All right. Hello, everyone. Thank you for attending today's Ask the Expert webinar on Beyond Closed Loop Stimulation, New Research Avenues and Clinical Insights Utilizing Ambulatory ECOG from RNS. Please note this webinar does offer continuing education credit after completing the evaluation. You'll be taken to the evaluation immediately after the webinar. You'll receive an email when the on-demand version will be available. Before we get started, I would like to take a moment to acquaint you with a few features of this web event technology. At any time, you may adjust your audio using any computer volume settings that you may have. And on the right-hand side of your screen, you'll see the Q&A window. There is a large window which holds all your sent messages and a smaller text box at the bottom where you'll type in your questions. To send a question, click the text box and type in your text. And when finished, click the Send button. All questions that you submit are only seen by today's presenters, and your questions will be responded to in the order in which they were received and will be addressed at the end of the presentation. I'd like to introduce today's moderator, Dr. Wolfgang Mulhofer, AES Online Education Committee member. Thank you so much for the kind introduction, Ashley, and good afternoon and welcome, everyone, to today's webinar. Before I introduce our speaker here, I have a quick disclosure to make due to unfortunate circumstances of Memorial Day weekend travel. I'm actually joining you right now live from Atlantic Airport, so I didn't want to miss any of this important educational event, so I really apologize for any audio issues or background noises that you might hear from my end. With that, though, it's my pleasure and honor to introduce Dr. Vikram Rao. Dr. Rao is an associate professor and chief of the Epilepsy Division at UCSF, where he also holds a distinguished professorship in neurology from the Ernest Gallo Foundation. He obtained his M.D. and Ph.D. in biomedical science from UCSF by a neurology residency at MGH and returned to UCSF for his fellowship in clinical neurophysiology and epilepsy. Dr. Rao's research focuses on seizure forecasting using chronic electrocardiography in patients who are implanted with the brain-responsive neurostimulation device, or short, RNS system. He's a nationally and internationally recognized expert in use of the RNS system. As a physician scientist, he has led multi-center collaborations and published extensively in high-impact journals on clinical and research applications of this device. For his research work, he was recently awarded the Dreyfus-Penry Epilepsy Award of the American Academy of Neurology and has been appointed as a fellow of the American Epilepsy Society. I'm very excited to have him as a guest speaker on the AAS Ask the Webinar series today. His presentation is titled, Beyond Closed-Loop Stimulation, Clinical Insights and Research Avenues with RNS Chronic EEG. And with that, Dr. Rao, the stage is yours. Thanks for that nice introduction, Wolfgang. Thanks for everybody for being here. Wolfgang was a fellow with us a few years back, and he has way more embarrassing material on me than that, so I appreciate you leaving all of that out. It is really an honor and a pleasure to be here today. This topic is near and dear to my heart, and I'm excited to share this with you. Really, what I'm going to be telling you about today are diagnostic applications of a therapeutic device. And my goal is to convince you that long-term recordings of brain activity that are possible with this device can be useful to those of us in clinical epilepsy. So let's jump right in. These are my disclosures. I have consulted for Neuropace in the past, but today's content is entirely my own. These are our learning objectives. So we're going to be talking about electrocorticography, or ECOG, data that's captured by the Brain Responsive Neurostimulation System, or RNS system. We'll talk about how this data can inform different dimensions of our management in clinical epilepsy. We'll specifically talk about how RNS ECOG can reveal temporal patterns of brain activity, like cycles, how those patterns can be leveraged for applications like seizure forecasting. And I'll highlight some emerging research avenues that I think are particularly exciting that really highlight the power of the chronicity of recordings from the RNS system. We're going to frame our discussion today in terms of how these data can address common challenges in clinical epilepsy. But before we do that, let me just sort of take the temperature of participants here, and we'll do a poll to gauge the level of experience that folks have with the RNS system already. So I'll ask you to vote. My level of experience with the RNS system is A, I've never heard about this device. B, I've heard about it, but I have not personally managed patients treated with RNS. C, I manage a few, maybe five or fewer patients with RNS. D, I manage a moderate number, six to 20 patients, or E, I'm a pro, and I have managed more than 20 patients treated with RNS. I'll give folks a second to vote here. All right. So a couple of votes still trickling in. But it seems like there's a good distribution of experience, and at least everybody's heard about the device, so that's good. And I see that a lot of people have heard about it but maybe haven't used it yet. And so hopefully this will be useful to people at all levels of experience. And I'll be curious to get perspectives during our discussion as well. Let me start, though, by telling you some things that I'm sure everybody already knows, which is that epilepsy is one of the most dynamic, maybe the most dynamic disorder in all of neurology. Its hallmark, of course, is recurrent, unprovoked seizures. Seizures are brief, typically self-limited events, and they occur sporadically. They seem to occur at random. And that temporal dynamism of seizure occurrence means that our evaluations of individuals who suffer from seizures are intrinsically prone to temporal undersampling. Temporal, I mean time, not temporal load. Most of our evaluations occur here in a place like this, an epilepsy monitoring unit. This is a workhorse of our presurgical evaluation where we learn so much about our patient's epilepsy. But for all of its virtues, we know that the EMU is also a bit of a limited environment, isn't it? We monitor patients by video EEG for a handful of days, relatively brief recording duration. We monitor them in an inpatient environment, which is not exactly a natural place to be. And we certainly apply some unnatural seizure-provoking maneuvers like sleep deprivation, like abruptly stopping anti-seizure medications like photostimulation and so on. The concept of temporal undersampling is illustrated by this cartoon here. If we imagine somebody who has bitemporal epilepsy, they make seizures from both temporal lobes, and you brought them into the EMU for, let's say, five days on one week, and you recorded three left and two right-sided seizures, well, you would conclude that the person has bitemporal epilepsy, and you would be right. But what if just through a scheduling quirk, which we have a lot of these days during the pandemic, they just came in the next week for the same amount of time, five days, and you record, by chance, three left-onset seizures? Well, you might conclude that they are unilateral left, and in that case, you would actually be wrong. Temporal undersampling can also mislead us if we're trying to study a cyclical phenomenon. So I've just drawn a hypothetical signal of interest here, and let's say you wanted to characterize this and understand it. And what you did is you did a short timescale recording, like a 30-minute routine EEG. Well, it might be hard to see any cycles at all in such a short-duration recording, but perhaps over a few days in the EMU, you would say, oh, look at that, there's actually this fast red oscillation that I can see. But only with a very long-term recording, a wide temporal lens, such as chronic EEG, would you see that there's also a slow blue oscillation present in this signal. And actually, that blue curve was present all along. It was in your routine EEG. You just couldn't see it. You weren't looking over a sufficiently long period of time. Now, a second clinical challenge that we face commonly is that of monitoring disease activity. I am envious of our colleagues in endocrinology who can monitor glycemic control easily and quantitatively in people with diabetes by measuring hemoglobin A1c, or our friends in cardiology who have implanted loop recorders that can look for rare arrhythmias, or they can ask patients to serially monitor their blood pressure at home to monitor hypertension. In epilepsy, we rely heavily, almost exclusively, on what patients tell us, on their self-report. And that is fine. Self-reported seizures are a gold standard in clinical trials, but we know that they have limitations, right? People may write down events that they think are seizures that maybe don't have an electrographic correlate, and conversely, we know that electrographic events can occur that people don't remember or don't perceive or otherwise are unable to document. A final challenge that we'll talk about today is that of seizure unpredictability. These are the results of a survey that was done by the Epilepsy Foundation a few years back, and they asked over a thousand patients one question. They said, of all the things that are bad about epilepsy, what's the worst? And there were many answers, as you can imagine, but chief among them, the number one answer in this survey anyways, was unpredictability. And that is the simple fact that I really can't tell a patient that I'm seeing whether their next seizure is going to be five minutes from now, five days from now, or five weeks from now. I can't distinguish those possibilities for a patient. So I leave them in a state of looming uncertainty, and uncertainty is stressful and stress is disabling. And that really is our current clinical paradigm, isn't it? Somebody comes to my clinic, they do the best they can providing a self-report. They may or may not keep a diary of seizures. If they do, they might show me a timeline of events like this, and I would say, all right, well, that looks pretty random, so here's what we'll do. Why don't you take this medicine twice a day, every day? I'm offering a static treatment for what I know full well is a very dynamic disorder, and I know that I'm asking them to incur the side effects of that medicine, dizziness, fatigue, cognitive slowing, whatever, all the time, even though the risk of seizures is probably not uniform at all times. And that's where I think some recent technology might be able to help us, or at least that's the case that I'll make today. Judging by the poll, it sounded like most people had some familiarity with the RNS system, but just so that we're all on the same page, let me remind you that we call it the RNS system because it is a system of interconnected components that are shown here. So that starts with the cranially implanted neurostimulator. That is what houses the electronics in the battery. That's connected to two lead wires that have four electrode contacts. Each of the electrodes are placed intracranially at the site or sites where seizures come from. Those electrodes enable the device to do essentially two things. One is it can continuously sense or listen to brain activity, not continuously store brain activity, mind you, but continuously sense it and look for patterns of abnormal activity. And number two, in response to detections of patterns of abnormal activity, such as those that herald the onset of seizures, the device can respond by delivering electrical pulses to the seizure focus or foci to hopefully reduce seizures over time. That's the concept of closed-loop stimulation. Now, patients have a computer at home and a wand that they can use to download data stored on the device and transmit it over the Internet to a secure online data bank. Providers have a tablet that they can use to program settings on the device, detection settings, stimulation parameters. And patients also have a circular sort of donut-shaped magnet that they can swipe over the neurostimulator that forces it to store an ECoG at that moment. And that can be useful to sort of judge the electrographic correlate of a clinical symptom. How well does this work to reduce seizures? Well, the answer is it works pretty well. We have recently published nine-year long-term treatment trial data that showed us that the RNS system, like other neurostimulation devices, the efficacy tends to improve over time. By year nine in this trial, patients experienced a median reduction in seizure frequency of 75%. Subsequent to that trial, a real-world outcome study that was done across eight academic centers was completed, and that showed that in the real world, outside of the confines of a clinical trial, people are doing at least as well as they did in the clinical trials and perhaps getting those outcomes maybe even a little bit better than the trials and getting those outcomes a little bit earlier, maybe not having to wait for nine years to achieve those kind of outcomes. Perhaps that reflects our cumulative experience using this therapy now for many years. So we use the RNS system, of course, for what it's labeled for, for a therapeutic indication to reduce seizures. But part and parcel of normal clinical use of this device involves the acquisition of a lot of data, and that is really what confers us with this very powerful diagnostic function, which is what we're going to be talking about today. So what are these data that are collected by the RNS system? Well, it comes in many forms. So one form is that the device stores snapshots of brain activity in the form of four-channel, bipolar-montaged electrocorticograms sampled at 250 hertz. These eCOGs can show us lots of things. They can show us interictal discharges. They can show us recordings of seizures. They can be stored in response to patient magnet swipe, and so they can show us what brain activity correlates with clinical symptoms. We also get graphical representations of different forms of data, data like detections of interictal epileptiform activity, what I'll call IEA. That basically means detections made by the device of patterns that we tell it to look for. Those detections can be plotted to reveal patterns, trends, responses to other interventions. And all of these data, mind you, are from patients who are ambulatory in their home environment, normal activities on medicine, so maybe a more naturalistic environment than the inpatient world. So what I'll contend today is that all of this data, if used appropriately, can help us move beyond our clinical paradigm currently of patient self-report only and empiric treatments and maybe move towards more data-informed management. That is not to say that these data are perfect or that it solves every problem, just that I think it's really an unprecedented opportunity in our field to make decisions predicated on more quantitative data than we typically have access to. So we'll talk about some examples categorized as shown on this slide, and I'll bring in some examples from my own clinical practice. Maybe it will remind you of some patients that you look after also. We'll start by talking about monitoring disease activity, that first challenge we mentioned. So we don't have a hemoglobin A1c for epilepsy, do we? But we have something, so let me give you two examples. One is this gentleman. He's 69 years old. He's a veteran. He has bitemporal epilepsy, and he did very well with RNS. He actually surprised all of us by the outcome that he achieved. He came in and would report saying, I don't think I've had a seizure for the last 10 months, and he was a little sheepish about that because he couldn't be 100% sure that he was in fact seizure-free. So we looked at his data together, and sure enough, detection counts of IEA were downtrending during this time, and we saw no stored electrographic seizures, no what we call long episodes during that time frame. So that was a nice sort of corroboration of his clinical report and gave him some confidence that in fact he wasn't missing events at home. This can go the other way too. This is a young woman, 19 years old. She had seizures coming from Broca's area, and she was definitely not seizure-free with RNS. I would say she had a moderate response, but she had achieved some modicum of stability until her life took a turn for the worse, and she had a lot of psychosocial challenges. She started using drugs, stopped taking her anti-seizure medications, et cetera. And in that setting, detections of IEA went through the roof, more detections of electrographic seizures. And when she would come to clinic, we could look at these data together, and I wouldn't claim that this is what helped her turn her life around necessarily, but I think it was meaningful to her to see, I've been doing well for a long time, and then sure enough, I can very clearly see that this is a turn for the worse. And perhaps that helped her right the ship over time. Now, RNS chronic EEG or CEG can also be a useful adjunct to seizure diaries and people who keep diaries. We use this term long episodes, which is a jargon term. All that means is a detection of abnormal activity made by the device that is sustained beyond a pre-specified threshold duration, say 30 seconds or 40 seconds. And in many patients, although not all, those long episodes, those sustained detections of abnormal activity are a very good proxy for electrographic seizures. So even if we don't store every seizures because we don't store continuous ECOG, we can store a fairly continuous count of electrographic seizures in many individuals. Sometimes long episodes are an excellent proxy for clinical seizures. Sometimes it's one-to-one. Even if it's not, even if long episodes, say, outnumber clinical seizures, they tend to co-vary. This is work from Mark Quigg and colleagues showing that long episodes co-vary with seizures from diaries and long episodes have a high negative predictive value, meaning that periods of time when you don't see long episodes are very likely to be periods of time when a patient is clinically seizure-free. It turns out that being able to quantitatively keep tabs on disease activity and epilepsy during a time when it's hard for people to come in for in-person visits, such as a pandemic, is very useful. This is a nice essay penned by Casey Halpern, where he makes that case, that remote care options are increasingly needed in our field given the challenges of seeing patients in person. So the fact that I can fire up that online data bank and look at trends or long episodes or magnet-triggered ECOGs and communicate those results to a patient, even from afar, is very useful. What about measuring the effects of all the other therapies that you do to improve your patient's epilepsy? For example, adjusting medication timing. So the data here are all from one of my patients. Each row is a histogram that's averaged over one week, and it shows the distribution of detections of interactivity by hour of the day, from midnight to midnight. And each row here is one week, and it goes back over the last 12 weeks. You see that week over week, very consistently, interictal activity goes up around 10 p.m., peaks at nighttime, and then kind of abates again in the morning. This correlates almost perfectly with this individual's sleep patterns. So then this is a very common sort of circadian distribution that we see. We were able to use these data to asymmetrically dose his anti-seizure medicine, more medicine at bedtime, because presumably he needs higher blood levels of that medicine at bedtime, or during the night, and lower doses in the morning to mitigate side effects during the daytime. We can visualize also the effects of things like rescue medications. This is a different patient with a regional left temporal parietal seizure onset zone. Long episodes were a very good proxy for electrographic seizures for her, and when we looked at the distribution of long episodes over many months, we found that when she would have one seizure, she often would have multiple seizures, often four or five seizures at once. The horizontal dashed line here is a count of one, and you see that on many instances she doesn't just have one, she has many, so this is a tendency for seizure clustering. We gave her intranasal midazolam and instructed her to use it as a rescue medicine after the first seizure, and subsequently she continued to have sporadic seizures, but they would not often cluster, and you can see that very clearly in the data here, so it made us feel like this was quite effective. We can also look at the effectiveness of newly started anti-seizure medications. Sometimes we see a precipitous drop in detections of interactable activity in counts of electrographic seizures. These examples show Clobizam at the top and Levotiracetam at the bottom, just as examples where this very profound decrease in activity recorded by the device was seen very early after starting the medicine, and work from Larry Hirsch and Ron Qureshi and others has shown us that when you see this, this is a good prognostic sign that that medicine may be a good one for the patient for the long haul. We can also correlate detections of interactable activity with behavioral and lifestyle factors, like a young woman who drank huge amounts of caffeine on the weekend that was suspected of worsening her seizures, and the authors of this case report nicely showed that detections were higher on weekends when she would drink a lot of caffeine, lower when she would abstain. Looking at detections in relation to a medication that was suspected of worsening seizures, like mirtazapine, looking at detection counts before and after a dietary therapy, ketogenic diet, and even looking at effects of cortical desynchronization from VNS using RNS recordings. Now, we had a patient who was initially implanted with RNS and had good benefit, but very much wanted to be seizure-free, and so went on to have also implanted thalamic DBS, and so that afforded us a very unique opportunity, there aren't many of these patients, to use RNS-CEEG to gauge the effects of thalamic DBS, and we found some really interesting things. We had electrodes with RNS in the hippocampus and temporal neocortex, and we found that during DBS stim on times, this was an anterior nucleus of the thalamus, that hippocampal local field potentials were suppressed, that the functional connectivity between hippocampus and overlying temporal neocortex was modulated, and that hippocampal spike rate was suppressed during DBS on times. So, kind of a neat window into the effects of one neurostimulation device using chronic ECOG from another. As epileptologists, one of our principal charges is sort of pinpointing seizures in the brain where seizures come from, and that means we have to lateralize and localize seizures. Some of the earliest work on seizure lateralization using RNS data came from David King-Stevens, who's now at Yale. David looked at 82 patients from the RNS system clinical trials who were implanted with bilateral mesial-temporal leads, and he asked a simple question. He said, from the moment you record one seizure from one side, how long on average is it until you record at least one seizure contralaterally, in other words, to prove that the person is bilateral? And the answer is, it took a surprisingly long amount of time, 13 days, longer than most EMU admissions, although, admittedly, these patients were not having their medications tapered. In the patients, the 11 patients for whom there was suspicion that seizures were actually unilateral, two-thirds of them ended up having seizures recorded bilaterally with long-term recordings. And conversely, of the remaining patients, 71 who were thought to be bilateral, that's where they were implanted bilaterally, a small but significant minority, 13% of them, only ended up having seizures coming from one side. So that's an important group of individuals, and taken together, the originally presumed seizure lateralization changed 20% of the time. So we looked at this before, a patient who comes in on one instance and has bilateral seizures by chance, on another instance, seizures only from one side. But you can also imagine a person who gets taken off medications and makes seizures from all over the place in that very unnatural setting, but perhaps on medicines at home in an ambulatory environment, only has seizures from one side. I call this the functionally unilateral patients. Medicines are more effective at suppressing seizures on one side. Now, the beauty of chronic ECOG is that we can see all of these dynamics. And it's really important to do that, because sometimes you can actually take people on to a definitive, potentially curative therapy like resection. So for example, if you saw very asymmetric seizure lateralization, and you then proceed with a unilateral resection on the worst side, Larry Hirsch followed a series of 24 patients treated in this way and found that 70% of them were seizure-free with this, and even of those who were not seizure-free, the worst among them still had a halving of their seizure frequency. So this is pretty profound, because this group of patients, by definition, were not deemed to be great candidates for resection on the front end. But were able, by virtue of establishing a seizure lateralization ratio, over time were able to move on to a curative resective surgical procedure. That begs the question, though, how long you have to follow somebody to feel like you have faithfully established the seizure laterality ratio. So how long is that? Sharon Chang, a very talented neurology resident here at UCSF, a soon-to-be graduated, asked this question and looked at 13 patients from our center who had bilateral mesial temporal leads, and what she found is that the seizure laterality ratio that you observe converges to its long-term value after about eight months of recording, or after capture of about 50 seizures. Now, there's some nuance to that. That's not a hard and fast rule that we should monitor for eight months and then move on to surgery. This provides some statistically-based bounds for a ballpark of how long you should record before you can feel like you have faithfully estimated the long-term value. Probably a month or two is too short, maybe three years is unnecessarily long. Now, seizure localization is nothing that I would oversell with RNS, because our spatial sampling is incredibly limited, right? We have two forecontact leads, so I don't want to make more of this than need be. We have had two instances where RNS ended up being quite useful for this. These were patients who had either subdural grids or stereo EEG for phase two monitoring. They sat in our EMU for three weeks, lots of spikes, but no seizures. Without ICTL recordings, we didn't feel like we could take them to a definitive resective surgical procedure. In these instances, we implanted them with RNS and we put the two leads that we had in the two areas where we had the highest pretest suspicion that seizures arose from. We sent them home. One patient had their first seizures 49 days later, and the other 225 days later, and RNS allowed us to get ICTL recordings and we could see which of the structures that we suspected the seizure started in. Now, you might say, well, how do you know that a seizure didn't start right next door and spread into one of the leads? I would say, I don't know, but I still think that this was our hypothesis anyways and these recordings helped us distinguish between the most likely possibilities, and it is certainly better than nothing. In fact, the patient on the left went on to have a resective surgical procedure and became seizure-free. So, that would be like this, going back to our cartoon, where you don't record seizures in the EMU during phase two, but with chronic ECOG, you can observe a rare seizure. Okay, spell characterization, also a bread and butter kind of enterprise for us in clinical epilepsy. Here are a few vignettes that kind of highlight the power of RNS-CEG in this regard. So, we had a young man with autism who had RNS and actually did very well with this therapy. His convulsions had been controlled for years with medicine, but his teachers called us from school and said that he had uncharacteristically a flurry of convulsive spells in the classroom and this kept happening. We eventually gave them an RNS magnet and said, take care of him first, but if you can, go ahead and swipe the magnet over the device during one of these episodes, which they did, and we saw essentially nothing. And so that, plus a number of important pieces of history, such as the presence of a lot of psychosocial stressors in the classroom and so on, helped us determine these were likely non-epileptic spells. This young woman was implanted with RNS and called us 10 days later to say that her right hand was numb for hours and we were worried about a vascular event like a perioperative stroke, but we asked her to swipe the magnet and send us recordings, and when she did, we realized that she was actually in focal sensory status epilepticus, arising from perirulandic cortex in the hand knob. Pretty amazing. So we treated her with benzos over the phone and these symptoms resolved and never recurred. And then this young woman, who was implanted with RNS and underwent a concurrent right anterior temporal lobectomy, did very well, seizure-free seven months, and then her mom came home one day and found her down, took her to the ER, and she actually had a cardiac history so they did a syncope versus seizure evaluation, but her mother sent us RNS recordings from the ER, and we could see clearly this was actually a seizure, so we told the ER docs that, and it turned out, with some more digging, that she had actually stopped her seizure medicines because she had been doing so well. There's a nice publication from Issa Roach and colleagues that describes the use of magnet-triggered ECoGs to help distinguish between patients with psychiatric comorbidities whose psychiatric symptoms closely mimic their seizure semiology, for example, panic attacks versus periictal anxiety, and they showed that sort of judicious use of the RNS magnet could elegantly help them separate which clinical spell was epileptic in etiology and which was something else, and they could treat them accordingly. Okay, and then finally, going back to seizure unpredictability, which we talked about in the beginning, I showed you already one example of kind of circadian cycles with my patient who had higher interictal activity at nighttime. At the bottom now, I'm showing you one year of data. These are daily counts of interictal activity made by the device over one year. This is a male patient, and you see that it's quite obvious that the counts are not the same every day. They actually fluctuate in these cycles that have an average periodicity, in this case, of about 14 days. We call these multi-day or multidian cycles, and on the one hand, it's just sort of fascinating that this happens and nobody really knows what drives these cycles, but on the other hand, it's incredibly useful because it turns out that these cycles help determine seizure timing. These are data from a different patient showing seizures as black dots, and the size of the dot means how many seizures they had at that time, and you see that cycle over cycle, the seizures tend to occur on the rising phase, on the upswing of these cycles, near the peak, but typically just before it. When we looked in a large cohort of patients in this way, we found, and here I'm representing these cycles in a circular, so-called polar plot, where if you go from the top and you go clockwise around, that's peak to trough to peak of one average cycle. Each patient is an arrow. The direction the arrow points is the phase of that underlying cycle where seizures are most likely to occur, and the length of the arrow indicates the extent of phase locking to that preferred phase. For both electrographic seizures and self-reported seizures, most patients cluster on this rising phase of these cycles, which means that they help determine periods of highest seizure risk. That means then that you can maybe characterize those cycles and extrapolate them forward in time and ask, well, when's the next time somebody's going to be on the rising phase of these cycles, given that they're relatively stable within individuals? We did this in a pseudo-prospective way, meaning we took a retrospective data set, 10-year data set from the RNS system clinical trials. We built models and trained them on the initial portion of the data, and then we asked them to make forecasts over the next 24 hours of how likely it was that a seizure would occur based on the phase of circadian and multidian cycles. The models spit out forecasts, and then we could see when seizures, clinical and electrographic seizures, actually occurred, and we could ask, well, how well did these models perform? We could quantify their performance in a variety of ways. This is a busy slide, but I wanted to give you a sense of what this actually looks like. Starting at the top in Panel C, that's the raw data. That's the daily IEA counts from the device. From that, we can extract with wavelets the multidian cycle. From that, we can determine the phase of that cycle at any moment, and we know the timeline of seizures that these patients actually had because they wrote them down in seizure diaries. These models would then forecast over the next 24 hours how likely it was that a seizure would occur, and we could ask, did seizures actually occur during a high-risk time, or did they fall outside of that time? This is what that looks like in one of these participants. We looked at 157 participants, but here's data from one. Here's an average forecasted probability. The red is when seizure risk was forecasted to be high, and at the bottom, you see what was actually observed, where the seizures actually fell, and most of them fall during periods of forecasted high risk. A couple of seizures fall during periods of low risk, but that's okay, too. If there's a 10% chance of rain, 10% of the time, it is going to rain. Here's what the data looks like over the entire cohort. We looked at self-reported seizures with a forecast horizon of one day and found that two-thirds of patients could be forecasted better than chance, so IOC is improvement over chance. Even extending out to three days in advance, still 40% of individuals had forecasts that performed better than chance, and in a smaller cohort of 18 patients, where we knew their electrographic seizures very precisely, at a forecast horizon of one hour, we could do better than chance in all of them, and even up to 14 hours, almost half the patients could be forecasted better than chance. That means that this is not the be-all, end-all of seizure forecasting, but it does mean that this sort of proof of principle, that this approach, leveraging cycles of interictal activity can be useful to anticipate future seizure risk. We've talked about a lot of strengths of RNS-CEG, but I want to also be sure we highlight some limitations, so let's do this poll, and I'll ask you, the limitations of chronic electrocorticography with the RNS system include all of the following except, so that means pick which one is not a limitation of the RNS system. So A, spatial sampling in the brain is highly limited, B, CEG recordings are not stored continuously, C, patients are not able to trigger storage of CEG themselves, D, CEG stored by the device does not stream automatically to an online data bank, or E, due to memory constraints, data on the device can be overwritten, if not downloaded regularly. So which of these is not a limitation? Give folks a moment, okay, perfect. So I agree with you, it looks like the winner here is C, so C says patients are unable to trigger storage of CEG, that is not true, they have a magnet that they can swipe over the device to force the device to store an ECOG at that moment. All of the other four answers are current limitations of the RNS system. All right, we have covered a lot of ground, we have talked about a lot of challenges in clinical epilepsy, hopefully I've convinced you that the data from RNS can really change the way we approach these challenges. Time to pivot now to sort of some forward-looking things. I really believe everything I've just talked about is the tip of the iceberg in this field. We are imagining things that didn't seem possible before, so our current clinical paradigm, maybe we will increasingly understand seizures as being related to underlying cycles of seizure risk, and instead of offering static treatments to patients, bathing them in medicine all the time, maybe we could do something smarter, more dynamic, and risk-stratified like chronotherapy. We and several other groups around the world are working towards what has been called a seizure gauge, which would be sort of akin to a weather forecasting app on your phone that you would fire up and look at the chance of rain for the next week. Maybe if our patients had access to an application like that, that involved data from an implanted device perhaps that was processed in a cloud-based environment and then issued directly to them, maybe they could use that information to their advantage. There are other potential applications of this as well. It's been shown that timing EMU admissions during periods of predicted high seizure risk can maximize diagnostic yield of inpatient video EEG. Sharon Chang here at UCSF has done some nice work showing that the brain state determines the efficacy of stimulation parameters by a device, meaning that the same stimulation parameters that are helpful to reduce seizures when the brain is in one risk state might actually be counterproductive to suppressing seizures when the brain is in a different risk state. Our current devices don't have a way of sort of assaying the seizure risk state. They sort of do the same thing all the time that we tell it to do. But you can imagine a next-generation device that's smarter or more adaptive in that way. We are also interested in understanding outcome variability. I've showed you clinical trial data that is really reassuring as to how well this works on average. This is data from our own center, which shows you that, on average, our median seizure frequency reduction is quite good, 79%, very much in keeping with data from clinical trials. But the spread of how patients do is disappointingly broad sometimes. And we have patients who do amazingly well, seizure-free or nearly seizure-free. We call them super responders. We have patients who maybe don't do as well as we want. And the question is, why is that? And I can tell you that it is nothing obvious. But we have looked at this using RNS eCogs themselves and asked, well, what's different about people who do great with this versus people who don't do as well? And to make a long story short, what we find is that the people whose networks can functionally reorganize the best are the ones who respond best to RNS therapy over time. I think that directly supports a neuromodulatory or sort of plasticity-based mechanism of action for this device. Those differences can emerge as early as the first year of therapy, where you can start to see that people whose networks are being remodeled are the ones who do the best over time. Of course, we want biomarkers of treatment response also before you put a computer in somebody's head. So we have looked at things like network synchronizability from phase 2 monitoring as a biomarker of treatment response to subsequent RNS therapy. And then we've even looked at non-invasive biomarkers, such as functional connectivity metrics derived by magnetoencephalography, or MEG, prior to implant collected non-invasively that can also stratify likely responders from non-responders. And then finally, it's worth mentioning that RNS now, I think, is firmly established in epilepsy but is being explored on an investigational basis for other things, like treatment-resistant depression, PTSD, even obesity, really amazing potential applications that are in various stages of exploration. I think we are increasingly thinking of RNS as a platform technology, right? It's a way to continuously monitor a neural signal of interest to deliver electrical stimulation to normalize activity in relevant brain circuits. That's why it's been used for a lot of really incredible neuroscience research, looking at cognition, looking at different forms of memory. Very elegant work from Nathia Suthana, Anli Liu, Barbara Yopes, many others. And RNS has been used to study biomarkers as well. I showed you one application, biomarkers of treatment response, but many labs are using this sort of treasure trove of data to discover biomarkers that track with clinically relevant phenomena. So, here's a summary of everything we talked about. We talked about many potential clinical and research applications of RNS-CEG. I hope that you found all of that useful. Maybe it reminded you of some patients from your own practice or some patients that you think might benefit from this. And I'll just conclude with a couple of quick take-home points. So, we talked about how temporal undersampling is a problem in clinical epilepsy and we need a wide temporal lens and RNS-CEG can provide that to study the dynamics of epilepsy. We reviewed many applications of RNS-CEG, clinical and research applications. We talked about cycles in epilepsy, something near and dear to my heart, and how they can determine seizure timing. We showed that it's possible to leverage those cycles to forecast seizure risk over future horizons, often accurate up to days in advance in some individuals. And then we concluded by saying that RNS is really thought of now as a platform technology with many potential research applications that I think you'll hear about increasingly in the coming years. So with that, I'm going to thank everybody back at home. We have a terrific group here at UCSF, and all of them have contributed to our understanding of this technology, and I'll thank all of you for your attention as well. Well, thank you so much, Dr. Rao, for such a comprehensive and really amazing talk. I think a lot of information to digest, but I wanted to give the audience now a chance to pose some of their questions that might arise during the talk in that little chat box, and I will be trying to relate all of those to Dr. Rao. And while these questions are trickling in, maybe I can ask one myself. I know that this might be a pretty complex discussion here, but can you take us briefly, tell us briefly your take on how chronic ECOG data gained by the RNS device from subcortical structures such as the thalamic nuclei could change the way we see and treat epilepsy in the future? Yeah, that's probably an Ask the Expert webinar on its own, but I like the question because it's really at the bleeding edge of what we are doing with RNS, isn't it? Putting electrodes in thalamus. There's nothing in the labeling of RNS that precludes you from doing that. People are increasingly doing stereo EEG recordings, for example, in thalamus. Thalamic DBS obviously is a thing, and now more and more we're seeing uses of corticothalamic RNS, meaning one lead in cortex, one lead in thalamus. You can detect or stimulate on one or both of those leads, and it has been very eye-opening to see what the thalamus is doing during seizures. Definitely you can see ictal patterns in thalamic recordings. Sometimes they precede what we see in cortex, sometimes they follow that. There are now small series that are encouraging, showing that stimulation of thalamic nuclei, like the central median nucleus, which projects broadly to frontoparietal networks, can be effective for treating things like regional neocortical epilepsy, even perhaps primary generalized epilepsy. I think it's early days, but very promising technology to not only understand the physiology of what's happening in thalamus during seizures, but also to leverage its diffuse connectivity for therapeutic benefits. Great, thank you. I think we're still waiting on some other questions coming in. Here we go. Dr. Polis is writing, Dr. Rao, thanks for a great talk. Quick question, what percentage of your patients achieve the gold standard of one year seizure-free after RNS implant, and do those that achieve seizure freedom achieve it within the first year or two, or is it a gradual increasing function? Yeah, great question. You know, seizure freedom really is the ultimate goal, isn't it? We always describe RNS as a palliative therapy. I really never tell anybody to expect to become seizure-free with RNS, but that said, some people do. In our hands, we have about almost a hundred patients now, so one of the larger cohorts in the country. I can tell you that about 10 to 15 percent of patients achieve at least one year of seizure freedoms in our hands somewhere along the way, and that's actually pretty consistent across centers that I've chatted with. Now, I would love to know what makes those folks different than others. Is it lead placement? Is it something about their physiology? I think the latter, but the second part of your question is, do they achieve it within the first year or two, or is it gradually increasing? To be honest, I have seen it both ways. I have seen an early and sustained response to RNS. I have also seen people, maybe a bit less commonly, who have just a sort of progressive running down of their seizures and eventually achieve long periods of seizure freedom. I think it's more common that people respond early and very well for some reason, reasons that we don't fully understand, but definitely it can happen both ways. I do sort of counsel patients to just be patient with us, though, you know, because I hate folks to think that this is like a switch where, you know, the device gets implanted, flip the switch, turn it on, and no more seizures. I usually ask people to give me at least a year of sort of iterative tinkering with programming settings before we kind of cast judgment on how effective this is going to be for them. That at least calibrates their expectation to not think of it in terms of days to weeks, but more months to years. And I have seen people respond even at the two-year post-implant mark, where we've been trying for two years, we try a million things, and nothing seems to work. And then we chance upon some stimulation parameters that seem to have to be much more effective than anything else we tried. Now, that is quite maddening because I would love a way to sort of know what those great parameters are on the front end and to spare mostly the patient the trouble of, you know, this iterative, empiric, black box kind of tinkering, which is what the state of affairs is currently. I would love a rational way to do that, where we sort of have a way to kind of gauge the right stimulation parameters, so to speak, for somebody on the front end and get them their outcomes even faster. I think the real-world outcome study data that I showed you at least gives us a sense that we have learned something along the way since 2004 when the first patient was implanted. We are getting good outcomes earlier than we used to, but still I wish we could do it better and more rationally. So that's an area of active research for us. I think that answers Dr. Post's question. Are there any other questions by the audience? Dr. Dings was asking, for the patients being seizure-free with RNS, would you try to wean off or simplify their AEDs? Yeah, great question. In the RNS clinical trials, a minority of patients were able to reduce their seizure medicines, I believe about 9%. So in a way, it's like anything else. It's like people who do very well with epilepsy surgery. You get them to the one or two-year seizure-free mark, and first of all, there's sort of two different kinds of patients. There's the kind of patient who's the don't rock the boat, can't do better than seizure-free kind of patient, and then there's a person who's just eager to get off meds at the first moment that you tell them it's reasonable to think about that. If somebody very much wants to come off medicines, I definitely have tried slowly tapering off seizure medicines, as we always do, and one advantage that you have in this scenario compared to after resective surgery is that you can actually follow metrics. Maybe I should include this as yet another use of RNS ECOG, which is that you can actually follow detection counts, and if you see that they've been low and stable and you start peeling off medicines and they're climbing up, it may be a sort of electrographic writing on the wall that this is not going well, that things are revving up and becoming more active. Or conversely, if things remain stable, I saw somebody last week who we tapered off Clobazam and nothing changed on their data, and they're perfectly fine. So I think we just spared them a medicine that maybe wasn't doing a lot of heavy lifting. You have that ability to sort of quantitatively gauge the response to withdrawal of medicines just as you do the initiation of medicine. So I do try to take people, maybe not completely off medicines, I'm not sure I've ever gotten somebody totally seizure-free off meds with RNS, I have to think, but I don't think so. I definitely have taken people from three medicines down to two or two down to one, and that can be really meaningful too, but it is nice to kind of follow along electrographically and see the effects of that taper. Great, there is another question. How long do you wait to see if a patient has failed RNS, and what is your approach for treating these patients if they do not respond to RNS? Yeah, this is a really important question because it really requires us to balance many factors. So the good thing and the challenging thing about RNS is that the parameter space for what's possible with stimulation settings is incredibly vast, right? You have all the waveform parameters, you have all the stimulation pathways, you have, when you start multiplying those, the combinatorial, you know, possibilities are almost endless. And so on the one hand, that's great, you have a lot of options, but on the other hand, it means that there always will be something else to try. How do you know that somebody who doesn't respond to 200 Hertz stimulation won't respond to 250 Hertz stimulation? You know, how could you know without trying? So there is a tendency to want to keep trying and trying and trying. No, there's something else we can try. There's one more combination we should try. And I think that serves people well to a point, but I do think there does come a point when it's worth thinking for the patient's sake, okay, if this is not the answer that we had hoped it would be, what else can we do? We have other neurostimulation devices, we have other surgical options. For example, I showed the data on taking people to resection if they are very asymmetric in their temporal lobe seizure lateralization. At what point does that come? I think it is variable for different patients. I like to push on and try everything I can for like two years because, you know, we see people back every three months or so and in two years that's, you know, eight to ten visits. So it gives me a good shot at trying a lot of the sort of combinations of settings that in my experience, you know, have helped different people. And at that point, it is not to say that nobody has ever responded past that point. It is to say, though, that maybe there's diminishing returns and we should at least think about is there something else that could be used as an adjunct. I showed an example of a patient who got RNS first. He actually did very well. He had a big reduction in seizure frequency, but for him, he was one of these seizure-free or bust kind of people and, you know, who can blame him? So he went on to have thalamic DBS and did even better with that. And that is that's an amazing thing. It taught me that, you know, your response to one device does not necessarily, if you don't respond as well as you want to one device, it doesn't mean you won't respond to a different device that works differently. So I think it does behoove us at a certain point to draw that line and kind of take a step back and think about all the possibilities, given that we have so many choices, maybe around two years. Great. We still have about 10 minutes left. If there are any other questions in the audience, feel free to type them in the little chat box here. And don't make Wolfgang tell jokes to fill time, because he will. That's right. Let's see. Have you ever combined RNS and DBS together in the same patient? I think you were referring to that just now. Yeah, yeah. Yeah, once we have. And I think it's an important thing. We wrote a case report about it because, and I've actually chatted with folks at other institutions who are thinking about doing this, but there's a lot of unknowns, like are the devices compatible with each other? You know, will artifact from one device influence detection for the other device, et cetera? And we didn't see any of that, at least with the lead configurations that were present in our patient. I certainly can't say whether the devices are truly synergistic. I mean, this is N equals one experience and hard to know. But I can tell you that in this individual, he did great with the combination of devices. So I can definitely say it is possible to use both devices. I would love to see a larger series of patients treated with multiple devices. There is a publication recently that we were involved with looking at outcomes in people who are treated with both RNS and VNS. And there's actually a lot of publications now looking at multiple device approaches. I think that's natural. You know, we have three devices. They work in slowly orthogonal ways. Why not combine them and get the best of both worlds from multiple devices, given that it seems like they're compatible and given that failure of one doesn't necessarily predict failure of another? So yes, I see a question also popping up about VNS after RNS implantation. That is definitely possible. It is not wrong. The only thing I would say about that is that that one feels like, I always hate to do the more invasive thing first and then move to a less invasive thing. You know, if you were going to do VNS, you could have done that without localizing seizures and without doing a brain surgery like you have to do for RNS. But it's possible. And who's to say that somebody who fails RNS wouldn't do even better with the addition of VNS? I think that's possible. In my experience, that is a less common pathway than maybe some of the other things. I tend to see VNS, and if that doesn't work, then RNS. Or I tend to see RNS. If that doesn't work, then we did DVS. Maybe a little less common to go to a less invasive modality like VNS, but totally possible. And I wouldn't doubt that people can do well with that strategy too. Yeah, it's funny that you mentioned that. We do have actually a patient, it's a patient of mine, who's going to have the trifecta. You have the VNS, RNS, and now it's receiving the DVS. It was just implanted, so I don't have any outcome data yet. There's a patient here in the Bay Area at a different center who has all three, and that's a lot of hardware to have. But if it helps, that's great. If nothing else, we will learn from these early experiences of combining devices. But I think it's something we'll see more and more over time. Let's see, there's one more. On seizure forecasting, which sounds very promising, can you comment a little bit more about the predicting seizures or other parameters that you have noticed to be helpful? Yeah, so seizure forecasting is really something that I'm quite keen on. And it's nothing that I ever planned to study, but just in the course of taking care of patients, we started seeing these cycles that, you know, almost everybody has circadian cycles, and about two-thirds of people have these multi-day cycles. And the association of seizures with these cycles is very striking. And for many individuals, their cycles are quite stable over time. So it's very tempting to say, well, can't we just extrapolate these cycles forward in time and alert people when they're going to enter sort of a high-risk period, proictal state, as we call it. And we have not yet done that prospectively. In fact, we are just starting a prospective trial to do exactly that, literally as we speak. So I hope to have prospective data to share with you that would hopefully validate this as a, you know, a clinical possibility. But I can tell you that from our retrospective or pseudo-prospective analysis that this can work very well. And it's an approach I think a lot of people are taking. In terms of parameters that we've noticed to be helpful, I'm not totally sure if you mean detection parameters or stimulation parameters, but in any case, there are people who are so regular with their cycles and their seizures are so periodic that it's almost possible I just sort of anticipate an email from a patient because I'm like, oh boy, it's been about two weeks, it's it's, you know, time for a seizure, I think. And you get an email the next morning saying, I just had a seizure, and it's striking. It really, there have been a few instances where I've actually asked people to take a medicine like a, you know, a little Klonopin or something for a couple of days on their cycle to try and reduce seizures because it's so predictable. I won't say everybody is quite that regular, but it has, you know, it has gone well so far. But we need prospective data and larger cohorts to really make that claim. I can also imagine the processing of data in real time and giving this forecasting to the patient in an easy-to-digest fashion. I mean, that's another step that needs to happen to make this a reality in clinical everyday life, right? Exactly, yes, because it's a sort of disjointed process currently where patients have to download data, transmit it to PDMS, the secure online data bank. We have to get the data from there and analyze it and then give a forecast back to a patient. So that is not trivial because it's not so streamlined. We've developed some data pipeline workarounds for that that will facilitate it. But yeah, it's not like data streams off their devices and shows up on our servers that we can analyze so readily. There's one more question that came in. When you talked about the long episodes, can you talk a little bit more about utilization of rescue medications? How long does it take to establish a pattern and then your advice on education of the patient on recognizing these patterns? Yeah, well in the example I showed, that young woman had very obvious clusters of seizures. You know, her long episodes were always clinical seizures and when she would have one, she would tend to stack them back-to-back. So it wasn't so hard to determine her tendency for clustering. And when we gave her a rescue medicine, it was quite obvious clinically and electrographically that it was doing what we wanted it to do, which is to sort of stave off seizures to through. I think it's important also to make a claim. This is something, the feedback that I hear when I present data like this, sometimes people say, well do you really need data? Like if somebody was clustering, you don't need RNS data to give them a rescue medicine, right? I mean most people don't have RNS. So what do you do then? I totally agree. I wouldn't claim that these data are essential. At the end of the day, you're going to do what you think is best clinically. You're going to use all the clinical information available to you. I really think this is just really a complementary source of information that can help reinforce patterns that you observe, sometimes reveal patterns that are not so obvious to discern clinically, corroborate clinical reports that you get that you're not sure how accurate they may be. So this is adjunctive information that doesn't supplant your clinical intuition. So whether you want to give somebody a rescue medicine, that really depends on what you would have done anyways. But sometimes it is nice to see the rescue medicine doing what it's supposed to do in a quantitative way. Great. I guess we have time for one more question. So whoever wants to be the final one. Otherwise, we can probably end this webinar on time. We all agree that we could probably go on and on and Dr. Rao has probably a lot of more insights to share, which are we kind of been under time constraints here. So and I think behind there's a plane about to deboard, so it's going to get really loud here in just a second. Thanks everybody for joining. We really appreciate it. Well, yeah, thank you again Dr. Rao for joining us for this Ask the Expert webinar and we definitely want to make sure that everyone will get, I think, some form of evaluation, a link to fill out. Definitely give us your feedback about this session and future sessions and thank you again for participating. And again, thank you Dr. Rao for a great talk and all these great answers. Take care everyone.

ambulatory electrocorticography

Responsive Neurostimulation

epilepsy management

diagnostic applications

seizure forecasting

chronotherapy

brain activity monitoring

neurostimulation devices

epilepsy surgery outcomes

Vikram Rao