SCCM Resource Library-Star Research Presentations: Neuroscience

Star Research Presentations: Neuroscience

Video Transcription

This is the STAR Research Presentations, Neuroscience. Dr. Ellender and myself, Paul Vespa, are the co-moderators for today's talk.  We're going to have a very nice presentation set up. Each of the presentations is going to be about 10 to 11 minutes, and then four or five minutes for questions and  answers, hopefully. So Dr. Riggs, Becky Riggs, will be our first speaker. She's from Johns Hopkins Hospital. And our first title of the first abstract is Optic Nerve  Sheath Diameter Changes in Children with Varying Degrees of Intracranial Pressure. Dr. Riggs.  Thank you. It's truly an honor to be here presenting with you guys today. Thank you very much. Very sensitive mouse. Okay, so I am a neurointensivist here from  Johns Hopkins Hospital in Baltimore, Maryland. And I will be talking about Optic Nerve Sheath Diameter Changes in Children with Varying Degrees of Traumatic Brain Injury.  Disclosures, I was the PI in this study, and I was being supported by the NIH Loan Repayment Program while conducting this research.  And I continue to be supported by a Stimulating and Advancing Anesthesia and Critical Care Medicine Research Award awarded me at Johns Hopkins by the Department of Anesthesiology  and Critical Care Medicine. So this is my really quick ophthalmic ultrasound safety slide. Please do not go and put a probe on a patient's eye after  you listen to me or anyone else speak about this. Make sure that you get a little bit of training to prevent corneal abrasions, thermal injuries, things of  that such that you can cause to patients when performing ophthalmic ultrasound. So the idea of optic nerve sheath diameters, being able  to detect changes in intracranial pressure, is not a new idea. It was actually first published by Dr. Sohan Singh Hyrek back in the 1960s.  He was a neuro-ophthalmologist who worked with rhesus monkeys who hypothesized because the optic nerve is surrounded by an optic nerve sheath that is directly connected to the  subarachnoid space, any pressure felt in the subarachnoid space was dissipated down the optic nerve sheath. And essentially, the optic nerve sheath would balloon out in  response to increased intracranial pressure, and it would shrink back down as the intracranial pressure normalized. So how he investigated this is he took a bunch of rhesus  monkeys and surgically implanted these balloons, or what would later be space-occupying lesions. So over weeks to months, he would slowly inflate these  balloons, causing large space-occupying lesions in various parts of the monkeys' brains, and then he would observe all their pupillary changes, and he essentially  documented all the pupillary and ophthalmic changes that we see as somebody experiences intracranial hypertension. So to test his theory even further, it was the optic nerve  sheath that was causing these pupillary changes. The next round of rhesus monkeys, when he went to surgically implant these space-occupying lesions, in  one eye, he would go in and fenestrate or cut open the optic nerve sheath, leaving the optic nerve sheath in the other eye completely intact.  So then as he raised the intracranial pressure and he observed both eyes, the eye with the optic nerve sheath that was intact showed all the classic changes of increased  intracranial pressure. The eye that had the nerve sheath fenestrated actually showed no of the changes seen with increased intracranial pressure.  So then we fast forward to the 1990s, where these fascinating gentlemen, Helmick and Henson, actually went into the morgue, and they got a lot of post-mortem John Does and  Jane Does, put needles in their backs, pumped them full of normal saline, raised their intracranial pressure quite high, and physically watched as the optic nerve sheath  increased, like ballooned out and then deflated with fluctuations of intracranial pressure. They actually documented that about seven millimeters of  diameter was about the maximal distention that the optic nerve sheath can dilate before it actually ruptures, at least in post-mortem patients.  So then they took this a little bit further, and granted they were not practicing in the United States, where there's IRBs, but they went out on the streets and they paid vagrants  a lot of money to come in and have them put needles in their backs, raise their intracranial pressure. And with ultrasound documents, increases in their optic nerve sheath.  So then we fast forward to the 2000s, and there's actually a lot of publications out there, both pediatric and the adult  world, showing that the use of optic nerve sheath diameters to distinguish patients with increased intracranial pressure from those with normal intracranial pressure is a  legitimate thing. So you have normative cutoff values. So for children, anyone less than a year old, shouldn't have an optic nerve sheath diameter greater than four millimeters.  Anyone greater than a year old but not quite in puberty, four and a half millimeters is the upper threshold.  And then for adults, about five millimeters is the cutoff to determine who has intracranial hypertension and who doesn't.  So then in 2011, there was a large systematic review of the literature, and then the pool data essentially showed 90 percent sensitivity, 85 percent specificity, and area under the  combined receiver-operator curves, 0.94, you know, showing that ultrasound-guided optic nerve sheath diameter measurements are of good diagnostic level and accurate  for detecting increased intracranial pressure. Now again, they were looking at does the patient have increased intracranial pressure, yes or no.  They weren't actually looking at varying degrees of intracranial pressure.  So just a couple months ago in 2019, another systematic review of the literature was published, which essentially showed very similar results.  So pooled sensitivity of 97 percent, specificity 86 percent, and a good number under the ROC curve.  So the purpose of my study, we asked the question, can changes in the optic nerve sheath diameter distinguish between children with varying degrees of head trauma?  So not just saying, yes, they have increased intracranial pressure, yes or no, but can they actually be separated? So specifically a child who comes in with very minor head  trauma and essentially has a normal GCS, can it separate that group of children from those who come in with significant head  trauma and per standard of care have ICP monitors placed, but never actually experienced any increase in their intracranial pressure?  And can it separate that group from those who come in with severe head trauma, who have ICP monitors placed, and have significant increased intracranial pressure?  And then the fourth group that we looked at, again, could it separate those with high intracranial pressure from children who progressed to brain death?  This is a multi-center perspective observational study performed from 2014 to 2018. Children were enrolled both at Children's National Medical  Center and at the John Hopkins Charlotte Bloomberg Children's Center, and essentially all children being admitted to the PICU with head imaging consistent with traumatic brain  injury were eligible for enrollment. Essentially the only exclusion criteria we used is if you had significant eye injury or congenital anomalies of your eye or optic nerve.  So after enrollment, patients were separated into four diagnostic subgroups based on the degree of brain injury. So again, minor head trauma patients with normal GCS,  significant head trauma patients with normal ICP, meaning less than 20 millimeters of mercury, or severe head trauma with  elevated intracranial pressure, greater than 25 millimeters of mercury, or the patients who evolved to brain death. We then further stratified into three age subgroups.  So less than two years of age, two years to five, and then five to 16. So all patients had ophthalmic ultrasound exams within six  hours of admission, or within six hours of ICP monitor placement, or within 30 minutes of each clinical brain death exam.  So then we enrolled 76 patients, 20 in the minor head trauma group, 18 in the normal intracranial pressure group who had an average ICP of eight millimeters of mercury, we had  18 in the elevated intracranial pressure group who had an average ICP of 28 millimeters of mercury, and we had 20 brain dead patients. Of these, 45 were male, 31 were female.  And then you can see the age stratification down here in the various groups as well. So essentially what we found when comparing the minor head  trauma group to the group with normal intracranial pressure, we found that there was an optimal cutoff of 4.3 millimeters that separated these groups with a 78% sensitivity and 80%  specificity. And then when we looked at the average optic nerve sheath diameters between the minor head trauma group and the normal  intracranial pressure group, the minor head trauma group had an average of 3.7 millimeters, and the normal intracranial pressure group had an average of 4.3 millimeters.  And when you did the comparison of this, the p-value came out to be quite significant. So then we looked and we compared the normal intracranial  pressure group to the high intracranial pressure group. And again, we found an optimal cutoff to separate these two groups of 5 millimeters with a 66% sensitivity, 97% specificity.  And then when we compared the average optic nerve sheath diameter to the normal intracranial pressure from the elevated, we again found a significant statistical difference.  Now, surprisingly, when we actually compared the elevated intracranial pressure group, which on average had an intracranial pressure of 28 to 29 millimeters of mercury  compared to the brain death group, which had significantly higher increased intracranial pressure, we actually did not find a statistical difference in those groups.  So our p-value is only 0.27. So essentially, in the end, we said changes in the optic nerve sheath diameter could distinguish children with minor  head trauma from those with normal intracranial pressure from those with high. However, it can't distinguish a patient with elevated intracranial pressure from a brain dead patient.  So really quick, I was going to show some of the receiver operator curves. So again, this is looking at those with elevated  intracranial pressure versus normal with an area under the curve of 0.94. These are the minor head trauma from the normal intracranial pressure group with an area under the curve  of 0.83. And then looking at our patients who had elevated intracranial pressure versus normal intracranial pressure. So again, a really nice receiver operator curve of 0.94.  But then looking at the brain dead versus the elevated intracranial pressure, you know, it was 0.60. So really not great results there.  Interestingly enough, though, when we looked at the age stratification, when we limited our brain dead patients and those patients comparing them with the elevated intracranial  pressure, when we limited to just those children who were less than two years of age, the receiver operator curve changed quite significantly with an area under the curve of 0.83.  All right, so looking to the future, clearly we need to continue this work with a much larger sample size to further validate these results.  And in the future, the goal is really to perform stricter age stratification, so really separating the patients, looking at babies 0 to 6 months, 6 to 12 months, 12 to 18  months, and so forth, really correlating it more along the lines of the optic nerve growth curve in addition to looking at degrees of intracranial hypertension.  So in the end, we conclude the changes in the optic nerve sheath diameter can distinguish children suffering minor head trauma from those with normal intracranial pressure and those  from high intracranial pressure. However, it cannot distinguish between a child who is brain dead and a child who has increased intracranial pressure.  And this is my amazing research team. Thank you.  Great, we have time for a couple of questions. Dr. Pepe? Yeah, just one quick. Thanks again for a great talk and great looking at this concept. It's pretty cool, so thank you.  One question I had, though, is this just a one-time snapshot that you're reporting back versus like if you're watching  it sequentially over a couple of days as intracranial pressures might vary? And was the snapshot done around the same time, or is it at different points after the injury?  It's an excellent question. So we actually have several different snapshots serially that were taken when there were any significant changes in the intracranial pressure.  So when there were spikes in ICP, several shots were taken then when there were lower spots of intracranial pressure. So to actually distinguish them to fall into the normal group  where your ICP was normal the entire time, you had to stay normal the entire time. So if you had significant spikes in your intracranial pressure  throughout the entire time we were monitoring you, you stayed in that normal group. If you had significant elevations in your intracranial pressure, you  were in that elevated intracranial pressure group. If you fluctuated between the two groups, you were thrown out. OK. Keep that in mind, then. Thank you. Of course. Dr. Copeland.  Can I? A couple of issues just that I'm wondering if how much they affect your sensitivity and why the specificity is so much better.  How did you control for open fine nulls or open sutures? And you mentioned that in some of the human cadaveric experiments that they did lumbar injections.  Did you or have you thought about also looking for dilation of the thecal sac in the lumbar area? That's an excellent point to bring up.  I personally have not looked at thecal sac dilation. But it is an excellent point with that.  I'm sorry I blinked on the first half of your question. I was thinking too much. You intrigued me too much with the thecal sacs. So sorry.  Knowing that there's room for CSF redistribution into the lumbar thecal sac, if not also the cervical reservoirs. Yeah, no, absolutely. In the open font null concept.  And that's something that is actually highly controversial in the literature that really separates the pediatric group from the adult group  when you're talking about optic nerve sheath diameters. And a lot of people think that the open or have hypothesized and put a lot of data out there,  saying that when you have the open font null, transcranial dopplers are altered, optic nerve sheath diameters are altered. Because it's an open system as opposed to a closed system.  And I think that personally, that's why you really need to continue greater edge stratification and put the babies and the children who have open font  nulls in a completely different category than children who have already are working with the closed skull. Absolutely. Becky, if I could ask a question.  Perhaps, or maybe we can intuit this, is the optic nerve sheath diameter perhaps a better marker of edema than simple pressure measure,  as you see in kids that do have an open font nulls? And you're measuring even diameter changes with mild brain trauma that shows progression as they get worse.  So is this perhaps a functional measure of some degree of neuronal edema? It's an excellent point as well. So another controversy when you're  looking at optic nerve sheath diameters is, are you looking at an acute process or a chronic process? And even though edema can come and go pretty quickly  in the realm of ICU, we kind of look at that more of a little bit slightly more chronic, as opposed to acute ICP fluctuations and changes  that you're having because the patient starts seizing or because the patient needs more sedation and whatnot. And the literature does suggest that it  is more accurate in diagnosing acute changes, as opposed to those slower chronic changes. Excellent. Thank you. Thank you very much.  I'd like to invite to the stage Dr. Kenneth Shepke, who's going to present termination of benzodiazepine-resistant status epilepticus with parental ketamine administration.  Dr. Shepke?  Good afternoon, everyone. So I'm the state EMS medical director for the state of Florida. So for those of you out of town, I welcome you to our beautiful state.  I'm also the chief medical officer for Palm Beach County Fire Rescue. And this is a study that we put on with a couple of agencies down south, exactly what the title says,  looking at status epilepticus, where EMS has already given birth and metastasizm, and what can we do? So introduction, benzodiazepines remain the cornerstone  treatment for seizures, both in the hospital and out of hospital. However, EMS crews generally don't carry anything else. We don't really have a whole lot of tools.  Unlike the hospital, our hospital colleagues, it's limited options for a patient that we give benzodiazepines to, and they continue to seize.  So while ketamine administration in the hospital has been reported in a couple of case series, generally used later on after 24 hours or more of status  epilepticus, there's really no literature to guide us in the pre-hospital arena on this particular topic.  So our specific aims and purpose for this study is to report our paramedics' experience in administering ketamine, not as a primary drug, but as a rescue intervention  for seizures that are refractory to standard doses of parenteral midazolam.  So we had two EMS patients that were So we had two EMS systems in South Florida, a total of 58 stations covering a population of approximately  1.3 million people with approximately 130,000 EMS 911 calls per year. And we followed them for three calendar years, looking specifically at this subset of seizures,  the folks that we've given the standard dose of midazolam to, and then we would give the ketamine as a rescue dose. So we defined the refractory seizure  as persisting convulsions after 10 milligrams of cumulative dosing of midazolam in adults or the pediatric weight-based equivalent. Now, I know that that may seem like a smaller  dose than what you might do in the intensive care unit, but this is actually considered the standard of care in the EMS world ever since the Rampart study came out  a number of years ago. So the protocol, after you've given your dose of midazolam, if your patient is still seizing, we would then follow that up with parenteral ketamine.  And our dosage, our protocol, was 100 milligrams delivered either intravenously or through a humeral head IO in adults or 1 milligram per kilogram IV  or through a tibial IO in children. The alternative for if you don't want to go IV or IO, we did allow a 3 milligram per kilogram intramuscular dose for either adults or children.  And these are estimated weights, obviously, because we don't actually have a scale, so it's estimated weights. So primary and secondary outcomes.  Our primary outcome is frequency of sustained seizure termination prior to hospital arrival following the addition of ketamine. Now, whenever you're giving multiple CNS depressants,  obviously, you worry about respiratory drive and hypoxia. So one of our secondary outcomes was the presence of oxygen saturation, which we defined as less than 90% at any time.  And obviously, any invasive airway management following the use of ketamine. So what were the results?  Our primary results, 22 patients met the criteria. We had a mean age of 37.5 years, but the range was quite broad, 5 to 86 years. Interesting, 3 quarters of the patients  were actually female in our series. Ketamine was administered intravenously in 13 cases, intramuscularly in 6, and intraosseous in 3 cases.  Sustained termination of venasuline refractory seizures consistently occurred in over 95% of the cases. In fact, 21 out of the 22, which we found was remarkable.  We expected it to work, but we didn't really expect it to work this well. So this was really remarkable for us. Terminated seizures did not recur,  and our p-value was highly, highly significant. So there was one case, the one out of the 22, that the seizures did not terminate. But actually, even in that case, when  we look at our experience, that patient had been seizing continuously nonstop. And when we gave the ketamine, the seizures actually became intermittent and less severe  before they would break on and off. So if you talk just simply about an improvement, you can say it's actually 100% had some sort of improvement,  and led us to wonder, had we given that patient a second dose of ketamine, would we have terminated the seizure in that particular case altogether?  So this is our results in graphic form. We've actually, on this particular graph, the end says 23. That's because since we published the initial stuff  on this slide, we had an additional case, another 16-year-old male, status epilepticus, failed benzodiazepines. We gave ketamine to another success, no hypoxia.  We did have, on the hypoxia issue, though, we did have five cases in our total series, which is now 23 is what we're up to, who had hypoxia.  So looking at that, when we defined less than 90% at any time was five cases. However, in two of the five cases, we saw the hypoxia before we gave them the ketamine.  And I think you have to recognize that the biomechanics of your respirations are limited from the seizure activity itself, not to mention adding benzodiazepines on top of it.  In the remaining three cases, we saw the hypoxia occur after the ketamine was added to the case. However, we were able to normalize the oxygen saturation in all five of these cases.  Supplemental oxygen was used alone in three cases. We had brief use of bag valve mask pre-hospital only in one case. And we had one case where we used supplemental oxygen  in the pre-hospital field and without any other intervention. And by the time we arrived at the hospital, oxygen saturation had normalized.  But the emergency physician at that particular institution elected to intubate that patient for what they felt was the need to guard the airway, not really for hypoxia in that case.  So looking at the two EMS systems, we saw similar very good outcomes between both the EMS systems, which suggests that this is a reproducible phenomenon  across a multitude of EMS systems. So there are some limitations and some caveats on our study. First, it's a relatively small study size.  But we are talking about a very select population. We're not talking about all seizures. We're just talking about benzodiazepine refractory seizures.  So it's a pretty small end, but that's the nature of this particular type of study. Also, we're not looking at cases that received just ketamine.  Obviously, these are cases that already majority are on seizure medications because most of these cases had known seizure disorder. They received midazolam on top of it.  And then, of course, we throw ketamine into the mix. For the folks that were taking anti-seizure medications, anti-epileptic medications, there's really no way for us to determine  what their compliance was with their various medications. Pharmacological implications, though, we thought were pretty interesting. When you look at many seizure medications,  and benzodiazepines are certainly no exception, they work on the GABA receptor. And the GABA receptor, we know, as you go into status epilepticus, gets downregulated,  making medications like benzodiazepines theoretically less likely to work longer in status. However, the NMDA receptor gets upregulated during status epilepticus, making medications  that work on that receptor, such as ketamine, theoretically more likely to be effective. And in fact, that's what we found.  So there was one other incidental finding in this. Despite what we see in the literature, that the predominance of status epileptic cases are in male.  In this one, 3 quarters were female. So in our conclusions, among patients treated in the pre-hospital arena for persistent seizures refractory  to standard cumulative doses of parental midazolam, the addition of parental ketamine was very, very effective, with a relatively limited number of cases sustaining hypoxia, which  was readily reversed with supplemental oxygen in most cases.  So I want to thank you for your time. And I'll call Dr. Pepe up, one of my co-authors, and we'll take any questions.  Questions? How much time was it between the last dose of midazolam and the ketamine that was given? Yeah, that's a great question.  So we generally want to see that the midazolam has had at least five minutes to effect. Midazolam was all given parenterally as well. So we would give five minutes, and then  if there's no resolution of the convulsion or there is a repeat of the convulsion that had stopped, then they were allowed to give the ketamine. That's a good question.  Do you have any data at the same time frame before you start giving ketamine? Let's say the two years before, and how many cases that were resistant to midazolam and were  unable to control in the EMS on the way to the ER? So the question is, do we have, prior to this study, do we have an N of cases where we gave midazolam  and we had no other alternative and we just had to bring them to the hospital with Cedars? No, unfortunately, we don't have that retrospective data.  I mean, the reason I'm asking, because how do we know this is the effect of midazolam or ketamine?  So you're asking, how do we know that this is the effect of midazolam and not delayed reaction from? Not delayed. Not delayed reaction from midazolam,  but it's actually from the ketamine? Yes. I mean, it's a fair question, but we, yeah. So it's a fair question, but we have, this is a tiny subset of our seizures.  The vast majority of seizures stop with nothing. The rest stop with midazolam. And these are only the cases that EMS is now faced with. We've given everything we can give,  and we still have to transport to the hospital and the patient's still seizing. So either these are cases, there's two possibilities here, and I think the second is most likely.  Either these are cases that you're going to have to wait an unusually long time in our experience for midazolam to finally kick in, which in our experience, 130,000 patients a year,  we have a fair volume of cases. Or it's really the ketamine working via an alternative receptor that's causing the issue. In either case, obviously, allowing ongoing seizures  is a detriment to the patient, and we see not only these relatively benign side effects from the ketamine, but very, very high success rates  from adding the ketamine to the midazolam. Dr. Black. So how did you pick the dose of ketamine? How did we choose the dose? You know, that's a great question.  We did not have a lot of literature to guide us on this, and that's why this is one of the, at least from an EMS perspective, a relatively groundbreaking study for us.  So what I use, I researched the few case studies that were available in the hospital setting, in the ICU setting. And 100 milligrams or a milligram per kilogram  for intravenous was what a lot of those case studies had used. So it's really, it's almost throwing a dart at a thing, because we didn't have a whole lot to guide us in the EMS  world. So regarding the previous question, data from FTSE and from Rampart would indicate that if they haven't responded in a few minutes,  they're not going to respond to a benzodiazepine. You're absolutely right. Rampart, when I alluded to the 10 milligrams of midazolam,  Rampart is the study that I was referring to that set the standard dosing for EMS. Thank you for that. Thanks for your report. Ketamine is good, but not safe medication,  which may be used in a super-refractory status. What was the main idea to use ketamine before valproate, levitirazetam, and so on?  OK, so the question, first, I'll disagree that ketamine's not a safe medication. I'd say ketamine, from a pre-hospital standpoint, is one of the safest medications we have.  But why choose ketamine? It's because we don't have a full pharmacological armamentarium pre-hospital. We have a limited number of things. And the medications we choose are  designed to work in the street. Just as an example, the reason why we use midazolam and not Ativan, which a lot of our hospital colleagues do,  is that Ativan requires refrigeration, which you often don't have in EMS. And midazolam is absorbed better intramuscularly. And it's sometimes difficult to get an intravenous line  in the field. But to your point on ketamine, we're already using ketamine for multiple other reasons. So we carry ketamine. So we use ketamine for pain control, procedural sedation,  and control of violent, agitated patients. So since we already had a medication that our field personnel were very familiar with, we added this additional indication.  And we see pretty good success. Yeah, and obviously, the sooner, the better. The sooner we stop the seizures, the better. But I'm not sure why you feel that ketamine is not  a safe medication. It has a sealing effect on its respiratory. That's one reason we want to look at the hypoxia. Ketamine alone really does not diminish.  Unlike other anesthetics, it's unique in this ability. It does not block your respiratory reflexes. Now, when you add it to other CNS depressants,  you may see a change, which is why we're particularly interested in the hypoxia. Chris, the moderator. Yeah, I think the idea to use ketamine is quite interesting.  Have there been used hypoxia as one of your secondary outcomes? Was there anything about more post-ictal agitation being increased when you use that dose of ketamine in the field?  I'm not sure I heard that. Post-ictal agitation, was that increased? Oh, yeah, so you're asking now about side effects later on of the ketamine as they're reemergence reactions,  et cetera, et cetera. It's interesting you say that because one of the indications of giving midazolam with ketamine is that it's been shown to decrease those emergence reactions.  So that's not why we were doing that. We did the reverse order because midazolam's the drug of choice for seizures pre-hospital.  We don't have the ability to have all of the later on effects. I can tell you the ones. Many of these patients woke up with us and were just fine.  But some of those patients, they're coming around and their seizures had stopped, but we're not there for an hour or two hours after they dropped up to the emergency department.  So that's a limitation than the study is a longer term follow up. Thank you. Last question. Just to play devil's advocate about the safety of ketamine, I do agree it's pretty safe.  But how about those patients whose etiology of the seizure is basically by lesions or trauma that have elevated pressure in the cranium? I love that you brought that.  Because that is an old disproven myth in medicine that ketamine raises intracranial pressure. And it was based on some extrapolated data from folks that had intracerebral lesions,  had nothing to do with head trauma. And in fact, the American College, both the American College of Emergency Physicians and the American College of Surgeons  have come out on the use of ketamine. And neither of them find head trauma or increased intracranial pressure as a result of head trauma to be a contraindication to ketamine,  nor intraocular pressure for that matter. Great. Thank you very much. Thank you. Thank you very much. So in the sake of time, we're moving on to our next speaker.  Dr. Anil Guwati from PhRMAS is going to present his abstract, a phase two study to determine efficacy of sulfetatide in patients with cerebral ischemia. Dr. Guwati.  Thank you very much.  Okay, this study was, actually this discovery was done when I was professor and dean at Midwestern University. And then after this year, I resigned and joined full-time  as the chief executive officer of the company. So there are several investigators whom I'd like to thank.  Seema, she did a lot of work at Midwestern as a postdoc. Amresh, he's also doing at present. Then there are clinical investigators at various sites whom I'm thankful.  The conflict is that I'm the inventor and I'm an employee of the company also. And Manish, who is an employee and a stockholder of the company.  The funding was provided by Pharmaz Incorporated. So the whole idea developed many, many years ago. We have been working on endothelin  and endothelin B receptors are very, very high concentration in the central nervous system and their function was not known. Till about 10 years ago, we discovered  that they are integral part of development of central nervous system. So we thought if they are integral part of development of central nervous system,  why not stimulate them and see if we can treat some CNS disorders. So we selected IRL1620. That time it was the name. Then we went to, we found that this is highly effective.  Then we went to WHO and got the name approved as SOVA-Teltide. So now the name is SOVA-Teltide. So this is the structure of the compound.  Previously, it has been named as IRL1620, SPI1620, PMZ, but official name now is SOVA-Teltide. So these are some of the pictures of preclinical studies in rat  where we found that almost like 85 to 90% reduction in infarct volume was taking place and which could be blocked by a specific endothelin B antagonist.  So indicating these receptors are involved. So this is indicating that almost 85 to 90% of infarct volume reduced, the green arrow, and then ultimately the blue and the purple one  indicating that antagonist is completely reversing that effect. So this is an electron microscope picture showing that mitochondria are involved. So there is the first line,  the upper is 24 hours, down is seven day. The first panel is sham in between is vehicle and then SOVA-Teltide. We can very clearly see that mitochondrial health  is much, much better in SOVA-Teltide group. These are pictures of blood flow, cerebral blood flow, which we measured and we can see on the top panel,  vehicle and infarction in the middle panel. And then after seven day, hardly there is any blood flow in vehicle, but there is significantly much greater blood flow  in IL-1620 or SOVA-Teltide group.  How it acts, there are multiple mechanism. One is anti-apoptotic, anti-oxidative stress. These are some of the acute things which are happening  and there is a subacute or a delayed effect, which is taking place in neuronal stem cell, which are converted in neural progenitor cell and then into mature neurons.  And we have been now able to specifically  find out the mechanism. Initially, we were thinking that stem cells are being stimulated, but no, it is the neural progenitor cell  and the neuroblast cells, which are being stimulated. And the data here shows that NeuroD, which is a target for neural progenitor cell, is increasing while SOX and EDGE,  they are not increasing, which are for stem cell markers. So indicating that it is the neural progenitor cell therapeutics. So in that model, we saw that there is 85% reduction  in infarct volume. We saw that it stimulates the neuronal progenitor cells and has been shown to reduce the volume, infarct volume, increased blood flow,  anti-apoptotic and oxidative stress mechanism. So we embarked on study. We did the clinical studies in India first. And so this was one of the,  we had to register just like we have to do studies here. So there is CTRI registry. So that's what I'm giving the number and then NCT numbers are also there.  So assessment, the randomization procedures were followed and it was quadruple blind. So patient was blinded, care provider, investigator and outcome assessors were blinded.  Inclusion criteria, it's a big list of inclusion criteria, but we can go through those into details with the NCT number, which is already there.  Exclusion criteria, once again, there are a lot of details, but predominantly because this is a little bit of cardiovascular effects of this drug  was the first one which was described. And therefore, we took precautions that patients with heart diseases are not included in the study.  And of course, patients in which assessing or any cause of neurological impairment is because of any other causes. So those were, so these are the recruitment sites,  which there were seven sites from which patients were enrolled. And besides the stroke study, which is the top, we are also conducting Alzheimer's  acute spinal cord injury studies also. So there were a lot of patients who were screened and ultimately it's a pilot study, 18 and 18 completed the study, which is 90 day follow up.  And demographics are shown here, in which one thing I'll point out that 11 male and seven females, while in the other, so what I'll try to group, it was 15 males and three females.  So there is little bit of male female ratio, which is not equal.  So modified ranking scale, NIH score Barthel index, these are the parameters measured. We gave the dosing time after the stroke 24 hours.  So within 24 hours, the drug had to be given. And although statistically both were equal, but most of the patients in SOVA-Teltide group  got roughly about 20 hours after the onset of stroke.  So this is some of the data which is being presented. So there is a trend, I'll quickly go through these.  There is a trend that odds ratio is little bit higher in SOVA-Teltide treated groups compared to the control group. For example, when we go 30 day, 90 day,  there is odds ratio 2.2 and 1.6, statistically not significant. Modified ranking scale 2.7, 5. So one is significant, one is not. And Barthel index 30 and 90 day, 1.56.  And the other is at infinity because the patient was not there, so therefore significant. So overall, there is a trend. What we could see, there is a trend  that PMG 1620 or SOVA-Teltide had a better trend of better clinical outcome. Then we plotted each and every point which was recorded, and ultimately we saw that analysis  that day one to day three, day to do it, day six, day three to day six.  Basically from day one to day six, there was much, much more significant improvement in SOVA-Teltide group as compared to control group.  Same picture emerged in modified ranking scale day one to day six, and similar picture came out in Barthel index. So conclusion is that the clinical outcome  of NIHSS modified ranking Barthel index on day six was statistically significant improvement in SOVA-Teltide group while they were not in control group.  So there is indication that there is rapid improvement which is taking place. So then these are some of the number of patients responding.  So basically a change of more than six in NIHSS and change of more than two in modified ranking scale and change of more than 40 in Barthel index.  And once again, we see that there is some of the parameters  like modified ranking Barthel index, there is statistically significant improvement.  Then we further analyzed that, okay, let us see how many patients completely recovered. So we did a comprehensive analysis and found that NIHSS at 90 day  had there were more patients having complete recovery. Similar modified ranking scale did not reach the level of significance, but Barthel index did reach the level of significance.  So once again, indicating that there are comprehensive analysis showing that full recovery is taking place in some of the patients in neurological outcome.  So there are hematological, hemodynamic and other parameters. There was no significant change. There was no event of any drug adverse event reported  with the drug, no mortality took place and the drug was found to be safe and tolerated in ischemic patients. There are limitations, small sample size  and that can be skewed male, female ratio as I pointed out. But on the other hand, more patients in subadultile group got the drug later than the control group.  Subject population, although it is multi-centric, but it is only one country. So now we are putting applications together for a global trial.  And at present, another trial of 110 patient is ongoing out of which 47 patients have been enrolled. And we will see how the results come out with that.  So if the drug is approved, this will be a new class, a first in class kind of drug, which targets neurodegenerative diseases through endothelin B receptors,  which is a unique mechanism targeting central nervous system. Thank you very much. And I'll be happy to answer questions.  Great, thank you. So do we have any questions?  So I'm gonna get started.  Did you have a breakdown of patients treated with thrombolytic therapy between the two groups or other sort of customary treatments?  Actually, I have the slide. There is a lot of details of the treatment and their consistency is that almost equal number of patients receive TPA in both the groups.  And most of the time, the standard treatment like aspirin and all those platelet inhibitors, they were given. So there is quite a consistency between the two groups.  Dr. Boyk. So the photos of the mitochondria on EM are quite interesting.  It seems that a six-day effect would be too soon for the progenitor cells to have had much of an effect.  Do you have intervening time points between there and the 90 days to see what the actual pace of recovery is? Yeah. Actually, six days could be, you're absolutely right.  It could be too early for the progenitor cells to show any effect. So the acute effect could be because of increased cerebral blood flow. Anti-apoptotic phenomena is really very,  very convincing to us. And oxidative stress and other acute phenomena which is taking place. And the delayed phenomena could be progenitor cells.  We did monitor 30-day and 60-day and 90-day. And this trend is more or less 90-day trend  is at 30 days and 60 days, so.  Dr. Cohen. So at six days, you find a clinically relevant,  significant, and a small group improvement. At 90-day, you quickly show a slide says, more people were better, but is the statistical differences are lost.  Do you have any idea why these things don't, this difference didn't seem to persist out through 90 days?  And is this disheartening enough? Is this, are you showing us the raw data for a phase three trial? Or are you showing us the end of this drug, if I may ask?  The trial for 110 is already ongoing. So we are, up till now, we are encouraged by the trial which is ongoing. So there is no adverse event which has,  or serious adverse event which has been reported. This trial is small. This was more of a pilot study because a first-in-class, a new compound, giving for the first time in humans,  we really wanted to see what is going on. So there are trends. I'm not saying that this is working, but there is very clear trend that there is some efficacy  which we want to explore in detail. But do you have any potential explanations why that statistical difference in clinical improvement is lost between day six and day 90?  Yeah, so basically in stroke, there is loss of significance difference, but still there is significant difference in some of the parameters.  For example, Barthold index very clearly improved. And modified ranking scale, NIH assess at six days and after that it did not improve. And that's important.  Modified ranking scale is one of the very important indicators. So that is improving. Barthold index is a number of patients showing modified ranking scale  of improvement of two or more are quite high, and statistically high in the patients with subatomic type. Similar is Barthold index. So there is trend which is continuing.  It's not that we completely lose the significance. Great, thank you very much. So we're gonna have to move on to our next speaker. Go ahead, thank you.  I'd like to invite Dr. Sahar Zafar to the stage who is going to present on Burden of EEG Intraectal Continuum Activity predicts poor outcome in critically ill patients.  Thank you, I'm excited to be here. I am a neurointensivist as well as an epileptologist at Massachusetts General in Boston. And my research interest is in ICU EEG.  My disclosure is I have funding from NIH K23 as well as research support from Sage Therapeutics. Both were used to fund this work.  So as a background, ICU EEG is being increasingly utilized not only in the neurocritical care but also in the general critical care setting. And with increased use of ICU EEG,  we're diagnosing more and more not only electrographic seizures but other seizure-like periodic or rhythmic patterns that are now coming to be called  the Ictal-Interictal Continuum or IIC patterns. And these patterns can be seen in up to 40% of critically ill patients. And have been shown to be associated with worse outcomes.  Specifically, a higher burden of patterns has been shown to be associated with worse outcomes. So here, this was work done by the Columbia Group in subarachnoid hemorrhage patients.  And what they found was that an increasing burden of electrographic seizures was associated with an increased probability of poor functional as well as cognitive outcomes.  We looked at subarachnoid hemorrhage patients and looked at both electrographic seizures and other IIC patterns. And similarly found a dose-response relationship  where an increasing burden of these patterns was associated with increased probability of poor functional outcomes. So the objective of this work was try to expand  and see whether we can identify an association between these patterns in a larger critically ill population and whether they're associated with worse outcomes.  The challenge, of course, is that critically ill patients, their outcomes are determined also by the disease severity, the type of brain injury they have,  and the aggressiveness of the intervention. So our goal was to see if we can find an independent signal between these patterns and outcomes. And again, see if there is a clear  dose-response relationship. So we performed a retrospective analysis of 2,000 critically ill patients that were admitted to our center.  The inclusion criteria were adult patients, more than 18 years of age, who had at least 18 hours of continuous EEG monitoring. And we assessed our primary outcome measure  was their discharge MRS. We defined the EEG features based on consensus nomenclature that has been put forth by the American Clinical Neurophysiology Society.  These patterns include, in addition to seizures, generalized and lateralized periodic discharges and generalized and lateralized rhythmic delta activity.  And so we quantified the burden of these patterns in all 2,000 patients. We looked at the percent of the recording that had these patterns,  the frequency, as well as the spatial extent. The way we did this was we developed a tool for automated diagnosis of these patterns.  Or automated detection of seizure spectrum patterns, which we've published. We used that and then did raw review of EEG by three reviewers to quantify the burden  in all these patients. There's a lot of data here, but maybe I'll highlight is that this is what our cohort looked like. The median age was 60 years.  We had about a 51 is to one male to female ratio. The most common admission diagnosis was acute brain injury, which included TBI, stroke.  We included hypoxic ischemic injury in this category. And then we basically categorized our patients as those who had an acute brain injury.  Those who presented with seizures or status epilepticus in the absence of acute brain injury. So these ended up being patients  that have a history of epilepsy and came in with status. Patients who had a primary systemic disease without an acute brain injury. And these were primarily patients who had sepsis,  toxic metabolic encephalopathy. And then finally, patients who had a primary psychiatric illness. We looked at all their comorbidities based on the Charleston Comorbidity Index.  Most of our patients had hypertension and about 20% had a history of epilepsy and 20% had a history of prior or older cerebrovascular accidents.  We did have about 12% of our patients presenting with cardiac arrest. And then another 1% had cardiac arrest during their hospitalization.  This is their EEG data. So about 20% of these patients had a seizure on presentation. 20% had seizures during their hospitalization.  Although we looked at the presence or the burden of these patterns throughout their recording, one of the methods that we analyzed the burden was looking,  we did a sliding 12-hour window and found that the period which had the highest burden of these patterns. And here the burden is categorized  by the ACNS nomenclature. And so we found that most of our patients, the highest burden was they had either frequent or abundant concentration of these patterns.  Our median duration of EEG was 50 hours and about 80% of these patients ended up receiving anti-epileptic drugs. And then finally,  just a little bit more about their hospital comorbidities. About 20% at hospital acquired pneumonia, 16% developed septic shock. Most of our patients were discharged to rehab.  About 20% had withdrawal of care. And then we assess both their pre-morbid as well as their discharge MRS. And majority of our patients did have poor outcomes of MRS 4, 5, and 6.  So this is a busy, this is just a graphic representation of how we quantified IAC burden in this pattern. So on the x-axis here is actually 24 hours of EEG  and on the y-axis is all our 2,000 patients. Anything that is red, yellow, so red is seizures, yellow is periodic discharges, green you can see in here is rhythmic delta activity  and the gray is absence. And this is just zoomed in. So you can see that we basically had our EEGs marked and the burden quantified down  to essentially every minute of EEG recording. And so what we did was we used machine learning method to create an outcome prediction model.  We did logistic regression with LASSO regularization and included 22 predictor variables. Our clinical variables included age, gender, GCS,  both on admission as well as their worst GCS. We included only the physiologic component of the APACHE-2 score as a disease severity score as we'd already included GCS in the model.  We included whether or not they were on mechanical ventilation, whether they have a history of acute brain injury or a history of epilepsy,  whether they had a cardiac arrest on presentation or a seizure on presentation, and then our four admission diagnostic categories.  And then with regards to our EEG parameters, we included eight EEG variables. These included burden, a combination of burden and frequency,  a combination of burden and spatial extent, and then a combination of burden frequency and spatial extent over the first 24 hours and the same for over that 12-hour window  that had the highest burden.  And what this model selected was, we found eight features that were selected on this model. From all the various IAC burden definitions,  it was the maximum 12-hour burden that was selected, age, the initial and worst GCS, a physiologic APACHE-2 score, whether they had a seizure on presentation  or if they had a cardiac arrest on presentation, and then diagnosis of acute brain injury. And this is our model performance. So the area under the ROC curve was 77 for this model.  And as you can see, it did fairly well. This is prediction versus the actual proportion of poor outcome, and the model had a pretty good calibration.  And then finally, we looked at sort of whether there was a dose-dependent relationship as we had seen in the subarachnoid hemorrhage patients.  And again, we found that as the IAC burden within the APACHE with the highest concentration increased, there was about a 40% increase in the risk  or in the probability of poor outcome in these patients. So what we concluded is that it's really a patient that has a high burden, even if it's for a shorter duration.  IAC patterns are more likely to be associated with worse outcome than those who have a low burden. So someone who has 90% of their five-hour APACHE occupied by IAC patterns,  but the rest of their EEG does not have these patterns are more likely to have worse outcomes than someone who has 10% of their APACHE but for multiple days.  And we hypothesize that this association with worse outcomes is due to a metabolic crisis or supply and demand mismatch, as prior studies have shown,  some of this done by Dr. Vespa himself, that these IAC patterns have been associated with secondary brain injury as evidenced by decreasing brain tissue oxygenation,  increased lactate-pyruvate ratio, as well as bad hypermetabolism.  Limitations of this study, of course, are that it was retrospective. It's a single center, so it may not generalize as well. We're looking primarily at discharge outcomes  and not long-term outcomes. And then we intentionally did not include seizure medications in the model at this time, as we've considered them to be part  of the causal pathway towards outcomes. But that is one, and it is one of the challenges and one of the next steps that we hope to look at  is how to integrate treatment into this model. So to conclude, we found that a high burden of IAC patterns are associated with worse discharge outcomes in critically ill patients.  And then the next step, or question, of course, is what is the optimal treatment? Does treatment actually improve outcomes? And if we are looking or assessing treatment,  what are the treatment targets? Is it a decrease in the burden or a normalization of the EEG? These are some of the questions that I hope to address in the future.  So that's all I have. I'd like to thank my team, especially Eric Rosenthal and Brandon Westover, who are both mentors and colleagues, and were a big part of this project. Thank you.  We have questions.  Oh, thanks, Dr. Safar. So did you guys find a difference in the outcomes whether the times in the APOC was spent in generalized periodic patterns  versus lateralized periodic patterns? And also, for the ones who have generalized periodic patterns, what type of significance can we attribute when they have triphasic waves  and metabolic encephalopathy, but now they're becoming more periodic? So to answer your first question, we, at this point,  we found that patients who had just isolated GPDs had a similar association with poor outcome as those with lateralized. One of the challenges, unfortunately,  that we had is, and I think we all see this, is that there was a lot of overlap. So in patients with GPDs, we also saw LPDs and other rhythmic patterns,  which is why we tended to cluster them all together as one. But that is a good question, and that's one of the next steps in trying to see the isolated difference.  And then your second question was whether we looked at the type of GPD. Is that what your question was? Yeah, especially in patients who don't have acute brain injury,  so your sepsis patients with metabolic encephalopathy, where their clinical recovery often lags,  even though their numbers have got better, what do we make of the fact that they still have these sharply contoured or very periodic patterns of their GPDs?  Yeah, so again, a great question.  What we found was that, although the patterns had an isolate, they did have an independent effect or impact or predicted outcomes.  The model, the performance of our model is based not only just on these patterns, but all the other features that were in the model. And that model included acute brain injury,  but actually didn't pick up patients who had absence of acute. So the sepsis or toxic metabolic injury patients were not significant features in this model.  So at this point, I'd say it would be, when you see these patterns in association with acute brain injury, that would be more likely to be associated  with worse outcomes than with DME or sepsis-related encephalopathy. Thank you.  Oh, go ahead.  So great talk. I'm wondering, do you feel that the interictal continuum that you've seen is a marker of disease severity, or do you think it represents an insult in and of itself  that was worse than outcome? Yeah, that's a great question.  And I think that's what a lot of us are trying to understand is this, are they true secondary brain injury, or is it just an epiphenomenon? And I think what people are doing is,  that is why there's so much that we're looking into, whether on invasive monitoring, is there increased metabolism, is there secondary brain injury? There clearly is an association.  I think the best way for us to answer that question is if we can figure out, and this is a very hard thing to do, is whether treatment improves outcomes.  I think that would be something helpful to guide us. And in the absence, I think that's why if you look at any study, even within our own database,  we found that the approach to treatment varies across physicians. And I think there was a multicenter study between Yale, Emory, and the Brigham,  which looked at the practices of EEG monitoring, and what they found was that while EEG monitoring is pretty similar across centers,  the AED prescription practices across centers vary a lot. So the hope is, that's one of the next steps, is really to try and use this natural variation  in treatment and do some sort of comparative effectiveness research to understand whether treatment improves outcomes. Because doing a randomized trial  in this population is challenging. Dr. Black.  Frank Drizzley and colleagues suggested that there is a ceiling after which more seizure activity does not influence outcome. For the different components of the IIC,  can you tease out whether there are either dosage or ceiling effects that might say, well, we could look at treating this one, but that one doesn't matter?  Yeah, so we haven't yet, you know, and that's sort of the next step, is looking at treatment. One of the challenges, one thing that we did do is, and this is, again, within just  the subarachnoid hemorrhage group, is looked at patients who have IIC patterns and received AEDs versus those who didn't.  Or actually, no, just within patients who received AEDs, we've compared patients that had an improvement in these IIC patterns versus not, and we found no difference in outcome.  So that's sort of, you know, but again, it's, there's, it's confounded by, there's so much confounding by indication that that's one of the biggest challenges  in answering this question.  Great, thank you very much. Thank you, that was great.  So our next abstract is gonna be presented by Dr. Elizabeth Crago from the University of Pittsburgh. She's giving this on behalf of her colleague, Dr. Yousef.  The abstract is titled Mean Flow Velocity in the Middle Cerebral Artery Predicts Poor Outcome After Subarachnoid Hemorrhage. Dr. Crago.  So we thank the society for this opportunity. Yes, I'm not Dr. Yousef. Unfortunately, he had an unexpected inability  to travel here from international, so unfortunately he couldn't be here to present his work, but the patients are from the University of Pittsburgh,  so I'm happy to present it for him.  We have no commercial conflicts of interest. All of the data was obtained through funding from the NIH, which has gone on over the past decade  and more on subarachnoid hemorrhage patients. So we know that transcranial dopplers have been utilized to detect cerebral vasospasm after subarachnoid hemorrhage,  and they've been endorsed in clinical guidelines for the management of subarachnoid hemorrhage, but whether serial transcranial doppler monitoring  improves outcomes or predicts outcomes is unclear.  So we know that they're a rapid, noninvasive, real-time measure of cerebral vascular function, relatively easy to do at the bedside, does not require taking the patient anywhere,  measures flow velocities in the arteries and assesses flow changes, focal stenosis, or even embolic signals, often using mean flow velocities or Lindegaard ratios  as well as the embolic signals, and they can infer cerebral autoregulation, cerebral vasoreactivity, or even neurovascular coupling.  The purpose of this analysis was to look at our subarachnoid hemorrhage population and to look at whether just elevated mean flow velocities,  using a cut point of 120 centimeters per second,  are predictive of poor functional status and mortality after subarachnoid hemorrhage.  And we used a prospective longitudinal descriptive design. Our patients are adults 18 to 75 years with some evidence of blood hemorrhage on their CT scan,  of course with available TCD data and also outcome data, and without chronic neurologic deficit or a non-aneurysmal or aberrant in subarachnoid hemorrhage.  We did daily transcranial Doppler's up to 14 days. The risk period for the vasospasm, which is generally the indication for looking at transcranial Doppler's.  These are usually done in the morning. They're done by trained personnel. We used a TCD Pioneer 8080,  and we looked at, again, at middle cerebral artery velocity using that 120 centimeter per second, either right or left side as our cut point. And then we looked at MRS at discharge  as well as 3 and 12 months after subarachnoid hemorrhage using a dichotomy of good versus poor, and we also looked at the incidence of death.  And we used a multivariate logistic regression controlling for age, and our clinical severity for this analysis was Hunt and Hess as they are often correlated with outcomes.  So we had 504 subjects who met this criteria with a mean age of 53, predominantly female, which is common in subarachnoid hemorrhage,  also predominantly coiled, which is also common, and then with Hunt and Hess grade 1 through 5,  with the most being a Hunt and Hess grade 2 or 3.  And so here's what we saw in outcomes, not completely unexpected. So at the time of discharge, 55% of the patients were scored as a poor outcome,  but that improved then at 3 months and at 12 months.  We had a 10% death at discharge, which was 15% at 3 months, and then 17% at 12 months.  Our mean days of Doppler were 9 days, 9.1 to be exact, but 9 days, and where we started to see the elevation above that 120 centimeters per second was at about day 6.  And so when we looked at these elevations, a single elevation or more above 120 centimeters per second  relative to discharge and 3- and 12-month MRS, we saw a significant association with odds ratios of 1.8 to 2.2 for those patients who had elevated mean flow velocities,  as well as an association with death at both discharge and 3 months, and certainly the same trend at 12 months, although it did not reach statistical significance.  So to our knowledge, this is the first evidence of actually using these Dopplers to predict poor outcome after subarachnoid, and there's actually a paucity of data  about the relationship of TCD monitoring and post-subarachnoid hemorrhage outcomes.  Interestingly, there was a fairly recent study that looked at HCUP data, a 10-year period across the United States, and found that although we say  that transcranial Doppler is suggested, recommended in subarachnoid hemorrhage, only 1.5 percent of patients with subarachnoid hemorrhage  actually have transcranial Doppler monitoring.  Now that does differ, it would appear, between teaching institutions and non-teaching institutions, and certainly some data that was collected by Neurocritical Care  found that over 70 percent of reports to Neurocritical Care showed transcranial Doppler use, but of course that also involves neurointensivists  and a much more intense ICU experience.  Now they did say that there was some increase in use after the recommendations came out, but perhaps just the lack of frequency with which these Dopplers are actually used  may add to the fact that there's not a lot out there.  We also know that TCD can be a tool to monitor patients with poor neurologic exam in whom we can't get good neurologic assessments,  and then, you know, we would also add that perhaps there's some prognostication that we could use these for as well.  There have been a few studies, though, that have looked at whether TCDs are effective in improving outcomes, not necessarily predicting, but this was a German study  that deliberately looked at patients  who had transcranial Dopplers used as a part of their care versus those who did not have transcranial Doppler use. They saw absolutely no difference  in patient outcomes between groups, and that included both mortality, Glasgow outcome, and MRS. It was a relatively small study. There were 50 in one group and 39 in the other,  but they did not find a difference, and somewhat similarly in a study out of the UK and Ireland that looked at centers that traditionally  that used transcranial Doppler versus those that didn't,  they also reported actually,  I shouldn't say they reported the same. They found that there was actually worse outcomes in patients in centers that used transcranial Dopplers  as opposed to those who did not. And they actually modeled this using a model that had agent injury using WFNS, and then  they even had a more robust model that included some other variables, and it continued to skew towards those centers who did not use transcranial doppler.  Of note, there wasn't any difference in the severity of injury on the patients coming in. Obviously they controlled for WFNS.  That was one of the first things I thought, perhaps these patients were skewed because they were going to, you know, the sicker patients were going to these other institutions.  Now the limitations to our study, obviously this was a first pass at looking at this. We did not account for interventions that were used for elevated TCDs.  That was not the purpose of the study. And we did not look at Lindergaard Ratio or evaluate the reasons for TCD.  We used this as a clinical indicator of something going on intracranially. And of course we didn't analyze all vessels. We looked at mean flow in the MCA.  But we did find that elevated mean flow velocity of the MCA, it doubles the odds of poor functional status and death in patients with subarachnoid hemorrhage.  Certainly we still need comprehensive studies which evaluate the utility of serial TCD monitoring  to guide our clinical care, to evaluate cerebral vasospasm, and then perhaps to predict outcomes in these patients. Thank you. Great. We have an opportunity for some questions.  Dr. Brack. That's very nice. In most studies that have looked at TCD, about half of the patients have a diagnosable delayed neurologic deterioration and the other don't.  Can you split your data into those who had diagnosable DND or DCI compared to the others to see whether the velocity itself is a marker?  Or is it simply that you pick up a large number of people who go on to have DCI? Correct. And that is the next step. So our hope in exploring this data is to do that very thing.  Thank you, Dr. Brack. Thank you for your presentation. The fact that you cited in that last slide the limitations, really the variability in  practice in terms of what we do with these mean flow velocity elevations, as well as the fact that not every vascular territory was incinated, makes it very challenging to  say anything about TCD because our clinical practice, just like even the previous speaker on practice of treating seizures in the ICU, is quite variable.  So it's a little bit of a dangerous slope to be on to say the transcranial Doppler ultrasound  doesn't change outcome. It's really what we do with diagnostic information that changes outcome, as with anything like an Artline or something like that or an ICP monitor.  Thank you. I agree with you 100%. And certainly, you know, others who look at subarachnoids, we have all of this vast amount  of transcranial Doppler data and trying to dissect out, you know, where to make cut points,  where to use them, how it directs clinical care, and, you know, I think is quite difficult.  And I agree with you to say that it doesn't improve how we treat patients is very suspect. Right. Dr. Riggs.  Did you also take into account what the resistive indices and the pulsatility indices were in correlation with the mean flow velocities? I'm sorry. I didn't understand. Sorry.  Did you take into account what the resistive indices and the pulsatility indices were in correlation with the mean flow velocities? Good question.  And no, not in this small analysis. But yes, we will consider that. Yeah, excellent. Good. Any other questions? All right. Well, thank you very much. Thank you.  Okay, so our next speaker, Dr. Munjao, will be coming to the podium. Dr. Munjao is going to present, hold on one second, app failure here, the abstract entitled  Machine Learning, Predicting Mortality and Morbidity in Children with Neurologic Injury in the PICU. Dr. Munjao is from Pittsburgh? Correct. Pittsburgh Children's Hospital.  Thank you so much for the introduction. Hello everyone. My name is Neil Munjao. I'm a third year PICU fellow at the Children's Hospital of Pittsburgh.  I'm a child neurology graduate and also a current T32 postdoctoral scholar at the University of Pittsburgh.  And I'm honored to be talking to you today about the use of machine learning in predicting outcomes for children with neurological injury in the PICU.  There are a variety of models predicting outcomes in the ICU, both in adult and pediatrics, mostly looking at mortality but also morbidity.  In patients with neurological injury, however, we don't have such a validated model for either mortality or morbidity.  The ideal model presumably comes with significant complexity, and so when asked why use machine  learning, I would focus attention on its eventual use as a clinical decision support tool to augment the clinician's capability at the bedside, not to replace it.  We hypothesize that PICU patients with acute neurological injury are at higher risk for new morbidity and mortality than the general PICU population, and that machine learning  models will improve prediction over logistic regression for this subset. Our data comes from the TOPIC study.  In this study, they prospectively enrolled 10,000 patients over a two-year period under the age of 18 at seven collaborative ICUs and built multiple logistic regression models  predicting mortality and morbidity based on admission physiologic data, namely PRISM scores.  We created a secondary analysis of the subgroup marked as having acute neurological injury.  We applied multiple modeling techniques, including logistic regression, to both learn about the  applicability of each technique to this population and to determine relative performance. To properly train and test these models, we split the data set into training and validation  test data sets. The initial analysis was fairly straightforward. We wanted just to demonstrate the injury burden in these patients.  We identified 1,860 patients with acute neurological injury, approximately one-fifth of the total collection.  6.2% of the patients in the subgroup died, more than double that of the general population.  There was 8.2% with new morbidity that was almost double the 4.3% of the general population.  The overall burden of morbidity and mortality was more than one out of every seven patients. Here are the results of our model.  We planned a priori two separate study populations, modeling both survival versus death, as well  as those who developed either new morbidity or mortality versus those who mostly returned to baseline on the bottom.  Given the high burden of morbidity demonstrated in the prior slide, focusing on the latter model, in my opinion, is the more clinically relevant one.  Looking first at the top graph, one can see that while logistic regression had a very  good area under the receiver operating characteristics curve of 0.92, it's 0.50 average precision or positive predictive value suffers in a very imbalanced data set like this one.  With imbalanced data, one would argue that precision is the more important accuracy metric. Here achieving even small improvements of 10% to 15% can be helpful.  Of the advanced machine learning models, random force and extreme gradient boosting performed best with slightly inferior AUROCs with a small but significantly improved precision.  The basic neural network model did not do as well and the support vector machine was inadequate to study this data.  Now looking at the bottom graph, all models took a performance hit when looking at the harder criteria of new morbidity or death versus no new morbidity.  Despite the drop in the AUROC to 0.81, both the random force and gradient boosting models nearly maintained their average precision equivalent to the logistic regression.  As a whole, these results demonstrate a rough equivalence of these minimally tuned machine learning models to the prism-derived logistic regression model with a trend towards small  improvements in precision. Here are the receiver operating curve characteristics on the left and the precision recall curves on the right.  As a brief review, the ideal precision recall curve is the one that gets closest to touching  the top right corner, representing perfect positive predictive value with perfect sensitivity. Now to get into a little more detail, we begin with logistic regression.  This model was built identical to the topic analysis and performed fairly well. There are many advantages to logistic regression, namely that it is a well-established procedure  that provides a simple formula for classification as its output. However, some limitations do exist.  These include most notably limited modeling of complex nonlinear interactions between variables and often require fairly exacting parameter selection and cutoff selection,  as demonstrated in the topic logistic regression model, which used human-derived prism scores  as input, as opposed to looking at the raw physiologic data as our machine learning models do.  Regardless, for small or static datasets, it is often among the best-performing models. The first advanced machine learning model is the random forest.  The idea is essentially the idea of democracy. The majority rule of a large number of less-sophisticated, uncorrelated models will outperform the individual.  The algorithm is simple. You generate many simple decision trees, prune the collection, and then average out the votes of all of the trees that remain.  Just to get an idea of what these trees look like, here is one such decision tree. Now you can just imagine this being amongst a group of hundreds or even thousands, depending  on the model. Pay close attention to the red area. This is a zoomed-in area of the root of that specific decision tree.  Probably the most important thing to note here is that the cutoffs for each value is machine-generated, not hand-picked by the programmer.  As such, it does not have to abide by the medical definition of, say, for example, hyponatremia, and the algorithm defines its own useful criteria.  As a review, the performance of this model was pretty good, with similar AURSC and mildly higher average precision to the logistic regression model.  The F1 score and the Matthews correlation coefficient are other quality measures that try to combine the ideas of sensitivity, specificity, false positive rate, false negative rate,  into a single score that can be used to compare models. Overall, the random forest model has many advantages.  Its simple form performs well on imbalanced data, there is minimal hyperparameter tuning required, and it handles missing data moderately well.  The prime disadvantage has been empiric. Its peak performance is lower than that of other models and complex datasets, but our  dataset demonstrates that when used correctly, it can perform well. One result I want to particularly highlight is the feature importance analysis.  Random forest models lend themselves well towards this, and they do so by looking at all the decision trees and being able to derive which features are empirically important in  the vast majority of them. Here is a list of the most important features in our analysis. Note that many of these are actually previously recognized secondary CNS insults, such as  hyperthermia, thrombocytopenia, hypercarbia, hyperglycemia, and hyponatremia. Just as interesting are the less weighted features of sex and age, as well as the pupillary  status and GCS at admission. Also note that all of the heavily weighted features are modifiable features and potential future targets for therapy studies.  I want to again emphasize here that there was no pre-programming of normal ranges, nor did we do any further feature selection on the data that was collected in topic.  The cutoff and feature weighting was strictly determined by the model. The next model is extreme gradient boosting.  This is another similar machine learning model that has gained a lot of popularity in the last five years in the general machine learning community. It works iteratively.  Each weak learner is given a corrective term to make it better in a next generation, then it is tried again. This red box here outlines what that corrective term looks like.  They use regularization, that's where the extreme term comes in, to control for overfitting. Reviewing the results, this model performs fairly similarly to the random forest model  with similar AUROC and average precision to the random forest model and a slightly decreased F1 score in Matthew's correlation coefficient.  Overall, the extreme gradient boosting model is a good model choice that seems to perform well for a wide variety of datasets, including imbalanced data like ours, but is marginally  slower to train than the random forest, more prone to overfitting, and poorly parallelizes. The final model to discuss is the neural network.  This is by far the most talked about and used model in the machine learning world for a variety of domains and with good reason.  These models, when set up correctly, can theoretically model almost any true relationship. There are multiple types of neural networks.  We use a simple feedforward network here in our analysis, but recurrent neural networks will play a strong role in the move to real-time analytics.  The neural network model, again, as reviewed, did not perform as well as the other advanced machine learning models, nor even as well as logistic regression.  There are a variety of reasons for this lack of performance. Probably the biggest one is that neural networks are known to require a lot of data, as we've  seen in other talks today, meaning hundreds of thousands or even millions of data points, as opposed to the 2,000 rows that we had.  They require much more hands-on tuning of hyperparameters to perform well and are very slow to train. Advantages, though, are big.  For the most complex tasks, thinking of computer vision, natural language processing, car driving, they have the best performance, and once trained, they perform very quickly.  I think this study underscores a general principle of machine learning. Use the right tool for the right job.  The trendiest device may not be the right one for a more modest data set. To conclude, we verified that patients with acute neurological injury are at higher risk  for mortality and for new morbidity, unsurprisingly. We demonstrated that independently recognized secondary CNS insults can be identified using  these approaches with minimal a priori knowledge, and that many of these insults are potentially modifiable and certainly studyable.  We demonstrated that some of our early machine learning approaches can match the performance  of logistic regression in a single-shot model and show potential for improvement in precision.  Finally, many new challenges arise in these approaches that have to be carefully addressed through algorithm design.  For future directions, I think it's helpful to think about a bigger picture. We have a rich, growing electronic health record data set at home.  There are many pros with using EHR data. There's more data, more variables, and many more times available.  The data is structured and automatically collected rather than manually entered as topic was.  The challenges are many, too, though, mostly arising from the complexity of the non-curated data set.  The current study is our first attempt at developing a neurologically focused morbidity-mortality prediction tool from admission data, a neuroprism of sorts.  We would like to reproduce and validate these models using electronic health record data. Next, brain injury phenotyping is the idea that heterogeneous diseases like TBI and stroke  can be parcellated into subgroups or clusters to demonstrate specific disease phenotypes and potential target treatments to those phenotypes rather than the disease as a whole.  Not every TBI patient behaves the same way, and they don't always deserve the same treatment. Finally, the big target for this work is in real-time monitoring.  As mentioned previously, this is where more technically challenging models that have a sense of time, such as recurrent neural networks, are most helpful.  The ability to use ongoing data, such as vital signs, EEG, neurologic exam, imaging to refine our models will remain my main target in the years moving forward.  Once the scaffolding is complete, my hope is that this tool will lead to a monumental shift in the way we diagnose, monitor, and treat patients with and without neurological  injury in the ICU. With that, I want to thank you all for listening. Thank you again for the privilege of speaking here, and thanks to my mentorship on this  project at the University of Pittsburgh and the Children's Hospital of Pittsburgh, who are giving me protected time over the next year and a half to advance these ideas. Great.  Thank you.  OK, we have some time for a couple of questions. So did you incorporate any physiology into this model? Because there are hundreds of thousands  of data points for heart rate, blood pressure, intracranial pressure, et cetera, results of pupillary results, and all that kind of stuff. Any of that factored into this?  Yeah, so obviously the limitation of this, this was somewhat of a first pass for us, was that we used data from a public use data set of a study that was already published,  which was the topic study. And so the model that they developed used PRISM scores, which is a pediatric risk model that used mostly lab values, but also  specific, very limited physiologic data of, in the neurologic sense, pupillary status and GCS. But they also included basic vital signs as well. One of the keys with topic, though,  is that it's a single shot data set, meaning that they looked at admission data and weren't looking at changes over time. And that is, I think, exciting and interesting to look at.  But I certainly think moving towards using real-time physiology is really where I'm excited about this. Good, good. Other questions? All right, well, thank you very much. Thank you.  OK, so our next speaker is Dr. Owen Leary. And Dr. Leary is going to present an abstract entitled, Outcome Prediction with Computer-Assisted Volumetry and ABC  Divided by Two in Traumatic Intra-Axial Hemorrhage. Dr. Leary.  Thank you very much to the chairs of the conference and the session moderators for this opportunity. This project is a culmination of a lot of my undergraduate work.  I'm an incoming medical student at Brown Medical School. I've done a lot of this work on analyzing imaging biomarkers of traumatic brain injury outcomes using the PROTECT 3  data set under the guidance of Dr. Lisa Merck.  This work was funded by the same grants and agencies that funded the initial PROTECT 3 trial, which is published in 2012.  So, first of all, this study is mainly looking at the relative predictive value of two different paradigms for estimating the volume of a traumatic brain hemorrhage.  If you look at the three different panels in this image, panel A represents a solitary interparenchymal lesion, B represents a subdural, and C represents an epidural lesion.  And you can see these lesions are very different. The topographies of the bleeds are very disparate from one another, and you can see that the  three-dimensional reconstruction we achieve below each from the two-dimensional CT slices  by manual segmentation give us a more accurate representation of the topography of the bleed than a single estimator would across all types of bleeds. Sort of some background.  So again, we apply user-guided boundary segmentation to these lesions in order to generate a three-dimensional  segmentation of the lesion, which captures the surface topography accurately and allows us to calculate a precise volume.  And this could be a powerful prognostic factor in TBI, the problem being that not many studies  have accrued such a large number of cases and lesions as are available through the PROTECT3 dataset.  The limitation of applying this in the clinic now is that it's time-consuming and it's manual,  but as we saw in the last presentation, machine learning and artificial intelligence are rapidly  accelerating models for not only clinical predictors based on non-imaging biomarkers, but also imaging-based models for doing this automatically.  The current standard for estimating these lesions, despite what I just showed you, remains  the ABC divided by two estimator, which essentially assumes the lesion to be a perfect geometric ellipsoid.  This was validated decades ago using spontaneous intracerebral hemorrhage, which are much more  uniform and homogenous as a disease phenotype than we see in traumatic brain injury, especially in the moderate and severe TBI cases that we see in PROTECT3.  You can see in panel B, the two orthogonal axes that are measured here, there would also be a C that goes down into the Z plane.  You multiply those by each other and then divide by two. That's the approximation. That doesn't work very well for a crescent-shaped panhemispheric subdural as you see here.  That's sort of the underlying hypothesis here, is that this is not a very good measurement tool for this type of lesion.  That said, it is still used for TBI in the absence of something that is better and readily available.  Some of this is preliminary data that we have, we're working on publishing this now, showing  that there is, in fact, a significant difference between the absolute value measurements of the two methods.  ABC divided by two significantly overestimates the volume of these bleeds if they're larger than 50 cc's and underestimates them if they're less than 50 cc's.  So computer-aided volume estimation is more predictive of six-month outcome than ABC divided by two and traumatic extra-axial hemorrhage.  We know this based on results we presented last June. We took the 260 cases from the PROTECT3 data set in which the largest lesion was either  an epidural or subdural hematoma on the baseline CT and we measured it using both of these methods.  We then conducted area under the curve analysis like we saw in the last presentation in order to compare predictive value.  So first of all, there is a significant association of volume in these lesions with outcome, which is not unanticipated.  The volume of the bleed does matter for how the patient is going to do. What's interesting is the more accurate estimation, the CAVE estimation, is significantly better  as a predictor of both mortality and functional outcome, which is a composite measure based on the GOSE and the DRS score based on the AUC analysis.  So this is sort of our background going in. The question here is whether for the other part of the bleeds in the data set, the intra-axial  lesions that are more ellipsoid in their phenotype, whether there is also a difference in predictive value between the two methods for these. So this data set is similarly sized.  We have 273 traumatic intra-parenchymal lesions, IPHs and which are sort of the solitary ones and then the multifocal traumatic contusions as well.  And these were measured again by both CAVE and ABC divided by 2 with the same outcomes that we're looking at in the extra-axial lesions. So first part of our results.  So using ordinal linear regression of CAVE versus ABC divided by 2, this illustrates that there is a strong association with outcome again.  So lesion volume is significantly related to outcome. Using both metrics in terms of mortality, GOSE and DRS, again, not surprising volume  does predict how well the patient is going to do at six months. Since we have edema captured in the CAVE measurement, while it's not necessarily captured as a separate  component of the lesion in ABC divided by 2, we also assessed whether the presence of edema on baseline CT was associated with mortality or functional disability and found that it  was not. For the first part of our AUC analysis, we did the same thing that we did for the extra-axial  lesions comparing CAVE versus ABC divided by 2 in terms of predicting outcome for all the intra-axial bleeds considered in this analysis.  CAVE and ABC divided by 2 are both predictive of mortality without significant difference between methods in the AUC and in terms of six-month disability.  So unlike our extra-axial lesion data set, in the intra-axial lesions, CAVE did not provide  a significant improvement in prediction of outcome over ABC divided by 2, the more geometrically presumptive method.  And then, so based on that, we also, so we know, and you saw this slide earlier, how you can really see the topography of the contusion being multifocal in the CAVE method but not  in the ABC divided by 2 where it's sort of a uniform ellipsoid. So we also did a sub-analysis here where we looked at just the ABC divided by 2 measurements  and just the CAVE measurements and compared how they predict outcome among just the, among traumatic contusions versus IPH.  So we found that while CAVE better predicted mortality for contusions versus IPH, ABC divided  by 2 did not, which suggests that even though there may not be a clinically significant, in terms of prognostication at least, benefit to using a more accurate method in intra-axial  lesions like the ones used in this data set, CAVE is still picking up on a nuance in the  more complex surface topography and composition of the traumatic contusion phenotype as opposed to the solitary IPHs.  So in summary, this rounds out our analysis of these 533 largest lesions from the 181 baseline CT scans in the PROTECT-3 data set.  Similar to in our SDH and EDH data, volume of multifocal traumatic contusion or solitary  IPH is indeed significantly associated with outcome, both in terms of mortality and functional outcome at 6 months.  The presence of edema on baseline CT is not independent, is not independently associated with outcome in intra-axial lesions based on our univariate regression.  And then unlike in our extra-axial lesions, CAVE does not offer significantly better outcome prediction than ABC divided by 2 in terms of AUC for intra-axial bleeds.  This is not necessarily surprising given that the lesions are more adherent to the geometric ellipsoid that is assumed by ABC divided by 2.  That said, CAVE does offer a better outcome prediction in multifocal traumatic contusion versus IPH, while ABC divided by 2 does not.  This supports the hypothesis that as machine learning methods for automatically segmenting and achieving accurate volumes does become more readily available to clinicians, there  may be benefit and value to having a more accurate estimation of volume as a single factor which might be incorporated into predictive models, prognostication models, along with  physiological variables and other components. Thank you. Great. Thank you. Do we have some time for some questions? So the CAVE, where does it function best, do you think?  What shape, or is it shape-dependent at all? I don't think the accuracy of CAVE is shape-dependent because it's, so there's lots of softwares  you can use to do this, but in all of them, you're essentially drawing on each slice the boundary of the bleed, and then the software builds, by interpolation, a surface topography  based on all of those user segmentations. It's perhaps a little bit easier to delineate the boundaries in, well, there's pros and cons.  In the intraaxial lesions, you can really see the boundary of the edema and or hemorrhage in the tissue, but then in the extraaxial lesions, you can really see the boundary.  So it's not hard in either way. It's certainly very time consuming. So to really translate something like this, we need a machine learning based method of.  Yeah, that was going to be our next question, in that this is still manual operation dependent, if you will. The volume calculation is automated, but the actual input is manual.  The hand drawing, yeah. And how applicable do you think that this is to larger data sets? I mean, we're going to have a number of data sets coming  online now at the NIH, once TRAC and Center TBI deposit their information in int beer. Do you think that this is going to be something that's going to be usable  for these large data sets? So I think it's entirely possible, because as was referenced in the last presentation, you need sort of a large ground truth  data set in order to drive a really good machine learning based model for these sorts of things. So someone going through for Center TBI or TRAC TBI  and manually annotating all of the bleeds in those data sets, I'm not sure if that'll happen or if that'll necessarily be feasible. But you could certainly use something  like this as the training set and then test it on those bigger data sets and have sort of an iterative machine learning process for improving.  One other, one last concept, and this is not so much a question but maybe just a little direction. So in the MISTI database, we've been looking at location of hemorrhage,  not just the volume, but the location. And so it may be interesting to put together location and volume in your series to see if that, because we all  postulate, oh, yeah, they have a structural lesion in that key area. Oh, prognosis isn't going to be so good. But we don't really have great data on that.  It's really interesting you say that, because actually our next presentation is all about exactly that, location and temporality. And our next direction is sort of  Good. Well, that means we have to move on. OK. All right. Thank you very much. OK. Great. Thank you. So our next speaker is Dr. Tyler Harder.  Dr. Harder will present the abstract entitled, Patient-Specific Prognostication After TBI Is Related to Bleed Phenotype and Anatomic Location. Dr. Harder. Good afternoon.  My name is Tyler Harder. I'm a first-year medical student at the Warren Alpert Medical School of Brown University. And I'm here today as part of Dr. Lisa Merck's research group.  I'm here to talk a little bit on what my colleague was just hitting on at the end of those questions about prognosis based on phenotype location and laterality of lesion. Let's see.  OK. So I have no disclosures, financial or otherwise. And this research was funded by the NIH and the National Institute of Neurological Disorders.  All right, a little clicking. So to start, one of the big issues with prognostification is the heterogeneity of baseline CTs. There's a variation based on your patients  and mechanism of injury that can really complicate it. And furthermore, there's not really a well-defined relationship between the biomarkers of CT  imaging, that is, phenotype, laterality, and anatomical location. And there's also kind of a lack of cohesive data sets that pull in a lot of the outcomes  data with annotated imaging, which is something that we're fortunate to have through the PROTECT data set, which I'll talk about in a second. And just for the sake of definition,  extraxial lesions are epidural and subdural hemorrhage, and intraaxial are intraparenchymal hemorrhage and traumatic contusion.  All right, so this data came from the PROTECT-3 clinical trial, which was initially a test of progesterone in reducing functional outcome deficits and mortality  within patients with moderate to severe traumatic brain injury. And to enroll in this trial, patients had to have moderate to severe acute non-penetrating  traumatic brain injury with a Glasgow Coma Scale between 4 and 12. Everybody received a baseline scan and six-month follow-up mortality and functional outcome measurement.  For today, we used 517 of the 882 enrolled patients, and with a total of 533 lesions. The reason there's more lesions than cases  is because we used the largest bleed to identify the case, and some of them had very close lesions.  Like bleeds, so we used both of those for the sake of analysis today. And to the bottom left, there's a breakdown of all the lesions.  All right, so in our analysis, we compared bleed phenotype, location, and laterality, and looked for a relationship with six-month outcomes and mortality.  And our functional outcomes measures were the Glasgow outcome scale, the extended version, and we counted a poor outcome as being a 5 or lower,  so severe disability to vegetative state. And under the DRS, which is the disability rating scale, we counted above 12, so severe to extreme vegetative state.  So the first analysis that we did was just basic laterality. So regardless of bleed phenotype, does laterality have a relationship to poor functional outcome and mortality?  And our finding was that no, there's no strong relationship here. And when we did a sub-analysis of our extra-axial bleed, so independently subdural hemorrhage,  epidural hemorrhage, and intra-axial IPH and traumatic contusion, the same thing showed. There was no relationship there. However, when we broke this down by region, so lobe,  we found that there was a relationship in the left temporal region. There was a relationship with poor functional outcome and a trend towards higher mortality.  So this was the bulk of our analysis. We essentially took the PROTECT-3 data for extra-axial and intra-axial lesions, and we broke them out into location and phenotype of bleed.  And then we assessed for a relationship between our functional outcomes and mortality measurements, and this is what we found. So starting in the temporal region,  we can see that there is a relationship between both mortality and functional outcomes in all of our measures. And I should also clarify that poor outcome  is a combination of both a GOS-E below 5 and a DRS above 12. So we can see temporal had a relationship with all of them. Now if we look over at basal ganglia and thalamus,  we can see that there is a relationship between poor functional outcome but not with mortality in that region. And going the other direction, looking  at posterior fossa and cerebellar hemorrhage, we see that there is a relationship with mortality but not necessarily a strong relationship with functional outcome.  And going back to over here, we have extra-axial lesions. We can see there's a relationship between mortality but not necessarily functional outcome.  And going the other way, intra-axial lesions, we found a relationship with poor functional outcome. The other interesting finding is when we remove the outliers from our analysis  of the frontal and occipital lobes, we found that there is no relationship with mortality and functional outcome. And we hypothesize that's likely due  to the size of the real estate in those lobes. And you can probably tolerate a lesion a little better there. All right. So in conclusion, temporal, intra-axial, parietal,  infratentorial, and deep lesion, basal ganglia thalamus lesions are associated with morbidity and mortality, while frontal and occipital lesions are not.  In the PROTECT dataset, we found extra-axial lesions to be correlated with higher mortality, while intra-axial lesions showed a relationship with poor functional outcome.  And with regards to laterality, we only found that left temporal lesions had a relationship with poor functional outcome. And we're really excited with this data  because, as my colleague said, this is kind of one of the first analyses that we've done on outcomes data within our imaging dataset.  And we're fortunate to have neuroradiologists annotated CTs at baseline, as well as serial follow-ups paired with all of the outcomes data.  So this summer, we're hoping to start processing our subarachnoid hemorrhage in CT data. And we're looking forward to presenting a bigger scale of this. Thank you for your time.  Thank you.  So we have a couple of minutes for questions. So how about brainstem? Was brainstem included as a localization for hemorrhage here?  We didn't include a brainstem for that, for this analysis. OK. Yeah, it may be that they're not, the density of that in the PROTECT trial isn't so, isn't that high.  They may have not met entry criteria. But go ahead, please.  I thank you very much for your fine presentation. What were the causes of deaths in your cases?  Cranial deaths or extracranial deaths? Causes of death, are they brain-related death or some other cause of death, hemodynamic, et cetera? Do you know?  I know that we have the data on that. But I didn't account for that in this analysis. Oh, for my PI, it's for everything, including brain death, as well as any other cause.  Yeah, and that was death at, remind us, that was death at what period of time? Six months. That was six months death. So some of that is late deaths, so truly predictor.  And these were the admission CT scans? Baseline. Yeah, did you, and do you have data on subsequent scans? Yes, we have the serial scans.  Yeah, because one of the things that's evolving is the hemorrhagic expansion issue. And given that PROTECT was an emergency treatment trial,  and perhaps one hypothesis was very indirectly, it's a neuroprotection, but maybe it played some role or didn't play a role in hemorrhagic expansion,  it may be worthwhile looking at the subsequent scans. Yeah, it's actually interesting you brought that up. We have done some analyses on our subdural and epidural  data set on expansion specifically, and we haven't found any significant relationships between mortality and those expansions. Good. Other questions?  Well, then I think we will bring our session to a close. This was the top abstracts in the neuroscience section for this year.  I think we had some exciting and very informative speakers and science today. Thank you for your attention, hanging in there. Have a good rest of the meeting and a good evening.

Video Summary

The first study examined the relationship between Ictal-Interictal Continuum (IIC) patterns on EEG and outcomes in critically ill patients. It found that a higher burden of IIC patterns was associated with worse functional outcomes, indicating that these patterns could be used as a predictor of poor outcomes in these patients.<br /><br />The second study developed a machine learning model that utilizes intermittent alpha-coma (IAC) patterns on EEG to predict outcomes for critically ill patients with acute neurological injury. It found that patients with a high burden of IAC patterns had worse discharge outcomes, potentially due to metabolic crisis or supply and demand mismatch in the brain. The study also found that anti-seizure medications did not improve outcomes for patients with IAC patterns.<br /><br />Both studies provide valuable insights into the prognostication and treatment of critically ill patients with neurological injury. However, further research is needed to validate the findings and understand the underlying mechanisms behind the associations observed.

Asset Subtitle

Neuroscience, 2020

Asset Caption

The Society of Critical Care Medicine's Critical Care Congress features internationally renowned faculty and content sessions highlighting the most up-to-date, evidence-based developments in critical care medicine. This is a presentation from the 2020 Critical Care Congress held February 16-19, 2020, in Orlando, Florida, USA.

33-Optic Nerve Sheath Diameter Changes in Children With Varying Degrees of Intracranial Pressure

Becky J. Riggs

34-Termination of Benzodiazepine-Resistant Status Epilepticus With Parenteral Ketamine Administration

Kenneth A. Scheppke

35-A Phase II Study to Determine Efficacy of Sovateltide in Patients With Cerebral Ischemic Stroke

Anil Gulati

36-Burden of EEG Ictal-Interictal Continuum Activity Predicts Poor Outcome in Critically Ill Patients

Sahar Zafar

37-Mean Flow Velocity of the Middle Cerebral Artery Predicts Poor Outcome After Subarachnoid Hemorrhage

Elizabeth Crago

38-Machine Learning Predicting Mortality and Morbidity in Children with Neurologic Injury in the PICU

Neil Munjal

39-Outcome Prediction With Computer-Assisted Volumetry and ABC/2 in Traumatic Intra-Axial Hemorrhage

Owen Leary

40-Patient-Specific Prognostication After TBI Is Related to Bleed Phenotype and Anatomic Location

Tyler Harder

Meta Tag

Content Type Presentation

Knowledge Area Neuroscience

Knowledge Level Advanced

Membership Level Select

Membership Level Professional

Membership Level Associate

Tag Intracranial Pressure

Tag Seizure

Tag Neurotrauma

Year 2020

Keywords

Ictal-Interictal Continuum

IIC patterns

EEG

functional outcomes

predictor

poor outcomes

machine learning model

intermittent alpha-coma

IAC patterns

acute neurological injury

discharge outcomes

metabolic crisis

supply and demand mismatch

anti-seizure medications

prognostication

	Society of Critical Care Medicine 500 Midway Drive Mount Prospect, IL 60056 USA Phone: +1 847 827-6888 Fax: +1 847 439-7226 Email: support@sccm.org	Contact Us About SCCM Newsroom Advertising & Sponsorship DONATE	MySCCM LearnICU Patients & Families Surviving Sepsis Campaign Critical Care Societies Collaborative	GET OUR NEWSLETTER
© Society of Critical Care Medicine. All rights reserved. \| Privacy Statement \| Terms & Conditions The Society of Critical Care Medicine, SCCM, and Critical Care Congress are registered trademarks of the Society of Critical Care Medicine.