Patient-detectable responsive neurostimulation as a seizure warning system

Summary

Many people with epilepsy are not aware of their seizures or do not have reliable auras. The responsive neurostimulation system (RNS) delivers stimulation triggered by intracranial epileptiform activity. If an epileptiform pattern continues, RNS repeats stimulation up to five times per event. RNS can cause acute stimulation-related symptoms that can be avoided by reducing stimulation. Because each of the five therapies can be programmed independently, it may be possible to program the latter therapies to induce a seizure warning. The goal of this study was to determine what proportion of patients could have tolerable symptoms safely elicited by stimulation, ultimately for the purpose of subjective seizure recognition. Of 18 patients, 12 (67%) had induced symptoms, tolerable in 11. Phosphenes were most common. We also present one patient in whom the fifth therapy was set to induce a symptom for early recognition and treatment of clusters of focal impaired awareness seizures, which were previously unrecognized and had led to days of disabling cognitive impairment. This protocol prevented disabling clusters successfully for several years. The findings suggest RNS can provide a seizure warning, potentially improving safety and quality of life, and leading to prevention of clinical seizures or clusters in select patients.

Introduction

Many patients with epilepsy are not aware of their seizures,¹ and more than half of seizures go unreported.² Contributing factors include ictal amnesia (especially with dominant temporal seizures), more chronic memory issues, and occurrence during sleep.³ Some patients report an aura before a seizure, but it is a matter of debate whether these auras are reliably associated with seizures, and whether they provide enough of a warning for the patient to take action.⁴ Modern neurostimulators, including the responsive neurostimulation system (RNS) and the latest iteration of the deep brain stimulator (DBS), include intracranial EEG recording capabilities, which have shown us that many seizures go unnoticed by patients,⁵ whether because they are “subclinical” or because the patient loses awareness during them. The RNS System uses stimulation that responds to abnormal epileptiform activity, programmed in an individualized manner. This activity may happen hundreds or thousands of times per day. Only a small fraction of these episodes are prolonged and trigger repeated stimulations. Up to five sequential therapies based upon re-detection of abnormal activity are permitted for a prolonged event with the current system.

There have been infrequent reports of acute symptoms, considered adverse effects, time-locked to the stimulation from the RNS System.⁶ In all reported cases, it has been possible to adjust stimulation such that these effects go away.⁷ If these symptoms occur reliably, are tolerable, can be induced at stimulation levels that do not induce after-discharges or further seizures, and if there is a way to stimulate them preferentially for seizures rather than for other more frequent activity, then it may be possible to use the symptoms to provide patients with an indicator of seizure activity—i.e. to provide them with an artificial aura. The warning might be a signal that seizures are occurring, have recently occurred, or are about to occur. Advantages of such a warning are that they can be implemented with the currently available system, and they do not require ancillary equipment such as a phone or watch but rely completely on the implanted system in these patients. We reasoned that if we could identify stimulation settings that are symptomatic and tolerated, then we could program them into the last (fifth) therapy, before stimulation times out. The last therapy is reached in a small minority of detected events, and only when abnormal epileptiform activity persists after four prior stimulations.

The goal of this study was to determine what proportion of RNS patients could safely be provided with symptoms from stimulation, for the ultimate purpose of potentially using these stimulation parameters for a future seizure warning system. We also incorporate a case report as a proof of concept of how such a warning system can be implemented to help prevent seizures or seizure clusters and improve quality of life.

Methods

We recruited patients age 18 and older who are followed at the Yale Comprehensive Epilepsy Center for RNS management. The study was approved by the university’s institutional review board. These patients had two, four-contact RNS electrode leads (hereafter referred to simply as leads) implanted at their seizure onset zone(s) for the management of medically refractory focal epilepsy. Each lead contains a linear array of four contacts in a strip or depth configuration with a tail connecting the lead to the neurostimulator.

Participants who consented underwent testing in clinic to see if a tolerable sensation could be elicited with a single stimulation burst in the contacts of either or both of their implanted leads. Patients were blinded to stimulation (seated such that they could not see the programming tablet and could not tell when stimulation was being delivered). Testing began with the less active lead (based on the frequency of detected epileptiform activity in recent weeks to months) and then moved to the more active lead. Stimulation started with a current of 0.5 mA and was increased gradually while estimated charge density (calculated from lead current, pulse width, number of contacts, and contact surface area) was monitored. Stimulation was increased in increments of 0.5 mA, but in participants with RNS model 320, it was sometimes increased more slowly (as low as 0.1 mA) to find the symptomatic threshold. We limited the charge density to approximately 12 μC/cm² for neocortical contacts and approximately 6 μC/cm² for hippocampal contacts. The other parameters (pulse width, frequency, burst duration) were left at the patients’ pre-existing settings, but in cases where no symptoms were identified they were modified (frequency up to 333 Hz, pulse width up to 200 μs, burst duration up to 300 ms) after reaching the maximal current to see if symptoms could be elicited.

The initial stimulation montage, which describes the endpoints of the stimulation pathway, was based on the originally programmed, clinically determined settings. The stimulation montage is reported with notation indicating the polarity (of the first phase of the stimulation pulse) of each of the four contacts in the tested lead in the first set of parentheses, starting with the tip (most distal contact), followed by the polarity of the neurostimulator case (zero for bipolar stimulation or opposite to the polarity of the stimulated contacts for monopolar stimulation) in a second set of parentheses. This was typically either a monopolar montage between the electrode contacts of a lead and the neurostimulator case or a bipolar montage between contacts of a single lead (Figure). Stimulation was tested separately in each lead.

Figure.

Example stimulation montages used during testing of stimulation-induced symptoms. The RNS System includes one or two electrode leads with four contacts on each lead, and with each lead connected to a neurostimulator. The polarity of each contact as well as the neurostimulator case are set individually as an anode, cathode, or neutral. In routine clinical care, an lead is usually configured with all contacts at the same polarity and referenced to the generator case (monopolar) or with alternating polarity (bipolar). In this study, stimulation in each lead was first tested with the original clinical settings, which typically involved all of the contacts. The stimulation field was then narrowed to identify the lowest number of contacts with which the symptoms still occurred.

If the original montage for a lead was monopolar across all contacts, either (−−−−)(+) or (++++)(−), then this montage was used with varying parameters (primarily increasing current, then modifying other parameters) until symptoms were elicited. If symptoms could be elicited, then the field was narrowed by removing one contact at a time from the stimulation to identify the smallest number of contacts needed to still produce symptoms. For example, symptoms might continue to be elicited with (-−−0)(+), then with (−−00)(+), but then not with either (−000)(+) or (0–00)(+). In this example the final montage would be (−−00)(+). If the original montage for a lead was bipolar, either (−+−+)(0) or (+−+−)(0), as was often the case for hippocampal contacts, then this montage was similarly tested with increasing stimulation levels until symptoms were elicited. If no symptoms could be elicited with this bipolar montage, then monopolar stimulation was tested instead, as described above. If symptoms could be elicited with bipolar stimulation, then smaller fields with two contacts were attempted, for example following (−+−+)(0) we would next test (−+00)(0), (0−+0)(0), and (00−+)(0). Again, the smallest field that reliably produced symptoms is reported. If no symptoms could be elicited in either lead, then as a last step we attempted stimulation of both leads simultaneously with monopolar stimulation of the same polarity. This stimulation montage typically limited the available charge density, however.

Patients were asked to report any symptoms. For those with leads in or near the occipital region, including those with longitudinal hippocampal leads inserted posteriorly, testing was done with the lights dimmed to facilitate identification of visual symptoms. The live intracranial EEG signal was monitored for after-discharges or seizures. If symptoms were elicited, the montage was narrowed (and current adjusted accordingly to maintain the same charge density) to find the smallest field necessary to produce them, and the final settings were given multiple (at least five) times to make sure the symptoms occurred reliably and did not induce after-discharges. If after-discharges were elicited, stimulation was tried with a narrower field (fewer contacts) and/or decreased amperage, duration, or pulse width. Seizures were never elicited. After testing in the office was completed, in all but one patient, settings were returned to baseline. In one patient, the identified settings were programmed into the last stimulation therapy for home use (out of five sequential therapies) as a trial of an ambulatory seizure warning system.

Results

18 patients were tested in the office, one of whom was also tested at home. In the office, symptoms were elicited with the described protocol in 12 out of 18 (67%) patients (Table 1). The most common symptoms were visual (6 patients), typically phosphenes in the contralateral upper visual field (5 patients, all while stimulating the posterior contacts of a longitudinal HC depth lead) and in one case a change in color vision from an occipital strip electrode. The next most common type of symptom was sensory (2 patients with scalp paresthesias from ipsilateral inferior frontal strips, one with hand paresthesia from contralateral thalamic contacts). One patient each had an eye twitch sensation that could be felt by the patient but was not observable to the examiner (ipsilateral posterior temporal lead), epigastric discomfort (medial temporal lead, placed orthogonally), and hot flash (longitudinally placed medial temporal lead). There were 6 patients in whom no symptoms could be elicited with our protocol, 2 of whom were limited by after-discharges, both medial temporal. The charge density required to elicit symptoms was between 2–6.6 μC/cm². In patients in whom symptoms were not seen at their usual stimulation frequency and pulse duration despite higher current, modifying the frequency and pulse duration did not yield any additional symptoms. In most cases, symptoms could be narrowed to two contacts out of the four on the lead in either monopolar or bipolar configuration. There was one case in which symptoms were only elicited by stimulating the entire lead (all four contacts). One patient reported the symptom (vision change from occipital strip) was not tolerable due to increased anxiety. Two patients (both with medial temporal leads) had after-discharges as stimulation was increased beyond their clinical settings. These patients were excluded from further testing due to risk of provoking seizures, but they are left in the result denominator to provide an unbiased estimate of how many patients might be eligible to use stimulation-triggered symptoms as a seizure warning in a prospective fashion.

Table 1.

Parameters and results of acute symptomatic stimulation testing.

Patient	Electrode lead locations*	Lead types	Symptomatic lead	Symptom	Reported as tolerated	Montage for contacts on symptomatic lead(s)†	Current (mA)	Pulse width (μs)	Charge density (μC/cm2)	ADs
1	Left medial temporal and right medial temporal-occipital	Depths	Left temporal	Ipsilateral eye twitch (subjective)	Yes	(00+−)(0)	2.5	160	5.1	No
2	Bilateral medial temporal	Depths	Right temporal	Contralateral phosphene	Yes	(00−−)(+)	4.5	160	4.5	No
3	Bilateral medial temporal	Depths	Left temporal	Contralateral phosphene	Yes	(00−−)(+)	7.0	120	5.3	No
4	Bilateral medial temporal	Depths	Both temporal‡	Contralateral phosphenes	Yes	(00+−)(0)	1.0	160	2.0	No
5	Bilateral medial temporal	Depths	Right temporal	Hot flash (“heat in upper body”)	Yes	(00−+)(0)	1.0	160	2.0	No
6	Bilateral medial temporal	Depths	Both temporal‡	Contralateral phoephenes	Yes	(−+−+)(0)	2.2 left, 2.5 right	160	2.3 left; 2.6 right	No
7	Bilateral medial temporal	Depths	Right temporal	Contralateral phosphene	Yes	(00−−)(+)	2.5	160	2.5	ADs only when stimulating all contacts
8	Right temporal depth (orthogonal) and right frontal strip	Strip and depth	Right temporal depth	Epigastric sensation (“twinge in stomach”)	Yes (similar to habitual aura)	(000−)(+)	3.0	160	6.1	No
9	Left medial temporal, anterior thalamic nucleus	Depths	Neither	None						Yes in temporal lead at 5 μC/cm2, still present with posterior contact only
10	Left medial temporal (longitudinal hippocampal, longitudinal entorhinal)	Depths	Neither	None						Yes in hippocampal lead (2 μC/cm2, still present with posterior contacts only)
11	Right medial temporal (longitudinal hippocampal, longitudinal entorhinal)	Depths	Neither	None						No
12	Right medial temporal (longitudinal hippocampal, longitudinal entorhinal)	Depths	Neither	None						No
13	Left orbitofrontal	Strips	One of two parallel strips	Ipsilateral scalp paresthesia (“tingling”)	Yes	(00–0)(+)	3.0	160	6	No
14	Left lateral frontal	Strips	Both frontal‡	Ipsilateral scalp paresthesia (“nerve pinch”)	Yes	(−−00)(+)	6.5	160	6.6	No
15	Inferior frontal and temporal	Strips	Neither	None						No
16	Right parietal	Strips	Neither	None						No
17	Left superior and inferior occipital	Strips	Inferior occipital	Colors change, entire visual field	No (anxiety)	(00++)(−)	4.0	160	4.1	No
18	Right centromedian thalamus depth and lateral parietal strip	Depth and strip	Right thalamic	Contralateral hand dysesthesia	Yes (if brief)	(−+00)(0)	2.0	80	2.0	No

^*Medial temporal depth leads were implanted posteriorly along the long axis of the hippocampus except for patient #8, who had it placed orthogonally.

^†Montage for the symptomatic leads(s) shows the polarity of the four contacts, starting with the most distal contact (tip) and proceeding to the most proximal contact, with + or – indicating positive or negative charge and 0 indicating no stimulation, followed in separate parenthesis by the polarity of the neurostimulator case. Refer to the Methods section and Figure for further explanation about this nomenclature.

^‡Stimulation settings and montages that induced symptoms were the same in each of the two leads except as noted.

Case Report

In one patient (#1 in Table 1), one of us (LJH) programmed the symptomatic stimulation settings for clinical care and long-term use. The patient was a young woman with medically refractory bitemporal epilepsy with focal impaired awareness seizures. After implantation of bitemporal RNS, more than half of her seizures came from the right (non-dominant) hippocampus, where she had near-continuous spiking interictally. After Wada testing demonstrated poor function on that side and intact function on the left, she underwent a right anteromedial temporal resection. She continued to have occasional prolonged clusters from the remaining temporal lobe, but because she was unaware of discrete seizures from that side, these were not recognized until she had days of marked amnesia. This was highly disabling as she was a college student at the time. Review of her RNS diagnostic data showed that device detections that were long enough to receive the maximum of five therapies occurred only on days when she also had definite seizures, whereas her more common detections only triggered 1–2 therapies before stopping. Stimulation testing in clinic induced a sensation of an ipsilateral eye twitch (not visible on examination) at 5.1 μC/cm² (160 μs, 333 Hz, 1 burst); this setting was left in place to provide her a warning that she was having prolonged runs of epileptiform activity and likely entering a seizure cluster. When she felt several eye twitches in one day, she would take rescue medications (a benzodiazepine plus increased doses of her regular anti-seizure medication) and minimize certain activities on those days. This completely prevented the prolonged clusters and the days of severe amnesia, allowing her to complete her undergraduate studies successfully.

These settings have now been in place for nine years, though she now rarely needs the cluster protocol as her seizures have continued to decrease over the years. Subsequent review of her device activity showed that she typically had multiple events that triggered five therapies within a few hours prior to the much longer and higher amplitude electrographic seizures known to correlate with her clinical seizures.

Discussion

We demonstrated that in most patients (at least 12/18, or 67%) it is possible to deliver a stimulation burst through the RNS System that causes symptoms and is not associated with after-discharges. In all but one of those cases, the symptoms were tolerable. These settings can be identified safely and quickly during a routine clinical visit for RNS programming. Although this type of symptomatic stimulation has been considered an adverse effect that can be avoided, we propose that it can be used as a form of warning signal (providing an aura or prodrome) to give patients a better sense of impending or ongoing seizures that might otherwise go unrecognized. We also tested a warning system at home in one patient, by programming the warning to occur during the fifth therapy, which is the final one before therapy is exhausted. This allowed her to recognize impending seizure clusters very early and enabled her to take rescue medication to prevent clusters and the associated prolonged memory dysfunction associated with them. This warning and rescue protocol was used successfully for several years and is still in place.

As seizure detection and prediction via implanted devices improves, this type of process could be made more specific, but we have seen in general that event duration, including reaching the fifth and final therapy for a given event, is a fair surrogate for days in which patients have seizures. This correlation needs to be determined in each individual patient, preferably quantitatively. Our proof-of-concept demonstration has some key limitations that need to be addressed to address real-world feasibility in further patients. One important concern is the rate of false positives, which is likely to vary between patients. Another related question is whether these symptoms are reliably triggered around the time of, rather during, seizures, particularly seizures with loss of awareness. The stimulation testing presented here was done interictally during a clinic visit, and the settings were only tested at home in one patient. This type of stimulation will need to be confirmed prospectively in additional patients to confirm that symptoms would be triggered and noticed in the real world during or around the time of seizures.

It remains to be seen whether this type of seizure warning remains reliable and well tolerated in the long term across multiple patients. The concept of providing patients with an indicator of likely imminent seizure activity has been piloted before,⁸ but our proposed method has the advantage that it can be used in patients with the existing device that is FDA approved in the US and does not require incorporating ancillary devices such as mobile phone or wrist-based notification.

Key Points.

1. In 12 out of 18 patients with the responsive neurostimulation system, symptoms could be elicited safely by adjusting stimulation parameters.

2. Repeated stimulation was tolerated by all but one patient in whom symptoms could be elicited.

3. In one patient who was unaware of her focal seizures, a stimulation-induced symptom was delivered for activity lasting long enough to trigger five sequential therapies.

4. This patient has used symptomatic stimulation as an indicator of impending seizures that she had otherwise not recognized, triggering rescue medication use and successfully preventing seizure clusters for years.

5. The currently available RNS System can be used to deliver seizure warnings via symptomatic stimulation.