Neurological Sequelae Signal Clusters in VAERS for COVID-19 Vaccines 2020–2025

John Murphy, M.P.H.,DPH, The COVID-19 Long-haul Foundation

Scope and analytical approach

This analysis focuses on neurological adverse-event signal clusters reported to the U.S. Vaccine Adverse Event Reporting System (VAERS) following COVID-19 vaccination from late 2020 through 2025. It does not assert causality or efficacy. It interrogates patterns consistent with signal detection, emphasizing how neurological phenotypes group, when they occur, who they affect, and what in-database fields hint at severity or persistence. Where useful, it outlines methods to quantify disproportionality and temporal clustering directly within VAERS, recognizing the system’s constraints.

Key framing points:

VAERS is a passive, open system. Reports are unverified and subject to underreporting, stimulated reporting, misclassification, and duplication. It lacks a reliable, vaccine- and stratum-specific denominator. Therefore, VAERS is best for signal generation, not rate estimation or causal inference.
The question here is narrowly: Which neurological signal clusters are prominent in VAERS for COVID-19 vaccines, how are they characterized, and do fields within VAERS point to potential longer-term harms or negative consequences?
“Neurological sequelae” spans central, peripheral, autonomic, and neurovascular manifestations. Phenotypes were organized by MedDRA Preferred Terms (PTs) that commonly co-report, clinical logic, and time-to-onset shape.

Signal detection methods you can implement on VAERS alone:

Disproportionality metrics:
- Proportional reporting ratio (PRR)

PRR=a/(a+b)c/(c+d)\text{PRR} = \frac{a/(a+b)}{c/(c+d)}

where aa = counts of the neurological PT (or cluster) with COVID-19 vaccines; bb = other PTs with COVID-19 vaccines; cc = the same PT with all other vaccines; dd = other PTs with all other vaccines.

Reporting odds ratio (ROR)

ROR=a/bc/d\text{ROR} = \frac{a/b}{c/d}

Empirical Bayes shrinkage (e.g., EBGM) if implementing a Bayesian disproportionality approach to stabilize sparse counts.
Temporal signal characterization:
- Time-to-onset distribution (days from vaccination to symptom onset) to detect early acute clusters vs delayed patterns.
- Dose-sequence effects (dose 1 vs dose 2 vs booster) for recurrence or amplification.
- Age–sex stratification, recognizing confounding by vaccine rollout phases.
Co-reporting and network structure:
- Symptom-network analysis to identify PTs that co-occur (e.g., tinnitus with dizziness and headache; paresthesia with palpitations and orthostatic intolerance).
Outcome fields as severity proxies:
- Hospitalization, life-threatening, disability, death, “not recovered,” “prolonged hospitalization,” and “permanent disability.”

Data hygiene essentials:

Deduplicate near-identical reports and merge multiple submissions about the same person/event whenever feasible.
Normalize symptom terms across MedDRA PT spelling variants.
Treat changing data release practices over time as potential sources of structural breaks in counts.
Implement sensitivity analyses excluding reports lacking a plausible temporal link (e.g., onset months before vaccination) or missing critical fields.

Phenotype clusters observed in VAERS

The clusters below synthesize commonly reported neurological presentations and their characteristic patterns in VAERS. For each, the emphasis is on what VAERS can show: co-reported PTs, onset windows, severity proxies, and recurrent patterns upon re-exposure. None of this implies causality without validated follow-up.

Guillain–Barré spectrum (GBS/CIDP, variants)

Defining PTs: GBS, acute inflammatory demyelinating polyneuropathy, Miller–Fisher syndrome, polyradiculoneuropathy, chronic inflammatory demyelinating polyneuropathy (CIDP), albuminocytologic dissociation (when specified), areflexia.
Time-to-onset: Often within 2–30 days, with a concentration in the first 2 weeks consistent with post-immune phenomena.
Severity/persistence signals: Frequent hospitalization; some reports marked “life-threatening,” ventilatory support for autonomic or bulbar involvement; a subset with “permanent disability” and long recovery arcs. Presence of “not recovered” in follow-ups flags persistence.
Rechallenge patterns: Recurrence or worsening after subsequent dose in a fraction of narratives, which strengthens a signal but remains anecdotal in VAERS.
Mechanistic plausibility: Post-vaccinal immune-mediated demyelination via molecular mimicry or bystander activation (hypothesized, not proven within VAERS).

Demyelinating CNS events

Defining PTs: Optic neuritis, transverse myelitis, acute disseminated encephalomyelitis (ADEM), multiple sclerosis (MS) relapse/exacerbation, neuromyelitis optica spectrum disorder (NMOSD) events.
Time-to-onset: Typically within days to a few weeks.
Severity/persistence signals: Hospitalization common; “disability” fields sometimes positive. Narrative fields may mention MRI lesions or CSF oligoclonal bands.
Rechallenge patterns: Some reports suggest relapse after boosters in previously stable patients; causality cannot be established from VAERS.
Notes: Misclassification risk is high; case validation is essential in any follow-up design.

Peripheral neuropathies and small fiber neuropathy (SFN)

Defining PTs: Paresthesia, burning pain, dysesthesia, neuropathy peripheral, small fiber neuropathy, allodynia, erythromelalgia, numbness, tingling.
Time-to-onset: Frequently within hours to days; some with delayed onset over 1–4 weeks.
Severity/persistence signals: Many marked “not recovered” at time of reporting; prolonged courses in narrative follow-ups. Co-reporting with dysautonomia and tinnitus is common in network analyses.
Mechanistic plausibility: Potential immune-mediated injury to small fibers or dysautonomic involvement; VAERS cannot adjudicate mechanism.

Dysautonomia and POTS-like syndromes

Defining PTs: Postural orthostatic tachycardia syndrome (POTS), orthostatic intolerance, palpitations with dizziness/lightheadedness, syncope, tachycardia with fatigue and brain fog, exercise intolerance.
Time-to-onset: Often within 0–14 days; sometimes after a viral-like prodrome.
Severity/persistence signals: High prevalence of “not recovered”; functional impact on activity of daily living noted in narratives; some “disability” flags.
Co-reporting: Overlaps with SFN, anxiety, chest pain, and migraine. Distinguishing primary dysautonomia from anxiety is nontrivial in a passive system; cluster-level analysis benefits from requiring multiple dysautonomic PTs.

Seizure and convulsive events

Defining PTs: Seizure, tonic–clonic seizure, partial seizure, status epilepticus, loss of consciousness with postictal confusion (narrative).
Time-to-onset: Often within 0–3 days.
Severity/persistence signals: Hospitalization and ER visits frequent; small subset marked life-threatening.
Confounding: Fever and metabolic precipitants may be present; narratives sometimes implicate prior seizure disorders.

Tinnitus and vestibular syndromes

Defining PTs: Tinnitus, hypoacusis, hearing loss, hyperacusis, vertigo, dizziness, Meniere’s disease.
Time-to-onset: Hours to days; tinnitus often reported within 48 hours.
Severity/persistence signals: A large fraction not recovered at initial report; chronicity is a recurring theme in follow-ups.
Co-reporting: Headache, neck pain, anxiety, insomnia; network analyses often connect this cluster to dysautonomia symptoms.

Headache and migraine phenotypes with neurological features

Defining PTs: Headache, migraine, hemiplegic migraine, visual aura, photophobia, phonophobia, brain fog.
Time-to-onset: Within 0–3 days, frequently same day.
Severity/persistence signals: Usually self-limited, but a subset indicates prolonged impairments (“not recovered,” disability claims in long narratives).
Note: Headache is ubiquitous and nonspecific; signal value lies in atypical neurological accompaniments or persistence.

Movement disorders and tremor

Defining PTs: Tremor, internal vibrations, myoclonus, dystonia, parkinsonism, akathisia.
Time-to-onset: 0–14 days.
Severity/persistence signals: Tremor and “internal vibrations” often persist; functional impairment reported in narratives; hospitalizations less common than for GBS/CNS demyelination.
Adjudication challenge: Distinguishing functional neurological disorder (FND) from organic movement disorders requires clinical examination, not possible in VAERS.

Encephalitis and encephalopathy

Defining PTs: Encephalitis, autoimmune encephalitis, encephalopathy, confusion, delirium, altered mental status.
Time-to-onset: Days to weeks.
Severity/persistence signals: High rates of hospitalization; some ICU admissions; variable recovery.
Notes: Heterogeneous etiologies; narratives sometimes note CSF/EEG findings.

Cerebrovascular and thrombotic neurovascular events

Defining PTs: Cerebral venous sinus thrombosis (CVST), ischemic stroke, hemorrhagic stroke, TIA, thrombosis with thrombocytopenia syndrome (TTS).
Time-to-onset: Often 5–20 days for CVST/TTS-type patterns; ischemic stroke clustering earlier.
Severity/persistence signals: High severity including death; “life-threatening” flags frequent for CVST/TTS.
Mechanistic note: Platelet-activating antibodies described in clinical literature for TTS; VAERS does not contain assay data.

Facial nerve palsy and cranial neuropathies

Defining PTs: Bell’s palsy, facial paresis, abducens palsy, trigeminal neuralgia.
Time-to-onset: Days to weeks.
Severity/persistence signals: Mixed; many recover, some persistent deficits noted.
Signal value: Disproportionality can be tested due to relatively distinct PT coding.

Functional neurological disorder (FND)

Defining PTs: Functional neurological symptom disorder, psychogenic seizures, conversion disorder; often inferred from narratives describing fluctuant signs with normal imaging.
Time-to-onset: Immediate to days.
Signal nuance: Important to recognize as a real, disabling condition; its presence complicates attribution and requires careful clinical validation in any follow-up framework.

Characterizing signals with VAERS-only analytics

While VAERS cannot provide incidence or causality, it supports robust within-database signal characterization.

Disproportionality analysis

Construct 2×2 tables for each cluster using bundled PTs to minimize dilution by term variability.
Compute PRR and ROR overall and by strata:
- By product (mRNA vs adenoviral vs protein-based).
- By age bands (e.g., 12–17, 18–29, 30–49, 50–64, 65+).
- By sex.
Shrinkage estimators (e.g., empirical Bayes) can mitigate spurious extremes for rare outcomes.

Mathematically, for each stratum:

PRRstratum=as/(as+bs)cs/(cs+ds),RORstratum=as/bscs/ds\text{PRR}_{\text{stratum}} = \frac{a_s/(a_s+b_s)}{c_s/(c_s+d_s)}, \quad \text{ROR}_{\text{stratum}} = \frac{a_s/b_s}{c_s/d_s}

where subscript ss indicates the stratum (e.g., male 18–29).

Interpretation heuristics (not hard thresholds):

Elevated PRR/ROR with narrow confidence intervals and internal consistency across related PTs supports a signal.
Divergent results between single PT vs cluster-bundled PTs often reflect coding noise; prefer clusters.

Time-to-onset and hazard-shape analysis

Kernel density plots for onset days show:
- Acute immunologic patterns: peaks at 0–3 days (e.g., seizures, severe headache).
- Post-immune phenomena: peaks around 7–14 days (e.g., GBS, CVST/TTS).
- Chronic/persistent trajectories: flatter starts with long tails (e.g., tinnitus, SFN, dysautonomia).
Comparative onset curves across doses may reveal:
- Earlier onset with subsequent doses in sensitized individuals.
- Diminished early peaks in populations with lower cumulative reactogenicity.

Dose-sequence and rechallenge

Fields: Dose number and narrative re-exposure descriptions.
Signal markers: Same-phenotype recurrence or exacerbation after subsequent dose strengthens association but is vulnerable to recall/reporting bias.
Analytic step: Compute within-person conditional probability of recurrence (subset with multiple-dose reports), acknowledging selection bias.

Outcome severity gradients

Composite severity index: Weight hospitalization, life-threatening, disability, and death indicators to rank clusters.
Persistent outcome flags: “Not recovered,” “prolonged hospitalization,” “permanent disability” provide crude persistence signals.
Sensitivity: Recalculate after excluding reports missing outcome fields to check robustness.

Co-reporting networks

Approach: Build a bipartite graph of reports and PTs; project to PT–PT network weighted by co-occurrence.
Insights:
- Tinnitus–dizziness–headache cluster shows strong internal connectivity.
- Paresthesia–tachycardia–brain fog–fatigue suggests dysautonomic overlap.
- Distinct separation between GBS-spectrum and migraine-like clusters.

Lot and calendar-time checks

Lot clustering: Overrepresentation of specific lots in severe events may suggest handling/manufacturing issues but is confounded by lot size.
Temporal breaks: Identify spikes coinciding with media coverage (stimulated reporting) vs rollout phases.

Indicators of potential longer-term harms and negative consequences

VAERS does not provide longitudinal follow-up; nevertheless, several in-database markers and patterns suggest potential for chronicity or delayed consequences that merit prospective study.

Persistence proxies within VAERS

Not recovered at time of report: Highly prevalent in tinnitus, SFN, dysautonomia, internal tremor clusters; less so in acute headache or simple syncope.
Disability flags: More frequent in GBS/CIDP, demyelinating CNS events, and encephalitis; present but less frequent in vestibular/tinnitus and dysautonomia clusters.
Prolonged hospitalization: Signals severity for GBS, encephalitis, and stroke clusters; relatively rare for SFN/dysautonomia despite functional burden.

Chronic neuropathic phenotypes

SFN/dysesthesia: Narrative mentions of biopsy-confirmed SFN exist; many report months-long burning pain and sensory dysfunction. The co-reporting with POTS-like symptoms indicates possible shared pathophysiology.
Dysautonomia/POTS: Reports describe sustained orthostatic intolerance, exercise intolerance, and brain fog; some include ineffective short-term management, hinting at longer trajectories.
Tinnitus and vestibular disorders: Large subset reports enduring symptoms; a fraction describe constant intrusive tinnitus with quality-of-life impacts consistent with chronicity.

Transition from acute to chronic neuroimmunology

GBS to CIDP evolution: A minority of reports describe ongoing or relapsing deficits beyond the typical GBS course, consistent with CIDP evolution; this is a hypothesis-generating pattern requiring clinical confirmation.
MS relapses and new demyelination: VAERS narratives sometimes indicate relapse temporally linked to vaccination; persistence by definition is expected in MS, but whether vaccination precipitates relapses at supra-baseline rates requires controlled designs outside VAERS.

Rechallenge and cumulative exposure

Same-phenotype recurrence after boosters: Reported across several clusters (tinnitus, paresthesia, dysautonomia, headache with aura, Bell’s palsy, and rarely GBS); while still anecdotal, recurrence on re-exposure is a classic signal-strengthener.
Dose-spacing effects: Some narratives suggest mitigation with longer intervals; VAERS cannot quantify this, but it informs hypothesis generation for trial design.

Neurocognitive complaints

Brain fog, memory issues, attention deficits: Often co-reported with fatigue, headache, dysautonomia; persistence flags are common. This cluster’s non-specificity and overlap with post-infectious syndromes complicate interpretation but justify prospective cognitive testing in follow-up cohorts.

Data-quality constraints and bias management

Interpreting neurological signals in VAERS demands explicit strategies to mitigate bias and misclassification.

Underreporting and denominator void: Mild events are less likely reported; severe or media-salient events may be overrepresented. Incidence cannot be computed without reliable denominators and case verification.
Stimulated reporting: Media cycles (e.g., around tinnitus, myocarditis, CVST) can generate transient spikes. Calendar-time models help distinguish organic patterns from stimulated bursts.
Diagnostic misclassification: Neurological diagnoses require examination and testing. Many VAERS reports reflect symptom-level descriptors rather than confirmed syndromes. Bundling related PTs and performing clinician review on a sample improves validity.
Duplication and serial updates: Multiple reports for the same event can inflate counts. Deduplication keys (age, sex, date, location, lot, narrative) reduce multiplicity but may miss less obvious duplicates.
Competing risks and background illness: Older adults and those with comorbidities have high background rates of stroke, neuropathy, and encephalopathy. Comparative designs outside VAERS are needed to attribute excess risk.

Recommendations for rigorous next steps

Below is a practical roadmap you can execute to turn VAERS neurological signals into decision-grade evidence, tailored to fit regulatory, investor, and clinical audiences.

Case definition and curation

Define cluster-level case criteria combining PTs and narrative keywords:
- GBS-spectrum: GBS/CIDP PTs plus “areflexia,” “CSF albuminocytologic dissociation,” “NCS/EMG demyelinating pattern.”
- CNS demyelination: “optic neuritis,” “transverse myelitis,” “ADEM,” “new MRI lesions.”
- SFN: “burning pain,” “allodynia,” “skin biopsy,” “IENFD,” “QSART.”
- Dysautonomia/POTS: “standing HR increase ≥30 bpm,” “orthostatic intolerance,” “tilt-table.”
- Tinnitus/vestibular: bilateral/unilateral tinnitus, “sudden sensorineural hearing loss,” “ENG/VNG.”
- Seizures: ictal features plus postictal descriptors; exclude syncope-only narratives.
- Encephalitis/encephalopathy: “CSF pleocytosis,” “EEG slowing,” “autoantibodies” (if reported).
Clinician adjudication of a stratified sample to estimate positive predictive value (PPV) of your case-finding rules within VAERS.

VAERS-only analytics

Compute PRR/ROR for each cluster overall and by product, age, and sex; perform sensitivity analyses:
- Exclude reports with onset >42 days.
- Exclude reports missing outcome fields.
- Down-weight media-stimulated weeks.
Time-to-onset models:
- Fit mixture models to differentiate immediate reactogenic vs post-immune peaks.
- Compare dose 1 vs 2 vs booster hazard shapes.
Co-reporting networks:
- Identify subclusters (e.g., SFN–dysautonomia–tinnitus) for targeted follow-up protocols.

Validation beyond VAERS (minimal viable)

Chart-review cohort: Partner with systems willing to review a sample of VAERS-linked patients to confirm diagnoses and outcomes.
Self-controlled designs: Implement self-controlled case series (SCCS) in EHR datasets to estimate relative incidence within individuals, minimizing confounding.
Targeted registry build: Establish a prospective registry for high-signal clusters (GBS-spectrum, CNS demyelination, persistent tinnitus/SFN/dysautonomia), with standardized follow-up at 1, 3, 6, and 12 months.

Outcomes and assays to capture

Functional outcomes: mRS for CNS events, INCAT/ONLS for neuropathies, COMPASS-31 for dysautonomia, THI for tinnitus, PROMIS-Cog for cognitive symptoms.
Objective testing: NCS/EMG, skin biopsy for IENFD, tilt-table/active stand, audiometry, MRI/CSF as indicated.
Biomarkers: Autoantibodies (e.g., ganglioside panels for GBS), inflammatory markers, platelet-activating assays when thrombotic neurovascular events are suspected.

Safety signal governance

Triaging criteria: Elevate clusters with high severity (hospitalization, life-threatening, death), strong temporal clustering, and recurrence on rechallenge.
Communication: Present VAERS-derived signals as hypotheses with quantified uncertainty and a concrete validation plan. Avoid claims about absolute risk without validated denominators.

Summary table of neurological signal clusters

Cluster	Defining PTs (examples)	Typical onset window	Severity/persistence markers	Frequent co-reporting
GBS/CIDP	GBS, polyradiculoneuropathy, CIDP	2–30 days	Hospitalization, disability, ventilatory support	Dysautonomia, pain
CNS demyelination	Optic neuritis, transverse myelitis, ADEM, MS relapse	3–28 days	Hospitalization, disability	Headache, visual symptoms
SFN/neuropathic pain	Paresthesia, dysesthesia, SFN	0–28 days	Not recovered, chronicity	Dysautonomia, tinnitus
Dysautonomia/POTS	POTS, orthostatic intolerance, syncope	0–14 days	Not recovered, activity limitation	Palpitations, brain fog
Seizures	Seizure, tonic–clonic, status epilepticus	0–3 days	ER/hospitalization	Fever, headache
Tinnitus/vestibular	Tinnitus, vertigo, hearing loss	0–7 days	Not recovered, disability (subset)	Headache, insomnia
Movement/tremor	Tremor, myoclonus, dystonia	0–14 days	Persistent symptoms	Anxiety, dysautonomia
Encephalitis/encephalopathy	Encephalitis, altered mental status	3–21 days	ICU/hospitalization	Seizure, fever
Neurovascular thrombotic	CVST, stroke, TIA, TTS	5–20 days	Life-threatening, death	Headache, thrombocytopenia
Cranial neuropathies	Bell’s palsy, trigeminal neuralgia	3–21 days	Variable recovery	Taste change, ear pain

Sources for this table are the VAERS report structures and symptom co-reporting patterns; it is intended for signal orientation rather than rate estimation or causal claims.

Other Considerations

Based on VAERS, what neurological side effects are most commonly reported? Headache and dizziness dominate numerically but are nonspecific. Among more distinct neurological phenotypes, tinnitus/vestibular symptoms, paresthesia/neuropathic pain (including SFN-like presentations), and dysautonomia/POTS-like syndromes are frequently reported. Seizures, Bell’s palsy, and movement disorders (tremor/internal vibrations) appear regularly but at lower absolute counts than nonspecific headache. Severe but rarer clusters include GBS-spectrum, CNS demyelination, encephalitis/encephalopathy, and neurovascular thromboses.
Are there signals that might predict longer-term harms? VAERS persistence proxies (“not recovered,” “disability,” narrative chronicity) are enriched in tinnitus/vestibular, SFN/neuropathic pain, and dysautonomia clusters. GBS-spectrum and CNS demyelination show high severity and documented prolonged recovery in narratives and outcome flags. Rechallenge recurrences in a subset (e.g., tinnitus, paresthesia, Bell’s palsy) and evolution from GBS to CIDP in rare cases are additional warning markers. These are signals—not proof—and they justify prospective, controlled follow-up.