Back in 2021, we wrote a post with the subtitle “A personalizable seizure monitoring wearable for home use that extends the capabilities of specialized hospital equipment”.Back in 2021, we wrote a post with the subtitle “A personalizable seizure monitoring wearable for home use that extends the capabilities of specialized hospital equipment”. That post focused on the “why”, meaning we defined several unmet clinical needs in epilepsy and discussed innovations that we considered essential to meet those needs.

‍

We identified a need for improvements in:

1. The delivery of state-of-the-art care for people with epilepsy.
2. The development of new treatment strategies, especially for treatment resistant epilepsy.
3. Reducing the unpredictability of seizures.
4. Managing stress, treatment side effects, and epilepsy comorbidities.

Since then, we have stayed passionately committed to the mission of tackling them on two fronts:

1. Leveraging the suite of Byteflies remote patient monitoring (RPM) tools to build a fit-for-purpose seizure monitoring service for use in clinical care and drug development.
2. Identifying like-minded people and organizations. RPM and virtual care solutions in neurology are still very much a “zero to one” story despite some important innovations, such as neurostimulation and -modulation devices. In contrast, fields like cardiology have a well-established history of using innovative technology to monitor patients (ultra)long-term. In neurology and epilepsy, it will take a village as the -somewhat tired- cliché goes. Consequently, we also focus on identifying the inhabitants of that village.

‍

For a more in-depth discussion on the unmet needs in relation to the current standard of care, have a look at the original post.

‍

This post will focus on what we have built since then, and how it is being used in practice to help people with epilepsy and their families right now.

‍

Table of Contents

‍

• Crucial technological enablers
• Which patients and seizure types?
• Clinical acceptance and reimbursement
• The village

‍

Crucial Technological Enablers

‍

Byteflies Care@Home products are bundles of hardware and software medical devices, combined with data annotation and logistical services, which are purposefully put together to cover a specific disease indication or care path. The figure presents that same idea more schematically:

‍

‍

EpiCare@Home combines medical wearable devices that acquire a low-montage electroencephalogram (EEG), which can be further augmented with cardiorespiratory and motion data. We have covered this set up and how it compares to a traditional in-hospital video-EEG in a previous post.

‍

Since then, we added two important technological enablers that make EpiCare@Home fit-for-purpose in its intended dual use environment, namely the patient’s home and in clinical neurology workflows:

‍

1. User-friendly adhesive electrode “patches”; and
2. Data annotations services

‍

Electrode Patches

‍

Because scalp EEG is a finicky signal to record and prone to technical measurement artifacts, high quality “cup” electrodes are typically used in the hospital or for ambulatory EEG recordings. These electrodes are glued to the scalp and filled with a conductive paste to generate a low resistance path from the scalp to the recording device. Although EpiCare@Home can use these types of electrodes, they come with a major usability downside. No one likes to have electrodes glued to their scalp for long unless absolutely necessary and they are almost impossible to reapply without help from someone trained to do so.

‍

Therefore, we set out to develop an entirely new EEG “patch”, which we call “EEG Adhesives”, with embedded electrodes that use a discreet “behind-the-ear” design so that most of the components can be hidden out of sight, to minimize any visual stigma.

‍

Back in 2019, Byteflies together with several academic and industrial partners were awarded a research grant (Plug ‘n Patch) to explore the design, production, and validation of various patch designs with embedded biopotential electrodes.

‍

When developing these EEG Adhesives, or any type of medical skin adhesive for that matter, it is important to understand how long users will be expected to wear them. Based on conversations with epileptologists, we identified multi-day up to multi-week continuous EEG monitoring as a good trade-off between patient comfort and diagnostic yield (see “Which Patients and Seizure Types” for more details).

‍

‍

In early 2020, that led to the first small scale prototype production of several designs, some of which are pictured below. Volunteers and academic researchers extensively tested each design to assess usability with a particular focus on wear comfort, and -of course- signal quality.

‍

‍

Specifically, we needed the find a balance for the following quality attributes:

‍

• Biocompatibility, and in particular breathability that allows a monitoring period of up to 4 weeks, with no or as little skin irritation as possible.
• Inter-electrode distance as it relates to the ability to record a meaningful EEG signal and compatibility with a broad range of ear shapes.
• Electrode technology: dry, solid gel, hydrogel as a function of signal quality and biocompatibility.
• Adhesion: strong enough to provide good EEG signal quality for 2 days while light enough to not impede easy removal without glue residue.

‍

Although they look deceptively simple, each EEG Adhesive is actually a stack of ten components that must seamlessly come together, and our manufacturing partners need to be able to produce them at scale.

‍

Based on the collected feedback, we went through eight (!) more production iterations to identify the optimal balance of user comfort and signal quality (hardware is hard … anyone?!). Each iteration went through extensive user testing and signal quality assessments, compared to reference cup electrodes under activities of daily living (sleeping, walking, working, etc.), and had to pass strict medical device requirements.

‍

‍

Although we’ve had great success with dry electrode technology for recording cardiac signals, for the much lower amplitude EEG signal, this turned out to be too complex. Therefore, one of the design decisions made was to use hydrogel electrodes. This common type of electrode has a skin-friendly gel that permeates the surface layer of the skin, thereby lowering resistance, similar to the function of the conductive paste in the cup electrodes.

‍

In late 2022, we settled on the following hydrogel design:

‍

‍

Of course, technology never stops, and we are actively working on next-gen EEG Adhesives that further improve usability by doing away with the wires and snap connectors. That said, our current generation of EEG Adhesives works very well, and we are beginning to roll these out to our users.

‍

We are grateful to all partners in the Plug ‘n Patch consortium who tirelessly worked to make this possible. In fact, one of our research partners attempted to use investigational prototype EEG Adhesives for up to 8 months per patient. This research came with many technical challenges but is nevertheless an interesting data point towards a future where material sciences keep improving skin-compatible materials for ultra long-term usage (multiple months and up).

‍

Data Annotation Services

‍

A second technological enabler has to do with data annotation. Wearable devices and other RPM solutions are collecting valuable data over increasingly longer time periods. The major benefit is that this should greatly improve the diagnostic value of the data. This is especially true in epilepsy, where seizures can be far and few between (but therefore no less debilitating). With the ability to collect long-term high density clinical-grade data, comes the need to convert that data into clinically actionable information and do it in a way that does not increase the data review burden for healthcare professionals, who are already strapped for time.

‍

If you are now thinking “well, that’s what algorithms are for… duh”, you’d be right but that’s only part of the story. Algorithms to be used in this context are “Software as a Medical Device” (SaMD) and come -rightfully so- with strict performance and safety requirements dependent on their intended use.

‍

Machine learning (ML) has made great strides in the last half decade or so, including for the processing of physiological signals such as EEG and electrocardiography (ECG). It is a powerful set of tools to develop highly performant algorithms relatively quickly on the condition that you have access to high quality labeled data. In a supervised setting, to allow the “machine” to learn from the data, it first needs to know which patterns it is expected to identify. In the case of EpiCare@Home, this labeled data needs to be generated by our system as we use an atypical reduced behind-the-ear EEG montage, augmented with other cardiorespiratory and motion signals. No historical databases are available that we could use retrospectively to train an ML model, at least not one that would perform well in the real world. Generating high quality labeled data is time intensive, immensely costly, and requires a lot of technical expertise but, in this case, there is no way around it.

‍

That is why we decided to roll out EpiCare@Home combined with a human-driven annotation service. This means that data from the wearable devices is reviewed by certified annotators. The events they annotate as “potential seizure events” are shared with the neurologist for final validation. This has two advantages:

‍

1. People with epilepsy can benefit from EpiCare@Home now because the data annotation service allows a healthcare professional to use it without increasing their data review burden.
2. We continuously generate high quality labeled data based on real-world data collection, which is needed to continue to develop the certified algorithms mentioned above.

‍

Generating high quality labeled data is time intensive and immensely costly but there is no way around it.

‍

And of course, we are working on that second step. While trained annotators are very good at identifying relevant events, they are still human and can only process so much data in a day. A performant algorithm can process data exponentially quicker; not with the intention of removing human annotators from the equation completely, but to allow them to focus on the more complex cases. In other words, once our ML model is ready for prime time, it will become the annotator's best friend.

‍

Over the years, we have collected data on more than 600 people with epilepsy, accounting for more than 4000 seizures as part of the academic SeizeIT1 and SeizeIT2 studies via long-standing collaborations with partners in the EU and UK. In these studies, data was collected in-hospital with video-EEG and behind-the-ear EEG simultaneously, to identify the seizure types that can be reliably detected in this configuration. This data was also used to develop seizure detection algorithms which resulted in multiple publications (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14) with the following high-level takeaways:

‍

• A behind-the-ear low-montage EEG set up can in many cases be used to identify epileptic seizures reliably, especially when a prior diagnosis of epilepsy exists. See for instance references # 3 (focal seizures), 9 (focal temporal lobe seizures), 12 (absence seizures), and 13 (focal and absence seizures).
• The ability to identify seizures (by algorithms and/or human annotators) can be improved by including additional signal modalities, such as heart rate and motion. See references # 6 (cardiac, muscle, and physical activity) and 10 (cardiac and brain activity) for examples. These so-called multimodal approaches hold a lot of promise for performant real-world seizure detection with unobtrusive wearables, especially when no video feed is available.

‍

So … problem solved, right?

‍

Not quite.

‍

First, most of these studies developed and validated their algorithm on a single and small dataset which always runs the risk that the results will not translate well to the real-world. In a recent example, an algorithm that was developed based on a subset of the in-hospital SeizeIT1 data was used “as is” to identify seizure events on ambulatory home data without accounting for the difference in setting (hospital vs. home) and the use of different electrode technology to acquire the data. The algorithm underperformed, while visual inspection of the same dataset by a trained annotator would have yielded much better results. Similarly, upgrading the original algorithm to “learn” from more relevant data could have improved the results.

‍

Secondly, recall that we are working in a medically certified space and algorithms that were developed for academic purposes (and thus publication), usually were not designed to meet the patient safety and performance requirements of a SaMD.

‍

Some work has also been done on “personalized algorithms”. The idea is that you start from a base algorithm that is then tweaked with prior knowledge about the patient’s seizure signature to increase detection performance. A major downside of this approach is that it tends to “overfit” an algorithm to a known set of historical data which may be hard to obtain in the real-world. In addition, it may not account for the fact that people with epilepsy often experience changes in seizure types and burden.

‍

Medical algorithms are “Software as a Medical Device” (SaMD), which comes with strict quality, performance and safety requirements.

‍

The TL;DR so far is that the usability and clinical performance of EpiCare@Home has improved significantly by introducing EEG Adhesives and adding data annotation services. Combine that with the exploratory and validation research that happened over the last couple of years, and we are ready to make the technology available to a wider range of people with epilepsy and gradually scale up the volume of data it can handle.

‍

Which Patients and Seizure Types?

‍

The number one question we receive about EpiCare@Home is:

‍

Which seizure types can it monitor?

‍

Excellent question. And the answer? Many, but … it depends. Before you get your gallows out of storage, hear us out.

‍

The prescribing neurologist should consider the following two questions before using EpiCare@Home:

‍

1. What do I already know about my patients’ seizures?
   - Is their seizure signature generalized, focal or a combination?
   - How frequent are their seizures?
   - In the case of focal seizures, is the location of the EEG electrodes (behind-the-ear) likely to record them (e.g., consider temporal vs frontal lobe seizures)?
   - Are there other physiologic or contextual signatures (e.g., cardiorespiratory, motion) that EpiCare@Home may be able to pick up?
   ‍
2. What diagnostic question(s) am I trying to answer?
   - Am I trying to establish an initial diagnosis of epilepsy? If so, is a reduced EEG montage a good option?
   - Am I trying to acquire additional diagnostic information to reduce the seizure burden of a refractory patient with a long-documented history of epilepsy?
   - Am I evaluating a change in treatment? For instance, to reduce side effects or due to a change in circumstances (e.g., pregnancy).

‍

Epileptologists are taught “don’t treat the EEG”, as it may lead to overinterpretation, misdiagnosis, and even unnecessary treatment. Because EpiCare@Home does not include a video feed, its other signals (heart rate, respiratory rate, motion, seizure diary events …) provide important contextual and physiologic information that assist with reading the EEG. Our previous post discusses in more detail how we can compensate for the lack of video.

‍

In conversations with many epileptologists, we have refined its position in the care path for people with epilepsy as follows:

‍

‍

Consider the following two examples, loosely based on EpiCare@Home patients:

‍

Based on reports from a parent, a pediatric patient is suspected to have typical absence seizures. A brief (1 hour) video-EEG in the hospital does not yield a conclusive result. The epileptologist decides to send the patient home with EpiCare@Home for 3 days. From the data, eight electrographic absence seizures of more than 3 seconds are identified which provides enough information for the epileptologist to start a treatment plan. One month later, the patient is monitored again for 3 days and this time no seizures are identified which matches the reports from the parents.

‍

‍

An adult patient with a decade long history of refractory focal impaired awareness (FIA) epilepsy self-reports a recent increase in seizures. He has spent many weeks in video-EEG monitoring units over the years and is not looking forward to doing it again. Because the epileptologist knows that his seizures are primarily occurring in the left temporal lobe, she suggests EpiCare@Home for one week, followed by a dosage adjustment and another week of monitoring. She does make it clear that if the data is inconclusive, he may need to come into the monitoring unit after all. The first week of monitoring yields three seizures, the second only one. The patient also reports an improvement. The epileptologist suggests maintaining this treatment regimen for a couple more months and then check-in again, potentially with another one week of EpiCare@Home monitoring.

‍

‍

These are just a few representative examples, and we have more. In fact, we presented retrospective data at the Antiepileptic Drug and Device Trials (AEDD) and American Epilepsy Society (AES) meetings in 2022 that the patients, caregivers, and neurologists using EpiCare@Home, reported a positive clinical or quality of life impact in 93% of cases. The diagnostic data generated by EpiCare@Home was used as follows in 22 case studies (including 8 children):

‍

• ~30% reported an existing treatment regimen was adjusted, or a new treatment was started.
• ~30% reported confirmation of a known diagnosis or a referral to another specialist.
• ~15% maintained a treatment plan.
• ~15% reported an improved care experience.

‍

Two more video impact stories from parents with children with epilepsy who were helped with EpiCare@Home can be found here and here.

‍

Of course, we need systematic and preferably prospective data analysis to assess the value and clinical acceptability of EpiCare@Home at scale in the care trajectory of people with epilepsy.

‍

So, what has already happened on that front and what is ongoing?

‍

As mentioned previously, the SeizeIT studies acquired a large body of data. The clearest example to date of a clinical workflow assessment based on that data is a paper that evaluated both algorithmic and blind reading by annotators of 284 absence seizures. The results demonstrated that EpiCare@Home provides a clinically viable method to monitor absence seizures as compared to video-EEG. A major limitation is that all data was recorded in the hospital and not yet at home.

‍

Similarly, the results of small-scale blind reading studies were presented at the Epilepsy Foundation Pipeline Conference in 2020. For absence seizures, 96% sensitivity was reached based on 155 seizures. For focal seizures, 83% sensitivity was reached based on 46 focal seizures. When ECG and electromyography (EMG) data were added, the sensitivity increased to 93%. In both cases, the recorded data vastly outperformed self-reporting via seizure diaries.

‍

Additional analysis of the SeizeIT datasets by academic partners is ongoing.

‍

Finally, we have recently started a prospective clinical trial at Thomas Jefferson University (TJU) to focus specifically on focal epilepsies.

‍

Clinical Acceptance and Reimbursement

‍

In Belgium and Germany, we and epileptologists have witnessed the impact of EpiCare@Home in clinical practice firsthand, in pediatric and adult patients, and in generalized, focal and rare epilepsies (including so-called developmental epileptic encepalopathies or DEEs).

‍

A retrospective analysis of EpiCare@Home case studies found that in 93% cases, a positive impact was reported by either patient, caregiver, and/or neurologist.

‍

Now, we are at an important crossroads since EpiCare@Home is not yet reimbursed in Belgium or abroad. We are working hard to collect reimbursement evidence, while continuing to roll out EpiCare@Home. This results in a typical digital health “catch 22”: the lack of reimbursement means that it is difficult for healthcare professionals to implement it in their workflows, while we need to be looking for other sources of funding to bankroll evidence collection.

‍

Fortunately, we are far from alone in realizing this. Health Technology Assessment (HTA) frameworks for rapid evaluation of new and promising digital health technologies are popping up everywhere and many regulators are realizing that the dogma of the Randomized Clinical Trial (RCT) does not apply everywhere. Even more encouraging is that harmonization efforts are under way in the EU to recognize HTAs across member states.

‍

The digital health catch 22: how to fund the roll out of promising and safe technologies while moving through the slow reimbursement process?

‍

Nevertheless, these efforts will not result in concrete results overnight and much work remains to be done, so until then finding the right partners is crucial, which brings us to … the village.

‍

The Village

‍

Over the years, we have built up a diverse network of partners and like-minded individuals who share the mission: to improve the standard of care for people with epilepsy, with a particular focus on virtual care solutions that are patient-centric and value-based.

‍

In neurology and epilepsy, this is imperative as no single solution will “fix” the unmet clinical needs we outlined above. Instead, we must work together to bring a diverse and complementary set of digital health solutions to the patient.

‍

One concrete example would be to provide several seizure RPM tools to neurologists and epileptologists so they can progressively “ramp up” the sophistication of the monitoring tools, knowing that that usually goes hand in hand with the obtrusiveness of the technology. For instance, solutions like EpiCare@Home could provide a logical steppingstone towards sub-scalp ultralong term EEG monitoring devices.

‍

We must work together to bring a diverse and complementary set of digital health solutions to the patient.

‍

We already mentioned some of the partners and it is beyond the scope of this post to mention everyone that has contributed to this mission over the years. We do want to include a shoutout to the following partners, who have gone above and beyond to rally stakeholders:

‍

The Epilepsy Foundation, in particular the Antiepileptic Drug and Device Trials and Pipeline conferences, which always bring together a critical mass of technology enablers. And last but most certainly not least, UCB Pharma’s digital health focused teams for their leadership in recognizing the value of EpiCare@Home and similar solutions and their continued efforts to bring them to market and innovate the clinical trial space.

‍

The information contained in this article represents the views and opinions of the writer(s) and does not necessarily represent the views or opinions of other parties referenced or mentioned therein.

The article is not intended to be a substitute for professional medical advice, diagnosis, or treatment. Always seek the advice of your physician or other qualified healthcare professional with any questions you may have regarding a medical condition. Never disregard professional medical advice or delay in seeking it because of something you read in this article.

‍