Expert-level detection of acute intracranial hemorrhage on head computed tomography using deep learning.

Computed tomography (CT) of the head is used worldwide to diagnose neurologic emergencies. However, expertise is required to interpret these scans, and even highly trained experts may miss subtle life-threatening findings. For head CT, a unique challenge is to identify, with perfect or near-perfect sensitivity and very high specificity, often small subtle abnormalities on a multislice cross-sectional (three-dimensional [3D]) imaging modality that is characterized by poor soft tissue contrast, low signal-to-noise using current low radiation-dose protocols, and a high incidence of artifacts. We trained a fully convolutional neural network with 4,396 head CT scans performed at the University of California at San Francisco and affiliated hospitals and compared the algorithm's performance to that of 4 American Board of Radiology (ABR) certified radiologists on an independent test set of 200 randomly selected head CT scans. Our algorithm demonstrated the highest accuracy to date for this clinical application, with a receiver operating characteristic (ROC) area under the curve (AUC) of 0.991 ± 0.006 for identification of examinations positive for acute intracranial hemorrhage, and also exceeded the performance of 2 of 4 radiologists. We demonstrate an end-to-end network that performs joint classification and segmentation with examination-level classification comparable to experts, in addition to robust localization of abnormalities, including some that are missed by radiologists, both of which are critically important elements for this application.

Prognostic Value of Hemorrhagic Brainstem Injury on Early Computed Tomography: A TRACK-TBI Study.

BACKGROUND: Traumatic brainstem injury has yet to be incorporated into widely used imaging classification systems for traumatic brain injury (TBI), and questions remain regarding prognostic implications for this TBI subgroup. To address this, retrospective data on patients from the multicenter prospective Transforming Research and Clinical Knowledge in TBI study were studied. METHODS: Patients with brainstem and cerebrum injury (BSI+) were matched by age, sex, and admission Glasgow Coma Scale (GCS) score to patients with cerebrum injuries only. All patients had an interpretable head computed tomography (CT) scan from the first 48 hours after injury and a 6-month Glasgow Outcome Scale Extended (GOSE) score. CT scans were reviewed for brainstem lesions and, when present, characterized by location, size, and type (traumatic axonal injury, contusion, or Duret hemorrhage). Clinical, demographic, and outcome data were then compared between the two groups. RESULTS: Mann-Whitney U-tests showed no significant difference in 6-month GOSE scores in patients with BSI+ (mean 2.7) compared with patients with similar but only cerebrum injuries (mean 3.9), although there is a trend (p = 0.10). However, subclassification by brainstem lesion type, traumatic axonal injury (mean 4.0) versus Duret hemorrhage or contusion (mean 1.4), did identify a proportion of BSI+ with significantly less favorable outcome (p = 0.002). The incorporation of brainstem lesion type (traumatic axonal injury vs. contusion/Duret), along with GCS into a multivariate logistic regression model of favorable outcome (GOSE score 4-8) did show a significant contribution to the prognostication of this brainstem injury subgroup (odds ratio 0.08, 95% confidence interval 0.00-0.67, p = 0.01). CONCLUSIONS: These findings suggest two groups of patients with brainstem injuries may exist with divergent recovery potential after TBI. These data support the notion that newer CT imaging classification systems may augment traditional clinical measures, such as GCS in identifying those patients with TBI and brainstem injuries that stand a higher chance of favorable outcome.

Connectome mapping with edge density imaging differentiates pediatric mild traumatic brain injury from typically developing controls: proof of concept.

BACKGROUND: Although acute neurologic impairment might be transient, other long-term effects can be observed with mild traumatic brain injury. However, when pediatric patients with mild traumatic brain injury present for medical care, conventional imaging with CT and MR imaging often does not reveal abnormalities. OBJECTIVE: To determine whether edge density imaging can separate pediatric mild traumatic brain injury from typically developing controls. MATERIALS AND METHODS: Subjects were recruited as part of the "Therapeutic Resources for Attention Improvement using Neuroimaging in Traumatic Brain Injury" (TRAIN-TBI) study. We included 24 adolescents (χ=14.1 years of age, σ=1.6 years, range 10-16 years), 14 with mild traumatic brain injury (TBI) and 10 typically developing controls. Neurocognitive assessments included the pediatric version of the California Verbal Learning Test (CVLT) and the Attention Network Task (ANT). Diffusion MR imaging was acquired on a 3-tesla (T) scanner. Edge density images were computed utilizing fiber tractography. Principal component analysis (PCA) and support vector machines (SVM) were used in an exploratory analysis to separate mild TBI and control groups. The diagnostic accuracy of edge density imaging, neurocognitive tests, and fractional anisotropy (FA) from diffusion tensor imaging (DTI) was computed with two-sample t-tests and receiver operating characteristic (ROC) metrics. RESULTS: Support vector machine-principal component analysis of edge density imaging maps identified three white matter regions distinguishing pediatric mild TBI from controls. The bilateral tapetum, sagittal stratum, and callosal splenium identified mild TBI subjects with sensitivity of 79% and specificity of 100%. Accuracy from the area under the ROC curve (AUC) was 94%. Neurocognitive testing provided an AUC of 61% (CVLT) and 71% (ANT). Fractional anisotropy yielded an AUC of 48%. CONCLUSION: In this proof-of-concept study, we show that edge density imaging is a new form of connectome mapping that provides better diagnostic delineation between pediatric mild TBI and healthy controls than DTI or neurocognitive assessments of memory or attention.

Concordance of common data elements for assessment of subjective cognitive complaints after mild-traumatic brain injury: a TRACK-TBI Pilot Study.

OBJECTIVE: To determine characteristics and concordance of subjective cognitive complaints (SCCs) 6 months following mild-traumatic brain injury (mTBI) as assessed by two different TBI common data elements (CDEs). RESEARCH DESIGN: The Transforming Research and Clinical Knowledge in Traumatic Brain Injury (TRACK-TBI) Pilot Study was a prospective observational study that utilized the NIH TBI CDEs, Version 1.0. We examined variables associated with SCC, performance on objective cognitive tests (Wechsler Adult Intelligence Scale, California Verbal Learning Test, and Trail Making Tests A and B), and agreement on self-report of SCCs as assessed by the acute concussion evaluation (ACE) versus the Rivermead Post Concussion Symptoms Questionnaire (RPQ). RESULTS: In total, 68% of 227 participants endorsed SCCs at 6 months. Factors associated with SCC included less education, psychiatric history, and being assaulted. Compared to participants without SCC, those with SCC defined by RPQ performed significantly worse on all cognitive tests. There was moderate agreement between the two measures of SCCs (kappa = 0.567 to 0.680). CONCLUSION: We show that the symptom questionnaires ACE and RPQ show good, but not excellent, agreement for SCCs in an mTBI study population. Our results support the retention of RPQ as a basic CDE for mTBI research. ABBREVIATIONS: BSI-18: Brief Symptom Inventory; 18CDEs: common data elements; CT: computed tomography; CVLT: California Verbal Learning Test; ED: emergency department; GCS: Glasgow coma scale; LOC: loss of consciousnessm; TBI: mild-traumatic brain injury; PTA: post-traumatic amnesia; SCC: subjective cognitive complaints; TBI: traumatic brain injury; TRACK-TBI: Transforming Research and Clinical Knowledge in Traumatic Brain Injury; TMT: Trail Making Test; WAIS-PSI: Wechsler Adult Intelligence Scale, Fourth Edition, Processing Speed Index.

DRD2 C957T polymorphism is associated with improved 6-month verbal learning following traumatic brain injury.

Traumatic brain injury (TBI) often leads to heterogeneous clinical outcomes, which may be influenced by genetic variation. A single-nucleotide polymorphism (SNP) in the dopamine D2 receptor (DRD2) may influence cognitive deficits following TBI. However, part of the association with DRD2 has been attributed to genetic variability within the adjacent ankyrin repeat and kinase domain containing 1 protein (ANKK1). Here, we utilize the Transforming Research and Clinical Knowledge in Traumatic Brain Injury Pilot (TRACK-TBI Pilot) study to investigate whether a novel DRD2 C957T polymorphism (rs6277) influences outcome on a cognitive battery at 6 months following TBI-California Verbal Learning Test (CVLT-II), Wechsler Adult Intelligence Test Processing Speed Index Composite Score (WAIS-PSI), and Trail Making Test (TMT). Results in 128 Caucasian subjects show that the rs6277 T-allele associates with better verbal learning and recall on CVLT-II Trials 1-5 (T-allele carrier 52.8 ± 1.3 points, C/C 47.9 ± 1.7 points; mean increase 4.9 points, 95% confidence interval [0.9 to 8.8]; p = 0.018), Short-Delay Free Recall (T-carrier 10.9 ± 0.4 points, C/C 9.7 ± 0.5 points; mean increase 1.2 points [0.1 to 2.5]; p = 0.046), and Long-Delay Free Recall (T-carrier 11.5 ± 0.4 points, C/C 10.2 ± 0.5 points; mean increase 1.3 points [0.1 to 2.5]; p = 0.041) after adjusting for age, education years, Glasgow Coma Scale, presence of acute intracranial pathology on head computed tomography scan, and genotype of the ANKK1 SNP rs1800497 using multivariable regression. No association was found between DRD2 C947T and non-verbal processing speed (WAIS-PSI) or mental flexibility (TMT) at 6 months. Hence, DRD2 C947T (rs6277) may be associated with better performance on select cognitive domains independent of ANKK1 following TBI.

Temporal profile of care following mild traumatic brain injury: predictors of hospital admission, follow-up referral and six-month outcome.

OBJECTIVE: To investigate the clinical management and medical follow-up of patients with mild traumatic brain injury (mTBI) presenting to emergency departments (EDs). METHODS: Overall, 168 adult patients with mTBI from the prospective, multicentre Transforming Research and Clinical Knowledge in TBI (TRACK-TBI) Pilot study with Glasgow Coma Scale (GCS) 13-15, no polytrauma and alive at six months were included. Predictors for hospital admission, three-month follow-up referral and six-month functional disability (Glasgow Outcome Scale-Extended (GOSE) ≤ 6) were analysed using multivariable regression. RESULTS: Overall, 48% were admitted to hospital, 22% received three-month referral and 27% reported six-month functional disability. Intracranial pathology on ED head computed tomography (multivariable odds ratio (OR) = 81.08, 95% confidence interval (CI) [10.28-639.36]) and amnesia (>30-minutes: OR = 5.27 [1.75-15.87]; unknown duration: OR = 4.43 [1.26-15.62]) predicted hospital admission. Older age (per-year OR = 1.03 [1.01-1.05]) predicted three-month referral, while part-time/unemployment predicted lack of referral (OR = 0.17 [0.06-0.50]). GCS < 15 (OR = 2.46 [1.05-5.78]) and prior history of seizures (OR = 3.62 [1.21-10.89]) predicted six-month functional disability, while increased education (per-year OR = 0.86 [0.76-0.97]) was protective. CONCLUSIONS: Clinical factors modulate triage to admission, while demographic/socioeconomic elements modulate follow-up care acquisition; six-month functional disability associates with both clinical and demographic/socioeconomic variables. Improving triage to acute and outpatient care requires further investigation to optimize resource allocation and outcome after mTBI. ClinicalTrials.gov registration: NCT01565551.

COMT Val 158 Met polymorphism is associated with nonverbal cognition following mild traumatic brain injury.

Mild traumatic brain injury (mTBI) results in variable clinical outcomes, which may be influenced by genetic variation. A single-nucleotide polymorphism in catechol-o-methyltransferase (COMT), an enzyme which degrades catecholamine neurotransmitters, may influence cognitive deficits following moderate and/or severe head trauma. However, this has been disputed, and its role in mTBI has not been studied. Here, we utilize the Transforming Research and Clinical Knowledge in Traumatic Brain Injury Pilot (TRACK-TBI Pilot) study to investigate whether the COMT Val (158) Met polymorphism influences outcome on a cognitive battery 6 months following mTBI--Wechsler Adult Intelligence Test Processing Speed Index Composite Score (WAIS-PSI), Trail Making Test (TMT) Trail B minus Trail A time, and California Verbal Learning Test, Second Edition Trial 1-5 Standard Score (CVLT-II). All patients had an emergency department Glasgow Coma Scale (GCS) of 13-15, no acute intracranial pathology on head CT, and no polytrauma as defined by an Abbreviated Injury Scale (AIS) score of ≥3 in any extracranial region. Results in 100 subjects aged 40.9 (SD 15.2) years (COMT Met (158) /Met (158) 29 %, Met (158) /Val (158) 47 %, Val (158) /Val (158) 24 %) show that the COMT Met (158) allele (mean 101.6 ± SE 2.1) associates with higher nonverbal processing speed on the WAIS-PSI when compared to Val (158) /Val (158) homozygotes (93.8 ± SE 3.0) after controlling for demographics and injury severity (mean increase 7.9 points, 95 % CI [1.4 to 14.3], p = 0.017). The COMT Val (158) Met polymorphism did not associate with mental flexibility on the TMT or with verbal learning on the CVLT-II. Hence, COMT Val (158) Met may preferentially modulate nonverbal cognition following uncomplicated mTBI.Registry: ClinicalTrials.gov Identifier NCT01565551.

Visualization of nerve fibers and their relationship to peripheral nerve tumors by diffusion tensor imaging.

OBJECT The majority of growing and/or symptomatic peripheral nerve tumors are schwannomas and neurofibromas. They are almost always benign and can usually be resected while minimizing motor and sensory deficits if approached with the proper expertise and techniques. Intraoperative electrophysiological stimulation and recording techniques allow the surgeon to map the surface of the tumor in an effort to identify and thus avoid damaging functioning nerve fibers. Recently, MR diffusion tensor imaging (DTI) techniques have permitted the visualization of axons, because of their anisotropic properties, in peripheral nerves. The object of this study was to compare the distribution of nerve fibers as revealed by direct electrical stimulation with that seen on preoperative MR DTI. METHODS The authors conducted a retrospective chart review of patients with a peripheral nerve or nerve root tumor between March 2012 and January 2014. Diffusion tensor imaging and intraoperative data had been prospectively collected for patients with peripheral nerve tumors that were resected. Preoperative identification of the nerve fiber location in relation to the nerve tumor surface as seen on DTI studies was compared with the nerve fiber's intraoperative localization using electrophysiological stimulation and recordings. RESULTS In 23 patients eligible for study there was good correlation between nerve fiber location on DTI and its anatomical location seen intraoperatively. Diffusion tensor imaging demonstrated the relationship of nerve fibers relative to the tumor with 95.7% sensitivity, 66.7% specificity, 75% positive predictive value, and 93.8% negative predictive value. CONCLUSIONS Preoperative DTI techniques are useful in helping the peripheral nerve surgeon to both determine the risks involved in resecting a nerve tumor and plan the safest surgical approach.

Magnetic resonance imaging improves 3-month outcome prediction in mild traumatic brain injury.

OBJECTIVE: To determine the clinical relevance, if any, of traumatic intracranial findings on early head computed tomography (CT) and brain magnetic resonance imaging (MRI) to 3-month outcome in mild traumatic brain injury (MTBI). METHODS: One hundred thirty-five MTBI patients evaluated for acute head injury in emergency departments of 3 LEVEL I trauma centers were enrolled prospectively. In addition to admission head CT, early brain MRI was performed 12 ± 3.9 days after injury. Univariate and multivariate logistic regression were used to assess for demographic, clinical, socioeconomic, CT, and MRI features that were predictive of Extended Glasgow Outcome Scale (GOS-E) at 3 months postinjury. RESULTS: Twenty-seven percent of MTBI patients with normal admission head CT had abnormal early brain MRI. CT evidence of subarachnoid hemorrhage was associated with a multivariate odds ratio of 3.5 (p = 0.01) for poorer 3-month outcome, after adjusting for demographic, clinical, and socioeconomic factors. One or more brain contusions on MRI, and ≥4 foci of hemorrhagic axonal injury on MRI, were each independently associated with poorer 3-month outcome, with multivariate odds ratios of 4.5 (p = 0.01) and 3.2 (p = 0.03), respectively, after adjusting for head CT findings and demographic, clinical, and socioeconomic factors. INTERPRETATION: In this prospective multicenter observational study, the clinical relevance of abnormal findings on early brain imaging after MTBI is demonstrated. The addition of early CT and MRI markers to a prognostic model based on previously known demographic, clinical, and socioeconomic predictors resulted in a >2-fold increase in the explained variance in 3-month GOS-E.

Semi-supervised Learning for Generalizable Intracranial Hemorrhage Detection and Segmentation.

Purpose To develop and evaluate a semi-supervised learning model for intracranial hemorrhage detection and segmentation on an out-of-distribution head CT evaluation set. Materials and Methods This retrospective study used semi-supervised learning to bootstrap performance. An initial "teacher" deep learning model was trained on 457 pixel-labeled head CT scans collected from one U.S. institution from 2010 to 2017 and used to generate pseudo labels on a separate unlabeled corpus of 25 000 examinations from the Radiological Society of North America and American Society of Neuroradiology. A second "student" model was trained on this combined pixel- and pseudo-labeled dataset. Hyperparameter tuning was performed on a validation set of 93 scans. Testing for both classification (n = 481 examinations) and segmentation (n = 23 examinations, or 529 images) was performed on CQ500, a dataset of 481 scans performed in India, to evaluate out-of-distribution generalizability. The semi-supervised model was compared with a baseline model trained on only labeled data using area under the receiver operating characteristic curve, Dice similarity coefficient, and average precision metrics. Results The semi-supervised model achieved a statistically significant higher examination area under the receiver operating characteristic curve on CQ500 compared with the baseline (0.939 [95% CI: 0.938, 0.940] vs 0.907 [95% CI: 0.906, 0.908]; P = .009). It also achieved a higher Dice similarity coefficient (0.829 [95% CI: 0.825, 0.833] vs 0.809 [95% CI: 0.803, 0.812]; P = .012) and pixel average precision (0.848 [95% CI: 0.843, 0.853]) vs 0.828 [95% CI: 0.817, 0.828]) compared with the baseline. Conclusion The addition of unlabeled data in a semi-supervised learning framework demonstrates stronger generalizability potential for intracranial hemorrhage detection and segmentation compared with a supervised baseline. Keywords: Semi-supervised Learning, Traumatic Brain Injury, CT, Machine Learning Supplemental material is available for this article. Published under a CC BY 4.0 license. See also the commentary by Swimburne in this issue.

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

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.

Clinical and Imaging Characteristics, Care Pathways, and Outcomes of Traumatic Epidural Hematomas: A Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury Study.

BACKGROUND AND OBJECTIVES: Guideline recommendations for surgical management of traumatic epidural hematomas (EDHs) do not directly address EDHs that co-occur with other intracranial hematomas; the relative rates of isolated vs nonisolated EDHs and guideline adherence are unknown. We describe characteristics of a contemporary cohort of patients with EDHs and identify factors influencing acute surgery. METHODS: This research was conducted within the longitudinal, observational Collaborative European NeuroTrauma Effectiveness Research in Traumatic Brain Injury cohort study which prospectively enrolled patients with traumatic brain injury from 65 hospitals in 18 European countries from 2014 to 2017. All patients with EDH on the first scan were included. We describe clinical, imaging, management, and outcome characteristics and assess associations between site and baseline characteristics and acute EDH surgery, using regression modeling. RESULTS: In 461 patients with EDH, median age was 41 years (IQR 24-56), 76% were male, and median EDH volume was 5 cm3 (IQR 2-20). Concomitant acute subdural hematomas (ASDHs) and/or intraparenchymal hemorrhages were present in 328/461 patients (71%). Acute surgery was performed in 99/461 patients (21%), including 70/86 with EDH volume ≥30 cm3 (81%). Larger EDH volumes (odds ratio [OR] 1.19 [95% CI 1.14-1.24] per cm3 below 30 cm3), smaller ASDH volumes (OR 0.93 [95% CI 0.88-0.97] per cm3), and midline shift (OR 6.63 [95% CI 1.99-22.15]) were associated with acute surgery; between-site variation was observed (median OR 2.08 [95% CI 1.01-3.48]). Six-month Glasgow Outcome Scale-Extended scores ≥5 occurred in 289/389 patients (74%); 41/389 (11%) died. CONCLUSION: Isolated EDHs are relatively infrequent, and two-thirds of patients harbor concomitant ASDHs and/or intraparenchymal hemorrhages. EDHs ≥30 cm3 are generally evacuated early, adhering to Brain Trauma Foundation guidelines. For heterogeneous intracranial pathology, surgical decision-making is related to clinical status and overall lesion burden. Further research should examine the optimal surgical management of EDH with concomitant lesions in traumatic brain injury, to inform updated guidelines.

Isolated Traumatic Subarachnoid Hemorrhage on Head Computed Tomography Scan May Not Be Isolated: A Transforming Research and Clinical Knowledge in Traumatic Brain Injury Study (TRACK-TBI) Study.

Isolated traumatic subarachnoid hemorrhage (tSAH) after traumatic brain injury (TBI) on head computed tomography (CT) scan is often regarded as a "mild" injury, with reduced need for additional workup. However, tSAH is also a predictor of incomplete recovery and unfavorable outcome. This study aimed to evaluate the characteristics of CT-occult intracranial injuries on brain magnetic resonance imaging (MRI) scan in TBI patients with emergency department (ED) arrival Glasgow Coma Scale (GCS) score 13-15 and isolated tSAH on CT. The prospective, 18-center Transforming Research and Clinical Knowledge in Traumatic Brain Injury Study (TRACK-TBI; enrollment years 2014-2019) enrolled participants who presented to the ED and received a clinically-indicated head CT within 24 h of TBI. A subset of TRACK-TBI participants underwent venipuncture within 24 h for plasma glial fibrillary acidic protein (GFAP) analysis, and research MRI at 2-weeks post-injury. In the current study, TRACK-TBI participants age ≥17 years with ED arrival GCS 13-15, isolated tSAH on initial head CT, plasma GFAP level, and 2-week MRI data were analyzed. In 57 participants, median age was 46.0 years [quartile 1 to 3 (Q1-Q3): 34-57] and 52.6% were male. At ED disposition, 12.3% were discharged home, 61.4% were admitted to hospital ward, and 26.3% to intensive care unit. MRI identified CT-occult traumatic intracranial lesions in 45.6% (26 of 57 participants; one additional lesion type: 31.6%; 2 additional lesion types: 14.0%); of these 26 participants with CT-occult intracranial lesions, 65.4% had axonal injury, 42.3% had subdural hematoma, and 23.1% had intracerebral contusion. GFAP levels were higher in participants with CT-occult MRI lesions compared with without (median: 630.6 pg/mL, Q1-Q3: [172.4-941.2] vs. 226.4 [105.8-436.1], p = 0.049), and were associated with axonal injury (no: median 226.7 pg/mL [109.6-435.1], yes: 828.6 pg/mL [204.0-1194.3], p = 0.009). Our results indicate that isolated tSAH on head CT is often not the sole intracranial traumatic injury in GCS 13-15 TBI. Forty-six percent of patients in our cohort (26 of 57 participants) had additional CT-occult traumatic lesions on MRI. Plasma GFAP may be an important biomarker for the identification of additional CT-occult injuries, including axonal injury. These findings should be interpreted cautiously given our small sample size and await validation from larger studies.

Performance of the IMPACT and CRASH prognostic models for traumatic brain injury in a contemporary multicenter cohort: a TRACK-TBI study.

OBJECTIVE: The International Mission on Prognosis and Analysis of Clinical Trials in Traumatic Brain Injury (IMPACT) and Corticosteroid Randomization After Significant Head Injury (CRASH) prognostic models for mortality and outcome after traumatic brain injury (TBI) were developed using data from 1984 to 2004. This study examined IMPACT and CRASH model performances in a contemporary cohort of US patients. METHODS: The prospective 18-center Transforming Research and Clinical Knowledge in Traumatic Brain Injury (TRACK-TBI) study (enrollment years 2014-2018) enrolled subjects aged ≥ 17 years who presented to level I trauma centers and received head CT within 24 hours of TBI. Data were extracted from the subjects who met the model criteria (for IMPACT, Glasgow Coma Scale [GCS] score 3-12 with 6-month Glasgow Outcome Scale-Extended [GOSE] data [n = 441]; for CRASH, GCS score 3-14 with 2-week mortality data and 6-month GOSE data [n = 831]). Analyses were conducted in the overall cohort and stratified on the basis of TBI severity (severe/moderate/mild TBI defined as GCS score 3-8/9-12/13-14), age (17-64 years or ≥ 65 years), and the 5 top enrolling sites. Unfavorable outcome was defined as GOSE score 1-4. Original IMPACT and CRASH model coefficients were applied, and model performances were assessed by calibration (intercept [< 0 indicated overprediction; > 0 indicated underprediction] and slope) and discrimination (c-statistic). RESULTS: Overall, the IMPACT models overpredicted mortality (intercept -0.79 [95% CI -1.05 to -0.53], slope 1.37 [1.05-1.69]) and acceptably predicted unfavorable outcome (intercept 0.07 [-0.14 to 0.29], slope 1.19 [0.96-1.42]), with good discrimination (c-statistics 0.84 and 0.83, respectively). The CRASH models overpredicted mortality (intercept -1.06 [-1.36 to -0.75], slope 0.96 [0.79-1.14]) and unfavorable outcome (intercept -0.60 [-0.78 to -0.41], slope 1.20 [1.03-1.37]), with good discrimination (c-statistics 0.92 and 0.88, respectively). IMPACT overpredicted mortality and acceptably predicted unfavorable outcome in the severe and moderate TBI subgroups, with good discrimination (c-statistic ≥ 0.81). CRASH overpredicted mortality in the severe and moderate TBI subgroups and acceptably predicted mortality in the mild TBI subgroup, with good discrimination (c-statistic ≥ 0.86); unfavorable outcome was overpredicted in the severe and mild TBI subgroups with adequate discrimination (c-statistic ≥ 0.78), whereas calibration was nonlinear in the moderate TBI subgroup. In subjects ≥ 65 years of age, the models performed variably (IMPACT-mortality, intercept 0.28, slope 0.68, and c-statistic 0.68; CRASH-unfavorable outcome, intercept -0.97, slope 1.32, and c-statistic 0.88; nonlinear calibration for IMPACT-unfavorable outcome and CRASH-mortality). Model performance differences were observed across the top enrolling sites for mortality and unfavorable outcome. CONCLUSIONS: The IMPACT and CRASH models adequately discriminated mortality and unfavorable outcome. Observed overestimations of mortality and unfavorable outcome underscore the need to update prognostic models to incorporate contemporary changes in TBI management and case-mix. Investigations to elucidate the relationships between increased survival, outcome, treatment intensity, and site-specific practices will be relevant to improve models in specific TBI subpopulations (e.g., older adults), which may benefit from the inclusion of blood-based biomarkers, neuroimaging features, and treatment data.

Prior traumatic brain injury is a risk factor for in-hospital mortality in moderate to severe traumatic brain injury: a TRACK-TBI cohort study.

Objectives: An estimated 14-23% of patients with traumatic brain injury (TBI) incur multiple lifetime TBIs. The relationship between prior TBI and outcomes in patients with moderate to severe TBI (msTBI) is not well delineated. We examined the associations between prior TBI, in-hospital mortality, and outcomes up to 12 months after injury in a prospective US msTBI cohort. Methods: Data from hospitalized subjects with Glasgow Coma Scale score of 3-12 were extracted from the Transforming Research and Clinical Knowledge in Traumatic Brain Injury Study (enrollment period: 2014-2019). Prior TBI with amnesia or alteration of consciousness was assessed using the Ohio State University TBI Identification Method. Competing risk regressions adjusting for age, sex, psychiatric history, cranial injury and extracranial injury severity examined the associations between prior TBI and in-hospital mortality, with hospital discharged alive as the competing risk. Adjusted HRs (aHR (95% CI)) were reported. Multivariable logistic regressions assessed the associations between prior TBI, mortality, and unfavorable outcome (Glasgow Outcome Scale-Extended score 1-3 (vs. 4-8)) at 3, 6, and 12 months after injury. Results: Of 405 acute msTBI subjects, 21.5% had prior TBI, which was associated with male sex (87.4% vs. 77.0%, p=0.037) and psychiatric history (34.5% vs. 20.7%, p=0.010). In-hospital mortality was 10.1% (prior TBI: 17.2%, no prior TBI: 8.2%, p=0.025). Competing risk regressions indicated that prior TBI was associated with likelihood of in-hospital mortality (aHR=2.06 (1.01-4.22)), but not with hospital discharged alive. Prior TBI was not associated with mortality or unfavorable outcomes at 3, 6, and 12 months. Conclusions: After acute msTBI, prior TBI history is independently associated with in-hospital mortality but not with mortality or unfavorable outcomes within 12 months after injury. This selective association underscores the importance of collecting standardized prior TBI history data early after acute hospitalization to inform risk stratification. Prospective validation studies are needed. Level of evidence: IV. Trial registration number: NCT02119182.

Data-driven distillation and precision prognosis in traumatic brain injury with interpretable machine learning.

Traumatic brain injury (TBI) affects how the brain functions in the short and long term. Resulting patient outcomes across physical, cognitive, and psychological domains are complex and often difficult to predict. Major challenges to developing personalized treatment for TBI include distilling large quantities of complex data and increasing the precision with which patient outcome prediction (prognoses) can be rendered. We developed and applied interpretable machine learning methods to TBI patient data. We show that complex data describing TBI patients' intake characteristics and outcome phenotypes can be distilled to smaller sets of clinically interpretable latent factors. We demonstrate that 19 clusters of TBI outcomes can be predicted from intake data, a ~ 6× improvement in precision over clinical standards. Finally, we show that 36% of the outcome variance across patients can be predicted. These results demonstrate the importance of interpretable machine learning applied to deeply characterized patients for data-driven distillation and precision prognosis.

Effects of age and time since injury on traumatic brain injury blood biomarkers: a TRACK-TBI study.

Older adults have the highest incidence of traumatic brain injury globally. Accurate blood-based biomarkers are needed to assist with diagnosis of patients across the spectrum of age and time post-injury. Several reports have suggested lower accuracy for blood-based biomarkers in older adults, and there is a paucity of data beyond day-1 post-injury. Our aims were to investigate age-related differences in diagnostic accuracy and 2-week evolution of four leading candidate blood-based traumatic brain injury biomarkers-plasma glial fibrillary acidic protein, ubiquitin carboxy-terminal hydrolase L1, S100 calcium binding protein B and neuron-specific enolase-among participants in the 18-site prospective cohort study Transforming Research And Clinical Knowledge in Traumatic Brain Injury. Day-1 biomarker data were available for 2602 participants including 2151 patients with traumatic brain injury, 242 orthopedic trauma controls and 209 healthy controls. Participants were stratified into 3 age categories (young: 17-39 years, middle-aged: 40-64 years, older: 65-90 years). We investigated age-stratified biomarker levels and biomarker discriminative abilities across three diagnostic groups: head CT-positive/negative; traumatic brain injury/orthopedic controls; and traumatic brain injury/healthy controls. The difference in day-1 glial fibrillary acidic protein, ubiquitin carboxy-terminal hydrolase L1 and neuron-specific enolase levels across most diagnostic groups was significantly smaller for older versus younger adults, resulting in a narrower range within which a traumatic brain injury diagnosis may be discriminated in older adults. Despite this, day-1 glial fibrillary acidic protein had good to excellent performance across all age-categories for discriminating all three diagnostic groups (area under the curve 0.84-0.96; lower limit of 95% confidence intervals all >0.78). Day-1 S100 calcium-binding protein B and ubiquitin carboxy-terminal hydrolase L1 showed good discrimination of CT-positive versus negative only among adults under age 40 years within 6 hours of injury. Longitudinal blood-based biomarker data were available for 522 hospitalized patients with traumatic brain injury and 24 hospitalized orthopaedic controls. Glial fibrillary acidic protein levels maintained good to excellent discrimination across diagnostic groups until day 3 post-injury irrespective of age, until day 5 post-injury among middle-aged or younger patients and until week 2 post-injury among young patients only. In conclusion, the blood-based glial fibrillary acidic protein assay tested here has good to excellent performance across all age-categories for discriminating key traumatic brain injury diagnostic groups to at least 3 days post-injury in this trauma centre cohort. The addition of a blood-based diagnostic to the evaluation of traumatic brain injury, including geriatric traumatic brain injury, has potential to streamline diagnosis.

Neuroworsening in the Emergency Department Is a Predictor of Traumatic Brain Injury Intervention and Outcome: A TRACK-TBI Pilot Study.

INTRODUCTION: Neuroworsening may be a sign of progressive brain injury and is a factor for treatment of traumatic brain injury (TBI) in intensive care settings. The implications of neuroworsening for clinical management and long-term sequelae of TBI in the emergency department (ED) require characterization. METHODS: Adult TBI subjects from the prospective Transforming Research and Clinical Knowledge in Traumatic Brain Injury Pilot Study with ED admission and disposition Glasgow Coma Scale (GCS) scores were extracted. All patients received head computed tomography (CT) scan <24 h post-injury. Neuroworsening was defined as a decline in motor GCS at ED disposition (vs. ED admission). Clinical and CT characteristics, neurosurgical intervention, in-hospital mortality, and 3- and 6-month Glasgow Outcome Scale-Extended (GOS-E) scores were compared by neuroworsening status. Multivariable regressions were performed for neurosurgical intervention and unfavorable outcome (GOS-E ≤ 3). Multivariable odds ratios (mOR) with [95% confidence intervals] were reported. RESULTS: In 481 subjects, 91.1% had ED admission GCS 13-15 and 3.3% had neuroworsening. All neuroworsening subjects were admitted to intensive care unit (vs. non-neuroworsening: 26.2%) and were CT-positive for structural injury (vs. 45.4%). Neuroworsening was associated with subdural (75.0%/22.2%), subarachnoid (81.3%/31.2%), and intraventricular hemorrhage (18.8%/2.2%), contusion (68.8%/20.4%), midline shift (50.0%/2.6%), cisternal compression (56.3%/5.6%), and cerebral edema (68.8%/12.3%; all p < 0.001). Neuroworsening subjects had higher likelihoods of cranial surgery (56.3%/3.5%), intracranial pressure (ICP) monitoring (62.5%/2.6%), in-hospital mortality (37.5%/0.6%), and unfavorable 3- and 6-month outcome (58.3%/4.9%; 53.8%/6.2%; all p < 0.001). On multivariable analysis, neuroworsening predicted surgery (mOR = 4.65 [1.02-21.19]), ICP monitoring (mOR = 15.48 [2.92-81.85], and unfavorable 3- and 6-month outcome (mOR = 5.36 [1.13-25.36]; mOR = 5.68 [1.18-27.35]). CONCLUSIONS: Neuroworsening in the ED is an early indicator of TBI severity, and a predictor of neurosurgical intervention and unfavorable outcome. Clinicians must be vigilant in detecting neuroworsening, as affected patients are at increased risk for poor outcomes and may benefit from immediate therapeutic interventions.