the severity of brain injury unites these patients. Biomarkers could potentially improve prognostic models in this critically ill population [12], [13], [14].

In this study we aimed to investigate the association between serum biomarker concentrations over time and hospital mortality in patients admitted to the ICU with SABI of various etiologies.

Materials and methods

Results

A total of 103 patients with SABI were included between January 3, 2023 and March 10, 2025 (Supplementary Figure S1). Patient characteristics are presented in Table 1. TBI was diagnosed in 48 patients (46.6 %) and SAH in 29 patients (28.2 %). Patients had a severely impaired consciousness at hospital admission, the median GCS score was 5 (IQR 3–7). In 37 patients (35.9 %) pupillary reactivity was absent in either one or both pupils.

At hospital discharge 52 patients (50.5 %) were alive, of which 8 were discharged to their homes, 43 to rehabilitation facilities and one to a nursing home. Of the 51 deceased patients, 34 patients died following withdrawal of life-sustaining therapy.

Blood sample availability and serum biomarker concentrations are depicted in Table 2. Collection of samples at all four timepoints for each patient was impossible as patients died, were transferred to other hospitals or discharged before the moment of sampling (n=106 timepoints). Additionally, some timepoints were missing due to logistical issues or inability to draw blood (n=33). Other samples were excluded because they were hemolytic (n=9) which hindered the measurement of NSE, or exceeded the upper limit of detection for GFAP and NFL (n=2). The number of available samples for survivors and non-survivors at each time point are presented in Table 2.

The highest median serum concentration of GFAP was found on day 1, 29,516 pg/mL (IQR 9,974–53,533 pg/mL). Serum NSE and NFL peaked on day 14, respectively 18 ng/mL (IQR 12–26 ng/mL) and 1,384 pg/mL (IQR 654–1,974 pg/mL). Serum concentrations of each biomarker at each timepoint between hospital survivors and non-survivors are visualized in boxplots (Figure 1). A statistically significant difference between survivors and non-survivors was found for all biomarkers on day 3, NSE (p=0.023), GFAP (p=0.041) and NFL (p=0.018).

Figure 1:

Serum concentrations of NSE, GFAP and NFL at each timepoint between survivors and non-survivors at hospital discharge. The y-axis are log-transformed (log10). Boxplots show the median and interquartile range. *Signifies a statistically significant difference between groups. GFAP=glial fibrillary acidic protein, NFL=neurofilament light, NSE=neuron specific enolase.

The serum biomarker trajectories over time stratified by survival are displayed in a lineplot (Figure 2), lines represent a second-degree polynomial regression with 95 % confidence intervals. Table 3 summarizes the findings of the mixed effects model assessing the association between biomarker concentrations over time and hospital mortality. Across all three biomarkers, time was a significant predictor of concentration, with each biomarker exhibiting distinct temporal patterns, NSE (β²=2.142, p≤0.0001), GFAP (β²=6.500, p≤0.0001) and NFL (β²=4.003, p≤0.0001), However, none of the biomarkers showed a statistically significant interaction between time and hospital mortality, indicating that the temporal trajectory of biomarker concentrations did not differ between survivors and non-survivors. Additionally, biomarker concentrations over time between the three most common admission diagnosis were analyzed (Supplementary Table S2 and Figure S3). The trajectory of NSE levels was significantly different for patients with SAH (β²=−2.744, p=0.0035) and intracranial hemorrhage (β²=1.700, p=0.012) compared to the reference group of patients with TBI. For GFAP levels only the trajectory of patients with intracranial hemorrhage differed from patients with TBI (β²=−8.448, p=0.006). Lastly, the serum biomarker trajectories over time stratified by survival were analyzed for the two main admission diagnoses, TBI and SAH. In TBI patients, temporal trajectories between survivors and non-survivors were not significantly different (Supplementary Table S4 and Figure S5). In SAH patients, a significant interaction between time and hospital mortality was observed for NFL (β²=−1.738, p=0.038), indicating temporal trajectories of this biomarker differ between survivors and non-survivors (Supplementary Table S6 and Figure S7). However, this finding should be interpreted with caution due to the small sample size in this subgroup (n=29).

Figure 2:

Serum biomarkers concentrations over time stratified by hospital survival. Lines represent a second-degree polynomial regression with 95 % confidence intervals. The y-axis are log-transformed. GFAP=glial fibrillary acidic protein, NFL=neurofilament light, NSE=neuron specific enolase.

The prognostic utility of individual biomarkers, all biomarkers combined and all biomarkers combined with factors known to be associated with hospital mortality (age, pupillary reactivity and GCS at admission) were used to predict hospital mortality on day 1 and 3 using ROC curves (Figure 3). On day 1, the AUC of the individual biomarkers ranged from 0.463 to 0.526, indicating poor performance [20]. The model including all biomarkers and other known predictors of mortality performed better (AUC 0.674). On day 3, prognostic accuracy increased for all individual biomarkers (AUC 0.619 to 0.652) and when all biomarkers and the other predictors were combined the highest predictive value (AUC 0.749) was reached indicating moderate performance.

Figure 3:

The prognostic utility (area under the receiver operating characteristics curve) of serum NSE, GFAP and NFL, all biomarkers combined and all biomarkers combined with other known predictors of hospital mortality (age, pupillary reactivity at admission and GCS total score at admission) in distinguishing survivors and non-survivors of severe acute brain injury at day 1 and 3. AUC=area under the curve, GCS score=Glasgow Coma Scale Score, GFAP=glial fibrillary acidic protein, NFL=neurofilament light, NSE=neuron specific enolase.

Discussion

In this prospective multicenter cohort study of critically ill patients with SABI, the highest serum biomarker concentrations were found on day 1 for GFAP, and on day 14 for NSE and NFL. A substantial decrease over time was found for serum GFAP concentrations and a substantial increase over time for serum NFL. The results further showed that all biomarker levels are significantly elevated in hospital non-survivors compared to survivors at day three, but the longitudinal slopes did not differ between groups. The prognostic utility of the individual biomarkers per timepoint (day 1 and 3) was poor, but when the biomarkers were combined with other known predictors of mortality a moderate performance was found on day 3.

Although our study population represents a rather heterogenous group of patients with SABI, the general trend of higher serum biomarker concentrations in patients with poor outcome aligns with findings from studies in patients with stroke, intracerebral hemorrhage, SAH or TBI [21], [22], [23], [24]. In patients with stroke, early assessment of both NFL and GFAP in the first 3 days after admission appear to be valuable biomarkers for prediction of poor outcome, especially when combined with the National Institutes of Health Stroke Scale (NIHSS) [25], 26]. A recent study in SAH and intracerebral hemorrhage patients by Gyldenholm et al. found elevated serum GFAP concentrations at admission were predictive for 30-day mortality and poor outcome defined as a mRS score ≥4 [27]. In the current study, we included patients with severely diminished consciousness and no recovery of consciousness within the first 24 h of admission. Due to the severity of brain injury at inclusion we found relatively high serum concentrations of all biomarkers in our study population. This affects our study results and hampers comparison of our results to many other studies on mild and moderate brain injury patients, but increases the relevance for clinical practice in the ICU [6], 28].

In patients with anoxic brain injury after cardiac arrest, high serum NSE concentrations are considered a robust predictor of poor outcome [29], 30]. Interestingly, we found serum NSE concentrations to be highest on day 14 in the overall population and day 1 in the non-survivors, whereas in patients with anoxic brain injury after cardiac arrest this biomarker typically peaks between 48 and 72 h [5]. This difference in timing likely reflects the different pathophysiological mechanisms involved in our study population, which included patients with SABI from various causes. Unlike post-anoxic brain injury, where the entire brain in affected, conditions such as cerebral hemorrhage involve more localized brain damage. This may influence the extent and the timing of NSE release into the blood [10]. Similarly, studies have shown that serum NFL concentrations measured 24 h post-cardiac arrest are strongly associated with poor outcome, while in other study populations of patients with SABI, NFL tends to perform better as a prognostication tool at later timepoints e.g. after 7 days [11], [31], [32], [33]. Thus, the dynamic patterns of both NSE and NFL release differ depending on the underlying cause of brain injury, emphasizing the need to consider disease-specific factors when interpreting biomarker trends. This is also demonstrated by the subgroup analysis conducted in the current study with SAH patients, where different temporal trajectories between survivors and non-survivors could be observed for NFL concentrations. However, it should be noted that this finding is based on an extremely small sample size and should be validated in a larger cohort.

Another interesting development in the field of blood-based biomarkers is the use of inflammatory biomarkers for outcome prediction. Although the biomarkers used in the current study are markers of structural brain injury, cell damage is typically followed by an inflammatory response with infiltration of immune cells and release of cytokines into the affected area. In recent literature, the potential of interleukin 6 (IL-6), IL-8, IL-15 and tumor necrosis factor α has been highlighted for monitoring disease progression and neuroprognostication in patients with severe TBI [34], [35], [36], [37], [38]. Yue et al. demonstrated that IL-15 concentrations could discriminate between patients with a favorable (Glasgow Outcome Scale Extended (GOSE) score ≥5) and an unfavorable outcome (GOSE ≤4) at 3 months follow-up with an AUC of 0.74 [38]. Moreover, Samanta et al. was able to identify two clusters of inflammatory phenotypes [39]. Patients assigned to the early-inflammation phenotype could be identified with IL-6, IL-15 and monocyte chemoattractant protein 1 and were more likely to have an unfavorable GOSE score at 6 months. These findings suggest future research should aim to develop and validate a biomarker panel of both structural and inflammatory markers for neuroprognostication.

In our cohort, higher concentrations of NSE and NFL were found for survivors compared to non-survivors on day 7 and 14, with a similar trend observed for GFAP on day 14. Although these differences were not statistically significant, it is an interesting finding. A possible explanation could be that the patients who ultimately did not survive, but were still alive at these timepoints, may have died due to complications following brain injury, rather than the severity of brain injury itself. In the overall cohort of non-survivors, one third died due to clinical deterioration which often encompassed a severe infection (e.g. pneumonia or meningitis). Among the non-survivors who were still alive at day 7, half eventually died from clinical deterioration and this proportion increased to 60 % for the non-survivors who were still alive at day 14. Additionally, a larger proportion of non-survivors at day 7 and 14 were scored as ‘Frail’ at admission compared to the overall cohort of non-survivors. These findings may partly explain the observed biomarker patterns.

While serum biomarkers may offer valuable insights into the pathophysiology of SABI and could potentially improve neuroprognostication, their utility as standalone predictors remains limited. A multimodal approach including modalities such as magnetic resonance imaging (MRI) and electroencephalography (EEG) with reactivity testing should be considered. Amiri et al. [40] used MRI and EEG in a combined prediction model to predict consciousness levels at study enrolment and ICU discharge, they found maximum AUCs of 0.78 (95 % CI 0.71–0.86) and 0.83 (95 % CI 0.75–0.89) respectively. Egbebike et al. [41] studied cognitive-motor dissociation (CMD), detected by EEG in clinically unresponsive patients with acute brain injury. The presence of CMD was an independent predictor of a shorter time to good recovery and that patients with CMD discharged home or to a rehabilitation setting consistently had better neurological outcome scores measures with the GOSE score compared to patient without CMD. Given these results, we would suggest the development of a comprehensive neuroprognostication model combining clinical parameters, serum biomarkers (both structural and inflammatory), MRI, EEG and disease specific scores (e.g. the IMPACT and CRASH for TBI patients and the SAFIRE for SAH patients), which may improve outcome prediction in SABI patients in the ICU [42], 43]. Identification of patients with a chance of recovery should be the main focus of such a prognostication model, as this could prevent unjust WLST [4].

Conclusions

In this prospective cohort of ICU patients with SABI serum concentrations of NSE, GFAP and NFL showed distinct temporal patterns, but the biomarker concentrations over time did not differ between survivors and non-survivors. Biomarker concentrations were higher in non-survivors during the early phase of admission, with the strongest discriminative performance observed on day 3. While individual biomarkers showed limited prognostic accuracy, in combination with known predictors of mortality prognostic accuracy improved to moderate performance. Future research should focus on implementing serum biomarkers in multimodal prognostication approaches to further improve care for neurocritical care patients.