Evidence of state-dependence in the effectiveness of responsive neurostimulation for seizure modulation
Sharon Chiang, Ankit N. Khambhati, Emily T. Wang, Marina Vannucci, Edward F. Chang, Vikram R. Rao
EEvidence of state-dependence in the effectiveness of responsive neurostimulation for seizure modulation
Sharon Chiang, MD, PhD a , Ankit N. Khambhati, PhD b , Emily T. Wang c , Marina Vannucci, PhD c , Edward F. Chang, MD b , Vikram R. Rao, MD, PhD a a Department of Neurology and Weill Institute for Neurosciences, University of California, San Francisco, San Francisco, CA, United States b Department of Neurological Surgery, University of California, San Francisco, San Francisco, CA, United States c Department of Statistics, Rice University, Houston, TX, United States *Corresponding author: Sharon Chiang, Department of Neurology and Weill Institute for Neurosciences, University of California, San Francisco, 505 Parnassus Avenue, San Francisco, CA 94143; E-mail: [email protected]
BSTRACT
Background
An implanted device for brain-responsive neurostimulation (RNS System) is approved as an effective treatment to reduce seizures in adults with medically-refractory focal epilepsy. Clinical trials of the RNS System demonstrate population-level reduction in average seizure frequency, but therapeutic response is highly variable.
Hypothesis
Recent evidence links seizures to cyclical fluctuations in underlying risk. We tested the hypothesis that effectiveness of responsive neurostimulation varies based on current state within cyclical risk fluctuations.
Methods
We analyzed retrospective data from 25 adults with medically-refractory focal epilepsy implanted with the RNS System. Chronic electrocorticography was used to record electrographic seizures, and hidden Markov models decoded seizures into fluctuations in underlying risk. State-dependent associations of RNS System stimulation parameters with changes in risk were estimated.
Results
Higher charge density was associated with improved outcomes, both for remaining in a low seizure risk state and for transitioning from a high to a low seizure risk state. The effect of stimulation frequency depended on initial seizure risk state: when starting in a low risk state, igher stimulation frequencies were associated with remaining in a low risk state, but when starting in a high risk state, lower stimulation frequencies were associated with transition to a low risk state. Findings were consistent across bipolar and monopolar stimulation configurations.
Conclusion
The impact of RNS on seizure frequency exhibits state-dependence, such that stimulation parameters which are effective in one seizure risk state may not be effective in another. These findings represent conceptual advances in understanding the therapeutic mechanism of RNS, and directly inform current practices of RNS tuning and the development of next-generation neurostimulation systems.
KEYWORDS:
RNS; Epilepsy; Electrocorticography; Seizure risk; Neuromodulation
ABBREVIATIONS
ECoG, electrocorticography; IED, interictal epileptiform discharge; LE, long episode; MCMC, Markov chain Monte Carlo; PDMS, patient data management system; RNS, responsive neurostimulation
NTRODUCTION
Epilepsy is a common neurological disorder characterized by recurrent seizures. Anti-seizure medications are first-line treatments for epilepsy, but up to 30% of patients have seizures that are incompletely controlled with medications. For these medically-refractory patients, resection of the seizure focus can be highly effective but may not be possible if seizure foci are bilateral, poorly localized, spatially-extensive, or overlapping with eloquent cortex. In recent years, responsive neurostimulation (RNS), brain electrical stimulation triggered by real-time detection of seizures, has emerged as promising therapy for medically-refractory patients who are not candidates for resective surgery. The RNS System (NeuroPace, Inc.) is a cranially-implanted device approved as a safe and effective treatment for adults with medically-refractory seizures arising from 1 – with ≥90% seizure frequency reduction (2), but about one-quarter of patients are non-responders with less than 50% reduction in seizure frequency (1). Investigations of factors that predict therapeutic response to RNS have focused primarily on stationary clinical features, such as the presence of mesial temporal sclerosis, seizure focus, prior resection, prior intracranial monitoring, or lead proximity to the seizure focus (3-5). However, no features have yet been found to explain this variability (3-5). Improving outcomes for all patients treated with RNS, particularly non-responders, may be facilitated by greater mechanistic understanding of this therapy and optimization of stimulation protocols based on individual temporal dynamics. revailing views on the mechanistic basis of RNS have shifted over time. Originally, the efficacy of RNS was thought to derive from direct inhibition of ongoing seizures by electrical stimulation (6-8); in other words, RNS was thought to abort seizures in the way that defibrillators terminate cardiac arrhythmias. However, a critical observation from long-term clinical trials is that the efficacy of RNS improves over time (1, 2), suggesting chronic, neuromodulatory effects in addition to acute seizure-terminating effects. Still, many questions remain about how best to harness the putative neuromodulatory effects of RNS. These effects likely depend on the stimulation parameters that are employed (9, 10), but aside from experience-based practice, there are virtually no objective methods for selecting these parameters rationally. For example, based on experience during clinical trials, charge density is often increased empirically to enhance therapeutic response, but a definitive correlation between charge density and seizure frequency has not been shown (5). Emerging research indicates that epilepsy is a cyclical disorder. Rates of interictal epileptiform discharges (IEDs) fluctuate with daily and multi-day periodicities (11-18). Electrographic seizures occur preferentially at certain phases of these IED cycles, which therefore help determine momentary seizure risk (18, 19). Periods of high and low seizure risk in IED fluctuations can be conceptualized as distinct brain states oscillating with hours-to-months long macro periodicities, which may have varying susceptibility to neurostimulation. In vitro and in vivo models suggest that neural modulation using electrical stimulation is state-dependent (20, 21). Despite these discoveries, RNS System programming in contemporary practice remains empiric and agnostic to brain state. Standardized titration schedules for stimulation parameters are applied to patients with diverse epilepsies (22), and parameters are adjusted relatively nfrequently. Thus, the dynamism of brain states in epilepsy based on macro oscillations of IEDs contrasts sharply with clinical management of RNS. Whether electrical stimulation protocols can be tailored in a patient-specific, time-varying manner to increase response rates is unknown. Here, we hypothesize that the ability of electrical stimulation to modulate seizures is state-dependent, and, specifically, that stimulation parameters that are effective in one brain state may not be effective in another brain state. To test this, we investigate the effect of electrical stimulation parameter changes on the incidence of electrographic seizures as a state-dependent function in a retrospective cohort of 25 adults with medically-refractory focal epilepsy treated with the RNS System. ATERIAL AND METHODS
Patient selection
We performed a retrospective review of patients with medically-refractory focal epilepsy who were implanted with the RNS System for clinical indications between August 2014 and June 2018 at the University of California, San Francisco (UCSF) Comprehensive Epilepsy Center. Data collection was approved by the UCSF Institutional Review Board and written informed consent was obtained from all subjects.
RNS System
The RNS System is comprised of a cranially-implanted programmable neurostimulator which is connected to two four-contact depth or cortical strip leads placed at the seizure focus/foci. The neurostimulator continuously senses electrocorticographic (ECoG) activity and is programmed by clinicians to detect specific patterns and deliver brief electrical pulses through the electrode contacts in response to those detections. Electrographic seizures are identified either through saturations (high amplitude ECoG) or Long Episodes (LE, sustained detections of abnormal brain activity that often represent seizures). Detection and stimulation parameters are adjusted based on clinical factors for each patient. Stimulation parameters that can be adjusted include the charge density (current amplitude x pulse width per phase, distributed over the number of cathodes), stimulation frequency (rate at which pulses are delivered), burst duration (amount of time for each train of pulses), and pulse width (duration of a single phase within a charge-balanced biphasic square-wave pulse). Most commonly, charge density, stimulation frequency, and burst duration are adjusted. Stimulation can be applied using various ‘stimulation pathways,’ designations of the eight electrode contacts as anodes or cathodes. The most common stimulation athways are a bipolar pathway (alternating anodal and cathodal contacts on each lead) or cathodal pathway (with one lead the anode and one lead the cathode; when leads are far apart, this approximates monopolar stimulation). The neurostimulator canister can also be used as an anode or cathode in a lead-to-generator montage within a cathodal/monopolar pathway. A grouped bipolar stimulation pathway (alternating pairs of anodal and cathodal electrode contacts) is less commonly used. Examples of bipolar, grouped bipolar, and cathodal/monopolar pathways are shown in Table 1. RNS System stimulation parameters (charge density, frequency, burst duration, pulse width) and stimulation pathway were determined by treating providers based on clinical information. All patients had two four-contact depth and/or cortical strip leads placed over the seizure-onset zone(s). Computed tomography brain imaging was performed post-operatively to confirm lead placement. For each patient, the number of LE detections, stimulation settings (charge density, frequency, burst duration, pulse width), and stimulation pathway at each hour were recorded. Stimulation parameters used in the first therapy delivered by the device during each detection episode were analyzed.
Data collection
RNS System stimulation settings and hourly LE counts were extracted from the Patient Data Management System (PDMS), a secure online repository of RNS System programming parameters and electrographic data. Stimulation pathways were analyzed separately. Less than 0.1% of the data had non-consecutive breaks in recorded hourly counts. All related clinical data were abstracted from the electronic medical record system.
State-dependent modeling of parameter associations with long episodes
LE detections were used to identify electrographic seizures. All ECoG recordings were manually reviewed by an experienced epileptologist (VRR) to confirm whether long episodes corresponded to electrographic seizures. Only patients for whom >90% of
LE’s detected by the RNS System corresponded to electrographic seizures (19) and for whom there were at least 20 time points in one of the stimulation configurations were included. A first-order non-homogeneous hidden Markov model for zero-inflated count data was used to identify whether stimulation parameters differed in effectiveness depending on current state (11, 12). This implies a future risk prediction horizon of one hour ahead, conditioned on the risk state of the previous hour. Non-homogeneous hidden Markov models are a class of latent process models that can be used to decode electrographic seizures into underlying latent (unobserved) levels of seizure risk, while simultaneously estimating the association of exogenous inputs at each time t (here, RNS stimulation parameters) with changes in latent risk from time t to t+1 (Figure 1). The temporal directionality of this inference allows for causal inference to be drawn. The latent process approach has several advantages: firstly, it reduces noise of count data; secondly, it explicitly accounts for conditional associations between RNS stimulation parameters and changes in risk conditional on current state; thirdly, the temporal directionality relating exogenous inputs to changes in risk allows for causal inference between RNS parameters and changes in risk. If the association between RNS stimulation parameters and seizure risk is independent of current state, then estimated conditional associations will not significantly differ. For a more in-depth overview of Bayesian inference, we refer the reader to (23). Figure 1. Hidden Markov modeling of effect of RNS stimulation on seizure modulation.
A hidden Markov model captures the temporal evolution of seizure risk states, consisting of a Markov chain with stochastic measurements on the hidden states (blue) and emission distribution conditional on the states (orange). Non-homogeneity of the hidden Markov process allows the transition matrix of the hidden states to depend on RNS parameter changes and other exogenous variables (grey). Abbreviations: RNS, responsive neurostimulation.
Markov chain Monte Carlo (MCMC) sampling was used to sample from the posterior distribution of parameters, which yields posterior estimates of the association of exogenous inputs with seizure risk state transitions. We analyzed the association of commonly adjusted RNS stimulation parameters, including charge density, pulse frequency, and burst duration, with state transitions. Pulse width was excluded from bipolar and cathodal analyses, as this parameter is less commonly manipulated than charge density, pulse frequency, burst duration, or stimulation pathway. Burst duration was included only in cathodal analysis, as the only value of burst duration used in patients employing a bipolar stimulation configuration was 100 ms. In addition, we controlled for several other baseline characteristics within the hidden Markov model, including age, sex, and seizure focus (mesiotemporal, neocortical, or both).(5) To account for potential influence of device detection sensitivity on long episode counts, the number of pisode starts was additionally included. For each configuration, we explored model fits with K between 2 and 3 states to find the number of states K yielding the best model fit. Model fit for each value of K was assessed using the deviance information criterion (24) and convergence of the state allocations to the stationary distribution. For clinical interpretability, we used K= ESULTS
From a total cohort of 49 patients implanted with the RNS System at our center between August 2014 and June 2018, we identified 25 patients with available data for whom LE reliably corresponded to electrographic seizures (see Methods). These patients represented a total accumulated experience of 286,473 patient-hours of treatment with the RNS System.
Demographic characteristics
Demographic characteristics are provided in Table 2. The mean age at time of RNS System implantation was 38.7±13.9 years (range, 18.2-72.0 years). A greater proportion of patients treated with bipolar stimulation had a mesiotemporal focus, whereas a greater proportion of patients treated with cathodal stimulation had a neocortical focus (p=0.003). Cathodal and bipolar cohorts were otherwise similar in age, sex, seizure frequency, age of epilepsy onset, epilepsy duration, and epilepsy etiology (Table 2). The average epilepsy duration was 21.6±13.0 years (range 2.0-61.0 years). The most common etiology of epilepsy was cryptogenic (54.2%).
RNS System parameters
Stimulation was enabled on the RNS System once reliable seizure detection was achieved. Initial RNS System stimulation parameters were empirically selected and iteratively adjusted based on the patient’s clinical course and experience from the RNS System Clinical Trials.
A cathodal stimulation pathway was used in 13/25 patients and a bipolar stimulation pathway was used in 16/25 patients. The grouped bipolar stimulation pathway was used temporarily in only one patient, with the same stimulation settings used during the entire duration of grouped bipolar stimulation (charge density of 2.0 μC/cm , frequency of 200 Hz, pulse width of 160 μs, and burst uration of 100 ms) and was thus excluded from analysis. Of the 25 patients, 8 patients were treated only with cathodal stimulation, 12 patients were treated only with bipolar stimulation, four were initially treated with bipolar stimulation and then switched to cathodal stimulation, and one patient was treated initially with cathodal stimulation and then switched to grouped bipolar stimulation. A total of 150,238 person-hours from 13 patients were available for analysis in the cathodal stimulation configuration and 130,266 person-hours were available from 16 patients in the bipolar stimulation configuration. Cathodal stimulation pathway
State allocations showed convergence to a stationary distribution for K =2. For interpretation, we thereon refer to the two ordered states as a high and low seizure risk state. Figure 2 shows an example of the posterior estimates of seizure risk states based on the number of ‘long episodes’ (electrographic seizures) per hour, for a patient with RNS in the cathodal stimulation configuration. The Markovian property accounts for memory when identifying seizure risk states. This permits, for example, continuous sequences of zero seizures per hour with only a few interspersed hours with one or two seizures to be identified as low risk despite the presence of a few hours with seizures; conversely, a continuous sequence of hours with many seizures occurring, with only a few interspersed hours with no seizures, can be identified as high risk despite the potentially random occurrence of a few hours without seizures. This reduces the noise inherent in seizure counting data by probabilistically discriminating between natural variability and sustained changes in risk. The distribution of seizures classified in high and low risk states is shown in Supplementary Figure A1. Figure 2. Estimated latent seizure risk states for a time sample from an example RNS patient.
Number of long episodes per hour is shown in black. Seizure risk state estimates, based on the mode of the posterior distribution of latent states, are shown in red. Abbreviations: RNS, responsive neurostimulation.
Using a cathodal stimulation configuration, the mean stimulation parameter setting over 150,238 person-hours was a charge density of 1.7+ +54.5 Hz, burst duration of 468.7+1234.7 ms, and pulse width of 155.5+
Stimulation parameters are shown in Table 3 and Figure 3A. Figure 4A shows the association of stimulation parameters with changes in seizure risk state for the cathodal configuration, with posterior means and HPD intervals in Supplementary Table A1. For time points when patients were currently in a low seizure risk state (Figure 4A, green circles), using a higher charge density, higher stimulation frequency, or shorter burst duration were associated with better chance of remaining in a low seizure risk state. For time points when patients were currently in a high seizure risk state (Figure 4A, red circles), using a lower stimulation frequency or shorter burst duration were associated ith a higher probability of transitioning from a high to a low seizure risk state (Figure 4A, Supplementary Table A1). Robustness of results in the cathodal stimulation configuration was compared to when the subset of four patients with preceding bipolar stimulation were excluded (i.e., with analysis restricted to patients treated only with cathodal stimulation). Subgroup analysis demonstrated consistent associations of higher charge density, higher stimulation frequency, and shorter burst duration with improved chance of remaining in a low seizure risk state with cathodal stimulation, as well as consistent associations between lower stimulation frequency and shorter burst duration with higher probability of transitioning from a high to a low seizure risk state. A near-zero (HPD interval, -0.42 to -0.06) inverse association between charge density and probability of transitioning from a high to low seizure risk state was present when analysis was limited to the subset of patients treated only with cathodal stimulation (Supplementary Figure A1, Supplementary Table A2).
Bipolar stimulation pathway
Using a bipolar stimulation montage, the mean stimulation parameter setting over 130,266 person-hours was a charge density of 2.1+ +12.1 Hz, burst duration of 100 ms, and pulse width of 157.6+
Figure 4B shows the association of stimulation parameter changes with electrographic seizure counts contingent on current state values for the bipolar stimulation configuration, with posterior means and HPD intervals in Supplementary Table A3. Higher charge densities were associated with better chances of remaining in a low seizure risk state (for times when patients were already in low risk states) as well as better chance of transitioning from a high to low seizure risk state (for times when patients were in high risk states). For times when patients were currently in a low seizure risk tate, higher stimulation frequencies were associated with remaining in low risk states. For time points when patients were currently in a high seizure risk state, lower stimulation frequencies were associated with transitioning from a high to low seizure risk state (Figure 4B, Supplementary Table A3).
Figure 3. RNS stimulation parameters : combinations of stimulation parameters used by clinicians within each stimulation pathway are shown. Abbreviations: RNS, responsive neurostimulation.
Figure 4. Association of increases in RNS stimulation parameters with low future seizure risk, for (a) cathodal and (b) bipolar stimulation configuration.
Posterior means and 95% highest posterior ensity (HPD) intervals for state-dependent log-odds of each stimulation parameter for transitioning to a low risk state are shown, conditional on current risk state (shown in red and green). Probabilities are relative to the probability of transitioning to (or remaining in) a high risk state. * = 95% HPD interval does not contain zero. Abbreviations: RNS, responsive neurostimulation.
ISCUSSION
In contemporary practice, RNS stimulation parameters are adjusted empirically and typically held constant for months, a timeframe during which there may be considerable fluctuation in underlying seizure risk. Here, we demonstrate that RNS charge density can modulate seizure risk and that, for the two most common stimulation pathways, the effects of changing other stimulation parameters depend on initial seizure risk state. This suggests that variability in clinical response to RNS may relate at least partly to counterproductive effects of stimulation parameter changes applied during inopportune brain states. These data also suggest that next-generation neurostimulation systems for epilepsy may benefit from real-time feedback and adaptive stimulation capabilities that continuously monitor brain state over prolonged periods of time, which may minimize interpatient variability and enhance clinical outcomes. Our results provide several conceptual advances in understanding the relationship between electrical brain stimulation and seizure risk. First, we show that the impact of responsive neurostimulation on modulating seizure frequency may exhibit state-dependence, and that stimulation parameters which are effective in one brain state may not be effective, or even have opposite effects, in another brain state. In vitro and in vivo animal models have revealed that the modulatory impact of electrical stimulation is frequency-dependent and timing-dependent (21, 26, 27), but, to our knowledge, this is the first demonstration in human neuromodulation protocols in epilepsy. A state-based view of electrical stimulation paradigms implies that the impact of neurostimulation extends beyond direct seizure termination, consistent with an emerging concept of RNS having long-term neuromodulatory effects (28) that facilitate transitions from high and low seizure risk states, which may bear an indirect relation to cortical up” and “down” states. We speculate that these effects may involve state-specific disruption of pathological cortical synchrony. Second, we discover several new state-dependent effects of various neurostimulation parameters which may catalyze future development of adaptive neurostimulation systems. Currently, the recommended initial settings for the RNS System are: frequency 200 Hz, burst duration 100 ms, charge density 0.5 μC/ cm , and pulse width 160 μs . Over time, current recommendations are to increase charge density stepwise in 0.5 μC/sq cm increments. (22) If this fails, burst duration and frequency are often increased, although specific guidelines for these adjustments are not available. Despite these recommendations, empirical evidence supporting these recommendations has not previously been shown. In clinical practice, the most commonly programmed stimulation parameters are an amplitude of 1.5 – of 160 μs, burst duration of 100 –
200 ms, and pulse frequency of 100 –
200 Hz (29, 30). For both bipolar and cathodal stimulation configurations, we found that higher charge densities were associated with improved outcomes, both for remaining in (bipolar/cathodal) or transitioning to (bipolar) a low seizure risk state. Bipolar stimulation exerts a more focal stimulation effect than cathodal stimulation, such that increases in charge density may make more of a difference for bipolar pathways. Within the range of 100 – strategy of differentially programming stimulation parameters for different “therapies” (bursts of current pulses) that are delivered within a given detection epoch if redetection of the electrographic seizure occurs. Up to five therapies can be delivered if redetection of the electrographic seizure occurs despite the preceding therapies. Because persistence of a seizure despite repeated therapies is more likely to reflect a high risk state, we postulate that a differential programming strategy may improve effectiveness, with parameters for the first therapy set to those effective in low risk states, and parameters for later therapies set to those effective in high risk states. There are several limitations to this study. The patient sample, while spanning a wide range of ages and localizations of epilepsy, was modest in size and may not be representative of all atients implanted with the RNS System. The minimum age evaluated in this sample was 18 years; given increasing interest in RNS use in the pediatric epilepsy population (31, 32), generalizability to pediatric populations is of interest. Due to storage limitations, the RNS System does not provide continuous ECoG tracings, so electrographic seizure detections were based on “ long episodes ,” which are an imperfect surrogate for seizures. However, the fact that we only included patients for whom >90% of long-episodes corresponded to electrographic seizures may mitigate this concern. Patients may differ with respect to the number of seizure states that they cycle through. The number of states used was chosen for interpretability and consistent interpretability in groupwise conclusions. However, if state-dependent electrical stimulation is used in an adaptive stimulation paradigm, it would be beneficial to allow for a varying number of states optimized to the individual patient. Larger samples will be useful for validating these findings across a greater variety of parameter changes, particularly for bipolar or grouped bipolar configurations. Although we speculate here on translatability to comparing bipolar and cathodal stimulation pathways, small sample sizes and limited matching characteristics caution against direct comparability of these montages and is outside the scope of this study. Larger matched samples are needed for direct comparison of stimulation pathways. Finally, randomized prospective trials to confirm differences among parameter effects and stimulation configurations are needed. Since bipolar montages tend to be trialed first, differences between bipolar and cathodal montages may reflect differences in timing. In our sample, 4 of the 13 patients with cathodal stimulation had bipolar stimulation first; however, sensitivity analysis demonstrated robustness to exclusion of these patients. n conclusion, our findings provide evidence to support potential state-dependence for the effectiveness of RNS in modulating seizure frequency. This finding provides a possible explanation for variations in efficacy of RNS electrical stimulation on modulating seizures and suggests the need for incorporating real-time state analysis into adaptive algorithms for future neurostimulation systems. CKNOWLEDGEMENTS
The authors thank Robert Moss for his collaboration on projects that contributed to code development.
FUNDING
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.
DECLARATION OF INTEREST
VRR has served as a consultant for NeuroPace, Inc., manufacturer of the RNS System, but declares no targeted compensation or other support for this study. All other authors declare no conflicts of interest relevant to this study.
AUTHOR CONTRIBUTIONS
SC: Conceptualization, methodology, software, formal analysis, investigation, data interpretation, writing-original draft preparation. ANK: Conceptualization, data curation, investigation, writing-review & editing. EW: Software, writing-review & editing. MV: Software, writing-review & editing. EFC: Data interpretation, writing-review & editing. VRR: Conceptualization, investigation, data interpretation, resources, data curation, writing-review & editing, supervision.
EFERENCES
1. Bergey GK, Morrell MJ, Mizrahi EM, Goldman A, King-Stephens D, Nair D, et al. Long-term treatment with responsive brain stimulation in adults with refractory partial seizures. Neurology. 2015;84(8):810-7. 2. Nair D, Morrell M. Nine-year Prospective Safety and Effectiveness Outcomes from the Long-Term Treatment Trial of the RNS® System (S36. 005). AAN Enterprises; 2019. 3. Geller EB, Skarpaas TL, Gross RE, Goodman RR, Barkley GL, Bazil CW, et al. Brain-responsive neurostimulation in patients with medically intractable mesial temporal lobe epilepsy. Epilepsia. 2017;58(6):994-1004. 4. Ma BB, Rao VR. Responsive neurostimulation: Candidates and considerations. Epilepsy Behav. 2018;88:388-95. 5. Ma BB, Fields MC, Knowlton RC, Chang EF, Szaflarski JP, Marcuse LV, et al. Responsive neurostimulation for regional neocortical epilepsy. Epilepsia. 2020;61(1):96-106. 6. Kossoff EH, Ritzl EK, Politsky JM, Murro AM, Smith JR, Duckrow RB, et al. Effect of an external responsive neurostimulator on seizures and electrographic discharges during subdural electrode monitoring. Epilepsia. 2004;45(12):1560-7. 7. Lesser RP, Kim SH, Beyderman L, Miglioretti DL, Webber WR, Bare M, et al. Brief bursts of pulse stimulation terminate afterdischarges caused by cortical stimulation. Neurology. 1999;53(9):2073-81. 8. Skarpaas TL, Morrell MJ. Intracranial stimulation therapy for epilepsy. Neurotherapeutics. 2009;6(2):238-43. . Chkhenkeli SA, Chkhenkeli IS. Effects of therapeutic stimulation of nucleus caudatus on epileptic electrical activity of brain in patients with intractable epilepsy. Stereotact Funct Neurosurg. 1997;69(1-4 Pt 2):221-4. 10. Cordeiro JG, Somerlik KH, Cordeiro KK, Aertsen A, Araujo JC, Schulze-Bonhage A. Modulation of excitability by continuous low- and high-frequency stimulation in fully hippocampal kindled rats. Epilepsy Res. 2013;107(3):224-30. 11. Chiang S, Vannucci M, Goldenholz DM, Moss R, Stern JM. Epilepsy as a dynamic disease: A Bayesian model for differentiating seizure risk from natural variability. Epilepsia Open. 2018;3(2):236-46. 12. Chiang S, Goldenholz DM, Moss R, Rao VR, Haneef Z, Theodore WH, et al. Prospective validation study of an epilepsy seizure risk system for outpatient evaluation. Epilepsia. 2019. 13. Baud MO, Proix T, Rao VR, Schindler K. Chance and risk in epilepsy. Curr Opin Neurol. 2020;33(2):163-72. 14. Proix T, Truccolo W, Leguia MG, King-Stephens D, Rao VR, Baud MOJM. Forecasting Seizure Risk over Days. 2019:19008086. 15. Karoly PJ, Cook MJ, Maturana M, Nurse ES, Payne D, Brinkmann BH, et al. Forecasting cycles of seizure likelihood. Epilepsia. 2020. 16. Karoly PJ, Goldenholz DM, Freestone DR, Moss RE, Grayden DB, Theodore WH, et al. Circadian and circaseptan rhythms in human epilepsy: a retrospective cohort study. Lancet Neurol. 2018. 17. Karoly PJ, Ung H, Grayden DB, Kuhlmann L, Leyde K, Cook MJ, et al. The circadian profile of epilepsy improves seizure forecasting. Brain. 2017;140(8):2169-82. 8. Anderson CT, Tcheng TK, Sun FT, Morrell MJ. Day-Night Patterns of Epileptiform Activity in 65 Patients With Long-Term Ambulatory Electrocorticography. J Clin Neurophysiol. 2015;32(5):406-12. 19. Baud MO, Kleen JK, Mirro EA, Andrechak JC, King-Stephens D, Chang EF, et al. Multi-day rhythms modulate seizure risk in epilepsy. Nature communications. 2018;9(1):88. 20. Wester JC, Contreras D. Differential modulation of spontaneous and evoked thalamocortical network activity by acetylcholine level in vitro. J Neurosci. 2013;33(45):17951-66. 21. Li G, Henriquez CS, Frohlich F. Unified thalamic model generates multiple distinct oscillations with state-dependent entrainment by stimulation. PLoS Comput Biol. 2017;13(10):e1005797. 22. NeuroPace. RNS system user manual. NeuroPace Mountain View, CA; 2015. 23. Gelman A, Carlin JB, Stern HS, Dunson DB, Vehtari A, Rubin DB. Bayesian data analysis: CRC press; 2013. 24. Spiegelhalter DJ, Best NG, Carlin BP, Van Der Linde AJJotrssSb. Bayesian measures of model complexity and fit. 2002;64(4):583-639. 25. de Bragança Pereira CA, Stern JMJE. Evidence and credibility: full Bayesian significance test for precise hypotheses. 1999;1(4):99-110. 26. Sadowski JH, Jones MW, Mellor JR. Sharp-Wave Ripples Orchestrate the Induction of Synaptic Plasticity during Reactivation of Place Cell Firing Patterns in the Hippocampus. Cell Rep. 2016;14(8):1916-29. 7. Tremblay SA, Chapman CA, Courtemanche R. State-Dependent Entrainment of Prefrontal Cortex Local Field Potential Activity Following Patterned Stimulation of the Cerebellar Vermis. Front Syst Neurosci. 2019;13:60. 28. Kokkinos V, Sisterson ND, Wozny TA, Richardson RMJJn. Association of Closed-Loop Brain Stimulation Neurophysiological Features With Seizure Control Among Patients With Focal Epilepsy. 2019. 29. Geller EB. Responsive neurostimulation: Review of clinical trials and insights into focal epilepsy. Epilepsy Behav. 2018;88s:11-20. 30. Sun FT, Morrell MJ. Closed-loop neurostimulation: the clinical experience. Neurotherapeutics. 2014;11(3):553-63. 31. Singhal NS, Numis AL, Lee MB, Chang EF, Sullivan JE, Auguste KI, et al. Responsive neurostimulation for treatment of pediatric drug-resistant epilepsy. Epilepsy Behav Case Rep. 2018;10:21-4. 32. Kokoszka MA, Panov F, La Vega-Talbott M, McGoldrick PE, Wolf SM, Ghatan S. Treatment of medically refractory seizures with responsive neurostimulation: 2 pediatric cases. J Neurosurg Pediatr. 2018;21(4):421-7.1. Bergey GK, Morrell MJ, Mizrahi EM, Goldman A, King-Stephens D, Nair D, et al. Long-term treatment with responsive brain stimulation in adults with refractory partial seizures. Neurology. 2015;84(8):810-7. 2. Nair D, Morrell M. Nine-year Prospective Safety and Effectiveness Outcomes from the Long-Term Treatment Trial of the RNS® System (S36. 005). AAN Enterprises; 2019. 3. Geller EB, Skarpaas TL, Gross RE, Goodman RR, Barkley GL, Bazil CW, et al. Brain-responsive neurostimulation in patients with medically intractable mesial temporal lobe epilepsy. Epilepsia. 2017;58(6):994-1004. 4. Ma BB, Rao VR. Responsive neurostimulation: Candidates and considerations. Epilepsy Behav. 2018;88:388-95. 5. Ma BB, Fields MC, Knowlton RC, Chang EF, Szaflarski JP, Marcuse LV, et al. Responsive neurostimulation for regional neocortical epilepsy. Epilepsia. 2020;61(1):96-106. 6. Kossoff EH, Ritzl EK, Politsky JM, Murro AM, Smith JR, Duckrow RB, et al. Effect of an external responsive neurostimulator on seizures and electrographic discharges during subdural electrode monitoring. Epilepsia. 2004;45(12):1560-7. 7. Lesser RP, Kim SH, Beyderman L, Miglioretti DL, Webber WR, Bare M, et al. Brief bursts of pulse stimulation terminate afterdischarges caused by cortical stimulation. Neurology. 1999;53(9):2073-81. 8. Skarpaas TL, Morrell MJ. Intracranial stimulation therapy for epilepsy. Neurotherapeutics. 2009;6(2):238-43. . Chkhenkeli SA, Chkhenkeli IS. Effects of therapeutic stimulation of nucleus caudatus on epileptic electrical activity of brain in patients with intractable epilepsy. Stereotact Funct Neurosurg. 1997;69(1-4 Pt 2):221-4. 10. Cordeiro JG, Somerlik KH, Cordeiro KK, Aertsen A, Araujo JC, Schulze-Bonhage A. Modulation of excitability by continuous low- and high-frequency stimulation in fully hippocampal kindled rats. Epilepsy Res. 2013;107(3):224-30. 11. Chiang S, Vannucci M, Goldenholz DM, Moss R, Stern JM. Epilepsy as a dynamic disease: A Bayesian model for differentiating seizure risk from natural variability. Epilepsia Open. 2018;3(2):236-46. 12. Chiang S, Goldenholz DM, Moss R, Rao VR, Haneef Z, Theodore WH, et al. Prospective validation study of an epilepsy seizure risk system for outpatient evaluation. Epilepsia. 2019. 13. Baud MO, Proix T, Rao VR, Schindler K. Chance and risk in epilepsy. Curr Opin Neurol. 2020;33(2):163-72. 14. Proix T, Truccolo W, Leguia MG, King-Stephens D, Rao VR, Baud MOJM. Forecasting Seizure Risk over Days. 2019:19008086. 15. Karoly PJ, Cook MJ, Maturana M, Nurse ES, Payne D, Brinkmann BH, et al. Forecasting cycles of seizure likelihood. Epilepsia. 2020. 16. Karoly PJ, Goldenholz DM, Freestone DR, Moss RE, Grayden DB, Theodore WH, et al. Circadian and circaseptan rhythms in human epilepsy: a retrospective cohort study. Lancet Neurol. 2018. 17. Karoly PJ, Ung H, Grayden DB, Kuhlmann L, Leyde K, Cook MJ, et al. The circadian profile of epilepsy improves seizure forecasting. Brain. 2017;140(8):2169-82. 8. Anderson CT, Tcheng TK, Sun FT, Morrell MJ. Day-Night Patterns of Epileptiform Activity in 65 Patients With Long-Term Ambulatory Electrocorticography. J Clin Neurophysiol. 2015;32(5):406-12. 19. Baud MO, Kleen JK, Mirro EA, Andrechak JC, King-Stephens D, Chang EF, et al. Multi-day rhythms modulate seizure risk in epilepsy. Nature communications. 2018;9(1):88. 20. Wester JC, Contreras D. Differential modulation of spontaneous and evoked thalamocortical network activity by acetylcholine level in vitro. J Neurosci. 2013;33(45):17951-66. 21. Li G, Henriquez CS, Frohlich F. Unified thalamic model generates multiple distinct oscillations with state-dependent entrainment by stimulation. PLoS Comput Biol. 2017;13(10):e1005797. 22. NeuroPace. RNS system user manual. NeuroPace Mountain View, CA; 2015. 23. Gelman A, Carlin JB, Stern HS, Dunson DB, Vehtari A, Rubin DB. Bayesian data analysis: CRC press; 2013. 24. Spiegelhalter DJ, Best NG, Carlin BP, Van Der Linde AJJotrssSb. Bayesian measures of model complexity and fit. 2002;64(4):583-639. 25. de Bragança Pereira CA, Stern JMJE. Evidence and credibility: full Bayesian significance test for precise hypotheses. 1999;1(4):99-110. 26. Sadowski JH, Jones MW, Mellor JR. Sharp-Wave Ripples Orchestrate the Induction of Synaptic Plasticity during Reactivation of Place Cell Firing Patterns in the Hippocampus. Cell Rep. 2016;14(8):1916-29. 7. Tremblay SA, Chapman CA, Courtemanche R. State-Dependent Entrainment of Prefrontal Cortex Local Field Potential Activity Following Patterned Stimulation of the Cerebellar Vermis. Front Syst Neurosci. 2019;13:60. 28. Kokkinos V, Sisterson ND, Wozny TA, Richardson RMJJn. Association of Closed-Loop Brain Stimulation Neurophysiological Features With Seizure Control Among Patients With Focal Epilepsy. 2019. 29. Geller EB. Responsive neurostimulation: Review of clinical trials and insights into focal epilepsy. Epilepsy Behav. 2018;88s:11-20. 30. Sun FT, Morrell MJ. Closed-loop neurostimulation: the clinical experience. Neurotherapeutics. 2014;11(3):553-63. 31. Singhal NS, Numis AL, Lee MB, Chang EF, Sullivan JE, Auguste KI, et al. Responsive neurostimulation for treatment of pediatric drug-resistant epilepsy. Epilepsy Behav Case Rep. 2018;10:21-4. 32. Kokoszka MA, Panov F, La Vega-Talbott M, McGoldrick PE, Wolf SM, Ghatan S. Treatment of medically refractory seizures with responsive neurostimulation: 2 pediatric cases. J Neurosurg Pediatr. 2018;21(4):421-7.