My name is Brett Sterk, and I'm a medical student at the University of Minnesota. I'm going to be talking about serum biomarkers GFAP, UCHL1, and S100B and their ability to predict brain death following both traumatic and hypoxic brain injury. These are the disclosures for the principal investigator, Dr. Uzma Samadani. There are no conflicts of interest relevant to this study. The background for this study is that there's a need for improved prognosis of severe brain injury. In particular, we looked at two distinct etiologies of brain injury that commonly result in a non-survivable injury. Diffuse axonal injury, or DAI, is a traumatic brain injury that results from shearing forces at the gray-white matter interface, particularly within the corpus callosum and the brain stem. Anoxic brain injury, which is typically atraumatic, is the result of global oxygen deprivation as a result of circulatory arrest, particularly within the highly metabolic cells of the hippocampus, cortex, and cerebellar Purkinje cells. Neuroimaging and clinical characteristics are the current gold standard for assessment in the acute phase of brain injury, but they don't provide the entire picture. CT imaging is frequently unrevealing or negative at 24 hours, even in some patients who will progress to brain death. MRI offers an increased degree of sensitivity, but has limitations including time in the scanner, availability, and may not be an option for patients with metal in their bodies. Clinically, the Glasgow Coma Scale is frequently used to monitor changes in neurological status, but it, too, lacks predictive value, cannot distinguish the underlying pathophysiology, and may be influenced by a number of subjective factors. Another complicating factor is that the clinical history provided is often incomplete. There will be a subset of patients who are found down with unknown primary mechanism further complicating their prognosis. As a clinician, in the first 24 hours of treating a patient with a severe brain injury and neurological deficits, you may have imaging that is unrevealing, and the question of whether or not this patient will recover, have permanent neurologic deficit, or progress to brain death is uncertain. Clearly, there is a compelling need for advanced objective measures to improve clinical decision making and provide realistic expectations for families. There is a growing body of evidence that biomarkers are a tool that have demonstrated utility in many aspects of TBI research, and are therefore a potential modality to aid in both diagnosis and prognosis of brain injury. The three serum biomarkers that we chose to look at were glial fibrillary acidic protein, GMP, ubiquitin carboxy terminal hydrolase L1, UCHL1, and S100 calcium binding protein B, S100B. Some of the benefits of biomarkers is that they are objective and quantitative, and may be implementable in a wide variety of settings. Our first hypothesis is that the levels of our three biomarkers, when drawn within the acute period, can predict which subjects will progress to brain death. Now, given the distinct mechanisms and underlying physiology of DAI and anoxic brain injury, it is possible that each injury may provide a unique biomarker profile. And therefore, our second hypothesis is that we may be able to differentiate the two types of injury based upon the levels of our three biomarkers alone. So this was a prospective cohort study that recruited subjects presenting to the emergency department with either traumatic or atraumatic brain injury, as determined by the clinical care team. For this study, we included both pediatric and adult populations. Blood was drawn within 32 hours of injury for analysis of biomarker levels. The patients were then followed for determination of primary outcome of either clinical brain death or survival to discharge. After being recruited, brain injury subjects were categorized into four main groups. All of the brain injury subjects that died as a result of their injuries prior to discharge formed the brain death group. And this included both anoxic and traumatic brain injury. All survivors were further categorized based on their mechanism of injury into either traumatic or atraumatic. And in the second row, you can see our non-traumatic brain injury groups, which were all survivors. Those with isolated TBI were further categorized based upon their CT findings upon their initial ED visit as to either negative with no abnormalities or positive with one or more abnormalities seen on their CT. We also recruited 41 healthy controls for the purposes of comparison. The mean age across all brain injury groups was similar, and only the isolated TBI with negative CT and the healthy controls included pediatric subjects. All of the brain injury groups were a majority male. The average time to blood draw ranged from 9 to 12 hours, but all average times fell within one standard deviation of each other. Each group had a sufficient number of sample results of each biomarker for statistical comparison. GFAP and UCHL1 were significantly higher in the brain death subjects compared to each other cohort. In the box and whisker plots, the x-axis represents the log concentration of each biomarker. The biomarker values were log transformed after assessing for abnormality, which was found to have a skewed distribution. Each of the individual boxes represents a subject group, and comparisons were made between brain death and each individual cohort. Unlike GFAP and UCHL1, S100B on the chart in the far right did not demonstrate any significant differences between either of the subject cohorts. All three serum biomarkers were then assessed in combination for three groups, healthy controls, all brain injury survivors, and those that died from the brain injury. As you can see in the 3D scatter plot, each group demonstrated a unique and differentiable pattern. Control subjects were tightly clustered around lower values for all three biomarkers, while brain injury survivors formed a group with intermediate values. Those in the brain death group generally had higher levels for all three biomarkers but didn't cluster as well as the other groups, which we'll go into in the next slide. Next, using machine learning, we assessed the ability of all three biomarkers measured within 32 hours of injury and its ability to predict which subjects will progress to brain death. This returned an area under the curve of 0.96, indicating a robust ability to predict brain death as an outcome. We then looked to see whether or not our model could differentiate brain death subjects based upon mechanism of injury. Here we have a 3D scatter plot focused on the brain death subjects subcategorized into three groups based on mechanism of injury. Cardiac and respiratory arrest, labeled CARA on the plot, forms the purely anoxic brain death group, while those with purely diffuse axonal injury form the DAI group. Subjects that were reported as found down, or FD, includes all brain death subjects with an unclear mechanism of injury. In order to capture a clearly atraumatic brain injury and compare it to any form of traumatic brain injury, we decided to group DAI and found down together for statistical comparison. This decision was supported by the limited history obtained on each found down subject in the medical record, which showed that all found down subjects had at least some component of trauma. For example, one subject was found at the bottom of the stairs with evidence of a fall. It was unclear whether a cardiac event precipitated this or not. However, all found down subjects, like this one, had some evidence of trauma in the medical record. Looking again at the plot, if we group DAI and found down together, which forms our group with at least some traumatic brain injury, they all appear to be clustered on the right, while the cardiac and respiratory arrest patients are clustered more on the left, providing a visual to a unique biomarker profile of each injury. In corroboration with this visual, our machine learning model was able to differentiate brain death subjects with either clearly atraumatic injury from those with traumatic injury with an AUC of 0.99 based upon biomarkers alone. In summary, if the model is provided with the biomarker levels of a particular subject, it can determine whether or not this subject fell into the anoxic or the traumatic injury group with a high degree of accuracy. In summary, serum levels of GFAP, UCHL1, and S100B, when measured in the acute setting following both traumatic and hypoxic brain injury, have utility in determining which patients will progress to brain death, and moreover, can provide information about the underlying mechanism of injury. Therefore, serum biomarkers may provide clinicians with objective measures to inform clinical decision making and set appropriate expectations. This study was limited by a relatively small sample size, and future directions would include a larger study representing a wider spectrum of severity and mechanism of injury. Thank you.

serum biomarkers

brain death

traumatic brain injury

hypoxic brain injury

glial fibrillary acidic protein