Seeing the Storm Before the Lightning

How Science is Forecasting Seizures

Article Navigation

The Unbearable Weight of Unpredictability

Imagine living with a neurological storm that strikes without warning. For over 50 million people worldwide with epilepsy, this unpredictability isn't just inconvenient—it's life-limiting. A 2016 Epilepsy Innovation Institute (Ei2) survey of over 1,000 individuals revealed a striking truth: regardless of seizure frequency or severity, unpredictability was the top concern. One respondent poignantly captured the collective fear: "Not knowing when the next seizure will hit means I can't trust myself to drive, swim, or even hold my baby" 1 6 .

Key Insight

This profound insight sparked a revolution. Ei2 launched the My Seizure Gauge initiative—a moonshot project to transform seizure forecasting from science fiction to reality 1 9 .

The Science Behind the Storms: Decoding Seizure Rhythms

The Hidden Clocks of the Brain

Seizures aren't random. Like ocean tides, they follow biological rhythms scientists are now mapping:

  • Circadian rhythms (24-hour cycles): Seizures clustering at specific daytime/nighttime windows.
  • Ultradian rhythms (<24-hour cycles): Patterns repeating every few hours.
  • Multiday cycles (7–35 days): Longer "seizure susceptibility" waves unique to each person 1 7 .
Beyond Brainwaves

Groundbreaking data from implanted devices like the NeuroPace RNS system showed 98% of patients had detectable circadian or ultradian seizure patterns. In one case, a patient's seizures peaked every 16.3 days like clockwork 1 3 . These rhythms became the first key to forecasting.

Multimodal Approach

If seizures are storms, brainwaves (EEG) are just one weather system. My Seizure Gauge pioneers argued: To forecast accurately, monitor the whole climate. This meant tracking:

Physiological

Heart rate, skin conductance, temperature

Biochemical

Cortisol (stress hormone), potassium/pH shifts in the brain

Behavioral

Mood, fatigue, gait changes 2 7

Table 1: Multimodal Parameters for Seizure Forecasting
Category Key Parameters Detection Method
Physiological Heart rate, skin conductance, temperature Wearables (Empatica E4, Fitbit)
Biochemical Cortisol, brain pH, inflammatory markers Sweat sensors, implanted probes
Behavioral Mood, fatigue, concentration difficulties Smartphone apps, patient diaries
Environmental Atmospheric pressure, humidity Weather APIs, IoT sensors

Personalization: Why One Size Fails

Epilepsy isn't a single disease. Causes range from genetic mutations to brain injuries—meaning forecasting must be as unique as a fingerprint. Early algorithms failed because they pooled data across patients. Ei2's insight: Effective forecasting requires "N-of-1" models trained on individual long-term data 1 6 .

The Breakthrough Experiment: My Seizure Gauge Study

Methodology: A Symphony of Sensors

In 2021–2022, Ei2's consortium (Mayo Clinic, King's College London, University of Melbourne) launched an unprecedented experiment:

  1. Participants: 39 patients with drug-resistant epilepsy (≥10 seizures/month)
  2. Wearables: Empatica E4 (heart rate/EDA), Fitbit Charge HR/Inspire (activity)
  3. EEG Systems: Subcutaneous devices (UNEEG SubQ, EpiMinder), implanted neurostimulators (NeuroPace RNS)
  4. Duration: 8+ months of continuous monitoring
  5. Data Types: Electrophysiological signals, movement, self-reported diaries via the Seer App 3 7 8 .
Results: Signals in the Noise

After collecting 12,500+ days (33.7 years) of data and 1,700+ seizures, machine learning models uncovered striking patterns:

  • Heart rate cycles synchronized with seizure likelihood in 53% of patients (10/19) using Fitbit 3 .
  • Electrodermal activity (EDA) from Empatica E4 predicted seizures above chance in 83% of patients (5/6) 3 7 .
  • Subcutaneous EEG enabled forecasting in 5/6 patients using patient-specific algorithms 3 .
Table 2: Wearable Performance in Seizure Forecasting
Device Key Metrics Tracked Forecasting Accuracy (AUC*) Notable Findings
Empatica E4 Heart rate, EDA, temperature >0.70 (5/6 patients) Detected pre-seizure autonomic changes
Fitbit Inspire HR Heart rate, movement 0.74 (mean across 11 patients) Heart rate cycles phase-locked to seizures
UNEEG SubQ EEG Brainwave patterns (subscalp) >0.75 (5/6 patients) Detected multiday seizure cycles
*AUC (Area Under Curve): 0.5 = chance, 1.0 = perfect prediction
Forecasting Performance Across Technologies
Technology Patients with Significant Forecasting Sensitivity (%) False Alarms/Day
Empatica E4 (wearable) 5/6 68–82 0.8–1.2
Fitbit (wearable) 11/11 71–89 0.7–1.5
SubQ EEG (EpiMinder) 1/1 (pilot) 83 1.1
Patient Self-reports 10/19 65 N/A
Analysis: The Forecasting Frontier

This study proved three radical ideas:

  1. Long cycles matter: Multiday rhythms (not just hourly changes) drive seizure risk.
  2. Peripheral signals reflect brain states: A wrist-based sensor can detect seizures via body-brain connections.
  3. Hybrid models win: Combining wearables + EEG + diaries outperformed any single source 3 7 .

The Scientist's Toolkit: Building a Forecast System

Long-term EEG Recorders

Track brainwave cycles over months/years

Example: NeuroPace RNS, UNEEG SubQ, EpiMinder

Wearable Biosensors

Capture heart rate, EDA, temperature

Example: Empatica E4, Fitbit Charge HR

Mobile Apps

Log seizures, mood, triggers

Example: Seer App, EpiDiary

Machine Learning Algorithms

Personalize forecasting models

Example: LSTM networks, SVM classifiers

Data Sharing Platform

These tools enabled a paradigm shift. For example, the EpilepsyEcosystem.org platform now shares anonymized data from My Seizure Gauge, accelerating global collaboration 8 .

The Road Ahead: From Forecasts to Freedom

Current Developments

The My Seizure Gauge initiative is now entering its clinical translation phase:

  • A free seizure diary app (Seer Medical) uses patient inputs to forecast risk 3 .
  • Prospective trials are testing real-time alerts via smartwatches.
  • A 2025 data science challenge aims to crowdsource better algorithms 5 7 .
Challenges Remaining
  • Reducing false alarms
  • Integrating forecasts with closed-loop therapies
  • Ensuring accessibility for all patients

"We're not just predicting storms. We're giving people back their blue skies."

Dr. Gregory Worrell (Mayo Clinic) 6 9

For more on seizure forecasting research, visit the Epilepsy Innovation Institute at epilepsy.com/Ei2.

References