NOTE: This podcast is an evolving collaboration between human and AI. While we strive for accuracy, AI hosts may misinterpret or oversimplify source material. Always refer to the original published articles for clinical decision-making. If you find any claims made by the AI hosts to be inaccurate, please let us know. Your feedback directly improves future episodes.

Episode 15: Fellow Bootcamp: OSA Is Not Just a Comorbidity

When should obstructive sleep apnea be treated as more than a passive comorbidity in epilepsy care?

This special uses the nine Bradford Hill viewpoints to examine whether OSA may modify epilepsy outcomes. The hosts keep three evidence lanes separate: seizure and interictal-discharge burden, incident or late-onset epilepsy, and surrogate SUDEP-risk markers. They also examine what the current CPAP studies can—and cannot—support.

AED Quiz for this episode:
https://audioepilepsydigest.com/episode-015-aed-quiz.html

Listen and follow Audio Epilepsy Digest:
https://audioepilepsydigest.com/

Key takeaways:
- Bradford Hill viewpoints organize causal reasoning; they are not a checklist, vote, or proof score.
- OSA is commonly reported in people with epilepsy, but prevalence estimates are heterogeneous and do not establish causation.
- Prospective and claims-based studies support an incident-epilepsy and late-onset-epilepsy hypothesis, with residual confounding and exposure measurement still important.
- Associations with revised SUDEP risk scores concern surrogate markers, not observed SUDEP events or evidence of prevention.
- Treat confirmed OSA for established sleep-medicine indications and track epilepsy outcomes without promising an epilepsy-specific response.

Sources include Rehim, Vendrame, and Devinsky 2026; Lin et al. 2017; Harnod et al. 2017; Carosella et al. 2023; Soontornpun et al. 2024; Pornsriniyom et al. 2014; Malow et al. 2008; and McCarter et al. 2018.

AI-assisted editorial/source review:
https://audioepilepsydigest.com/episode-015-ai-review.html

Editorial disclosure:
Two independent AI-assisted reviews cleared the episode with minor notes and no blocking source-fidelity or clinical-safety finding. Minor notes concern a few contextual certainty words, process-like language near the opening, an over-categorical effect-size heuristic, the ARIC sleep-substudy denominator, and author-name pronunciation. The clinician-editor listened to and approved the release candidate.

NOTE: This podcast is an evolving collaboration between human and AI. While we strive for accuracy, AI hosts may misinterpret or oversimplify source material. Always refer to the original published articles for clinical decision-making. If you find any claims made by the AI hosts to be inaccurate, please let us know. Your feedback directly improves future episodes.

Episode 14: Fellow Bootcamp: From EMU to SEEG to Surgery

How should an epilepsy fellow turn EMU findings, imaging, neuropsychology, and other phase 1 data into a defensible surgical hypothesis?

This special reviews the cortical concepts used in presurgical reasoning, how phase 1 findings become a testable hypothesis, when stereoelectroencephalography may resolve an actionable uncertainty, the limitations of sparse intracranial sampling, and how teams orient among resection, ablation, disconnection, neuromodulation, or no procedure.

AED Quiz for this episode:
https://audioepilepsydigest.com/episode-014-aed-quiz.html

Listen and follow Audio Epilepsy Digest:
https://audioepilepsydigest.com/

Key takeaways:
- Phase 1 evaluation should build a coherent electroclinical hypothesis and expose meaningful contradictions.
- Symptomatogenic, irritative, seizure-onset, epileptogenic-lesion, functional-deficit, eloquent-cortex, and epileptogenic zones may overlap but are not interchangeable.
- A visible lesion can strengthen a hypothesis but does not independently establish causality.
- S-E-E-G is spatially limited sampling and is most useful when it tests bounded alternatives that could change treatment.
- Treatment choice integrates targetability, network distribution, functional risk, expected benefit, residual uncertainty, and patient goals.

Sources include Schuele et al. 2025, Jayakar et al. 2016, Miller and Fine 2022, Simpson et al. 2022, and Jehi et al. 2022.

AI-assisted editorial/source review:
https://audioepilepsydigest.com/episode-014-ai-review.html

Editorial disclosure:
The clinician-editor selected this audio through an explicit human override after two independent AI-assisted reviews advised targeted revision. The central framework was substantially grounded, while some examples and mechanisms remain teaching simplifications. The final AI passage is intentionally speculative and does not describe current clinical capability.

NOTE: This podcast is an evolving collaboration between human and AI. While we strive for accuracy, AI hosts may misinterpret or oversimplify source material. Always refer to the original published articles for clinical decision-making. If you find any claims made by the AI hosts to be inaccurate, please let us know. Your feedback directly improves future episodes.

Episode 13: Temporal Encephaloceles: Lesion, Network, and How Much to Resect?

This monthly special debate asks how epilepsy surgery teams should reason about temporal encephaloceles: as a leading surgical target, as a clue to broader temporal network involvement, or both.

The episode uses surgical series, systematic reviews, imaging studies, EEG source-localization work, and a direct sEEG case to frame a conference-style question: when is a tailored mesial-sparing approach reasonable, and when should discordant data prompt broader mapping or a wider surgical hypothesis?

AED Quiz for this episode:
https://audioepilepsydigest.com/episode-013-aed-quiz.html

Listen and follow Audio Epilepsy Digest:
https://audioepilepsydigest.com/

AI editorial/source review for this episode:
https://audioepilepsydigest.com/episode-013-ai-review.html

Key takeaways:
- Temporal encephaloceles can be missed on routine review and may require deliberate skull-base re-review or additional imaging.
- Finding a temporal encephalocele is not the same as proving seizure onset from the herniated tissue.
- Concordant imaging, semiology, EEG, PET, neuropsychology, and invasive recordings can support a lesion-centered hypothesis in selected cases.
- Surgical extent should follow the full presurgical picture, not a default rule for or against hippocampectomy.
- The evidence base is mainly retrospective and selected, so the episode frames conference reasoning rather than a guideline.

Papers discussed include Smith 2023, Tsalouchidou 2025, Khoudari 2025, Zhou 2024, Garcia-Gracia 2024, Kondylis 2026, Ortiz 2025, Tsalouchidou 2023, Cox 2024, and Agashe 2023.

Caveats:
- Temporal encephaloceles are treated as potentially important surgical clues, not automatic proof of epileptogenicity.
- MRI metrics, PET patterns, CT defects, and postprocessing are supportive data, not direct proof of seizure onset.
- The direct sEEG case is illustrative and should not be generalized beyond a selected single-patient report.
- The practical framework is intended as conference reasoning, not a rigid surgical algorithm.

NOTE: This podcast is an evolving collaboration between human and AI. While we strive for accuracy, AI hosts may misinterpret or oversimplify source material. Always refer to the original published articles for clinical decision-making. If you find any claims made by the AI hosts to be inaccurate, please let us know. Your feedback directly improves future episodes.

Episode 12: Care Pathways Shape Epilepsy Outcomes

This weekly recent-studies episode asks what happens when epilepsy outcomes are shaped not only by medications, procedures, or devices, but by the care pathway around the patient.

The episode follows five care-delivery signals: rural inpatient outcomes, inpatient video-EEG access after repeated seizure-related emergency visits, pediatric-to-adult transition, brain tumor-related epilepsy coordination, and a UK roadmap for technology-enabled service redesign.

AED Quiz for this episode:
https://audioepilepsydigest.com/episode-012-aed-quiz.html

Listen and follow Audio Epilepsy Digest:
https://audioepilepsydigest.com/

AI editorial/source review for this episode:
https://audioepilepsydigest.com/episode-012-ai-review.html

Key takeaways:
- Care pathways are measurable parts of epilepsy care, not just background logistics.
- Rural-urban inpatient differences, vEEG access, transition care, and tumor-epilepsy coordination all point to pathway-level gaps.
- The studies are mostly observational or service-mapping work, so they support careful questions rather than practice-changing mandates.
- Technology can support redesign, but the roadmap source is not proof that specific digital tools improve outcomes.
- Repeated ED visits, transition handoffs, and multidisciplinary tumor care are moments to ask whether the patient reached the right diagnostic and longitudinal services.

Papers discussed include:
1. Rural-urban disparities in epilepsy hospitalizations in the United States. PMID: 42234954. PMCID: PMC13239426.
2. Kozak et al. Access to inpatient video-EEG monitoring for patients with frequent seizure-related emergency visits. Epilepsy Research (2026). PMID: 42349236.
3. Modi et al. Measuring transition success from paediatric to adult epilepsy services and its association with seizure outcomes. Archives of Disease in Childhood (2026). PMID: 42082329.
4. Anghileri et al. Mapping integrated care for brain tumour-related epilepsy in the Italian RIN-IRCCS network. Neurological Sciences (2026). PMID: 42350833. PMCID: PMC13303563.
5. Terry and Shankar. The future of epilepsy care in the United Kingdom: A roadmap for technology-enabled transformation. Epilepsia Open (2026). PMID: 41964582. PMCID: PMC13238867.

Caveats:
- Rurality, vEEG access, transition success, and pathway gaps are discussed as associations or service-design signals, not causal proof.
- Private-insurance examples in the audio should be heard as illustrative possibilities, not findings directly tested by the rural-urban study.
- The wearable false-alarm scenario is a hypothetical service-design concern, not an observed outcome in the technology-roadmap paper.

NOTE: This podcast is an evolving collaboration between human and AI. While we strive for accuracy, AI hosts may misinterpret or oversimplify source material. Always refer to the original published articles for clinical decision-making. If you find any claims made by the AI hosts to be inaccurate, please let us know. Your feedback directly improves future episodes.

Episode 11: Signals Before Practice Change: GLP-1 Drugs and the Temptation to Overread

This weekly recent-studies episode asks how epileptologists should talk about exciting signals before they become practice-changing evidence.

The main thread is GLP-1 receptor agonists and seizure-related outcomes in diabetes-linked populations, with pediatric levetiracetam and VNS predictor studies used as comparator examples of evidence restraint.

AED Quiz for this episode:
https://audioepilepsydigest.com/episode-011-aed-quiz.html

Listen and follow Audio Epilepsy Digest:
https://audioepilepsydigest.com/

AI editorial/source review for this episode:
https://audioepilepsydigest.com/episode-011-ai-review.html

Key takeaways:
- GLP-1 findings are hypothesis-generating signals, not seizure-treatment evidence.
- AbuAlrob et al. reported HR 0.82 for EHR-coded seizure recurrence; this is best described as an estimated 18% lower hazard with about a 2.1 percentage-point absolute recurrence difference.
- Pediatric levetiracetam is useful, but active-comparator evidence does not prove automatic superiority.
- VNS prognostic factors are research signals, not deterministic implantation rules.
- Clinical language matters when patient expectations can outpace the evidence.

Papers discussed include:
1. Eun Y, Bong S, et al. Semaglutide and adult-onset seizure target-trial emulation. Neurology (2026). PMID: 42308439.
2. Sindhu U, et al. Newer glucose-lowering drugs and late-onset seizure/epilepsy meta-analysis. Epilepsia Open (2024). PMID: 39487832.
3. AbuAlrob MA, et al. GLP-1 receptor agonist initiation and seizure recurrence in adults with epilepsy. Epilepsia (2025). PMID: 41251033.
4. Balestrini S, et al. Pediatric levetiracetam systematic review and meta-analysis. Neurology (2026). PMID: 42202238.
5. VNS prognostic-factor systematic review and meta-analysis. Neurosurgical Review (2026). PMID: 42307795.
6. Crutcher R, et al. VNS response predictors retrospective cohort. Neurosurgical Review (2025). PMID: 41042470.

Caveats:
- These GLP-1 data do not support prescribing for seizure control.
- Cardiovascular and renal outcome trials were not epilepsy-prevention trials.
- EHR-coded seizure recurrence is not seizure-diary adjudication.
- Crutcher et al. should not be turned into a 5-year VNS window rule.

NOTE: This podcast is an evolving collaboration between human and AI. While we strive for accuracy, AI hosts may misinterpret or oversimplify source material. Always refer to the original published articles for clinical decision-making. If you find any claims made by the AI hosts to be inaccurate, please let us know. Your feedback directly improves future episodes.

Episode 10: A Positive Signal With a Negative Primary Endpoint: The NAUTILUS Trial

This monthly special is a journal-club reading of NAUTILUS, the randomized sham-controlled trial of responsive centromedian thalamic stimulation for drug-resistant idiopathic generalized epilepsy with recurrent generalized tonic-clonic seizures.

The episode's main point is endpoint discipline: the primary safety endpoint was met, the prespecified primary effectiveness endpoint was not statistically significant, and the longer-term and post hoc signals remain clinically important but bounded.

AED Quiz for this episode:
https://audioepilepsydigest.com/episode-010-aed-quiz.html

Listen and follow Audio Epilepsy Digest:
https://audioepilepsydigest.com/

AI editorial/source review for this episode:
https://audioepilepsydigest.com/episode-010-ai-review.html

Key takeaways:
- NAUTILUS is best read as an encouraging signal, not a simple positive trial.
- The prespecified time-to-second-GTCS effectiveness endpoint was not statistically significant.
- Post hoc and 18-month seizure outcomes may still matter clinically, but they cannot erase the endpoint hierarchy.
- The narrow practice implication is a serious counseling and referral discussion for carefully selected drug-resistant IGE patients with recurrent GTCS.
- Thalamic RNS for IGE should not be framed as established standard-of-care therapy from this trial alone.

Papers discussed include:
1. Uysal et al. NAUTILUS randomized controlled trial, Epilepsia (2026). PMID: 42233958.
2. Haneef and Skrehot. Neurostimulation in generalized epilepsy, Epilepsia (2023). PMID: 36727550.
3. Alhourani et al. Bilateral centromedian thalamic neuromodulation systematic review and IPD meta-analysis, Neuromodulation (2026). PMID: 40498441.
4. Sisterson et al. Centromedian thalamic RNS in IGE case series, Journal of Neurology, Neurosurgery & Psychiatry (2022). PMID: 35217517.
5. Nanda et al. Centromedian-region thalamic RNS in IGE and multifocal epilepsy, Epilepsia (2024). PMID: 39052021.

Caveats:
- The Nanda 2024 82.6% average seizure reduction applies to the 17 patients with at least 1 year of follow-up, not all 21 implanted patients.
- NAUTILUS post hoc and 18-month outcomes are encouraging but cannot rescue the negative prespecified primary effectiveness endpoint.
- Centromedian thalamus should be understood as a plausible thalamocortical network node, not a proven circuit breaker.
- The episode supports careful discussion and referral consideration, not a guideline-level recommendation.

NOTE: This podcast is an evolving collaboration between human and AI. While we strive for accuracy, AI hosts may misinterpret or oversimplify source material. Always refer to the original published articles for clinical decision-making. If you find any claims made by the AI hosts to be inaccurate, please let us know. Your feedback directly improves future episodes.

Episode 9: From Paper to Podcast: The AED Assembly Line

This special episode explains how Audio Epilepsy Digest is made. Instead of covering a new set of epilepsy papers, it opens the production process itself: how sources enter the packet, how the episode angle is chosen, how AI helps create a conversational draft, how the transcript is checked, and where human editorial responsibility remains.

The central point is modest: useful AI in medicine does not have to mean diagnosis or prescribing. In AED, the goal is to help clinicians learn, organize evidence, and arrive at clinic better prepared, while keeping the human editor responsible for framing, caveats, and release.

AED Quiz for this episode:
https://audioepilepsydigest.com/episode-009-aed-quiz.html

Listen and follow Audio Epilepsy Digest:
https://audioepilepsydigest.com/

AI editorial/source review for this episode:
https://audioepilepsydigest.com/episode-009-ai-review.html

Key takeaways:
- AED starts from source material and a clinician-framed question, not from audio generation.
- AI can help draft, converse, transcribe, extract claims, and red-team a candidate, but it does not own clinical judgment.
- Transcript-first review helps make fluent audio auditable.
- Public source-review pages and listener feedback are part of AED's transparency loop.
- This episode is a process transparency special, not proof that the workflow has been externally validated.

Source review note:
This episode went through AI-assisted editorial review before release staging. The AI review recommended regeneration for process and branding issues in the spoken audio, including the exact closing language. Erafat listened to the candidate and explicitly cleared this version for release staging by human override.

Caveats:
- AI-assisted source review is not independent peer review.
- The episode describes AED's own documented process; it does not validate AED as a general medical education intervention.
- The public review page records the AI-review caveats and the human override decision.

NOTE: This podcast is an evolving collaboration between human and AI. While we strive for accuracy, AI hosts may misinterpret or oversimplify source material. Always refer to the original published articles for clinical decision-making. If you find any claims made by the AI hosts to be inaccurate, please let us know. Your feedback directly improves future episodes.

Episode 8: Status Epilepticus in Critical Care

This weekly recent-studies episode uses papers from roughly the last month, plus ESETT as landmark context, to ask what happens when status epilepticus moves beyond the emergency algorithm and becomes a critical-care problem.

The episode covers ICU practice variation, point-of-care AI seizure burden, perfusion imaging for seizure-vs-stroke-mimic uncertainty, VNS for SRSE/NORSE, and emergency resection in selected lesional SRSE.

The practical message is evidence-tier discipline: name the kind of evidence before naming the intervention.

AED Quiz for this episode:
https://audioepilepsydigest.com/episode-008-aed-quiz.html

Listen and follow Audio Epilepsy Digest:
https://audioepilepsydigest.com/

AI editorial/source review for this episode:
https://audioepilepsydigest.com/episode-008-ai-review.html

Key takeaways:
- ESETT anchors one second-line convulsive status decision point, but not ICU/SRSE/NORSE salvage care.
- ICU status epilepticus care is shaped by cEEG access, training, and workflow limits.
- Point-of-care AI seizure burden is a risk signal associated with discharge outcome, not proof that AI-guided treatment improves outcomes.
- Perfusion imaging can add context in selected cases, but EEG remains central.
- VNS and emergency resection remain salvage conversations with strict evidence boundaries.

Papers discussed include:
1. Kapur J, et al. "Randomized Trial of Three Anticonvulsant Medications for Status Epilepticus." New England Journal of Medicine (2019). PMID: 31774955.
2. Fanet et al. "Assessment of therapeutic management practices in generalized and focal motor status epilepticus." Journal of Intensive Medicine (2026).
3. Parvizi J, et al. "Point-of-Care Artificial Intelligence Measure of Seizure Burden Associates With Clinical Outcome at Discharge." Critical Care Medicine (2026). PMID: 42223304.
4. "Brain perfusion imaging in patients with status epilepticus, seizures, and IIC patterns." Clinical Neurophysiology Practice (2026).
5. Yoo JY, et al. "Vagus nerve stimulation as an adjunctive therapy for super-refractory status epilepticus including NORSE." Epilepsy & Behavior (2026). PMID: 41875755.
6. "Surgical resection as salvage therapy for super-refractory status epilepticus." Epilepsy & Behavior Reports (2026).

Caveats:
- ESETT should not be generalized to nonconvulsive, NORSE, SRSE, neuromodulation, or surgical salvage decisions.
- Survey data do not define an optimal ICU protocol.
- AI seizure burden is associative and does not replace expert EEG interpretation.
- Perfusion imaging is an adjunct, not a standalone diagnostic test.
- VNS and emergency resection evidence is not generalizable routine care.

NOTE: This podcast is an evolving collaboration between human and AI. While we strive for accuracy, AI hosts may misinterpret or oversimplify source material. Always refer to the original published articles for clinical decision-making. If you find any claims made by the AI hosts to be inaccurate, please let us know. Your feedback directly improves future episodes.

Episode 7: Real-World Evidence Meets Surgical Assumptions

This weekly recent-studies episode uses papers published in roughly the last month to ask how epileptologists should counsel patients with drug-resistant focal epilepsy when new evidence strengthens some assumptions, weakens others, and warns against turning plausible anatomy or AI output into shortcuts.

The episode covers real-world RNS postapproval outcomes, polymicrogyria epilepsy surgery, piriform cortex resection validation, and large-language-model diagnostic reasoning from seizure narratives.

The practical message is restrained: outcome data, visible lesions, anatomical targets, and AI models can all be useful, but each answers a different question. Surgical counseling should stay grounded in evidence quality, individualized mapping, and expert review.

AED Quiz for this episode:
https://audioepilepsydigest.com/episode-007-aed-quiz.html

Listen and follow Audio Epilepsy Digest:
https://audioepilepsydigest.com/

AI editorial/source review for this episode:
https://audioepilepsydigest.com/episode-007-ai-review.html

Key takeaways:
- RNS postapproval data strengthen confidence in long-term palliative neuromodulation for selected adults with focal epilepsy, but they do not define universal treatment sequencing.
- PMG-related epilepsy can be surgically treatable in selected patients, but the visible malformation is not automatically the full epileptogenic zone.
- A 2026 validation study found no association between connectivity-defined piriform cortex resection and seizure freedom, arguing against routine targeting while preserving patient-specific network evaluation.
- LLMs can organize seizure-semiology information above chance in a benchmark task, but reasoning and citation errors keep them in the decision-support category.

Papers discussed include:
1. Eliashiv D, et al. "Postapproval Study for Brain-Responsive Neurostimulation for Drug-Resistant Focal Epilepsy." Neurology (2026). PMID: 42030518.
2. Rinella K, et al. "Efficacy of neurosurgical interventions for epilepsy in polymicrogyria." Epilepsia Open (2026). PMID: 41689720.
3. "No link between piriform cortex subregion resection and seizure freedom in two cohorts with temporal lobe epilepsy." Journal of Neurology (2026). PMID: 42174251.
4. Poon MTC, et al. "Evaluating large language models for diagnostic reasoning from unstructured clinical narratives in epilepsy." Communications Medicine (2026). PMID: 42174195.

Source review note:
This episode went through AED's automatic two-reviewer source review. The final regenerated candidate cleared both reviewers with minor notes before human audio QA.

Caveats:
- RNS findings are noncomparative postapproval outcomes, not proof that neuromodulation should bypass resection or ablation evaluation.
- PMG surgery evidence is heterogeneous and does not make the visible lesion a universal target.
- The piriform cortex result is a negative routine-targeting validation, not proof that piriform involvement never matters.
- LLM benchmark performance is not clinical authorization for delegated surgical localization or triage.

NOTE: This podcast is an evolving collaboration between human and AI. While we strive for accuracy, AI hosts may misinterpret or oversimplify source material. Always refer to the original published articles for clinical decision-making. If you find any claims made by the AI hosts to be inaccurate, please let us know. Your feedback directly improves future episodes.

Episode 6: From Random Seizures to Risk States

This monthly special asks whether seizure timing is moving from clinical randomness toward individualized, time-varying risk states.

The episode covers recent work on epilepsy chronobiology, seizure cycles, past-only forecasting pipelines, chance-model pitfalls, diary and wearable approaches, home EEG feasibility, sleep-drive physiology, and computational medication-timing models.

The practical message is restrained: seizure timing is becoming a serious clinical-research variable, but current evidence supports better questions and better trials more than routine patient-facing forecasts or medication-timing changes.

AED Quiz for this episode:
https://audioepilepsydigest.com/episode-006-aed-quiz.html

Listen and follow Audio Epilepsy Digest:
https://audioepilepsydigest.com/

AI editorial/source review for this episode:
https://audioepilepsydigest.com/episode-006-ai-review.html

Key takeaways:
- Seizure timing can reflect circadian, sleep-wake, sleep-drive, and multidien risk rhythms, but that does not make seizures reliably predictable for routine care.
- Forecasting claims need past-only implementation, meaningful chance models, simple benchmarks, prospective validation, and attention to false alarms and patient burden.
- Wearable, diary, and home EEG studies show why the field is plausible, but feasibility and proof of principle are not the same as clinical effectiveness.
- Medication-timing models and sleep-drive experiments are useful for hypothesis generation, not patient-specific treatment advice.

Papers discussed include:
1. Baud MO, et al. "Timing is everything: Expert opinion on researching epilepsy rhythms by the ILAE Task Force on Chronobiology." Epilepsia (2026). PMID: 41483455.
2. Yang H, et al. "Seizure forecasting with epilepsy cycles: On the causality of forecasting pipelines." Epilepsia (2026). PMID: 41591752.
3. Andrzejak RG, et al. "Are seizure forecasts and cycles better than chance? What chance?" Epilepsia (2026). PMID: 41783988.
4. Chang CY, et al. "Rigorous evaluation of five models for e-diary-only seizure forecasting-retrospective and prospective datasets do not outperform the Napkin method." Epilepsia (2026). PMID: 41085335.
5. Xiong W, et al. "Forecasting seizure likelihood from cycles of self-reported events and heart rate: a prospective pilot study." eBioMedicine (2023). PMID: 37331164.
6. Cuddapah VA, et al. "Sleep drive, not total sleep amount, increases seizure risk." Nature Communications (2025). PMID: 40730814.

Source review note:
This episode went through AED's automatic two-reviewer source review. Both reviewers cleared the episode with minor caveats before human audio QA.

Caveats:
- Forecasting remains probabilistic and research-stage.
- The sleep-drive source is preclinical and should not be treated as human sleep advice.
- Medication-timing modeling is not a recommendation to change antiseizure medication schedules.
- Home EEG feasibility and forecasting protocols do not yet prove clinical effectiveness.