Diagnostic Utility of Glial Fibrillary Acidic Protein Beyond 12 Hours After Traumatic Brain Injury: A TRACK-TBI Study

Abstract

Blood levels of glial fibrillary acidic protein (GFAP) and ubiquitin carboxyl-terminal hydrolase-L1 (UCH-L1) within 12h of suspected traumatic brain injury (TBI) have been approved by the Food and Drug administration to aid in determining the need for a brain computed tomography (CT) scan. The current study aimed to determine whether this context of use can be expanded beyond 12h post-TBI in patients presenting with Glasgow Coma Scale (GCS) 13–15. The prospective, 18-center Transforming Research and Clinical Knowledge in Traumatic Brain Injury (TRACK-TBI) study enrolled TBI participants aged ≥17 years who presented to a United States Level 1 trauma center and received a clinically indicated brain CT scan within 24h post-injury, a blood draw within 24h and at 14 days for biomarker analysis. Data from participants with emergency department arrival GCS 13–15 and biomarker values at days 1 and 14 were extracted for the primary analysis. A subgroup of hospitalized participants with serial biomarkers at days 1, 3, 5, and 14 were analyzed, including plasma GFAP and UCH-L1, and serum neuron-specific enolase (NSE) and S100 calcium-binding protein B (S100B). The primary analysis compared biomarker values dichotomized by head CT results (CT+/CT-). Area under receiver-operating characteristic curve (AUC) was used to determine diagnostic accuracy. The overall cohort included 1142 participants with initial GCS 13–15, with mean age 39.8 years, 65% male, and 73% Caucasian. The GFAP provided good discrimination in the overall cohort at days 1 (AUC = 0.82) and 14 (AUC = 0.72), and in the hospitalized subgroup at days 1 (AUC = 0.84), 3 (AUC = 0.88), 5 (AUC = 0.82), and 14 (AUC = 0.74). The UCH-L1, NSE, and S100B did not perform well (AUC = 0.51-0.57 across time points). This study demonstrates the utility of GFAP to aid in decision-making for diagnostic brain CT imaging beyond the 12h time frame in patients with TBI who have a GCS 13–15.

Introduction

Blood-based biomarkers provide quantifiable measurements of traumatic brain injury (TBI). The diagnostic utility of circulating central nervous system (CNS)-specific biomarkers for TBI has expanded significantly over the past decade.^1-13 In 2018, the U.S. Food and Drug Administration (FDA) approved the use of benchtop assays for serum glial fibrillary acidic protein (GFAP) and ubiquitin c-terminal hydrolase L1 (UCH-L1) to rule out the need for a head computed tomography (CT) scan within 12h of injury in patients with mild TBI (mTBI), defined as having emergency department (ED) Glasgow Coma Scale (GCS) score of 13–15.⁸

In 2021, FDA indications were expanded to include the clearance of the Abbott i-STAT Alinity handheld device, which can provide results from plasma within 18 min. The rapid ability of blood test results may encourage more widespread adoption of this device and become a routine test in the evaluation of traumatic injuries.

Previous investigations have established the diagnostic utility of GFAP and UCH-L1 with presence or absence of intracranial pathology on head CT within 24h of injury.^{2–4; 6–13} Patients with head trauma, however, often present with symptomatology days after injury. In the military, delayed presentation may be further protracted by deployment and logistical challenges.

The diagnostic utility of blood-based biomarkers for TBI must be interrogated and validated for patients presenting for assessment beyond the current FDA indication within 12h of injury, particularly in patients with TBI severity GCS 13–15 who are more likely to be discharged home after injury.

Using a prospective, 18-center cohort of acute TBI participants with an initial GCS score of 13–15, the current study analyzed the diagnostic utilities of GFAP and UCH-L1, and the historical blood-based biomarkers neuron-specific enolase (NSE) and S100 calcium-binding protein B (S100B) for TBI diagnosis at days 1, 3, 5, and 14 post-injury.

Our findings provide evidence to extend the indications of use for blood-based biomarkers to 14 days post-injury and to evaluate the diagnostic accuracy of blood biomarkers in the detection of acute intracranial abnormalities in patients with mTBI based on CT imaging across these longitudinal trajectories.

Methods

Patient selection

The prospective, observational Transforming Research and Clinical Knowledge in Traumatic Brain Injury (TRACK-TBI) study (NCT02119182) recruited participants from the emergency departments (EDs) of 18 U.S. Level 1 trauma centers from February 2014 through June 2018.¹⁴ Inclusion criteria were presentation to a participating hospital after external force head trauma with, at minimum, alteration of consciousness per the American Congress of Rehabilitation Medicine (ACRM) criteria for TBI, completion of non-contrast head CT for clinical evaluation within 24h post-injury, and fluency in English or Spanish.¹⁵

Exclusion criteria included significant polytrauma associated with risk for early death that could interfere with follow-up or outcome assessments as determined by the site principal investigator, penetrating TBI, major debilitating psychiatric (schizophrenia or bipolar) or neurological disorders (stroke, dementia, or tumor) that would interfere with outcome assessment, current enrollment in an interventional clinical trial, and specific vulnerable populations (pregnant; incarcerated; involuntary psychiatric hold). The complete framework for inclusion and exclusion criteria standardized across TRACK-TBI sites is provided on Page 27 of the TRACK-TBI Clinical Protocol.¹⁶

Written informed consent was obtained either from participants or from their legally authorized representatives (LAR). The study was approved by the enrolling site Institutional Review Boards. Human subjects research conducted as part of this study followed the principles outlined in the Declaration of Helsinki.

The current objective was to assess the diagnostic accuracy of biomarkers on days 1 and 14. We included all participants aged ≥17 years presenting with a GCS score of 13–15, and having days 1 and 14 samples. A subgroup analysis was performed in hospitalized participants with samples on days 1, 14, and at least one time point on day 3 or 5.

Procedures

Demographic and clinical data were obtained by trained research assistants through participant interview and medical record review. Computed tomography (CT) imaging scans were transmitted to a central imaging repository (Laboratory of Neuro Imaging, Los Angeles, CA) and coded by one board-certified neuroradiologist, blinded to injury and biomarker information, in accordance with the National Institute of Neurological Disorders and Stroke (NINDS) TBI Common Data Elements for Neuroimaging.¹⁷

Blood samples were obtained within 24h of injury through peripheral venipuncture, processed, and stored as 500 μL aliquots at -80°C within 2h of collection according to TBI Common Data Elements for Biospecimens.¹⁸ Hospitalized participants underwent additional blood draws for proteomic processing on days 3 and 5, if available. Blood samples on day 14 were obtained from hospitalized inpatients, or outpatient for discharged participants. Coded biospecimen samples were batch shipped overnight on dry ice to a central repository (University of Pittsburgh Medical Center, Pittsburgh, PA) for permanent storage.

The guidelines for the processing, storage, and timing of sample collection are standardized across all 18 U.S. Level I trauma centers. A staffed research team and standardized equipment include blood specimen collection kits, containers for plasma/serum/buffy coat aliquots, centrifuge capable of ≥1500 rcf (1500 × g), and -80°C freezer.

Whole blood collected for each patient is collected into three different vacutainer tubes. It is then processed locally into plasma, serum, and buffy coat fractions, aliquotted, frozen at the study site and then shipped to the TRACK-TBI Biospecimen Repository (BR). Additional techniques to the collection, handling, and processing are provided in the TRACK-TBI BR Manual of Procedures.¹⁹

Biomarker analysis

In the current study, plasma aliquots were shipped from the central repository to Abbott Laboratories for analysis, and underwent one freeze-thaw cycle. The primary analysis included a total of 2284 blood samples (from 1142 TBI participants, matched days 1 and 14). In addition, each participant in the hospitalized cohort had at least three blood samples (days 1, 3, and/or 5, and 14).

Abbott Laboratories developed a point-of-care handheld device (i-STAT) as well as a core laboratory platform (ARCHITECT) for measuring biomarker levels and as such, GFAP and UCH-L1 were measured in plasma samples in two batches. Measurements in the first batch were performed with the prototype point-of-care i-STAT™ Alinity™ System (n = 1604) and measurements in the second batch were performed using the prototype core lab Abbott ARCHITECT^® platform (n = 680) for faster throughput.

The first set of plasma biomarker samples were analyzed for GFAP and UCH-L1 concentrations (n = 1604) using the prototype point-of-care i-STAT™ Alinity™ System, which is a sandwich enzyme-linked immunosorbent assay (ELISA) method with electrochemical detection of the resulting enzyme signal. Testing time for each assay was approximately 18 min.

The GFAP assay's calibration range was 0 to 50,000 pg/mL. The limit of detection (LoD) and limit of quantitation (LoQ) were <15 pg/mL and <25 pg/mL, respectively, resulting in a reportable range of 15 to 50,000 pg/mL. Within-laboratory precision, measured by the coefficient of variation (CV) was 2.8–14.2%. The UCH-L1 assay's calibration range was 0–20,000 pg/mL. The LoD and LoQ were <10 pg/mL and <20 pg/mL, respectively, resulting in a reportable range of 10 to 20,000 pg/mL. Within-laboratory precision was 5.0–10.0% CV.

The second set of plasma biomarker samples were analyzed for GFAP and UCH-L1 concentrations (n = 680) using the prototype core lab Abbott ARCHITECT platform for faster batch throughput, which is a two-step sandwich assay using chemiluminescent microparticle immunoassay (CMIA) technology. The prototype GFAP assay calibration range was from 0–50,000 pg/mL. The LoD and LoQ were 2 pg/mL and 5 pg/mL, respectively, for a reportable range of 2–50,000 pg/mL. The within-laboratory CV was 2.0–5.6%. Samples with values greater than 50,000 pg/mL were retested with a 10-fold automated dilution protocol.

The prototype UCH-L1 assay calibration range was 0-25,000 pg/mL. The LoD and LoQ were 10 pg/mL and 20 pg/mL, respectively, for a reportable range of 10–25,000 pg/mL. The assay had a CV of 2.0–5.7%. ARCHITECT values were converted to i-STAT equivalents using two previously derived equations: i-STAT = -12.36 + 1.02*ARCHITECT for GFAP (Spearman correlation coefficient = 0.985) and i-STAT = -3.29 + 0.72*ARCHITECT for UCH-L1 (Spearman correlation coefficient = 0.933).^6,20

Notably, the i-STAT device enabled assay processing and evaluation at the bedside, in contrast with benchtop assay platforms as methods for assay processing. As stated previously, the FDA provided clearance for plasma GFAP and UCH-L1 measurements on the Abbott i-STAT device on March 7, 2023, and it was selected as the platform for clinical analyses in our study because of its precision and adoptability in real-time emergency clinical settings.²¹

Serum aliquots were analyzed for S100B and NSE and were measured via electrochemiluminescence immunoassay (ECLIA), using the Roche Elecsys^® system. The reportable range for the S100B assay was 0.005-39–μg/L with an LoD of <0.005 μg/L and a CV of 20%. The reportable range of the NSE assay was 0.05–370 ng/mL with an LoD of <0.05 ng/mL and a CV of <20%.

Assays were performed neat, without dilution, in duplicates and blinded to participant's clinical and neuroimaging data. Readings greater than the calibration range were reported as greater than the reportable range and were not diluted.

Statistical analysis

Biomarker levels were summarized and compared by diagnostic groups using the Wilcoxon Rank Sum test. Receiver-operating characteristic (ROC) analysis was conducted to assess the performance of each biomarker in discriminating CT-positive (CT+) versus CT-negative (CT-) participants. Area under the ROC curve (AUC) was calculated with 95% confidence intervals.²² Statistical analyses were performed using R (version 4.1.2, R Core Team, 2013).

Results

The main analysis included 1142 participants with initial GCS 13–15 and biomarker samples at day 1 and day 14 (Fig. 1). The majority of participants were male (67%), white (77%), with mean age 39.8 years (standard deviation [SD] 17.1). Thirty-six percent were CT+. The median Injury Severity Score (ISS) on presentation was 10. Median ISS in CT- patients was 8 and in CT+ patients was 14. Detailed demographic/clinical data are presented in Table 1.

FIG. 1.

TRACK-TBI CONSORT flow diagram of included participants. TRACK-TBI, Transforming Research and Clinical Knowledge in Traumatic Brain Injury; CONSORT, CONsolidated Standards Of Reporting Trials; CT, computed tomography; ED, emergency department.

Table 1.

Demographics and Clinical Characteristics of Study Population

Participant characteristics	Overall cohort (n = 1142)	CT- cohort (n = 734)	CT+ cohort (n = 4 08)	Sig. (p)
Mean age in years (SD, range)	39.8 (17.1, 17-90)	36.5 (15.2, 17-88)	45.7 (18.8, 18-90)	< 0.0001
Mean years of education (SD, range)	13.6 (2.9, 0-20)	13.5 (2.8, 1-20)	13.8 (3.2, 0-20)	0.0391
Sex
Male	768 (67.25%)	474 (64.58%)	294 (72.06%)	0.0103
Female	374 (32.75%)	260 (35.42%)	114 (27.94%)
Race
White	880 (77.19%)	537 (73.36%)	343 (84.07%)	0.0004
Black	184 (16.14%)	146 (19.95%)	38 (9.31%)
Other	76 (6.67%)	49 (6.69%)	27 (6.62%)
Ethnicity
Non-Hispanic	924 (81.12%)	602 (82.24%)	322 (79.12%)	0.2066
Hispanic	215 (18.88%)	130 (17.76%)	85 (20.88%)
Mechanism of Injury
Road traffic incident	667 (58.56%)	488 (66.58%)	179 (44.09%)	0.0004
Incidental fall	293 (25.72%)	141 (19.24%)	152 (37.44%)
Violence/assault	70 (6.15%)	35 (4.77%)	35 (8.62%)
Other	109 (9.57%)	69 (9.41%)	40 (9.85%)
Psychiatric History
No	887 (77.67%)	556 (75.75%)	331 (81.13%)	0.038
Yes	255 (22.33%)	178 (24.25%)	77 (18.87%)
Initial GCS Score
15	889 (77.85%)	606 (82.56%)	283 (69.36%)	0.0004
14	202 (17.69%)	109 (14.85%)	93 (22.79%)
13	51 (4.47%)	19 (2.59%)	32 (7.84%)
Admission Status				< 0.0001
ED Discharge	337 (29.51%)	304 (41.42%)	33 (8.09%)
Admit to floor	496 (43.43%)	341 (46.46%)	155 (37.99%)
Admit to ICU	309 (27.06%)	89 (12.13%)	220 (53.92%)
Median Injury Severity Score (quartile 1, quartile 3)	10 (5, 17)	8 (4, 14)	14 (9, 18)	< 0.0001

CT, computed tomography; SD, standard deviation; GCS, Glasgow Coma Scale; ED, emergency department; ICU; intensive care unit.

Demographic, clinical and injury characteristics by initial CT imaging.

The average interval from the initial time of injury to the first blood draw for biomarker analysis was 14.5h (SD 6.8h). For day 1 biomarker analysis, comparisons were made based on the collection time by 4h intervals (0–4, 58, 9–12, 13–16, 17–20, 20–24). Between CT- and CT+ cohorts in the first 24h, GFAP values showed the greatest difference at 17–20h, while UCH-L1, S100B, and NSE showed the greatest difference between 0–4h (Table 2).

Table 2.

Biomarker Values by Collection Interval in the First 24 Hours

	GFAP in pg/mL	UCH-L1 in pg/mL	S100B in μg/L	NSE in ng/mL
0–4 h
CT-	n = 95 60.4 (13.4, 220.3)	n = 95 229.7 (127.5, 458.0)	n = 91 0.12 (0.09, 0.27)	n = 91 17.2 (13.7, 26.5)
CT+	n = 3 633.6 (344.6, 2085.8)	n = 3 835.2 (801.5, 997.0)	n = 3 0.48 (0.38, 0.52)	n = 3 29.7 (25.5, 34.7)
5–8 h
CT-	n = 135 97.0 (18.4, 430.0)	n = 135 212.4 (112.7, 346.0)	n = 127 0.12 (0.8, 0.19)	n = 127 16.9 (13.1, 25.8)
CT+	n = 31 423.3 (233.5, 1119.4)	n = 31 299.0 (217.5, 570.5)	n = 30 0.21 (0.13, 0.35)	n = 30 21.6 (14.5, 28.7)
9-–2 h
CT-	n = 96 182.3 (72.1, 591.8)	n = 96 204.1 (107.8, 360.8)	n = 89 0.14 (0.09, 0.22)	n = 89 22.9 (15.9, 36.9)
CT+	n = 49 744.3 (351.8, 2065.9)	n = 49 206.9 (118.4, 339.8)	n = 45 0.14 (0.09, 0.22)	n = 45 19.9 (12.7, 30.5)
13–16 h
CT-	n = 107 155.7 (26.9, 400.1)	n = 107 118.5 (71.6, 207.3)	n = 101 0.09 (0.06, 0.14)	n = 101 17.7 (14.0, 25.2)
CT+	n = 94 934.9 (276.2, 2060.9)	n = 94 214.1 (119.4, 366.4)	n = 92 0.14 (0.08, 0.19)	n = 92 23.9 (16.2, 41.2)
17–20 h
CT-	n = 132 143.3 (44.2, 419.3)	n = 132 122.4 (63.0, 219.4)	n = 129 0.09 (0.06, 0.13)	n = 129 17.4 (12.6, 24.9)
CT+	n = 93 1022.6 (422.0, 2279.1)	n = 93 177.0 (116.5, 332.0)	n = 92 0.10 (0.07, 0.16)	n = 92 18.5 (13.4, 31.9)
20–24 h
CT-	n = 157 143.0 (31.8, 400.7)	n = 157 121.4 (65.3, 221.6)	n = 151 0.08 (0.06, 0.13)	n = 151 16.3 (12.9, 23.0)
CT+	n = 133 815.0 (316.4, 2018.6)	n = 133 162.4 (98.4, 287.7)	n = 132 0.10 (0.07, 0.17)	n = 132 18.2 (12.9, 30.2)

GFAP, glial fibrillary acidic protein; UCH-L1, ubiquitin c-terminal hydrolase L1; S100B, S100 calcium-binding protein B; NSE, neuron-specific enolase; CT, computed tomography.

Biomarker values for Day 1 following injury and the collection time. Comparisons between CT- and CT+ cohorts were made. Values for the median (quartile 1, quartile 3) are reported for each biomarke,r respectively.

On day 14, 42 (3.7%) patients remained inpatient, and 32 of these patients were CT+. Blood samples in these 42 patients were collected while inpatient. The GFAP, UCH-L1, and NSE were higher in CT+ participants on days 1 and 14, while S100B did not differ by CT result on day 14 (Table 3; Fig. 2). The GFAP provided good discrimination of participants by CT on day 1 (AUC 0.82, Fig, 3A) and day 14 (AUC 0.72, Fig. 3B). The UCH-L1, S100B, and NSE did not provide good discrimination by CT on days 1 and 14 (AUC 0.47-0.59).

Table 3.

Day 1 and Day 14 Biomarker Values in Overall Cohort, by Head CT Result

	Overall cohort	CT-	CT+	Sig. (p)
GFAP in pg/mL
Day 1	n = 1142 277.2 (63.4, 794.8)	n = 734 132.6 (27.4, 398.4)	n = 408 836.1 (316.9, 2014.8)	< 0.0001
Day 14	n = 1142 16.7 (9.8, 30.0)	n = 734 13.7 (8.3, 22.0)	n = 408 25.7 (15.2, 45.3)	< 0.0001
UCH-L1 in pg/mL
Day 1	n = 1142 171.5 (92.3, 305.1)	n = 734 157.6 (82.9, 283.6)	n = 408 195.0 (116.1, 347.5)	< 0.0001
Day 14	n = 1142 68.9 (45.4, 106.3)	n = 734 66.9 (43.3, 100.2)	n = 408 74.8 (50.1, 117.0)	0.001
S100B in μg/L
Day 1	n = 1098 0.11 (0.07, 0.18)	n = 699 0.10 (0.06, 0.17)	n = 399 0.12 (0.07, 0.19)	< 0.0001
Day 14	n = 1098 0.05 (0.03, 0.07)	n = 699 0.05 (0.04, 0.07)	n = 399 0.05 (0.03, 0.07)	0.2388
NSE in ng/mL
Day 1	n = 1098 18.2 (13.5, 28.7)	n = 699 17.5 (13.4, 26.3)	n = 399 20.2 (13.5, 34.2)	0.001
Day 14	n = 716 13.6 (11.1, 17.3)	n = 417 13.1 (10.6, 16.5)	n = 299 14.3 (11.7, 18.1)	< 0.0001

CT, computed tomography; GFAP, glial fibrillary acidic protein; UCH-L1, ubiquitin c-terminal hydrolase L1; S100B, S100 calcium-binding protein B; NSE, neuron-specific enolase.

Biomarker values for Day 1 and Day 14 following injury. Comparisons between CT- and CT+ cohorts were made using the Wilcoxon Rank Sum Test at each timepoint. Values for the median (quartile 1, quartile 3) are reported for each biomarker respectively.

FIG. 2.

Blood biomarker concentrations (in log10 scale) are shown for GFAP, UCH-L1, S100B, and NSE, stratified by initial head CT result (negative/positive for traumatic intracranial injury). Medians, interquartile ranges, and overall ranges are shown by the boxplots. The Y-axis is marked in concentrations to facilitate clinical interpretation. (A) Biomarker concentrations are shown for the four biomarkers across 4h intervals within 24h post-injury. Of note, biomarker concentrations within the first 24h were extracted from unique patients and not from repeated sampling from the same patient. (B) Biomarker concentrations are shown for the four biomarkers on day 1 and day 14. CT, computed tomography; GFAP, glial fibrillary acidic protein; NSE, neuron-specific enolase; S100B, S100 calcium-binding protein B; UCH-L1, ubiquitin c-terminal hydrolase L1.

FIG. 3.

Area under receiver-operating curves (AUC) for day 1 to day 14 biomarker results for predicting initial CT+ versus CT- imaging results and the associated 95% confidence intervals (CI). An AUC is shown for each biomarker for the overall study cohort (N = 1142) and hospitalized cohort (N = 305). (A) The AUC of biomarker concentrations shown within 24h post-injury to predict CT+ versus CT-. (B) The AUC of biomarker concentrations shown on day 14 post-injury to predict CT+ versus CT-. (C) The AUC of biomarker GFAP on day 1 and 14 post-injury. CT, computed tomography; GFAP, glial fibrillary acidic protein; NSE, neuron-specific enolase; S100B, S100 calcium-binding protein B; UCH-L1, ubiquitin c-terminal hydrolase L1.

In the 305 hospitalized participants with serial samples, GFAP, UCH-L1, NSE, and S100B had their highest median values on day 1; with the exception of UCH-L1 and NSE, biomarker levels decreased iteratively until their nadir on day 14. The UCH-L1 values reached their nadir on days 3 and 14. The NSE values reached their nadir on day 3. The GFAP values were consistently higher in CT+ participants at all time points. However, the UCH-L1, S100B, and NSE were not significantly higher in CT+ participants on days 1, 3, 5 and 14 (Fig. 4).

FIG. 4.

Biomarker values for hospitalized cohort from day 1 to day 14, by head CT result. Blood biomarker levels (in log10 scale) are shown for GFAP, UCH-L1, S100B, and NSE, stratified by initial head CT result, for hospitalized TBI participants (n = 305). Each graph displays time points from injury (days 1, 3, 5, and 14). The Y-axis is marked in concentrations to facilitate clinical interpretation. CT, computed tomography; GFAP, glial fibrillary acidic protein; UCH-L1 = ubiquitin c-terminal hydrolase L1; S100B, S100 calcium-binding protein B; NSE, neuron-specific enolase; TBI, traumatic brain injury.

The GFAP provided good to excellent discrimination for CT across all time points (AUC 0.84, 0.88, 0.82, 0.74, respectively, Fig. 3C). The UCH-L1, S100B, and NSE did not provide good discrimination for CT across all time points (AUC 0.51–0.56, Supplementary Fig. S1). The converted values from ARCHITECT laboratory platform (mean 530.5 ± 1579.4 pg/mL; interquartile range Q1 53.2–Q3 370.6 pg/mL) using the derived equations were comparable in variability to that of the i-STAT hand-held platform (mean 396.2 ± 1504.6 pg/mL; interquartile range Q1 13.5–Q3 250.4 pg/mL).

Discussion

In this large, well-characterized cohort of TBI participants presenting to care with GCS 13–15 and serial blood-based biomarker values, we found that GFAP remains accurate for identifying patients likely to have a positive CT scan consistent with TBI up to 14 days post-injury. In the subset of hospitalized participants, day 3 GFAP levels demonstrated higher AUCs (AUC = 0.84) for distinguishing CT+ versus CT- compared with day 1 GFAP (AUC = 0.80).

These findings have immediate importance, because clinicians are frequently faced with the challenge of identifying which patients need further evaluation with a CT scan to diagnose structural injury consistent with acute TBI. The finding that GFAP has validated utility in the post-acute, compared with hyperacute time frame post-injury, gains additional significance, because some patients may delay seeking medical care beyond 24h post-injury. While these findings remain preliminary, the utility of biomarker analysis to risk-stratify patients with TBI can potentially provide prognostic value beyond the FDA-approved 12h after injury.

Although several studies have reported that day-of-injury levels of TBI biomarkers may be useful for decision-making for brain CT imaging, the evidence has been less clear as to whether these biomarkers remain useful for guiding such decisions at later time points. Our findings on GFAP are congruent with findings from Papa and associates²³ who reported an AUC 0.86 for distinguishing between CT+ (16.7%) versus CT- (83.3%) TBI participants on study enrollment, and AUCs from 0.90–0.97 at 36–144h after presentation, with the highest AUCs observed between 36–60h after presentation (AUC 0.96–0.97). Our larger sample size allows a more robust estimate of the diagnostic accuracy of the biomarkers measured at later time points, through 14 days post-injury.

It is worth noting that GFAP values from days 1 and 14 may have potential utility for identifying patients with TBI who have a negative CT with traumatic intracranial abnormalities on brain MRI scan. This finding extends our previous observation of the diagnostic accuracy of day 1 GFAP for identifying CT-/MRI+ patients with TBI.⁷ Approximately 30% of CT negative patients with TBI have MRI evidence of traumatic intracranial abnormalities at two weeks—e.g., microhemorrhages and axonal shear injuries, which increase the risk of poor outcome.^7,2,25

While the GFAP AUC of 0.69 for identifying CT-/MRI+ participants at day 14 falls just short of 0.70 (the accepted statistical threshold for “adequate” discrimination), these findings nevertheless suggest that GFAP values may be useful for this additional indication and should be considered for validation in larger cohorts at day 14. Identification of this particular cohort of patients with TBI may permit timely referral for: (1) targeted administration of symptom-based, cognitive, and behavioral therapy, which has been found to be efficacious when implemented during the acute phase of injury,^26–28 and (2) enrichment of clinical trial study populations who are at risk for protracted recovery.

An objective biomarker to aid in TBI diagnosis after the day of injury is imperative for special cohorts of TBI patients with delayed presentation. Older individuals have a high incidence of TBI with various comorbidities that confound neurological evaluations. Not only is the older adult population at higher risk for intracranial hemorrhage because of use of antiplatelet and anticoagulant drugs, but also presentation is frequently delayed by days, and trauma history may be incomplete.²⁹

Sports injuries are often underreported and delayed in presentation and can be compounded by multiple subconcussive injuries.³⁰ Athletes may self-delay, “tough it out,” or are instructed to rest before seeking care. It is only when concussive symptoms fail to resolve that they present to a healthcare facility.³¹

In addition, the potential impact of TBI biomarkers on patient care stretches beyond a modern trauma center. Rather, acute brain trauma often can occur in regions where imaging modalities are not immediately available. This can include various combat zones, as well as developing countries that still necessitate accurate triaging because of resource constraints, transportation barriers, and risk of disclosing location in combat operations.

Patients with severe TBI in the military who require emergent operative care have improved outcomes if timely triaged to level 1 trauma centers.³² A low-risk patient with TBI may be safely transferred to a community hospital or receive care in the field, however, so as to optimize resource utilization of cost and time.³² Patients in the military with milder TBI may not require extraction but will require accurate evaluation for return to duty, especially those concussed with cognitive and other symptoms impairing their ability to make strategic decisions.³³

A biomarker “cutoff” could be useful in these special populations, as well as the general trauma population. A study involving transfer of complicated patients with mTBI to a Level 1 trauma center posited a large proportion of “overtriage” to ICU care,³⁴ and lower-level care may result in the same patient outcome and reduce resource utilization. A biomarker cutoff could provide a point-of-care measurement to assist in transfer decisions in regions without cranial imaging resources.

Within both our overall and hospitalized cohort, UCH-L1, NSE, and S100B had poor diagnostic accuracy across all time points. This may be because UCH-L1 and S100B have shorter half-lives in blood than GFAP.³⁵ As a result, delayed measurement of these biomarkers may limit diagnostic accuracy. These biomarkers also do not appear to be useful in identifying CT- patients with TBI who have a positive MRI, at least when measured in this study time frame.

It should be noted that while S100B displayed poor diagnostic accuracy in our analysis, it has known historical utility in predicting CT findings and poor outcome after TBI.³⁶ Further, S100B and UCH-L1 have been shown to increase in patients with TBI in whom early circulatory shock developed.³⁷

Limitations

First, these findings are directly applicable only to patients with TBI presenting to a Level 1 trauma center within 24h of injury and completed day 14 follow-up (either outpatient or inpatient). Patients with other races accounted for 6.7% of the cohort; thus, generalizability of the findings to patients of other racial backgrounds is a limitation and would need further investigation. Sport-related mechanisms of injury were not specified in the present study.

Validation is needed in patients who delay presentation for medical evaluation after 24h post-injury. Our days 3 and 5 findings were biased toward hospitalized patients with more symptomatic TBI and may differ in patients solely discharged from the ED and/or exclusively under outpatient care. Second, 35.7% of our overall cohort were CT+, which is higher than previous reports for TBI in ED-based cohorts. This suggests the need for validation in lower-risk cohorts and to further characterize the injury profile of these patients in a future investigation.

Third, it is unknown whether at time points after 24h post-injury, as stated previously, GFAP may have utility in the identification of CT-negative participants with MRI-positive intracranial injuries attributable to their TBI at day 14, which will require further study in larger cohorts. Because of recent FDA clearance of the i-STAT platform, the ARCHITECT assay values were converted to i-STAT equivalents. Despite high correlation, the study nonetheless contains conversion of the ARCHITECT platform using previously published equations derived from Passing-Bablok regression.

Conclusions

Our results indicate the diagnostic utility of plasma GFAP for identifying structural TBI on CT on days 1, 3, 5, and 14 post-injury in the GCS 13–15 cohort. The GFAP provided good to excellent discrimination and outperformed UCH-L1, S100B, and NSE for CT-based TBI diagnosis at all time points. Our findings provide critical evidence for the utility of GFAP using a point-of-care device to evaluate patients with suspected TBI for up to 14 days post-injury and suggest the need to broaden the indication for this test beyond the current 12h time frame.

Transparency, Rigor, and Reproducibility Summary

The study was pre-registered as the prospective, observational TRACK-TBI study (https://clinicaltrials.gov/ct2/show/NCT02119182). The analysis plan was registered prior to beginning data collection at the Center for Open Science (https://www.cos.io/initiatives/prereg). A sample size included a total of 2284 blood samples collected from 1142 TBI participants, matched days 1 and 14, to evaluate biomarker levels by diagnostic groups using the Wilcoxon Rank Sum test. Each participant in the hospitalized cohort had at least 3 blood samples (days 1, 3 and/or 5, and 14). Area under the ROC curve (AUC) was calculated with 95% confidence intervals and statistical analyses were performed using R (version 4.1.2, R Core Team, 2013). Handling of biofluid samples and analysis was performed by team members blinded to relevant characteristics of the participants. GFAP and UCH-L1 were measured in plasma samples in two batches. The first set of plasma biomarker samples were analyzed for GFAP and UCH-L1 concentrations (n = 1604) using the prototype point-of-care i-STAT™ Alinity™ System. The second set of plasma biomarker samples were analyzed for GFAP and UCH-L1 concentrations (n = 680) using the prototype core lab Abbott ARCHITECT^® platform for faster batch throughput. All equipment and analytical reagents used to perform measurements on the fluid biomarkers are widely available from commercial sources. The key inclusion criteria and outcome evaluations are established standards. This article will be published under a Creative Commons Open Access license, and upon publication will be freely available at https://www.liebertpub.com/loi/neu.

Ethics

Consent for each patient in the study was obtained through the legal-authorized representative (LAR) for proxy consent.

Data Availability

The datasets generated during and/or analyzed during the current study are not publicly available but could be acquired from the corresponding author on reasonable request.

Funding Information

This work was supported by the following grants: National Institute of Neurological Disorders and Stroke Grant #RC2NS069409, #U01NS086090, #U01NS1365885 (to G. T. Manley); United States Departments of Defense Grant #W81XWH-13-1-0441, #W81XWH-14-2-0176, #W81XWH-18-2-0042 (to G. T. Manley); Neurosurgery Research and Education Foundation & Bagan Family Foundation Research Fellowship Grant (UCSF Award #A139203, to J. K. Yue).

Author Disclosure Statement

Kevin K.W. Wang, PhD, is a shareholder of Gryphon Bio, Inc., which was not affiliated with the work contained within this manuscript. For all other authors, no competing financial interests exist.

TRACK-TBI Investigators

Neeraj Badjatia, University of Maryland, Baltimore, MD; Jason Barber, University of Washington, Seattle, WA; Patrick J. Belton, University of California, San Francisco, San Francisco, CA; Yelena G. Bodien, Harvard Medical School, Boston, MA; Ann-Christine Duhaime, Harvard Medical School, Boston, MA; Brian Fabian, University of California, San Francisco, San Francisco, CA; Brandon Foreman, University of Cincinnati, Cincinnati, OH; Raquel C. Gardner, University of California, San Francisco, San Francisco, CA; Joseph T. Giacino, Harvard Medical School, Boston, MA; Shankar Gopinath, Baylor College of Medicine, Houston, TX; Ramesh Grandhi, University of Utah Health, Salt Lake City, UT; J. Russell Huie, University of California, San Francisco, San Francisco, CA; Britta E. Lindquist, University of California, San Francisco, San Francisco, CA; Debbie Y. Madhok, University of California, San Francisco, San Francisco, CA; Michael A. McCrea, Medical College of Wisconsin, Milwaukee, WI; Randall Merchant, Virginia Commonwealth University, Richmond, VA; Lindsay D. Nelson, Medical College of Wisconsin, Milwaukee, WI; Laura B. Ngwenya, University of Cincinnati, Cincinnati, OH; Gabriela G. Satris, University of California, San Francisco, San Francisco, CA; David M. Schnyer, University of Texas at Austin, Austin, TX; Murray B. Stein, University of California, San Diego, San Diego, CA; Nancy R. Temkin, University of Washington, Seattle, WA; Justin C. Wong, University of California, San Francisco, San Francisco, CA; Alex B. Valadka, University of Texas Southwestern Medical Center, Dallas, TX; Thomas A. van Essen, Leiden University Medical Center, Leiden, The Netherlands; Mary J. Vassar, University of California, San Francisco, San Francisco, CA; Ethan A. Winkler, University of California, San Francisco, San Francisco, CA; Ross Zafonte, Harvard Medical School, Boston, MA.