Significant advancements in genetics and the implementation of artificial intelligence have begun to carve a new, more personalized path for the diagnosis and treatment of seizure disorders.
Anup Patel, MD
The armamentarium of antiepileptic drugs has grown rapidly since 1990, more than doubling since the 1993 approval of felbamate for partial seizures with and without secondary generalization in adults.1 More recently, improvements in areas such as genetic sequencing have pushed the field of epilepsy care into a new domain, where the driving focus is no longer symptomatic treatment but proactive prevention and care designed with the individual patient in mind.2,3
Some experts have argued that research in this field is poised to deliver the precision medicine model—particularly in the develop-ment and deployment of targeted therapeutics—because of gains in the understanding of genetics in epilepsy, the wide availability of in vitro and in vivo model systems and their translation into drug-screening platforms, and an established base of collaborative research organizations.4
“We’re doing better in caring for patients with epilepsy, but we obviously still have gaps,” Anup D. Patel, MD, associate professor of clinical pedi-atrics and neurology at The Ohio State University and section chief of pediatric neurology at Nationwide Children’s Hospital in Columbus, told NeurologyLive®. “Specifically, from a treatment standpoint, 30% of patients still remain unable to become seizure free.”
Daniel Lowenstein, MD, professor of neurology, executive vice chancellor, and provost at the University of California, San Francisco (UCSF), noted that this number has remained relatively unchanged over the past 50 years.5 In an interview with NeurologyLive®, Lowenstein, who previously served as director of the UCSF Epilepsy Center and Epilepsy Research Laboratory, added that currently available drugs still carry with them limitations and drawbacks for patients whose seizures can be brought under control.
“There are certain syndromes where we know there is a selection or a portion of the [available] drugs that are more likely to be helpful and some that, in fact, can cause the epilepsy to get worse,” Lowenstein said. “But in most patients, doctors really are at best making an educated guess—and I’m being charitable when I use that phrase—in terms of picking which drug is most likely to work. We have to embark on this therapeutic odyssey of hoping that the first drug we choose is going to be the most effective. And when the patient has their seizures brought under control, they may have a certain amount of adverse effects.”
The ultimate goal of personalizing medicine for each patient requires that a number of pieces come together. The oncology community has already assembled several of them, with precision medicine allowing for more targeted and effective treatments for genetically driven cancers. Even with the broad phenotypic heterogeneity seen in epilepsy, the complexity of genotype-phenotype correlations, and the difficulties in translating the array of identified targets, investigators and clinicians are hard at work in identifying a path forward,2,4 led by advancements in genetics, artificial intelli-gence (AI), and other technology.
As investigators have identified more genes, advances in testing and next-generation sequencing have revolutionized the ability to provide genetics-based guidance in the laboratory and the clinic. In addition to providing answers to patients, genetic sequencing has opened the door for identifying potential protein targets. The best example of this may be in SCN1A-associated seizure disorders, which represent a spectrum of epilepsies that physicians can in part manage with a number of currently available pharmacologic agents, including clobazam, stiripentol, benzodiazepines, cannabidiol, topiramate, levetiracetam, valproic acid, and ethosuximide.6
However, large gaps in treatment remain for patients with the more severe syndromes associated with the gene, namely Dravet syndrome and intractable generalized tonic-clonic seizures. A group of investigators including Ian Miller, MD, director of the Epilepsy and Neurophysiology Program at Nicklaus Children’s Hospital in Miami, Florida, presented a novel approach to adeno-associated viral (AAV ) vector—based gene therapy for SCN1A-positive epilepsy at the 2019 American Epilepsy Society Annual Meeting. The study, which used a mouse model of Dravet syndrome, showed that a γ-aminobutyric acid–selective AAV vector significantly reduced hyperthermic seizures, decreased the frequency and duration of electrographic seizures, and significantly reduced the risk for sudden unexpected death in epilepsy by 89% when administered to SCN1A-positive and -negative mice.7,8
“Unlike the traditional model or the default mental picture that people have of taking an extra copy of SCN1A and putting it into the nucleus somewhere, the approach here is actually putting something into the nucleus, which, when translated, will increase transcription of SCN1A,” Miller told NeurologyLive®. “It’s a very regulation-centric model rather than a copy number model. That approach brings down the size of the payload that’s needed to induce the effect. It makes it possible to deliver this well-characterized, relatively safe virus and makes it really exciting for patients.”
Other research has illustrated this paradigm in epilepsy, evidenced for example by the early-stage success of quinidine in the treatment of patients with KCNT1-positive epilepsy. One of 3 patients self-re-ported an 80% reduction in seizures, and another saw similar results accompanied by improved psychomotor development.9,10 Likewise, GRIN2D-related epileptic encephalopathy has seen a gain-of-function effect with suppression of N-methyl-D-aspartic acid receptor activity, demonstrating potential for the patient population to derive benefit from treatment with memantine. An assessment of 2 children with gain-of-function GRIN2D mutations showed an overall mild to moderate improvement in seizures.11
Investigators have also achieved some success in loss-of-function or dominant-negative mutations, such as those observed with the potassium channel-activator retigabine (also known as ezogabine) in KCNQ2-related epileptic encephalopathy. Results of a retrospective study of 11 children showed that 75% of those treated before age 6 months had seizure improvement, whereas clinical response was lessened after 6 months, with only 28% experiencing improvement."12
We know from epidemiological studies that genes play a role in probably at least 60% of people who develop epilepsy,” Lowenstein said. “So even though the early wins are in relatively rare forms of severe epilepsy, I’m quite optimistic that this work is going to have an impact on a much, much broader number of patients in the world of epilepsy. Even with acquired epilepsies, for example, the question is: Why do they develop epilepsy? We really don’t have a very good answer to that. I can’t tell you why someone who gets a bad head injury goes on to develop lifelong epilepsy, but we do know that there is at least a mild genetic influence on the likelihood of developing epilepsy after an acquired injury.”
With more motivation than ever to uncover the underlying cause of seizure disorders, several pharmaceutical companies have stepped up to help fund genetic testing. Behind the Seizure, a program spear-headed by BioMarin Pharmaceutical and Invitae Corporation, is currently supported by 8 sponsor partners. The collaboration offers screening via Invitae’s comprehensive Epilepsy Panel at no charge to any child under age 8 who has had an unprovoked seizure. According to the initiative, participants in the program have received diagnoses 1 to 2 years earlier than reported averages.13
Access to testing has already helped propel the field by providing investigators with a better knowledge base in certain epilepsy syndromes associated with genetic variations. “We’ve really moved that field forward fast and are on the cusp of, potentially, some really cool studies that will look at that from a gene standpoint,” Patel said, noting that the epilepsy community has borrowed strategies and experiences from other neurology specialties, such as the neuromuscular field, which has found some success with genetically targeted therapeutics.
Ultimately, the use of genetic screening information will have a significant impact on not only diagnosis but also clinical management, in which a known mutation may, at the least, help a patient avoid unnecessary invasive testing and reduce the frequency of brain imaging3 and, at most, help simplify the proposed treatment course. One such example is the case of a patient who avoided epilepsy surgery upon the identification of a pathogenic SCN1A variant, after which a switch off of carbamazepine resulted in seizure control.14
Although genetic advancements have become a cornerstone of the pathway to precision medicine, another major player in the process is AI, particularly in its development of algorithms for clinical deci-sion making. Just as tapping genetic data can better inform treat-ment response for the individual patient, the incorporation of algo-rithmic data into the physician’s toolkit shows similar potential.
At the 2019 International Epilepsy Congress in Bangkok, Thailand, Lara Jehi, MD, associate professor of neurology at Lerner College of Medicine, codirector of Network Capacity for the Clinical and Translational Science Collaborative, and chief research information officer at Cleveland Clinic in Ohio, spoke with NeurologyLive® about the use of AI-developed algorithms as a step toward precision medicine, calling them “the ultimate example of personalized [treatment].”15
“We talk a lot about precision medicine as being the way of the future, and when we use the term precision medicine, it’s equated in our mind to the use of genetic data to target treatments,” Jehi said. “But we are more than our genetics. We are what we get exposed to, we are—in the world of epilepsy—our electrophysiological data, we are our imaging data. There are a lot of other variables that influence how we respond to any type of treatment.”
Following this philosophy, Jehi and colleagues created the Epilepsy Surgery Nomogram, an online risk prediction tool that generates the individualized likelihood of complete seizure freedom at 2 and 5 years postepilepsy surgery. The risk predictor was built through an analysis of more than 800 patients undergoing epilepsy surgery at Cleveland Clinic and validated in a group of more than 600 patients at Mayo Clinic and other centers worldwide. In the original cohort, the rate of complete seizure freedom was 57% at 2 years and 40% at 5 years, whereas results of the validation study showed a nomogram concordance statistic of 0.60 for complete seizure freedom, below the 0.80 threshold of strong concordance.16
In June 2017, the National Institutes of Health granted Jehi and colleagues a $3.4 million, 5-year grant to improve the model with additional variables such as electroencephalogram and magnetic resonance imaging (MRI) data, family history, and genetic information.17
The power that these data hold are greater than the sum of their parts, a dynamic demonstrated with the introduction of learning health care systems, which embed best practices—assembled and analyzed from various sources and means—into the care delivery process. Although these systems do not follow set standards and can vary in scale and manifestation, they have aided in the progress of intelligent automation, comparative effectiveness research, positive deviance, surveillance, predictive modeling, and clinical decision support.18
“It is an advance in technology that I really expect can help us. It’s a little bit of AI, but it’s also clinical decision support,” Patel said. “Can we craft our EHRs to nudge us in the right direction based on [existing] data on specific patients within the medical record? The answer to that is yes. We’re starting to do that now, and this learning health care system allows us to go even further.”
Machine learning techniques have also received increasing atten-tion for application in epilepsy, in, for instance, automated seizure detection from electroencephalography, video, and kinetic data; automated imaging analysis and presurgical planning; the prediction of medication response; and prediction of medical and surgical outcomes.19 One assessment of a support vector machine classifier revealed a peak diagnostic sensitivity of 82.5% and a specificity of 85.0% by evaluating the asymmetry of functional connectivity in homologous brain regions on resting‐state functional MRI in 100 patients with epilepsy and 80 controls.20
Similar techniques have suggested that machine learning from a range of clinical data can also deduce an epilepsy diagnosis. An examination that applied several classification algorithms to 105 scanning electromyography recordings from 9 juvenile patients with myoclonic epilepsy and 10 controls revealed a 100% diagnostic sensitivity and 83.6% specificity using an artificial neural network.21
Recently, digital health company Doc.ai announced the Epilepsy Digital Health Trial, led by Robert Fisher, MD, PhD, the Maslah Saul MD Professor in the Department of Neurology and Neurological Sciences at Stanford University Medical Center and director of the Stanford Epilepsy Center, both in California, to deploy and test AI in the development of a predictive model for epilepsy treatment.22
The trial aims to enroll up to 1000 participants by September 2020. Once enrolled via the Doc.ai app, participants will keep an online diary tracking their seizure episodes and adverse effects of medications for 3 months. The company will analyze a number of health data sets, with additional leveraging of Doc.ai technology, such as natural language processing, also planned.22
Although the advances made in genetics, AI, and other technologies are on their own enough to warrant a positive outlook, the key to achieving precision medicine is the melding of these facets. Similar to a combination medication approach, all these advancements, when working in tandem, will help to ultimately drive improvements in patient care. Although the medical community has already covered much ground, a vast amount of unknowns in the world of epilepsy remains.
“The challenge is that we’re dealing with such an unbelievably complex system that being able to come up with a therapy that doesn’t affect the normal function of neurons in terms of the various things that govern their excitability will be an ongoing challenge,” Lowenstein said. “I don’t think we have the tools yet to be able to understand the complexity of the human brain. That’s a pretty simplistic statement, but I really do think that history will show that in the 20th and 21st centuries, we made tremendous advances in understanding the genetics of epilepsy.”
Perhaps the only thing more complex than the brain itself that can stand in the way of adopting and applying these new tools is physi-cians’ own stubbornness, according to Patel.
“One of the things about the medical community is we’re very change averse. It’s going to take time, it’s going to take energy, and it’s going to take enthusiasm to actually help my peers and colleagues to be able to adapt and utilize these things,” Patel said. “We’ll have to be creative in how we’re going to overcome that because people are very set in their ways, and we’re now going to be potentially asking them to do things differently or to utilize other sources of data, whether it’s a genetic test or data from the EHR to help you provide better care.”
Despite these challenges, the field is already reaping some of the benefits of precision medicine. One example is the 10-year international collaborative effort led by the International League Against Epilepsy (ILAE), which in 2017 published new classifications of epilepsies and seizure types, including etiology together with clinical features as a principal development. Some of the new seizure types identified were focal myoclonic, focal tonic, focal spasms, and absences with eyelid myoclonia, marking the first official update to the ILAE classification since 1989.23
As this multipronged approach to precision care in epilepsy continues to improve, the large number of patients with epilepsy that is not well controlled can and will be reduced thanks to efforts to promote early genetic testing, assessment, and translation of therapeutic targets—coupled with the use of data-driven decision-making tools that are poised to push the epilepsy field further forward each year.
“Science has shown over and over again that it’s the tools that are created and discovered that really enable the major leaps forward,” Lowenstein concluded.
1. Cavalleri GL, McCormack M, Alhusaini S, Chaila E, Delanty N. Pharmacogenomics and epilepsy: the road ahead. Pharmacogenomics. 2011;12(10):1429-1447. doi: 10.2217/pgs.11.85.
2. Kearney H, Byrne S, Cavalleri GL, Delanty N. Tackling epilepsy with high-definition precision medicine: a review. JAMA Neurol. 2019;76(9):1109-1116. doi: 10.1001/jamaneurol.2019.2384.
3. Sheidley BR, Smith LA, Helbig KL. Genetics of epilepsy in the era of precision medicine: implications for testing, treatment, and genetic counseling. Curr Genet Med Rep. 2108;6:73-82 doi: 10.1007/s40142-018-0139-8.
4. EpiPM Consortium. A roadmap for precision medicine in the epilepsies. Lancet Neurol. 2015;14(12):1219-1228. doi: 10.1016/S1474-4422(15)00199-4.
5. Brodie MJ, Barry SJE, Bamagous GA, Norrie JD, Kwan P. Patterns of treatment response in newly diagnosed epilepsy. Neurology. 2012;78(20):1548-1554. doi: 10.1212/WNL.0b013e3182563b19.
6. Miller IO, Sotero de Menezes MA. SCN1A seizure disorders. In: Adam MP, Ardinger HH, Pagon RA, et al, eds. Gene Reviews. Seattle, WA: University of Washington; 1993-2020. ncbi.nlm.nih.gov/books/NBK1318. Accessed March 2, 2020.
7. Young AN, Tanenhaus A, Belle A, et al. A GABA-selective AAV vector upregulates endogenous SCN1A expression and reverses multiple phenotypes in a mouse model of Dravet syndrome. Presented at: American Epilepsy Society Annual Meeting 2019; December 6-10, 2019; Baltimore, MD. Abstract 3.1.
8. Segal E, Miller I, Perry MS, et al. Proposed designed for a natural history study of SCN1A-positive (SCN1A+) early-onset epilepsy. Presented at: American Epilepsy Society Annual Meeting 2019; December 6-10, 2019; Baltimore, MD. Abstract 3.085.
9. Bearden D, Strong A, Ehnot J, DiGiovine M, Dlugos D, Goldberg EM. Targeted treatment of migrating partial seizures of infancy with quinidine. Ann Neurol. 2014;76(3):457-461. doi: 10.1002/ana.24229.
10. Mikati MA, Jiang YH, Carboni M, et al. Quinidine in the treatment of KCNT1-positive epilepsies. Ann Neurol. 2015;78(6):995-999. doi: 10.1002/ana.24520.
11. Li D, Yuan H, Ortiz-Gonzalez XR, et al. GRIN2D recurrent de novo dominant mutation causes a severe epileptic encephalopathy treatable with NMDA receptor channel blockers. Am J Hum Genet. 2016;99(4):802-816. doi: 10.1016/j.ajhg.2016.07.013.
12. Millichap JJ, Park KL, Tsuchida T, et al. KCNQ2 encephalopathy: features, mutational hot spots, and ezogabine treatment of 11 patients. Neurol Genet. 2016;2(5):e96. doi: 10.1212/NXG.0000000000000096.
13. Behind the Seizure program further expands access to genetic testing for children to speed the diagnosis of genetic epilepsy [news release]. San Francisco, CA: BioMarin Pharmaceutical; January 8, 2020. ptcommunity.com/wire/behind-seizure-program-further-expands-access-genetic-testing-children-speed-diagnosis-genetic. Accessed February 29, 2020.
14. Perucca P, Scheffer IE, Harvey AS, et al. Real-world utility of whole exome sequencing with targeted gene analysis for focal epilepsy. Epilepsy Res. 2017;131:1-8. doi: 10.1016/j.eplepsyres.2017.02.001.
15. Jehi L. Algorithms in clinical practice — can they really help us predict outcomes? Presented at: 33rd Annual International Epilepsy Congress; June 22-26, 2019; Bangkok, Thailand.
16. Jehi L, Yardi R, Chagin K, et al. Development and validation of nomograms to provide individualized predictions of seizure outcomes after epilepsy surgery: a retrospective analysis. Lancet Neurol. 2015;14(3):283-290. doi: 10.1016/S1474-4422(14)70325-4.
17. Cleveland Clinic researcher receives $3.4 M NIH grant for epilepsy surgery research [news release]. Cleveland, OH: Cleveland Clinic; June 27, 2017. newsroom.clevelandclinic.org/2017/06/29/cleveland-clinic-researcher-receives-3-4-m-nih-grant-for-epilepsy-surgery-research. Accessed March 3, 2020.
18. Learning healthcare system. The Learning Healthcare Project website. learninghealthcareproject.org/section/background/learning-healthcare-system. Accessed March 3, 2020.
19. Abbasi B, Goldenholz DM. Machine learning applications in epilepsy. Epilepsia. 2019;60(10):2037-2047. doi: 10.1111/epi.16333.
20. Zhang J, Cheng W, Wang Z, et al. Pattern classification of large‐scale functional brain networks: identification of informative neuroimaging markers for epilepsy. PLoS One. 2012;7(5):e36733. doi: 10.1371/journal.pone.0036733.
21. Goker I, Osman O, Ozekes S, Baslo MB, Ertas M, Ulgen Y. Classification of juvenile myoclonic epilepsy data acquired through scanning electromyography with machine learning algorithms. J Med Syst. 2012;36(5):2705-2711. doi: 10.1007/s10916-011-9746-6.
22. doc.ai digital health trial aims to improve precision in epilepsy treatment [news release]. Palo Alto, CA: doc.ai; October 16, 2019. prnewswire.com/news-releases/docai-digital-health-trial-aims-to-improve-precision-in-epilepsy-treatment-300939688.html. Accessed March 3, 2020.
23. Zuberi SM, Brunklaus A. Epilepsy in 2017: precision medicine drives epilepsy classification and therapy. Nat Rev Neurol. 2018;14(2):67-68. doi: 10.1038/nrneurol.2017.190.