Vitals Knowledge Vault

Cluster Headache Detection Model

5 min read

Cluster Headache Detection Model

TL;DR

  • There is no validated consumer wearable that can detect, diagnose, or predict cluster headache attacks in real time. This is a hard implementation boundary.
  • The defensible wearable path for cluster headache is actigraphy-level sleep timing review, attack-time logging, and sleep-disruption correlation — not biometric attack detection.
  • HRV, heart rate, motion, and sleep-stage data cannot reliably distinguish cluster attacks from other nocturnal events in current data.
  • REM sleep is not a confirmed attack trigger; REM-stage wearable alerts are not supported.

What this note covers

This detection model defines the realistic detection and inference boundaries for cluster headache within a Vitals wearable context. It covers:

  1. What wearable signals are defensible
  2. What wearable signals are not defensible
  3. What logging and correlation support is useful
  4. What the evidence gap means for implementation

Defensible: Sleep-timing and attack-logging correlation

Attack-time logging (user-reported + timestamp)

  • The most defensible wearable-adjacent use is structured attack logging: timestamp, duration, side-locked pain, autonomic features, restlessness level, nocturnal wake time.
  • If a user logs attacks via Vitals, the system can correlate timestamps against sleep records to identify patterns: nocturnal timing, sleep-stage at onset, or sleep-disruption magnitude.
  • This is user-reported, not automatically detected.

Actigraphy sleep timing

  • Controlled actigraphy studies in cluster headache patients show real but complicated sleep-pattern deviations (PMID: 30470143).
  • The strongest defensible signal from actigraphy is sleep timing regularity and nocturnal disruption burden — not attack detection.
  • Sleep timing irregularity may be a contextual risk factor worth flagging if extreme, but interpretation requires clinical context.

Sleep-apnea association

  • Cluster headache has a recognized association with sleep apnea.
  • Vitals screening that flags possible sleep-apnea features (elevated RDR, repeated nocturnal desaturation events, very high AHI estimates from wearable) is within the defensible actigraphy-correlation envelope.
  • This is not a cluster-headache detector — it is a secondary flag worth human review.

Not defensible: Real-time attack detection

HRV-based attack detection

  • No study establishes a reliable HRV signature unique to cluster headache attacks.
  • HRV changes during pain states are variable and non-specific.
  • Attempting to detect attacks from HRV would produce high false-positive and false-negative rates.

Heart-rate-based attack detection

  • Cluster attacks produce autonomic activation (parasympathetic withdrawal, sympathetic surge) but the magnitude and timing of HR changes during attacks are not characterized well enough for reliable detection.
  • HR elevation during an attack is plausible but not validated as a detection pathway.

Motion-based attack detection

  • During attacks, patients are often restless (pacing, rocking) vs. migraine patients who tend to lie still. But motion patterns are too variable and non-specific for reliable detection.
  • Actigraphy motion data could theoretically detect the behavioral restless phenotype, but this is not validated as a cluster detector.

REM-stage prediction for attack timing

  • Earlier literature suggested attacks preferentially occurred during REM sleep.
  • Later work (PMID: 22337861) found attacks distributed across sleep stages rather than cleanly locked to REM.
  • A REM-alert wearable “to predict cluster attacks” is not supported by current evidence.

Evidence matrix for detection claims

SignalEvidence statusImplementation judgment
Attack-time logging + sleep correlationSupported (PMID: 30470143)Defensible — user-reported timestamps + actigraphy
Sleep timing irregularity as risk contextSupported (PMID: 30470143)Defensible — actigraphy review
Sleep-apnea flagSupported (epidemiologic association)Defensible with human review
HRV-only attack detectionGapNot defensible
HR-only attack detectionGapNot defensible
Motion-only attack detectionGapNot defensible
REM-alert attack predictionContested (PMID: 22337861)Not defensible
Oxygen-response-as-diagnosisContestedNot defensible

Confounders and limitations

What confounds attack-pattern correlation

  • Alcohol is a known attack trigger in many cluster patients; evening alcohol use confounds any sleep-attack correlation.
  • Sleep apnea events produce nocturnal awakenings that can mimic or mask attack-related wake events.
  • Daytime naps may trigger attacks in some patients, complicating timing-pattern analysis.
  • Preventive medications (verapamil, prednisone) affect sleep architecture independently of attack burden.
  • Shift work and circadian disruption confound timing-pattern analysis.

What limits real-time detection

  • The autonomic signature of a cluster attack is not uniquely separable from other nocturnal sympathetic surges (apnea events, nightmares, nocturia, GERD reflux).
  • Attack duration (15–180 minutes) means any real-time detection window would need to be broad, increasing false positives.
  • Many cluster patients do not have nightly attacks, which limits the ability to establish stable individual-level baselines.

Implementation boundary

Acceptable:
- User-reported attack logging with timestamp capture
- Timestamp correlation against sleep records (nocturnal timing)
- Sleep-disruption magnitude review (WASO, awakenings)
- Sleep-apnea feature flags (RDR, desaturation patterns) → human review
- Circadian timing pattern review across bout periods

Not acceptable:
- HRV-only attack detection
- HR-only attack detection
- Motion-only attack detection
- REM-stage attack prediction
- Automatic cluster headache diagnosis from wearable data
- Oxygen or triptan response as proof of diagnosis

Evidence status: Gap for real-time detection; Supported for actigraphy correlation and attack logging. Review: 2026-04-20.

Graph View

Backlinks