An automated blood test for glial fibrillary acidic protein (GFAP) and ubiquitin carboxy-terminal hydrolase L1 (UCH-L1) to predict the absence of intracranial lesions on head CT in adult patients with mild traumatic brain injury: BRAINI, a multicentre observational study in Europe Authors: Alfonso Lagares, Javier de la Cruz, Hugo Terrisse, Odile Mejan, Vladislav Pavlov, Celine Vermorel, Jean-François Payen, of the BRAINI participants and investigators Published: 2024 ABSTRACT SummaryBackgroundFollowing mild traumatic brain injury (mTBI), elevated concentrations of brain-specific blood proteins glial fibrillary acidic protein (GFAP) and ubiquitin C-terminal hydrolase-L1 (UCH-L1) may be indicative of intracranial lesions normally detected by head CT scans. We sought to validate the performance of this combination of biomarkers at predetermined cutoff values with an automated immunoassay to predict which patients did not have intracranial lesions.MethodsThis prospective, observational study was conducted in France and Spain at 16 emergency departments. Adult patients with mTBI were eligible if they had a head CT scan and gave a 10-ml blood sample within 12 h of injury. GFAP and UCH-L1 serum concentrations were measured and analysed, in less than an hour time, according to predefined cutoff values of 22 pg/ml and 327 pg/ml, respectively. Serum concentrations of S100B protein were concomitantly determined in a subset of patients. The primary outcome measures were the sensitivity and negative predictive value (NPV) of the combined GFAP–UCH-L1 test to rule out intracranial lesions on head CT scans. ClinicalTrials.gov (NCT04032509).FindingsBetween August 2019 and June 2021, 1508 patients were recruited, and 1438 were included in the main analysis. Median age was 69 years (IQR 44–83). Most patients (74%) presented 3 h after trauma. 179 (12.4%) patients were positive for intracranial lesions by CT. The sensitivity of the combined test was 98.3% (95% CI 95.0–99.7) and the specificity 24.9 (95% CI 22.6–27.4), with a NPV of 99.1% (95% CI 97.1–99.8). Three patients with a positive CT scan had negative biomarker test results. S100B had a sensitivity of 83.0% (95% CI 76.2–88.2) and a NPV of 94.2% (95% CI 91.6–96.0). Patients with higher biomarker values more frequently had poorer recovery at 3 months after injury.InterpretationTesting for GFAP and UCH-L1, using validated cutoffs obtained with a new, fast automated immunoassay platform, accurately predicted the absence of intracranial lesions on head CT following mTBI.FundingThis study is co-funded by the European Institute of Innovation and Technology (EIT) Health, a body of the European Union (Grant nº19474). Biomarkers tests were funded by bioMérieux.Keywords: Mild traumatic brain injury, Biomarkers, Diagnosis, Clinical laboratory techniques, CT scan, Prognosis Research in context Evidence before this study Before the initiation of this study (August 2019), S100B protein was the only biomarker included in a clinical algorithm for the management of mild traumatic brain injury (mTBI) in some national guidelines in Europe. Although there was evidence regarding a possible use of glial fibrillary acidic protein (GFAP) and ubiquitin carboxy-terminal hydrolase L1 (UCH-L1) for ruling out the need for CT scan after mTBI, a fast, robust and reliable means for measuring the blood concentration of these proteins in the emergency setting was not yet available. Moreover, the cutoff concentrations of these two biomarkers had not been externally validated, and only one study compared the diagnostic performance of their combination with S100B. Several prognostic models for neurological outcome in mTBI have been designed; however, only weak correlation was found. The relationship between biomarker levels on the day of injury and neurological outcome needs more studies. Added value of this study Our study used an automated analytical platform ascertaining blood concentrations of two brain biomarkers, GFAP and UCH-L1, within 1 h from blood sampling. The combined GFAP and UCH-L1 test, using the cutoff values suggested by the ALERT-TBI study, was able to rule out the need of CT scan in a cohort of 1438 adult patients with mTBI, with high sensitivity and negative predictive value. The study showed that test performance is stable during the 12 h time window from trauma. The test could outperform clinical decision rules such as the Canadian CT Head Injury Rule (CCHR) and New Orleans Criteria (NOC), but the evidence should be further confirmed as CCHR and NOC were retrospectively assigned in an adapted version. The GFAP–UCH-L1 test could better predict which patients were not harbouring intracranial lesions, resulting in fewer patients being prescribed a CT scan. The combined GFAP–UCH-L1 test and the GFAP test alone were superior to using S100B to rule out the need for CT scans beyond 3 h post-trauma. Higher levels of GFAP were associated with increased risk of neuroworsening. Biomarker blood levels were higher in patients with incomplete recovery or with persistent symptoms 3 months after mTBI. Implications of all the available evidence A blood test measuring GFAP and UCH-L1 using the predetermined cutoffs could reduce the need for brain CT scans after mild TBI in the adult population. The blood test maintains a good diagnostic performance over a 12-h time window from trauma. Further studies are needed to optimize cutoff values in the elderly population and in paediatrics, and to implement the biomarker test in a medical decision algorithm. Introduction Mild traumatic brain injury (mTBI) as defined by an initial Glasgow Coma Scale (GCS) score of 13–15, is the most common presentation of traumatic brain injury affecting more than 40 million patients annually worldwide.1, 2, 3 Although most patients with a mTBI do not develop intracranial complications, 8–10% of cases harbour an intracranial lesion, 1% of cases require a neurosurgical procedure and one in 1000 patients die.4 The main diagnostic challenge in the initial evaluation of mTBI patients is to determine which patients are at risk of developing complications. Computed tomography (CT) of the head is the reference method to determine the presence of intracranial lesions in the acute management of patients with mTBI. Clinical decision rules (CDRs) and guidelines have been developed as decision aids for the indication of a head CT scan based on patient characteristics and clinical symptoms.5, 6, 7 However, around 30% of prescribed head CT scans for mTBI do not follow recommendations.7,8 Several blood-based brain biomarkers have been studied for the prediction of absence of intracranial lesions on CT scans after mTBI. S100B protein has been the most extensively investigated biomarker. It has been recommended in the 2013 Scandinavian Guidelines for the initial management in the emergency care setting, as an option for reducing unnecessary CT scans in a subgroup of patients with mild-head-injury and low risk for intracranial complication or neurosurgical intervention and a blood sampling within the first 6 h after mTBI.5,9 However, it has not gained widespread acceptance.10 Two brain-specific proteins — glial fibrillary acidic protein (GFAP) and ubiquitin carboxy-terminal hydrolase L1 (UCH-L1) — having complementary kinetics of release following brain injury, provide a longer temporal diagnostic window for testing post-injury and have been proposed as candidate biomarkers. Blood concentrations of the two biomarkers were significantly higher in patients with CT-detected traumatic intracranial injuries compared with those with no intracranial injuries.11 Cutoff values of GFAP and UCH-L1 were therefore determined to predict the absence of intracranial injury on head CT at admission. In the ALERT-TBI study, a combination of GFAP and UCH-L1 had a sensitivity of 97.6% and a negative predictive value (NPV) of 99.6% in 1959 patients presenting within 12 h following mTBI.12 However, the technique to assess blood concentration of biomarkers was a manual ELISA technique that took 4 h to provide results — an excessive processing time for triaging patients with mTBI. The same ALERT-TBI cohort was subsequently used to validate a faster immunoassay with comparable performance characteristics.13 A new immunoassay using an automated platform has been developed to provide results within 1 h after blood sampling. The present study was designed to externally validate the cutoff values of the ALERT-TBI study in a multicentre cohort study in Europe using this immunoassay. The primary objective was to estimate the sensitivity and the NPV of the combined GFAP–UCH-L1 test results to rule out the absence of intracranial lesions on head CT after mTBI. Secondary objectives included comparing the performance characteristics of GFAP and UCH-L1 with S100B protein, and comparing the performance of the combined GFAP-UCH-L1 test with most common CDRs assigned retrospectively to understand if biomarkers may support medical decision for performing brain CT scan. In addition, neurological deterioration at 7 days and outcome 3 months after injury were assessed with regards to the biomarker tests as possible contributors to predict prognosis. Methods Study design BRAINI was a prospective, multicentre, observational study that was carried out in France and Spain. Patients were recruited between August 2019 and June 2021 in 16 participating sites (12 in France (eleven trauma centres and one community hospital) and 4 in Spain (three trauma centres and one community hospital)). Ethics In Spain, ethics approval was obtained on 20 July 2019 (Re#19/322) from the institutional review board (IRB) CEIm Hospital Universitario 12 de Octubre, and the study started on 5 August 2019. In Spain, written informed consent was obtained before inclusion from the patient themselves or a next of kin if the patient was unable to give consent. In France, the IRB Sud-Ouest III approved the study design on 14 November 2019 (Re#2019-A01525-52), and the study started on 29 November 2019. Given its observational nature, the French ethical committee waived the requirement for written consent from patients or relatives, once clear information had been given and a statement of non-opposition had been received. If a patient was considered eligible for the study and before any study-related procedure, he/she or relatives received information about the research, including study objectives, type and volume of the sample collected, nature of information collection, and the right to refuse to participate at any moment according to current French law. They were given enough time to review the “Information and non-opposition” form and asked whether they agreed or opposed to participate. The French Data Protection Authority (CNIL) gave its approval on 31 January 2020 (Re#919443). The study was registered with ClinicalTrials.gov (NCT04032509) on 25 July 2019, and the protocol was published.14 Participants Patients were included if they were aged ≥18 years in France and ≥15 years in Spain (as patients ≥15 years are managed in the same emergency room by the same physician as adults) and presented with a GCS score of 13–15 following TBI, defined as alteration in brain function, or other evidence of brain pathology caused by an external force. They had to have undergone a routine head CT scan and a blood sample drawn within 12 h after injury. Indications for CT followed recommendations from Canadian CT Head Injury Rule (CCHR)15 and other potential risk factors: neurological focal deficit; anterograde amnesia; GCS <15 after 2 h post TBI; suspicion of vault depression fracture; fracture of the basal skull; persisting nausea, vomiting or headache; post-TBI seizures; preinjury treatment with antithrombotic drugs; loss of consciousness or amnesia in patients over 65 years of age; fall more than 1 m or hit pedestrian; and any other condition requiring a CT according to the in-charge physician. Patients were excluded if they had at least one of the following criteria: GCS score of 3–12 on admission, unknown time of injury, time of CT scan or blood sampling unknown or exceeding 12 h after TBI, admission for a non-traumatic neurological disorder, penetrating head injury, neuropsychiatric and neurological comorbidities (dementia, Parkinson's disease, multiple sclerosis, seizure disorder, brain tumor, history of neurosurgery, stroke within the last 30 days or major psychiatric disease) that might interfere with the outcome assessment at 7 days and 3 months, venipuncture not feasible, brain CT scan not performed, subject under judiciary control, pregnant or breastfeeding woman, and participation in an ongoing interventional trial. Procedures A 10-ml blood sample was obtained and processed to serum, frozen and stored at −80 °C until shipment on dry ice to the central storage facility (bioMérieux R&D bioBank, Marcy l'Etoile, France). The samples were then randomly divided in three equally sized groups, based on the site of origin of the samples, CT scan evaluation, GCS score and patient age. Each sample group was shipped for analysis to one of three USA testing laboratories equipped with the automated platform. All measurement procedures were conducted independently and blinded to the clinical and CT findings of the patients. Serum samples were analysed for GFAP and UCH-L1 concentrations using VIDAS® 3 platform immunoassays (bioMérieux, Marcy l’Etoile, France) between September 2022 and June 2023. Testing on the VIDAS® 3 was performed in automatic pipetting mode unless manual pipetting was requested owing to insufficient volume or in case of retesting due to invalid results. The Analytical Measuring Interval (AMI) is the range of values corresponding to the acceptable performance limits (precision and linearity). The AMI of the VIDAS® TBI (GFAP) assay is 10.0 pg/ml to 320.0 pg/ml. The AMI of the VIDAS® TBI (UCH-L1) assay is 80.0 pg/ml to 2560.0 pg/ml. There were 142 samples below 10 pg/ml for GFAP and 42 samples below 80 pg/ml for UCH-L1. The values below the reportable range were assigned as <10 pg/ml for GFAP and <80 pg/ml for UCH-L1. Out of the 142 samples below 10 pg/ml for GFAP, there are 29 below the limit of detection (5.4 pg/ml). Out of the 42 samples below 80 pg/ml for UCH-L1, there is only one below the limit of detection (48.1 pg/ml) (VIDAS® 3 detection limits Supplementary Table S1). Although these samples had no individual values of biomarkers, their combined test result was ascertained as negative. When the GFAP result was found to be higher than 320 pg/ml and with a signal below 6500 RFV (relative fluorescence value), the sample was diluted 1/10 in a negative serum sample and retested. The biomarker tests had a coefficient of variation below 8% for GFAP and below 7% for UCH-L1 for “within-laboratory precision, between-lot and within-instrument”. The presence or absence of haemolysis was reported in blood samples. The concentrations of GFAP and UCH-L1 of each patient were combined using predefined cutoff values,12 i.e., 327 pg/ml for UCH-L1 and 22 pg/ml for GFAP. A positive combined biomarker test was defined as one or two biomarkers having equal or higher levels to the cutoff value, and a negative combined biomarker test was defined as both biomarkers having lower levels than cutoff values. The blood concentration of S100B protein was blindly conducted by a single laboratory (CHU Clermont-Ferrand, France) using the immunoassay Elecsys® S100 used clinically in Europe on the automated Cobas® system (Roche Diagnostics, Meylan, France). The cutoff value used for S100B protein was 0.1 μg/L, according to the Scandinavian guidelines.8 In France, S100B protein analysis was permitted once patients had gone through another specific information procedure. Testing of S100B was conducted on a subset of patients who had given their consent for this analysis and from whom we had collected at least three, 1 ml, serum tubes. Head CT scans were uploaded to a web-based central medical image repository (SHANOIR-INRIA, Rennes, France) and analysed by two central readers who were independent, board-certified neuroradiologists and blinded to clinical and biological results. Each CT scan was classified as positive or negative for intracranial lesions. A third independent neuroradiologist established the final head CT scan result in case of discrepancy between the two central readers. A CT-positive result was defined as the presence of one or more of the following injuries: (1) epidural haematoma, (2) acute subdural haematoma, (3) subarachnoid haemorrhage, (4) intraventricular haemorrhage, (5) intraparenchymal contusion, (6) petechial haemorrhage or (7) any finding related to diffuse axonal injury and depressed skull fracture. Linear skull fractures were recorded but were not included in the definition of a CT-positive result. Factors related to the decision of CT prescription by the attending physician, including clinical details related to the TBI, mechanism, symptoms and signs as well as reason for prescribing CT, were prospectively recorded during the first visit of the study in the electronic case report form. Patients were then retrospectively categorized according to the Canadian CT Head Injury Rule (CCHR)15 and the New Orleans Criteria (NOC)4 as to whether they would have required a head CT scan under these criteria applying the logical rule of the CDR.16 Some variables, not present in the case report form, were assigned as follows: Visible trauma above the clavicles as AIS face moderate or severe; dangerous mechanism when the TBI was caused by a pedestrian being hit by a car; amnesia was recorded whatever was anterograde or retrograde. Therefore, these were adapted versions of the original CDRs and will be named hereafter a-CCHR and a-NOC. Physicians taking care of these patients were in-training physicians: neurosurgeons, general practitioners and emergency medicine physicians. We applied retrospectively CDRs because most physicians did not make CT decisions using CDRs in their usual clinical practice. Sex data was collected at admission as sex at birth. All data was consigned into an eCRF. Inter-rater agreement was not assessed between the different site data abstractors (those who entered the data into the database). Outcome measures The primary outcome measures were the sensitivity and the negative predictive value (NPV) of the combined GFAP and UCH-L1 results to rule out intracranial injury on head CT. Secondary outcome measures included specificity, positive predictive value (PPV), and the diagnostic performance of each biomarker test: GFAP, UCH-L1, and S100B protein. Other outcome measures were the proportion of patients with neurological deterioration at 7 days and altered neurological status 3 months after injury. At 7 days, a neurological worsening was defined as a decrease in GCS score of greater than 2 points from the initial GCS, or a neurological deterioration sufficient to warrant intervention, i.e., osmotherapy, transfer to the intensive care unit, or neurosurgical intervention. At 3 months, trained outcome assessors blinded to the clinical and biological results did a central, structured telephone interview to assess extended Glasgow Outcome Scale Extended (GOSE) score ranged from 1 (dead) to 8 (no disability)17 and the Rivermead Post-Concussion (RPQ) scale ranged from 0 (no symptom) to 24 (severe symptoms).18 Patients with GOSE score of less than 8 were classified as having incomplete recovery. RPQ was dichotomized according to the International Classification of Diseases (ICD) criteria for post-concussion syndrome (PCS).17 PCS was deemed positive with a score ≥2 (i.e., mild to severe problems) on at least three of the following symptoms: headache, dizziness, fatigue, irritability, sleep disturbances, poor concentration, forgetfulness, poor memory, frustration, or depression.19 Statistics To determine the sample size, we set up a hypothesis of performance levels at least on par with the Banyan BTI, based on the ALERT-TBI cohort data for patients with a GCS score of 13–15.12 We aimed for a test sensitivity of 98%, which needed to be significantly higher than 93%. With a 2.5% type I error of falsely reject the null hypothesis, i.e., H0: Sensitivity ≤93%, and a 80% statistical power under alternative hypothesis, i.e., H1: Sensitivity >93% with sensitivity at 98% at least, a minimum sample size of 173 CT-positive patients was required. Given an 11% prevalence of positive head CT scans following mTBI, the total number of patients needed was 1573 (rounded to 1600). The total number of patients required was then estimated to be 1600 patients. We increased this total number by 10% to consider screening failures, setting the recruitment target to 1760 patients. Quantitative variables were described as median and first and third quartiles, qualitative variables were described as frequencies and percentages. Characteristics between patients with CT scan-positive and CT scan-negative were compared with Mann–Whitney U and Chi-square or Fisher's exact tests when expected frequencies were lower than 5. Primary and secondary outcomes measures were analysed from complete data with no replacement of missing data. The analyses were performed for the whole study sample and for the study sample with available S100B results. Pre-specified subgroup analyses were performed for the main outcomes according to age and sex, and sensitivity analyses were carried out according to GCS score and time from injury to blood sampling. Agreement between the first two central readers was assessed on positive and negative CT scans. Areas under the ROC curve were estimated with 95% confidence intervals and compared for S100B, GFAP and UCH-L1 using an algorithm suggested by DeLong.20 Sensitivity, specificity, NPV and PPV were calculated with 95% confidence intervals using Agresti-Coull method.21 The performance properties of GFAP & UCH-L1 combination and both clinical rules (a-CCHR and a-NOC) were compared using McNemar test for sensitivity and specificity and generalized score statistic proposed by Leisenring for NPV and PPV.22 The biomarkers blood concentrations of patients with and without neuro-worsening at 7 days and altered outcome at 3 months were compared with Mann–Whitney U test. Multivariate logistic regression analyses were used to test the association of biomarkers concentration with outcomes adjusted for age, sex, and GCS score at admission. Statistical significance was set at p ≤ 0.05. All analyses were done using Stata version 17.0 (StataCorp, College Station, Texas). Role of funders This clinical study is part of the BRAINI project, an Innovation project co-funded by the European Institute of Innovation and Technology (EIT) Health, a body of the European Union (Activity Grant nº19474). The BRAINI project was an innovation partnership between EIT Health, the industrial IVD partner bioMérieux, Université Grenoble Alpes (UGA), Madrid regional health service (SERMAS). The national coordinators for France and Spain were academic hospitals, respectively, Grenoble Alpes (CHUGA) and 12 de Octubre (FIB–H12O). The sponsor of the clinical study was CHUGA and conducted data collection, data management and statistical analysis independently. EIT Health had no role in the study design, data collection, data analyses and interpretation, or writing of the report. BioMerieux funded and performed the biomarker tests but did not have access to the patient's clinical data. Results Between August 2019 and June 2021, 1508 patients were assessed for eligibility in the BRAINI study. Of these, 40 patients did not meet inclusion criteria (Fig. 1). Blood samples were not available for 27 patients, and three patients had no CT scan results. Therefore, the study sample included 1438 patients with biomarker test results and a valid head CT scan. Characteristics of patients are shown in Table 1. Study flow chart. Table 1Characteristics of the study population according to the presence (CT positive) or absence (CT negative) of traumatic intracranial injury on head CT scan.VariablesCT positive N = 179CT negative N = 1259p-valueCountry, n (%)0.07 Spain85 (47)508 (40) France94 (53)751 (60)Age, years73 [55–85]68 [42–83]0.005Sex0.19 Male, n (%)106 (59)681 (54) Female, n (%)73 (41)578 (46)White skin, n (%)134 (75)880 (70)0.17GCS score, n (%):<0.0001 134 (2)5 (1) 1420 (11)47 (4) 15155 (87)1207 (95)ISS score6 [3–12]3 [1–6]<0.0001Comorbidities and treatments, n (%) Atrial fibrillation24 (13)198 (16)0.42 Anticoagulants34 (19)294 (23)0.19 Antiplatelet agents56 (31)261 (21)0.001Cause of injury, n (%)0.33 Road traffic accident38 (21)297 (24) Fall125 (71)801 (64) Sport4 (2)38 (3) Violence/assault7 (4)65 (5) Other3 (2)52 (4) Unknown2 (1)6 (1)CT scan indication, n (%)0.32 Focal deficit7 (4)14 (1) Loss of consciousness100 (58)418 (34) Post traumatic amnesia77 (44)272 (22) Suspected cranial fracture20 (11)60 (5) Cranial base fracture6 (3)10 (1) Persistent nausea, vomiting, headache50 (28)199 (16) Posttraumatic seizure9 (5)43 (3)Alcohol intoxication, n (%)16 (9)91 (7)0.42Time from injury to blood draw (h)4.9 [3.3–7.3]4.4 [3.0–6.5]0.05Time from injury to CT scan (h)4.0 [2.8–5.6]4.4 [3.0–6.4]0.04Time from injury to blood draw, n (%)0.57 ≤2 hrs13 (7)116 (9) 2–457 (32)445 (35) 4–647 (26)343 (27) 6–829 (16)179 (14) 8–1018 (10)95 (8) 10–1215 (8)81 (6)GFAP, pg/ml182.9 [71.8–320.0]35.2 [17.6–65.7]<0.0001UCH-L1, pg/ml360 [221–608]265 [166–462]<0.0001Data are expressed as the median [25–75th percentiles] unless otherwise specified.Italics indicates p-values relate to the differences between CT-positive and CT-negative patients.Abbreviations: GCS, Glasgow Coma Scale; ISS, injury severity score. Characteristics of the study population according to the presence (CT positive) or absence (CT negative) of traumatic intracranial injury on head CT scan. Data are expressed as the median [25–75th percentiles] unless otherwise specified. Italics indicates p-values relate to the differences between CT-positive and CT-negative patients. Abbreviations: GCS, Glasgow Coma Scale; ISS, injury severity score. 179 (12.4%) patients had a traumatic intracranial lesion as ascertained by head CT scan. The proportion of positive CT scans was not significantly different between France and Spain (11.1 vs 14.3%, respectively; p = 0.07). There was excellent inter-observer agreement between the two first CT reviewers for determining CT positivity and negativity (positive specific agreement, 85%; negative specific agreement, 98%, kappa coefficient 0.83). Patients with a positive CT scan were significantly older, more frequently had GCS scores of 13 and 14, more frequently had a higher median injury severity score (ISS) score and more frequently received antiplatelet drugs than patients with a negative CT scan (Table 1). The most frequent intracranial lesion was subarachnoid haemorrhage (104 patients; 58%). Other lesions were acute subdural haematomas (77 patients; 43%), intraparenchymal contusions (36 patients; 20%), epidural haematoma (26 patients; 14.5%), petechial haemorrhage (10 patients; 5.6%) and intraventricular haemorrhage (9 patients; 5%). Information regarding timing of blood sampling is shown in Supplementary Figure S1. Of the study population, 1121 patients (78%) had a positive GFAP and UCH-L1 combined test, and 317 (22%) had a negative result for these biomarkers. The proportion of negative tests varies with age, i.e., 38% for patients below 65 years of age vs 8% for older patients. The sensitivity of the test was 98.3% (95% CI 95.0–99.7) with a NPV of 99.1% (95% CI 97.1–99.8) (Table 2). Three patients had false-negative test results, i.e., a positive CT scan result for an intracranial lesion and a negative result for GFAP and UCH-L1 (see Supplemental Material Table S2 for detailed description). Two of these patients had subarachnoid haemorrhages due to a non-traumatic ruptured intracranial aneurysm. The third patient had a subarachnoid haemorrhage at the limit of visibility and did not present neurological deterioration during follow up. Table 2Diagnostic performance of the combined GFAP–UCH-L1, GFAP, UCH-L1 and S100B protein biomarkers to predict the absence of intracranial lesions in head CT scan in the study population (n = 1438 patients) and in the subset population with concomitant S100B protein measurements (n = 1210 patients).Cross tabulation of biomarker test results with head CTSensitivitySpecificityNPVPPVStudy population GFAP–UCH-L1CT+ (N = 179)CT− (N = 1259)98.3 (95.0–99.7)24.9 (22.6–27.4)99.1 (97.1–99.8)15.7 (13.7–17.9)Test+ (N = 1121)176945Test− (N = 317)3314 GFAPCT+ (N = 178)CT− (N = 1258)98.3 (94.9–99.7)31.3 (28.8–33.9)99.2 (97.7–99.9)16.8 (14.7–19.2)Test+ (N = 1039)175864Test− (N = 397)3394 UCH-L1CT+ (N = 179)CT− (N = 1258)57.0 (49.7–64.0)63.0 (60.3–65.7)91.1 (89.1–92.9)18.0 (15.0–21.4)Test+ (N = 567)102465Test− (N = 870)77793Subset with S100B S100B proteinCT+ (N = 153)CT− (N = 1057)83.0 (76.2–88.2)40.0 (37.1–43.0)94.2 (91.6–96.0)16.7 (14.2–19.5)Test+ (N = 761)127634Test− (N = 449)26423 GFAP–UCH-L1CT+ (N = 153)CT− (N = 1057)98.7 (95.1–99.9)25.6 (23.1–28.4)99.3 (97.2–100)16.1 (14.9–19.9)Test+ (N = 937)151786Test− (N = 273)2271 GFAPCT+ (N = 152)CT− (N = 1056)98.7 (95.0–99.9)31.8 (29.1–34.7)99.4 (97.7–100)17.2 (14.9–19.9)Test+ (N = 870)150720Test− (N = 338)2336 UCH-L1CT+ (N = 153)CT− (N = 1056)55.6 (47.6–63.6)65.4 (62.5–68.2)91.0 (88.8–92.9)18.9 (15.5–22.8)Test+ (N = 450)85365Test− (N = 759)68691Data are expressed as percent and (95% confidence interval).Abbreviations: NPV, negative predictive value; PPV, positive predictive value. Diagnostic performance of the combined GFAP–UCH-L1, GFAP, UCH-L1 and S100B protein biomarkers to predict the absence of intracranial lesions in head CT scan in the study population (n = 1438 patients) and in the subset population with concomitant S100B protein measurements (n = 1210 patients). Data are expressed as percent and (95% confidence interval). Abbreviations: NPV, negative predictive value; PPV, positive predictive value. With respect to secondary outcomes, performance characteristics of each biomarker and combined biomarker test are shown in Table 2. The specificity of the combined test was 24.9% (95% CI 22.6–27.4). The combined biomarker test had a lower specificity for patients aged ≥65 years compared with younger patients: 10.3% (95% CI 8.2–12.8) vs 42.1% (95% CI 38.1–46.1), p < 0.001 (Supplementary material Table S3). Specificity decreased to 6.0% in patients aged 80 and over (Supplementary material Table S4). The patient's sex, having a CGS of 15, and time from injury to blood sampling (i.e., <3 h, 3–6 h, and >6 h) did not affect the performance characteristics of the combined biomarker test (Supplementary material Table S5). Blood concentrations of GFAP and UCH-L1 were significantly higher in patients with a positive CT scan (Table 1). GFAP had a significantly greater area under the curve for detection of CT lesions than UCH-L1: 0.85 (95% CI 0.82–0.88) vs 0.63 (95% CI 0.58–0.67), p < 0.001(See Supplementary material Figure S2). S100B protein analysis In the subset of 1210 patients with concomitant S100B protein analysis, i.e., those who gave consent for this analysis and from whom we had collected at least three, 1 ml, serum tubes, 761 patients had a positive S100B protein test, and 937 patients had a positive combined GFAP–UCH-L1 test. GFAP had better diagnostic performances than UCH-L1 and S100B protein (Table 2 and Fig. 2). Results remained unchanged for patients with isolated TBI (n = 258 patients) and patients with multiple extra-cranial injuries (n = 82 patients) (Supplementary material Table S6). The Sensitivity and NPV of the S100B protein test were equal to that of the combined GFAP–UCH-L1 test during the first 3 h following trauma, but these became lower than the combined GFAP–UCH-L1 test beyond 3 h (Fig. 3 and Supplementary material Table S7). Blood samples within the first 2 h after mTBI were performed in 98 patients and showed no differences in the diagnostic performance of the biomarkers (Supplementary material Table S8). The area under the ROC curve to detect patients with intracranial injuries was significantly higher using GFAP than UCH-L1 or S100B protein: 0.85 (95% CI 0.82–0.88) vs 0.64 (95% CI 0.59–0.68) or 0.66 (95% CI 0.62–0.71), respectively (p < 0.001) (Fig. 2). Diagnostic performance of GFAP, UCH-L1 and S100B protein biomarkers to predict the absence of intracranial lesions in head CT scan (n = 1210 patients). a) boxplot showing the concentrations of biomarkers related to CT status. Outliers, defined according to Tukey (1977) reference in Stata software. The horizonal line is the cut off value for the different biomarkers (GFAP 22 pg/ml; UCH-L1 327 pg/ml, S100B 0.1 mg/ml). b) ROC curve for the different biomarkers. c) Comparison of AUC between the different biomarkers. Diagnostic performance (sensitivity and negative predictive value) according to the time from injury to blood drawn (n = 1210 patients). Clinical decision rules The NPV and specificity were higher for the combined GFAP–UCH-L1 test than for both a-CCHR and a-NOC (Supplementary Table S9), which were determined retrospectively. According to the a-CCHR, 1199 patients (83% of the sample) would have been classified as needing a CT scan (Supplementary Figure S3). Of the remaining 239 patients (17%) who would have been deemed as having no need for CT scans using the a-CCHR classification, 10 patients had detectable intracranial lesions associated with a positive combined biomarker test. Used independently from the CCHR, the combined GFAP–UCH-L1 test could have avoided 317 (22%) unnecessary CT scans; however, 3 patients presenting abnormalities at CT scan would have been missed. Neurological deterioration 11 patients had neurological worsening at 7 days post-injury. All causes of neurological deterioration were TBI related. GFAP blood concentrations were significantly higher in these patients compared with those without neurological worsening (320 pg/ml (Q1–Q3 42.5–630) vs 39.8 pg/ml (Q1–Q3 20.0–80.5), p = 0.004). No statistically significant association could be shown between S100B or UCH-L1 levels and neuroworsening. None of these patients were admitted to the ICU or had neurosurgical intervention. Neurological outcomes At 3 months post-injury, outcome information was available for 1062 (74%) patients for GOSE scoring and 1012 (70%) for RPQ scoring. Of these, 269 patients (25.3%) had a GOSE score of <8, and 246 patients (24.3%) had PCS-related symptoms according to RPQ. Patients with altered outcomes had significantly higher blood biomarker concentrations (GFAP and UCH-L1, but also S100B) on admission (GFAP 54.9 pg/ml (Q1–Q3 27.9–145) vs 39.0 (Q1–Q3 18.2–77.6); UCH-L1 327 pg/ml (Q1–Q3 194–552) vs 248 pg/ml (Q1–Q3 160–412); S–100 B 0.18 μg/ml (Q1–Q3 0.09–0.32) vs 0.12 μg/ml (Q1–Q3 0.07–0.20)) (Table 3, and Supplementary Figures S4 and S5) and more frequently had a positive combined GFAP–UCH-L1 biomarker test than patients with complete recovery. Increased concentration of any of the biomarkers was independently associated with incomplete recovery (GOSE <8) and PCS persistent symptoms (logistic regression adjusted for age, sex and admission GCS p-values for incomplete recovery <0.001 for the three biomarkers; for PCS persistent symptoms GFAP p = 0.004, UCH-L1 p = 0.001, S100B p = 0.18). The area under the curve for detecting patients with incomplete recovery or PCS persistent symptoms was similar for all biomarkers (0.61, 0.60 and 0.61 for GFAP, UCH-L-1 and S100B for GOSE and 0.56, 0.55 and 0.57 for GFAP, UCH-L-1 and S100B for PCS) (Supplementary Figures S6 and S7). However, their prognostic performance showed higher sensitivity for GFAP and the combined GFAP-UCH-L1 test (Supplementary Tables S10 and S11). Table 3Blood concentrations of biomarkers on admission to the emergency department and neurological outcome at 3 months (altered vs normal) after mild traumatic brain injury as assessed using GOSE score to identify patients with incomplete recovery (GOSE <8) and Rivermead post-concussion scale to identify patients with the presence of post-concussion syndrome.VariablesAlteredNormalp-valueAdjustedap-valueGOSEGOSE <8 N = 269GOSE 8 N = 793 GFAP, pg/ml54.9 [27.9–145]39.0 [18.2–77.6]<0.001<0.001 UCH-L1, pg/ml327 [194–552]248 [160–412]<0.001<0.001 S100B protein, μg/ml0.18 [0.09–0.32]0.12 [0.07–0.20]<0.001<0.001Rivermaid scalePost-concussion N = 246No post-concussion N = 766 GFAP, pg/mL47.4 [27.0–84.6]39.3 [18.7–84.6]0.010.004 UCH-L1, pg/ml280 [181–511]260 [162–431]0.0240.001 S100B protein, μg/ml0.16 [0.08–0.28]0.12 [0.07–0.21]0.0030.18Data are expressed as the median [25–75th percentiles].Abbreviation: GOSE, Glasgow outcome scale extended.aLogistic regression analysis adjusted for age, sex, admission GCS. Blood concentrations of biomarkers on admission to the emergency department and neurological outcome at 3 months (altered vs normal) after mild traumatic brain injury as assessed using GOSE score to identify patients with incomplete recovery (GOSE <8) and Rivermead post-concussion scale to identify patients with the presence of post-concussion syndrome. Data are expressed as the median [25–75th percentiles]. Abbreviation: GOSE, Glasgow outcome scale extended. Logistic regression analysis adjusted for age, sex, admission GCS. Discussion The combined GFAP–UCH-L1 biomarker test showed high sensitivity and high NPV to rule out the need for a CT scan in this European cohort of patients with mTBI using predetermined cutoff values. The diagnostic performance of the combined biomarker test was maintained through a variety of conditions, including time sampling up to 12 h after mTBI and sex. Patients with altered neurological outcome at 3 months had significantly higher, although marginally, blood levels of these biomarkers on admission. Despite efforts designed to specify which patients need a brain CT scan following mTBI, through the development of several CDRs and guidelines, a large variability in the prescription of brain CT scan exists in clinical practices.10 The assessment of S100B protein has been included in the Scandinavian TBI Guidelines for patients at low risk of posttraumatic intracranial lesions when used within the first 6 h after mTBI.9 However, this test has not gained wide acceptance due to several confounding factors such as skin colour and multiple fractures. GFAP and UCH-L1, two brain biomarkers have emerged as candidates in this setting. Their blood concentrations were elevated in mTBI patients with intracranial lesions.11,23 An assay measuring GFAP and UCH-L1 showed high sensitivity and high NPV for excluding the presence of intracranial lesions after TBI using predetermined cutoffs for these biomarkers.11,12 However, these cutoffs were not externally validated, and the benefit of the combined UCH-L1 and GFAP performance test over GFAP alone remained unclear.24,25 Diagnostic performance with the automated immunoassay The BRAINI project assessed the performance of an automated immunoassay blood test in the population of adults with mTBI. The assay was developed so that it would take less than 1 h from blood sampling to biomarker results using an automated platform. This technological advance means that biomarker analysis is suitable under emergency conditions, on par with a CT scan, which, if indicated, must be performed within 8 h from the trauma in patients at intermediate and medium risk, according to recent clinical guidelines.26,27 Under these conditions, the present study confirmed that the cutoff values for GFAP (22 pg/ml) and UCH-L1 (327 pg/ml), which were set in the ALERT-TBI study, could be used to rule out the presence of an intracranial injury detectable with CT in patients with mTBI. In addition, the combined GFAP–UCH-L1 biomarker test outperformed the S100B assay if blood sampling occurred after 3 h post-injury, which corresponded to 75% of the study population. These findings are consistent with other studies28,29 and indicate that the combined GFAP–UCH-L1 biomarker test could be extended to a wide sampling window in patients following mTBI. The recent French National Recommendations suggested using S100B within the first 3 h following injury or combined GFAP–UCH-L1 test within 12 h post-injury if they are available.27 Our study represents, to our knowledge, the largest series to directly compare two brain biomarker solutions available for clinical use in Europe. Regarding the diagnostic performance of each biomarker, GFAP had the highest sensitivity and NPV. We did not confirm the added value of UCH-L1 to the diagnostic performance of GFAP as shown in the ALERT-TBI study. This might be due to an older patient age or a longer time from trauma to blood sampling in the present study (see Supplementary material Table S12). Three patients had false negative results. Two of them presented with a fall due to ruptured intracerebral aneurysm. More research is needed to assess the performance of these biomarkers outside of trauma, such as intracerebral haematoma30 and subarachnoid haemorrhage.31 The third patient had a minimal traumatic subarachnoid haemorrhage, and their condition did not deteriorate. All cases would have had a CT performed according to the CCHR criteria. This study found that specificity of the biomarker tests diminished in patients older than 65 years, and consequently the proportion of negative tests, consistent with previous reports.32,33 Specificity further decreased in octogenarians. Several studies have pointed out the correlation between different biomarker levels (S100, GFAP, UCH-L1 and NF-L) with age and the presence of neurological or neurodegenerative conditions.34, 35, 36 This study suggests that cutoff values for these biomarkers should be revised in the older population, as planned in the ongoing BRAINI2 elderly study.37 Biomarkers and CDRs Our data show that the combined biomarkers test to predict CT features could be superior to the widely used CDRs CCHR and NOC. However, the CDRs were adapted and applied retrospectively Brain biomarkers are designed to capture almost any kind of posttraumatic abnormality on the head CT to maximize their sensitivity. The aim of CDRs is to not miss neurosurgical and clinically important lesions with a reasonably high specificity to limit a CT overuse, at the expense of missing small lesions without clinical significance. Several studies suggest that a combination of CDRs (CCHR) and GFAP would provide the highest accuracy to predict CT abnormality in adult mTBI.38,39 However, these studies either do not provide corresponding cutoffs for GFAP38 or do not propose an algorithm of their combined use.39 The study by Papa et al.39 has paved the way for combined evaluation of brain biomarkers and CDRs, where clinical factors from the CDRs were prospectively collected by the treating physicians during patient assessments. Such an approach may be considered as optimal. Our study does not include the positioning of biomarkers in an algorithm for managing mTBI, as CT prescription was performed independently of the biomarker result. The targeted population and an algorithm including clinical aspects and blood biomarkers is therefore warranted. Ongoing trials in France (NCT05885529) and in the USA (NCT06449183) will hopefully provide additional data on the added value and real-life performance of CDRs/blood biomarkers combination. Biomarkers and neurological outcome Although most patients with mTBI can return to normal life, some patients develop persistent symptoms that may have consequences on their professional and social life.2,38,40,41 The absence of CT-related intracranial lesions does not guarantee a lack of persistent symptoms post-TBI.40 The best-performing prognostic models for mTBI are those using information on symptoms and outcome at one week from the traumatic event.40 Therefore, the outcome of a patient with mTBI cannot be predicted in the emergency context. Protein-based blood biomarkers in the TRACK and CENTER-TBI studies, including S100B, GFAP and UCH-L1, showed growing interest for the prognostic use of biomarkers to predict the outcome of patients after TBI.42,43 In the current study, the median concentration of GFAP and UCH-L1 at admission was significantly higher in patients who subsequently had an incomplete recovery or persistent symptoms related to PCS. The adjustment for covariates showed that each biomarker was independently associated with incomplete recovery after TBI. However, the elevations were marginal. If this information is further confirmed, predictive models including biomarker assessment could be developed to identify patients at risk for incomplete recovery after mTBI. Limitations There are several limitations with this study. First, the BRAINI study was not an implementation study. CT prescription was performed independently of biomarker results in this validation study of an automated assay. The use of these biomarker tests in real life to triage the indications for brain CT scan is warranted. Second, although all patients had a CT scan, the indication for CT prescription varied between countries, centres or attending physicians for reasons related to the prescriber or to patient characteristics. Third, the number of intracranial lesions was high (12%), suggesting that CDRs might have been used by physicians in charge, and that there was no indiscriminate use of CT during the study. This higher incidence of CT-positive scans than expected prevented us from a potential lack of statistical power for the primary endpoint owing to a reduced sample size. Despite an expected unfavourable impact of a high CT positive frequency on NPV estimates, in our study the observed NPV reached the preplanned expectations. Fourth, no study regarding alternative cutoffs was performed as the main objective of the study was to validate previous cutoffs. Fifth, the accuracy of the decision rules relative to the biomarkers was performed on scored risk factors using an automated rule, not from physician judgement, and were derived from an adapted version to account for available data collected in the eCRF. This might have affected the performance of CDRs. This methodology differs from other studies, where emergency physicians prospectively completed data forms for each CDR before the patients’ CT scans.39 Results of the comparison between CDRs and the combined test should be considered with caution. Sixth, although we evaluated the CDRs and neurological deterioration at 7 days and outcome at 3 months after injury relative to the biomarkers, the study was not designed, nor powered to formally address these 3 secondary outcomes. Seventh, this is a relatively older group of patients with falls as a dominant mechanism of injury. The age spectrum of this cohort is reflective of the patients admitted in European emergency rooms. Accordingly, these findings may not be as generalizable to other emergency rooms with younger populations with different mechanisms of injury such as those in North America. However, TBI affecting elderly people is increasing in all areas of the world and in the future most countries will face a similar case mix of patients. Eighth, although we found significant differences in the concentration of biomarkers regarding neurological condition 3 months after TBI, the magnitude of differences was small. These results should be taken with caution and highlight the need for more studies. Finally, we could not collect the total number of possibly screened patients in the study due to the number of sites participating, the large number of patients admitted daily and limited logistics for such an endeavour. Conclusions The results of this study indicate the value of an automated blood test as an objective tool for screening patients with mTBI and rule out the need for brain CT in less than 1 h. The biomarker cutoff values used in this study could enable safe and accurate triage of patients with mTBI in acute care settings across different countries. The combination GFAP–UCH-L1 biomarker test might be more discriminant than S100B protein beyond 3 h following trauma. Contributors AL, J-FP, OM, VP, and JC developed the protocol. AL, J-FP, OM, VP and JC designed the study, AL, JF-P and BRAINI investigators did the data collection. HT, CV and JC did the statistical analysis. AL, J-FP, VP, OM and JC did the data interpretation. AL and VP wrote the first draft of the report. All authors reviewed the report and drafted the final version of the report. All authors attest to have full access to all the data in the study and had final responsibility for the decision to submit for publication. All authors approved the final manuscript. AL, J-FP, HT, CV and JC have directly accessed and verified the underlying data reported in the manuscript. Members of the Braini Study Group and collaborators contributed by the inclusion and follow up of patients as well as by obtaining blood samples for biomarker analysis. Data sharing statement Data collected for the study, including individual participant data, data dictionary defining each field in the set, and study protocol will be made available to others upon reasonable request. Data will be communicated as deidentified participant data according to French law, and will be available after publication of the manuscript. Data will be made available at the Direction de la Recherche Clinique, Grenoble Alpes University Hospital, Grenoble, France (drci@chu-grenoble.fr), with investigator support, after approval of a proposal from the consortium of the study with a signed data access agreement. Declaration of interests OM is an employee of bioMérieux. Dr. Pavlov was an employee of bioMérieux, and is currently an employee of Roche Diagnostics France. JFP has received payments or honoraria from lectures or educational events from Integra Lifesciences, Sedana Medical, IDD-CDM Lavoisier, IdMed. The consortium was co-funded as Innovation project by the European Institute of Innovation and Technology (EIT) Health, a body of the European Union (Activity Grant nº19474). AL, JdC and HT were recipients of the grant project. AL received payments for lectures and educational events from bioMérieux. Acknowledgements The authors declare no acknowledgements. Footnotes Supplementary data related to this article can be found at https://doi.org/10.1016/j.ebiom.2024.105477. Contributor Information Alfonso Lagares, Email: alfonlag@ucm.es. of the BRAINI participants and investigators: Maxime Maignan, Damien Viglino, Laurent Jacquin, Marion Douplat, Said Laribi, Philippe Pes, Patrick Ray, Jérémy Guenezan, Mustapha Sebbane, Frédéric Belen, Guillaume Durand, Cordelia Abric, Cédric Gil-Jardiné, Pudaru Mihai, Julián Morales, Ana Castuera, Ana María Castano-Leon, Juan Delgado, and Jose F. Alen Supplementary data Suppl.file Braini Investigators&Collaborators Supplemental file Braini Investigators and Collaborators Supplemental material Appendix 1 Appendix 1 References Associated Data Supplementary Materials Suppl.file Braini Investigators&Collaborators Supplemental file Braini Investigators and Collaborators Supplemental material Appendix 1 Appendix 1