Narcolepsy Revolution - Protocol and Methodology A diagnostic accuracy study protocol using the Dreem 3 headband for ambulatory diagnosis of narcolepsy in children and young adults

This prospective study protocol evaluates the diagnostic accuracy of the home-based Dreem 3 EEG headband, compared against standard in-patient polysomnography and MSLT, for the ambulatory diagnosis of narcolepsy in children and young adults to potentially reduce delays and improve access to care.

The Gist

Imagine you have a car that keeps stalling in the middle of the road. To figure out why, a mechanic usually has to tow the car into a special, expensive garage. They hook it up to massive computers, run it through a series of rigorous tests, and spend hours analyzing the data. This is how doctors currently diagnose narcolepsy (a condition where people fall asleep uncontrollably) in children and young adults. It's accurate, but it's slow, costly, and hard to get an appointment for.

This research paper is a proposal for a new, simpler way to diagnose the problem right in your own driveway.

Here is the story of the "Narcolepsy Revolution" study, broken down into everyday concepts:

1. The Problem: The "Garage" Bottleneck

Currently, to diagnose narcolepsy, a patient must stay overnight in a hospital sleep lab. They wear a "spaghetti monster" of wires (a polysomnography or PSG) and then take a nap test the next day.

• The Analogy: It's like towing your car to a high-tech lab just to see if the engine is misfiring. It works, but it's a hassle, it costs a fortune, and there are very few labs available. This causes long delays, leaving sick kids waiting years for answers.

2. The New Tool: The "Smart Headband"

The researchers are testing a device called the Dreem 3 headband.

• The Analogy: Instead of a tangled mess of wires, imagine a comfortable, sleek headband you wear like a sleep mask. It's a "dry" sensor (no sticky gels), so it's easy to put on and take off. It's designed to be worn at home, just like you would wear a smartwatch to track your steps.

3. The Experiment: The "Weekday vs. Weekend" Test

The study recruited 60 young people (ages 10 to 35) who were already scheduled for the traditional hospital tests. The researchers wanted to see if the headband could do the same job.

They asked the participants to wear the headband in two specific ways:

• The Weeknight Shift: Wear the headband for 5 nights while sleeping in your own bed.
• The Weekend Marathon: Wear the headband continuously for 48 hours over a weekend (sleeping and staying awake).
  • Why the weekend? The researchers asked young people what they thought. The kids said, "We don't mind wearing it to sleep, but we can't wear it to school or work!" So, they decided to do the continuous 48-hour test only on weekends when the kids were free.
  • The Trick: Since the headband needs charging, they used two headbands and swapped them back and forth, like a relay race, to keep the recording going non-stop.

4. The "Brain Detective" (Machine Learning)

The headband collects raw brainwave data, but a human can't read all that data quickly. So, the researchers built a digital detective (an AI/Machine Learning algorithm).

• The Analogy: Think of the AI as a super-smart detective who has read every single sleep pattern in the world. Its job is to look at the headband's data and spot a specific "fingerprint" called SOREM (sleeping and entering REM sleep too quickly).
• In narcolepsy, the brain skips the "slow sleep" phase and jumps straight into "dream sleep" (REM). The AI is trained to spot this jump instantly, whereas a human doctor might miss it or need hours to find it.

5. The Goal: The "Gold Standard" Checkup

At the end of the study, every participant also did the traditional, expensive hospital test (the "Gold Standard").

• The Comparison: The researchers will compare the AI's diagnosis (based on the home headband) against the doctor's diagnosis (based on the hospital test).
• The Question: Did the headband and the AI get the same answer as the expensive hospital?

6. Why This Matters

If the headband works as well as the hospital test, it changes everything:

• Speed: No more waiting years for an appointment.
• Comfort: Kids can sleep in their own beds, not in a strange hospital room.
• Fairness: It makes diagnosis available to families who live far from big cities or can't afford the travel costs.

The Bottom Line

This paper is the blueprint for a revolution. It's not the final result yet (the data is still being analyzed), but it lays out a plan to turn a complex, hospital-bound medical mystery into something as simple as wearing a headband at home. If successful, it could mean that in the future, diagnosing narcolepsy is as easy as checking your step count.

Technical Summary

Based on the provided preprint, here is a detailed technical summary of the Narcolepsy Revolution study protocol.

1. Problem Statement

• Clinical Challenge: Narcolepsy is a rare, lifelong neurological disorder often beginning in childhood. Diagnosis is frequently delayed (mean of 9.7 years in recent European studies) due to the complexity and inaccessibility of current diagnostic pathways.
• Limitations of Current Standard: The gold standard for diagnosis involves an overnight Polysomnography (PSG) followed by a Multiple Sleep Latency Test (MSLT), interpreted against International Classification of Sleep Disorders (ICSD-3) criteria.
  • Barriers: These tests are expensive, labor-intensive, require specialized inpatient facilities, and depend on scarce specialist staff for manual scoring.
  • Variability: MSLT results can suffer from biological and situational variability, leading to false positives/negatives.
• Technological Gap: While machine learning (ML) has shown promise in analyzing sleep neurophysiology for narcolepsy detection, previous studies relied on hospital-based PSG data. There is a lack of validation for wearable, home-based EEG devices in real-world clinical settings for diagnosing hypersomnolence.

2. Methodology

• Study Design: A prospective, single-centre, observational diagnostic accuracy study conducted at Guy's and St Thomas' NHS Foundation Trust (London, UK).
• Participants:
  • Target: 60 participants aged 10–35 years undergoing investigation for hypersomnolence.
  • Inclusion: Scheduled for routine inpatient PSG and MSLT.
  • Exclusion: Medical/psychiatric disorders independently causing sleepiness; regular use of medications affecting sleep architecture.
• Index Test (The Intervention):
  • Device: Dreem 3 (DH) headband, a dry-montage, wearable EEG device designed for home use.
  • Protocol:
    1. Weeknight Phase: 5 consecutive weekday nights (bedtime to rising).
    2. Weekend Phase: Continuous 48-hour recording over a weekend using two devices in rotation to manage charging.
  • Data Handling: Data uploaded to the manufacturer's (Beacon Biosignals) secure server, then downloaded for local analysis.
• Reference Standard:
  • Consensus clinical diagnosis by an experienced sleep physician based on clinical history, actigraphy, sleep diaries, and the standard inpatient PSG/MSLT (scored per AASM criteria and interpreted per ICSD-3).
  • Blinding: The diagnosing clinician is blinded to the DH data and ML outputs.
• Machine Learning Approach:
  • Goal: Develop classifiers to detect Sleep-Onset REM Periods (SOREMPs) and distinguish narcolepsy from other hypersomnolence causes.
  • Features: Time-domain (variance, Hjorth parameters), frequency-domain (band power), and non-linear timeseries features (entropy).
  • Algorithms: Random Forest, XGBoost, and CatBoost.
  • Validation: 80-10-10 data splitting (train-validate-test), K-fold cross-validation, and hyperparameter tuning. External validation planned on public datasets (e.g., MNC dataset).
• Patient and Public Involvement (PPI):
  • Young people with narcolepsy (ages 14–19) informed the protocol.
  • Key Feedback: Continuous wear during school/work is unacceptable; weekend-only continuous wear is acceptable. This shaped the 5-night + 48-hour weekend design.

3. Key Contributions

• Proof-of-Concept for Ambulatory Diagnosis: This is the first study to evaluate the Dreem 3 headband specifically for the diagnosis of narcolepsy in a real-world clinical setting, moving beyond hospital-based PSG.
• Hybrid Protocol Design: The study uniquely combines nocturnal weeknight data with continuous 48-hour weekend data. This aims to capture both standard sleep architecture and daytime sleep episodes/SOREMPs in ecologically valid conditions, which are often missed in short hospital stays.
• Algorithmic Innovation: The protocol outlines a comprehensive ML pipeline using interpretable machine learning to detect REM dysregulation signatures from a limited montage (single-channel EEG) compared to full PSG.
• Feasibility Assessment: It explicitly evaluates the acceptability and feasibility of long-term home wear in children and young adults, addressing a critical barrier to home-based diagnostics.

4. Results

• Status: No results are reported in this document.
• Study Phase: As of the posting date (March 27, 2026), recruitment was completed on January 30, 2026, and data analysis is ongoing.
• Protocol Iterations: The document details three protocol versions (v1.0 to v3.0) reflecting adjustments in recruitment age (extended to 35 years) and recording sequence flexibility to accommodate participant needs.

5. Significance and Implications

• Reducing Diagnostic Delays: If validated, this approach could significantly shorten the 9.7-year average diagnostic delay by enabling earlier, accessible testing outside of specialized sleep labs.
• Equity and Accessibility: A home-based wearable solution could democratize access to sleep diagnostics, reducing the burden on patients (no travel/hospital stay) and healthcare systems (lower cost, less specialist time).
• Scalability: Successful validation would provide the necessary parameters for a subsequent multicentre study, potentially establishing a new, scalable standard of care for narcolepsy diagnosis.
• Clinical Utility: The study aims not to replace clinical judgment but to determine if wearable EEG can provide clinically useful diagnostic information to support decision-making.

Note: This document is a study protocol published as a preprint on medRxiv. It has not yet undergone peer review, and the authors explicitly state it should not be used to guide clinical practice at this time.