Preprint
Review

This version is not peer-reviewed.

Vaccines, the Gut-Brain Axis, and Neurodevelopment: A Structured Narrative Review

Submitted:

29 September 2026

Posted:

30 September 2026

You are already at the latest version

Abstract
Objective. To systematically evaluate the interface between vaccination and the gut-brain axis (GBA) in both directions of effect: first, whether vaccines or their components alter GBA biology, including microbiome composition, barrier integrity, cytokine responses, and mitochondrial function; second, whether pre-existing GBA state modifies the response to vaccination; and third, whether either direction bears on neurodevelopmental outcomes or on the adverse events recorded in trials and surveillance. Data Sources. PubMed and Google Scholar were searched directly, and the Walden University Library federated discovery service was used to search its subscribed collections, including MEDLINE, Embase, Cochrane, CINAHL, Scopus and ScienceDirect, from inception to 5 September 2025. Eligibility Criteria. Human and animal studies examining licensed vaccines or vaccine components with measurable outcomes in GBA domains, neurodevelopmental endpoints, or post-marketing adverse events. Results. Of 3,047 records, 60 studies met inclusion criteria in the primary search, with a further 7 identified through a targeted supplementary search to September 2026, giving 67 included studies, listed in Appendix A. Microbiome composition influenced vaccine immunogenicity: higher Bifidobacterium abundance predicted stronger infant antibody responses, and neonatal antibiotic exposure reduced antibody titres to several infant vaccines measured at 7 months of age. Evidence on whether vaccination itself alters the microbiome is limited: one longitudinal adult cohort reported compositional stability, pediatric data are close to absent, and transient shifts have been observed in mice. Animal maternal immune activation models implicated IL-6 and IL-17A pathways in neurodevelopmental alterations, whereas maternal vaccination showed no such association. Cohorts exceeding one million children found no association between autism and the vaccine exposures they measured, principally measles-mumps-rubella vaccination, thimerosal and cumulative antigen count, consistent with a 2025 Danish cohort of aluminium-adsorbed vaccine exposure and a December 2025 WHO advisory reaffirmation. No study identified examined DTaP, hepatitis B or varicella in relation to neurodevelopmental outcomes. Few eligible studies directly assessed barrier or mitochondrial endpoints. Active surveillance quantified specific adverse events, including febrile seizures, intussusception, immune thrombocytopenic purpura, myocarditis (lower with 2023-2024 mRNA formulations), and anaphylaxis. Conclusions. The two directions of effect at this interface are supported to very different degrees, and that asymmetry is the principal finding. Evidence that gut microbiome state modifies vaccine response is moderate in certainty. Evidence on whether vaccination alters gut-brain axis biology, the direction the causal hypotheses in this field assert, is thin rather than negative: barrier, mitochondrial and paediatric data are essentially absent, and the single relevant longitudinal cohort concerns adults receiving COVID-19 mRNA vaccines. Large-scale epidemiology does not support vaccine-attributable increases in autism risk for the exposures those cohorts measured, principally measles-mumps-rubella vaccination, thimerosal and cumulative antigen count; it does not address the microbiome pathway. Several rare adverse events are quantifiable through active surveillance, whereas others remain uncertain signals. Continued active surveillance and transparent risk communication remain essential.
Keywords: 
;  ;  ;  ;  ;  ;  ;  ;  ;  

1. Introduction

1.1. The Gut-Brain Axis and Its Relevance to Vaccination

The gut-brain axis (GBA) is a bidirectional network that links the enteric nervous system, the central nervous system, the immune system, and the gut microbiota. This system has emerged as central to multiple physiological and pathological processes, ranging from digestion and metabolism to mood, cognition, and neurodevelopment (Cryan & Dinan, 2012; Carabotti et al., 2015). Vaccination, as a controlled stimulus of the immune system, can plausibly intersect with GBA pathways, raising the question of whether vaccines or their components have measurable effects on microbiota composition, barrier integrity, cytokine signaling, or neurodevelopmental outcomes.

Neural and Endocrine Pathways

The vagus nerve provides a direct line of communication between the gut and the brain, transmitting sensory information about luminal contents, microbial metabolites, and inflammatory states (Furness, 2012). Enteric neurons embedded in the intestinal wall modulate gastrointestinal motility and secretions, which in turn shape the gut environment and microbial composition. Endocrine pathways also contribute: gut microbes produce metabolites such as short-chain fatty acids (SCFAs), including acetate, propionate, and butyrate, which regulate host immune responses, influence neurotransmitter synthesis, and act as signaling molecules across the blood-brain barrier (Koh et al., 2016). Host stress hormones such as cortisol and catecholamines feed back on intestinal permeability and microbial ecology, reinforcing the bidirectional nature of the system (Mayer et al., 2014). Recent work has further highlighted that microbial metabolites including SCFAs and tryptophan-derived indoles can modulate neuroendocrine signaling through vagal and hypothalamic-pituitary-adrenal pathways, influencing stress responses and immune readiness (Ramadan et al., 2025).

Immune and Barrier Mechanisms

Pattern recognition receptors (PRRs) including toll-like receptors (TLRs) and nucleotide-binding oligomerization domain (NOD) proteins detect microbial signals and activate cytokine cascades. Pro-inflammatory mediators such as interleukin (IL)-6, IL-1β, tumor necrosis factor (TNF)-α, interferon (IFN)-γ, and IL-17A can influence blood-brain barrier permeability, cross into cerebrospinal fluid, or act indirectly via peripheral neural pathways to alter brain development and function (Belkaid & Hand, 2014). Barrier integrity is a critical element in this system. The intestinal epithelium and the blood-brain barrier regulate systemic passage of metabolites, immune cells, and microbial products. Dysregulation at these interfaces, often measured via tight junction proteins and circulating markers such as zonulin, has been linked to inflammatory bowel disease, multiple sclerosis, and autism spectrum disorder (Fasano, 2012; Fiorentino et al., 2016).

Neurotransmitter Interactions

Microbial species can directly synthesize or modulate neurotransmitters. Certain Lactobacillus strains promote gamma-aminobutyric acid (GABA) signaling, while gut enterochromaffin cells, influenced by microbial cues, produce the majority of peripheral serotonin (Yano et al., 2015). Dopamine, norepinephrine, and acetylcholine metabolism are likewise subject to microbial regulation. These mechanisms offer biologically plausible routes by which immune activation, whether from infection or vaccination, could reverberate through the gut-brain system.

1.2. Vaccination as a Controlled Immune Stimulus

Vaccines introduce antigens to the host immune system in formulations designed to generate adaptive memory without producing infection. Platforms include live attenuated viruses, inactivated pathogens, protein subunits, polysaccharide conjugates, recombinant viral vectors, and messenger RNA vaccines, often with adjuvants to enhance immunogenicity (Plotkin, 2014).
At the innate level, adjuvants provide “danger signals” that activate dendritic cells and prime T- and B-cell responses. Aluminum salts, oil-in-water emulsions such as MF59 and AS03, and saponin-based formulations such as AS01 and Matrix-M engage pattern recognition receptors and inflammasome pathways, resulting in cytokine release and adaptive immune priming (Lindblad, 2004). Transient systemic increases in cytokines, particularly IL-6 and TNF-α, have been observed following vaccination in both human trials and animal models (Mosca et al., 2008). Although typically short-lived, these signals can theoretically interact with gut permeability and central nervous system pathways.
Adaptive responses involve clonal expansion of antigen-specific B and T cells and generation of memory. While primarily systemic, these processes unfold in an environment shaped by the microbiome and by neuroimmune crosstalk. Thus, vaccination functions as a controlled immunological perturbation with potential to intersect with GBA biology.

1.3. Microbiome Influences on Vaccine Responses

The most consistent findings across human and animal studies concern the microbiome’s influence on vaccine immunogenicity. In a prospective Bangladeshi birth cohort, higher Bifidobacterium abundance in early infancy predicted stronger T-cell and antibody responses to several infant vaccines at 2 years of age (Huda et al., 2019), and in rural Ghana, pre-vaccination microbiome composition differed significantly between rotavirus vaccine responders and nonresponders (Harris et al., 2017). A United States longitudinal cohort extends this work to a high-income, high-uptake setting: among infants vaccinated with RotaTeq, higher alpha diversity at one month was associated with higher rotavirus-specific IgA at six months, and diversity at six months with IgA at twelve months, with no association between overall community composition and response at any timepoint. The analytic subsets numbered 47 to 56 of 353 originally enrolled, individual taxon associations were identified at a false discovery threshold of 0.2, and several directional findings, including those for Bifidobacterium and Escherichia-Shigella, run opposite to earlier cohorts. The taxon-level results are therefore treated as exploratory, while the diversity finding is consistent in direction with the low-income-setting literature (Narváez-Miranda et al., 2026). Praharaj et al. (2015) reviewed the evidence on whether probiotic and antibiotic exposures modify vaccine responses. In a two-phase randomized study of 33 healthy adults, a five-day broad-spectrum antibiotic regimen given around influenza vaccination reduced microbial richness and diversity in both phases. Antibody responses were largely unaffected in the first phase, in which participants had substantial pre-existing influenza immunity, whereas in the second phase, restricted to participants with low baseline titres, the antibiotic group showed impaired H1N1-specific neutralization at days 90 and 180 and reduced H1N1-specific IgG1 at days 7 and 30, confirmed by an independent laboratory using a separate assay (Hagan et al., 2019). Germ-free and antibiotic-treated mice fail to mount effective responses to viral vaccines, an effect reversible with microbial recolonization (Oh et al., 2014). Short-chain fatty acids derived from dietary fibre independently support host antibody responses, including plasma-cell differentiation and both mucosal IgA and systemic IgG, although that work examined homeostatic and pathogen-specific responses rather than vaccination (Kim et al., 2016). A 2025 prospective study by Ryan et al. further demonstrated that neonates exposed to antibiotics in the first week of life exhibited reduced antibody titres to pneumococcal conjugate, Haemophilus influenzae type b, and diphtheria toxoid antigens at 7 months of age, with reduced Bifidobacterium abundance at the time of vaccination correlating with those titres six months later.
Probiotic and synbiotic interventions reinforce these links. Probiotic supplementation around the time of vaccination has improved antibody responses in some randomized trials, though a systematic review of 26 such trials found a beneficial effect in only about half, with substantial variation by strain, dose, and timing, and the single trial of maternal supplementation during pregnancy reported lower vaccine responses in the infants (Zimmermann & Curtis, 2018). A 2024 systematic review and meta-analysis of probiotic supplementation reported continued modest benefits with significant heterogeneity across strains and vaccine types. Taken together these data support a moderate-certainty conclusion that microbial composition and metabolites shape vaccine immunogenicity, with the rating and its basis set out in Table 3.

1.4. Do Vaccines Alter the Microbiome?

The reverse question, whether vaccines themselves alter the microbiome, has been less extensively studied. A longitudinal study of healthy and immunocompromised adults receiving COVID-19 vaccines found that gut microbial diversity, composition, and functional capacity remained stable across vaccination timepoints (Boston et al., 2024). Pediatric data remain sparse. Animal models yield heterogeneous results, with some showing temporary shifts after adjuvanted vaccines but no durable dysbiosis (Borgognone et al., 2022; Khadka et al., 2023). Ardura-Garcia et al. (2024) noted that vaccines may generate changes in intestinal microbiota composition, though evidence specifically addressing microbiome stability remains limited. The balance of evidence suggests that vaccines do not cause persistent microbiome disruption, although more pediatric longitudinal data are needed.

1.5. Maternal Immune Activation Models

Maternal immune activation (MIA) provides a framework for understanding how immune stimulation can affect neurodevelopment. In rodent models, maternal exposure to viral mimetics such as polyinosinic:polycytidylic acid induces IL-6 and IL-17A-mediated pathways that disrupt cortical development and produce autism-like behaviors in offspring (Smith et al., 2007; Choi et al., 2016). Colonization with segmented filamentous bacteria predisposes dams to IL-17A production, linking microbiota composition to neurodevelopmental risk (Kim et al., 2017). The human evidence is weaker than the animal work and is set out in Section 3.4. In brief, maternal fever in the second trimester was associated with increased autism risk in one prospective cohort, while a registry analysis of maternal infection found no overall association, its positive estimates being confined to two of 33 trimester-specific comparisons made without correction for multiple testing (Atladóttir et al., 2010; Hornig et al., 2018). Maternal vaccination with influenza or Tdap has not been associated with increased autism risk in the cohorts examined (Zerbo et al., 2017; Becerra-Culqui et al., 2018). These findings underscore the importance of distinguishing between infection-driven immune activation and controlled vaccine-induced stimulation.

1.6. Historical Context and Controversies

Public debate about vaccine safety has historically centered on potential links to neurodevelopmental disorders, particularly autism spectrum disorder. The hypothesis that MMR vaccination could cause autism, first proposed in the late 1990s, has been comprehensively refuted by large epidemiological studies and meta-analyses involving millions of children (Madsen et al., 2002; Taylor et al., 2014; Hviid et al., 2019). Thimerosal, an ethylmercury preservative, was also hypothesized to contribute to autism risk; however, multiple cohort studies found no increased risk of autism, and autism prevalence continued to rise after thimerosal was removed from pediatric vaccines (Verstraeten et al., 2003; Andrews et al., 2004; Schechter & Grether, 2008). These cohorts did report isolated associations with other neurodevelopmental outcomes that were not consistent across sites, and those are described in Section 3.5.
Adjuvants have also attracted scrutiny. Aluminum salts are known to persist in phagocytes and distribute systemically in animal models, but epidemiological studies find no evidence of increased autism or neurodevelopmental risk associated with aluminum-containing vaccines (Hviid et al., 2003; Mitkus et al., 2011). The landmark 2025 Danish cohort study by Andersson et al. further strengthened this evidence base, examining 1,224,176 children born between 1997 and 2018 and finding no increased risk for 50 chronic disorders associated with aluminum-adsorbed vaccines. Squalene emulsions such as AS03 were linked to increased narcolepsy risk in Nordic countries following H1N1 vaccination, illustrating that rare, subgroup-specific risks can occur and that active surveillance systems are capable of detecting them (Partinen et al., 2012). More recently, myocarditis following mRNA COVID-19 vaccination has emerged as a quantifiable adverse event, especially in young males (Oster et al., 2022). These examples emphasize that extensive high-certainty epidemiological studies have not detected an association between vaccines and autism at the population level, while confirming that vaccines can, in rare cases, cause other well-characterized adverse events.

2. Methods

2.1. Protocol and Reporting Framework

This systematic review followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines, which emphasize transparency, reproducibility, and structured reporting (Page et al., 2021). The protocol was developed a priori and included decisions regarding eligibility criteria, outcomes of interest, and approaches to risk-of-bias and certainty-of-evidence assessment. This review is reported as a structured narrative review. It follows PRISMA 2020 reporting conventions, applied prespecified eligibility criteria, screened all records in duplicate, and assessed risk of bias with design-appropriate instruments. It is not reported as a systematic review, because retrieval cannot be reproduced: the search ran through a federated discovery service rather than independent native-interface searches, and the executed query strings were not retained. Registration with PROSPERO was attempted and was not completed, because the review did not fall within the eligible review categories: PROSPERO accepts systematic reviews of health-related outcomes in human populations and does not accept narrative or literature reviews, or reviews of animal studies, whereas the present review integrates human epidemiological evidence with animal and mechanistic evidence through the Navigation Guide method. The protocol, search strategy, screening records and appraisal materials are instead archived on the Open Science Framework at https://doi.org/10.17605/OSF.IO/GA2FM, under a CC0 1.0 Universal licence. That archive was created after the review was completed and records the methods as conducted; it does not establish that they were specified in advance, and the protocol was not separately published. The protocol defined the study population, interventions, comparators, outcomes, and study designs following the PICOS framework, and the absence of registration is recorded among the limitations in Section 5.
The overarching aim was to identify, appraise, and synthesize evidence on the interface between vaccination and gut-brain axis (GBA) biology. The interface admits two directions of effect, and the review addresses both, together with a third question that follows from them. The first question is whether vaccines or their components alter GBA biology, measured as microbiome composition, barrier function, cytokine signalling, or mitochondrial indices. The second is whether pre-existing GBA state modifies the response to vaccination, measured as immunogenicity, seroconversion, or antibody persistence. The third is whether either direction bears on neurodevelopmental outcomes or on the adverse events recorded in clinical trials and surveillance systems. Distinguishing the two directions is not a formality. Evidence bearing on one does not transfer to the other, and a substantial part of the confusion in this literature arises from treating a finding about the microbiome as a determinant of vaccine response as though it were a finding about vaccines as a disturber of the microbiome. The results are reported under headings that keep the directions separate, and the certainty ratings in Table 3 are assigned to each direction independently. The review also evaluated adverse events reported in both clinical trials and surveillance systems, distinguishing between confirmed events with quantifiable incidence and signal-generating reports where incidence remains uncertain. The synthesis avoided categorical framings of “safe” or “unsafe,” instead presenting quantified estimates where available and explicitly stating when incidence was unknown. Because safety is not a property an intervention can possess in the absolute, only a comparative one, the review frames each recognized harm against the outcome it is weighed against, namely the disease the vaccine prevents. This benefit-risk framing is developed in the subsection below and quantified in Table 13.

2.2. Information Sources and Search Strategy

The literature search was conducted from database inception to 5 September 2025 through three routes. PubMed was searched directly through the NCBI interface. Google Scholar was searched directly. The remaining coverage was obtained through the Walden University Library Multi-Database Search, a federated discovery service that queries the university’s subscribed collections simultaneously; those collections include MEDLINE with Full Text, Embase, the Cochrane Database of Systematic Reviews, the Cochrane Central Register of Controlled Trials, CINAHL Plus with Full Text, Scopus, ScienceDirect, and the Health and Medical Collection. Separate native-interface searches using database-specific controlled vocabulary were not run in Embase, Cochrane, Scopus or CINAHL, so recall within each individual database cannot be reported and hit counts are available for the combined search rather than by database. This is a departure from the conventional practice of searching each database independently and is recorded among the limitations in Section 5. In addition, Google Scholar was used to identify grey literature such as theses, preprints, and in-press manuscripts encountered alongside the indexed literature. This is distinct from the systematic grey-literature search listed as a dissemination-bias probe in Table 7, which would require searching trial registries, dissertation databases, and preprint servers against the full eligibility criteria and was not undertaken. Reference lists of included studies and relevant systematic reviews were also screened to ensure completeness.
The search strategy combined Medical Subject Headings with title and abstract free-text terms, structured as two concept blocks: vaccination combined with gut microbiome or gut-brain axis terms, and vaccination combined with neurodevelopmental and inflammation terms. The query executed in PubMed was:
("Vaccines"[MeSH Terms] OR "Vaccination"[MeSH Terms] OR "Immunization"[MeSH Terms] OR vaccin*[tiab] OR immuniz*[tiab] OR immunis*[tiab]) AND ("Gastrointestinal Microbiome"[MeSH Terms] OR "Brain-Gut Axis"[MeSH Terms] OR "gut microbiome"[tiab] OR "gut microbiota"[tiab] OR "gut-brain"[tiab] OR "gut brain"[tiab]) OR ("Vaccines"[MeSH Terms] OR "Vaccination"[MeSH Terms] OR "Immunization"[MeSH Terms] OR vaccin*[tiab] OR immuniz*[tiab] OR immunis*[tiab]) AND ("Neurodevelopmental Disorders"[MeSH Terms] OR neurodevelopment*[tiab]) AND ("Inflammation"[MeSH Terms] OR inflammat*[tiab])
The same concept blocks and search terms were applied in Google Scholar and through the Walden University Library discovery service, where Medical Subject Headings indexing is unavailable and free-text equivalents were used; the exact strings entered in those two routes were not retained. Reference lists of retrieved articles were additionally searched. No date or language restrictions were applied during the initial search, although only English-language publications were ultimately included. A targeted supplementary search was subsequently conducted to 27 September 2026 in four prespecified areas in which the evidence base was known to be active: aluminium adjuvant safety, myocarditis incidence and mechanism following messenger RNA vaccination, regulatory and advisory positions on vaccines and autism, and gut microbiome effects on infant vaccine response. This supplementary search was targeted rather than systematic, was not run across the full set of sources used in the primary search, and is reported here so that its contribution can be separated from that of the primary search. Sources identified through it are Andersson et al. (2025), Choi et al. (2025), Cao et al. (2025), Ryan et al. (2025), Pfeiffer et al. (2026), Doyon-Plourde et al. (2026) and Narváez-Miranda et al. (2026), together with the two narrative reviews Nirenberg et al. (2026) and Moser and Offit (2026), which are cited for context but are not counted among the included studies, and the regulatory and advisory documents cited in Section 3.5.

2.3. Eligibility Criteria

Eligibility criteria were established to include studies that directly addressed the relationship between vaccines or vaccine components and gut-brain axis or neurodevelopmental outcomes.
Inclusion criteria: Human and animal studies examining licensed vaccines or their components; outcomes relating to microbiome composition, metabolites, barrier function (intestinal or blood-brain), cytokine or chemokine levels, mitochondrial indices, or neurodevelopment; epidemiological studies of autism spectrum disorder or related neurodevelopmental conditions; and post-marketing surveillance data assessing adverse events following vaccination.
Exclusion criteria: Narrative commentaries and opinion pieces without primary data; ecological studies lacking individual-level outcome data; and case reports, except where they described novel adverse events not captured in active surveillance, in which case they were included only as signal-generating evidence. A defined exception applies to the first two of these categories. Ecological analyses, hypothesis papers, and commentaries are excluded from the evidence synthesis and contribute nothing to the certainty ratings in Table 3 and Table 16, but a subset is appraised separately in Table 15 and Table 19 and identified there as not weighted. These are the publications most frequently cited in support of a vaccine and neurodevelopment association, and a review that excluded them silently could neither explain why they do not support the inferences drawn from them nor be checked by a reader who has encountered them. They are therefore reported as appraised but excluded, and the distinction is carried explicitly in the weighting column of each table.
Study designs included randomized controlled trials, cohort studies, case-control studies, cross-sectional studies, systematic reviews, meta-analyses, and mechanistic animal experiments. Eligible populations encompassed all age groups, including pregnant individuals and neonates. Animal models were restricted to rodents and primates. Exposures included live attenuated, inactivated, subunit, polysaccharide-conjugate, mRNA, viral vector, and nanoparticle-based vaccines, as well as vaccine components such as aluminum salts.

2.4. Study Selection

Two reviewers independently screened all titles and abstracts using the eligibility criteria. Full texts of potentially eligible studies were retrieved and reviewed in detail. All records were screened independently by the two authors. Disagreements were resolved by discussion between them; no third adjudicator was available, and inter-rater agreement statistics were not calculated. Reasons for exclusion were documented. The seven eligible studies identified through the targeted supplementary search described in Section 2.2 were identified by directed searching in four areas rather than by the full database strategy, and were assessed against the same eligibility criteria by the same two reviewers. The process followed PRISMA standards: records identified, duplicates removed, records screened, full texts reviewed, and studies included.

2.5. Data Extraction

Data were extracted using standardized templates based on Cochrane Collaboration recommendations. Extracted variables included study characteristics (author, year, country, design, sample size, setting), population characteristics (age, sex, baseline health, comorbidities), exposure details (vaccine type, dose, adjuvant, route, schedule, timing), comparator details (placebo, unvaccinated, pre-vaccination baseline), outcome measures (microbiome indices, cytokines, permeability markers, mitochondrial metrics, neurodevelopmental diagnoses, adverse event incidence), effect estimates with 95% confidence intervals, and funding source and conflicts of interest disclosures. Where possible, denominators for incidence calculations were recorded. For passive surveillance data without denominators, results were treated as hypothesis-generating signals rather than incidence estimates.

2.6. Risk of Bias Assessment

Risk of bias was assessed according to study design. Randomized controlled trials were appraised using the Cochrane Risk of Bias 2 tool (Sterne et al., 2019). Observational studies were evaluated with ROBINS-I (Sterne et al., 2016). Animal studies were assessed using the SYRCLE tool (Hooijmans et al., 2014). Passive surveillance data were not subject to formal bias tools, since underreporting and stimulated reporting are inherent; instead, they were labeled as very low-certainty evidence and classified as signals only. These instruments structured the appraisal by fixing the domains considered for each design, namely randomisation and deviation from intended intervention for trials, confounding and selection into the study for observational designs, and sequence generation, allocation concealment, and selective outcome reporting for animal experiments. Domain-level ratings were not tabulated for every included study. The appraisal is reported narratively, at study level in Table 14 to 18 and at outcome-domain level in Table 3 and Table 7, and 16, with the specific design feature driving each judgement stated in the text. Study-level domain ratings are presented in Table 4.
For each included study, reported effect estimates, sample sizes, comparators, and outcome measures were extracted and appraised against the design that produced them. Where a study reported estimates across multiple sites, outcomes, or subgroups, each was recorded rather than only those the study emphasised.

2.7. Certainty of Evidence Assessment

Certainty of evidence was graded using the GRADE framework (Guyatt et al., 2008). Evidence quality was rated as high, moderate, low, or very low based on risk of bias, consistency, directness, precision, and publication bias. One starting rule governs every rating in this review and is applied without exception. Bodies of evidence composed of randomized trials start at high certainty and may be downgraded. Bodies of evidence composed of observational studies start at low certainty and may be upgraded only for a large effect, a dose-response gradient, or where plausible residual confounding would act against the observed result. Where a rating in Table 3 and Table 7 or 16 sits above low for an observational body of evidence, the upgrade factor is named in the basis column; where it sits below the starting level, the downgrade factor is named. Passive surveillance data were always classified as very low certainty. The factors driving each rating are stated alongside the rating in Table 3 and Table 7, and 16 rather than recorded in a separate domain matrix; across this evidence base the operative factors were most often indirectness, where the exposure or outcome measured differs from the one of interest, and imprecision, where confidence intervals remain compatible with effects in both directions.

2.8. Synthesis Approach

Given the heterogeneity across study designs, a mixed-methods approach was adopted. When multiple studies reported comparable outcomes, effect estimates were summarized narratively, and pooled results from existing meta-analyses were reported where appropriate. For rare adverse events, incidence was expressed as cases per denominator population when data were available from active surveillance. Where incidence was unknown, outcomes were explicitly described as signals. Subgroup analyses were highlighted when reported, including those by age, sex, vaccine platform, and geographic region.

3. Results

The results are reported under the three questions set out in Section 2.1, and the amount of evidence bearing on each differs sharply. That imbalance is itself a finding, and the structure below is designed to make it visible rather than to disguise it.
Section 3.2 addresses the first question, whether the state of the gut-brain axis modifies the response to vaccination. Section 3.3 addresses the second, whether vaccination alters the gut-brain axis, and reports that the endpoints named in the review title, barrier integrity and mitochondrial function, have not been measured after routine immunization in any study identified here.
Sections 3.4 to 3.7 address the third question, and their inclusion requires explanation, because none of them measures a gut-brain axis endpoint. They are reported as boundary conditions on the hypothesis rather than as evidence for or against it. Section 3.4 establishes that immune activation during development can alter neurodevelopment in animal models, which is the mechanism the hypothesis requires and without which it would not be worth examining. Section 3.5 bounds the magnitude of any such effect in humans: whatever a gut-brain axis pathway could contribute to neurodevelopmental outcomes must fit within the confidence intervals reported by cohorts exceeding one million children, so this literature sets an upper limit rather than testing the pathway. Section 3.6 and Section 3.7 supply the confirmed adverse outcomes whose mechanism any gut-brain axis hypothesis would have to explain, together with the adjuvant and preservative literature in which the causal claims in this field are usually located. Table 12 records how far each confirmed adverse event is gut-brain axis relevant. A reader who wants the direct evidence alone will find it in Section 3.2 and Section 3.3.

3.1. Study Selection

The database searches retrieved 3,047 records. After removal of 1,338 duplicates, 1,709 titles and abstracts were screened. A total of 1,576 records were excluded at the title and abstract stage for not meeting eligibility criteria. Full texts of 133 studies were retrieved and reviewed, of which 73 were excluded for not meeting the eligibility criteria and 60 were incorporated into the synthesis. A targeted supplementary search to 27 September 2026, described in Section 2.2, identified a further 7 eligible studies, giving 67 included studies in total, listed with their design and appraisal instrument in Appendix A. Publications cited in this report for context but not counted among the included studies, because the eligibility criteria exclude narrative reviews and opinion pieces without primary data, are Nirenberg et al. (2026), Moser and Offit (2026), Garg and Paliwal (2022) and Praharaj et al. (2015); also excluded from the count are Mitkus et al. (2011), a pharmacokinetic modelling study rather than a study of humans or animals, and Li et al. (2017), which has been retracted. Because many studies address more than one domain, individual records are assigned to multiple categories and the category counts therefore exceed the number of unique included studies. The included studies comprise microbiome and vaccine-response investigations, cohort and case-control analyses of vaccines and neurodevelopmental outcomes, active surveillance and registry-based adverse event studies, analyses drawing on passive surveillance, animal studies of adjuvant and preservative exposure, pharmacokinetic modelling, case series, and meta-analyses or systematic reviews. Methodological instruments, background mechanistic literature without vaccine exposure, narrative reviews, hypothesis papers, and regulatory or advisory documents are cited throughout for context but are not counted as included studies.
Table 1. PRISMA Flow Summary.
Table 1. PRISMA Flow Summary.
Stage Number
Records identified through database searching 3,047
Duplicates removed 1,338
Titles and abstracts screened 1,709
Excluded at title/abstract stage 1,576
Full texts retrieved and reviewed 133
Excluded at full-text stage 73
Studies included from primary search 60
Eligible studies added by supplementary search to 27 September 2026 7
Total studies included in synthesis 67

3.2. Question 1. Does the Gut-Brain Axis Modify the Response to Vaccination?

Microbiome Effects on Vaccine Responses

The most consistent evidence concerned microbiome modulation of vaccine immunogenicity. In rural Ghana, a nested case-control study of 78 infants found that pre-vaccination microbiome composition differed significantly between rotavirus vaccine responders and nonresponders, with response correlating with greater abundance of Streptococcus bovis and lower abundance of the Bacteroidetes phylum (Harris et al., 2017). In a prospective Bangladeshi birth cohort (291 infants with microbiota data, 249 followed to 2 years), Bifidobacterium abundance at 6 to 15 weeks was positively associated with CD4 T-cell responses to BCG, tetanus toxoid, and hepatitis B vaccines at 15 weeks, and with CD4 responses to BCG and tetanus toxoid, tetanus-specific IgG, and polio-specific stool IgA at 2 years (Huda et al., 2019). Praharaj et al. (2015) reviewed the evidence on whether probiotic and antibiotic exposures modify vaccine responses. In adults with low pre-existing influenza immunity, randomized antibiotic exposure around vaccination impaired H1N1-specific neutralizing and IgG1 responses, although the effect was not seen in participants with substantial pre-existing immunity (Hagan et al., 2019). Ryan et al. (2025) further demonstrated in a prospective design combining human cohorts with gnotobiotic mouse models that neonatal antibiotic exposure within the first week of life reduced antibody titres against polysaccharides in the 13-valent pneumococcal conjugate vaccine and against the Haemophilus influenzae type b polyribosylribitol phosphate and diphtheria toxoid antigens at 7 months of age, in a cohort of 191 healthy term infants followed to 15 months.
Randomized probiotic trials reinforced these associations. A meta-analysis of 20 randomized trials in 1,979 adults, of which nine trials with 623 participants contributed pooled data, found that probiotic or prebiotic supplementation improved seroprotection rates after influenza vaccination, with odds ratios of 1.83 for the H1N1 strain and 2.85 for the H3N2 strain, and 2.11 for B strain seroconversion (Lei et al., 2017). A separate systematic review of 26 randomized placebo-controlled studies in 3,812 participants, covering 40 probiotic strains and 17 vaccines, reported a beneficial effect in about half the studies, strongest for oral vaccines and parenteral influenza vaccination, and calculated no pooled estimate because of heterogeneity in strain, dose, and timing (Zimmermann & Curtis, 2018). In mice, germ-free conditions impaired antibody responses to vaccination (Oh et al., 2014), and short-chain fatty acids were shown separately to fuel host antibody responses in non-vaccine settings (Kim et al., 2016).

Do Vaccines Alter the Microbiome?

Evidence that vaccines alter the microbiome was weaker. Boston et al. (2024) found no significant changes in alpha diversity across vaccination timepoints. Ardura-Garcia et al. (2024) noted that vaccines may generate changes in intestinal microbiota composition, though evidence remains limited. Pediatric evidence is scarce. Murine studies show mixed results, with some suggesting transient taxonomic changes after aluminum-adjuvanted vaccines, though without lasting dysbiosis (Borgognone et al., 2022; Khadka et al., 2023). Overall, the certainty is moderate that vaccines do not alter adult microbiomes, but lower for infants and young children.

3.3. Question 2. Does Vaccination Alter the Gut-Brain Axis?

Two of the pre-specified gut-brain axis outcome domains, barrier integrity (intestinal and blood-brain) and mitochondrial function, were only sparsely addressed by eligible studies. Although barrier dysregulation, for example altered tight-junction proteins and elevated circulating zonulin, is well described in inflammatory and neurodevelopmental conditions, few studies directly measured barrier or mitochondrial endpoints in relation to vaccination itself, and those that did were small, heterogeneous, or purely mechanistic. No eligible study demonstrated a durable vaccine-induced change in intestinal permeability, blood-brain barrier integrity, or mitochondrial function. This constitutes an evidence gap rather than evidence of either harm or safety, and it is a priority for future mechanistic and longitudinal research.
The certainty of evidence across the principal outcome domains, assessed with the GRADE framework, is summarized in Table 3. Table 9 reports the same domains under a bias-adjusted appraisal and uses the identical ratings as its starting point; no domain is rated differently between the two tables under standard appraisal.
Table 3. GRADE Certainty of Evidence by Outcome Domain and Direction of Effect.
Table 3. GRADE Certainty of Evidence by Outcome Domain and Direction of Effect.
Outcome Domain Certainty Basis
Direction A (microbiome to vaccine): influence on vaccine immunogenicity Moderate Downgraded from high for inconsistency: infant cohorts conflict in direction at taxon level and probiotic trials reported benefit in about half of 26 studies with no pooled estimate calculable. The direction of the diversity association is consistent and mechanistic support is strong
Direction B (vaccine to gut-brain axis): alteration of the adult microbiome Low One longitudinal adult cohort of COVID-19 mRNA vaccination reporting compositional stability. Downgraded for imprecision, being a single study, and for indirectness, since oral live vaccines such as rotavirus and polio are the most plausible gut-active exposures and were not examined
Direction B (vaccine to gut-brain axis): alteration of the infant or child microbiome Low Sparse, small, mostly cross-sectional pediatric studies
Maternal infection or fever and offspring autism risk Low The largest registry analysis found no overall association, and its positive trimester-specific estimates arose from 33 comparisons without correction; no upgrade factor applies
Maternal vaccination (influenza, Tdap) and autism Moderate Upgraded from low on the strength of consistent null estimates across cohorts of more than one million children; no further upgrade applied, and residual confounding is the reason the rating is not higher
Childhood vaccination and autism spectrum disorder Moderate Upgraded from low on effect size and precision across multiple national cohorts and pooled analyses; residual confounding, including the healthy-vaccinee effect, is the reason no further upgrade is applied
Quantified adverse events (febrile seizures, intussusception, ITP, myocarditis, anaphylaxis) Moderate to high Active surveillance with defined denominators
Barrier integrity and mitochondrial endpoints Very low Few or no direct studies (evidence gap)
Passive-surveillance signals (e.g., Guillain-Barre syndrome, encephalitis) Very low No denominators; underreporting and stimulated reporting
The instruments named in Section 2.6 are applied below at study level, across both the positive-association and null literatures, so that the appraisal can be checked against a common rubric rather than inferred from the narrative. Domains are collapsed to the three that discriminate most in this evidence base. Ratings follow the conventions of each instrument: low, moderate, serious and critical for ROBINS-I; low, some concerns and high for RoB 2; low, unclear and high for SYRCLE. Passive surveillance analyses are not rated, since the instruments do not apply to them, and they are classified as signals throughout.
Table 4. Study-Level Risk of Bias Across Both Literatures.
Table 4. Study-Level Risk of Bias Across Both Literatures.
Study Instrument Confounding or Randomisation Exposure or Outcome Measurement Selective Reporting Overall Judgement
Madsen et al. (2002) ROBINS-I Moderate Moderate Low Moderate risk. Registry exposure and outcome; confounders adjusted; no chart validation beyond 40 cases
Hviid et al. (2019) ROBINS-I Low Moderate Low Moderate risk. Disease risk score adjustment; no individual chart review stated
Jain et al. (2015) ROBINS-I Moderate Serious Low Serious risk. Claims-based exposure with acknowledged 4 to 14 percent undercount of vaccination
DeStefano et al. (2013) ROBINS-I Moderate Low Low Serious risk. Participation 48 percent of cases and 32 percent of controls
Andersson et al. (2025) ROBINS-I Low Moderate Low Moderate risk. Exposure modelled from schedule records; sibling analyses diverge from main model
Verstraeten et al. (2003) ROBINS-I Moderate Moderate Moderate Serious risk. Site-specific findings not replicated; published erratum
Barlow et al. (2001) ROBINS-I Low Moderate Low Moderate risk. Ascertainment differed across sites; automated vaccination data only
Taylor et al. (2014) ROBINS-I (pooled) Moderate Moderate Moderate Moderate risk. Overlapping Danish cohorts across component studies
Wong et al. (2023) RoB 2 Low High Some concerns High risk on measurement. Self-reported outcomes; patent-holding investigators
Lal et al. (2015) RoB 2 Low Low Low Low risk
Cooper et al. (2005) RoB 2 Low Low Some concerns Some concerns. Small trial, surrogate endpoint
Hagan et al. (2019) RoB 2 Low Low Some concerns Some concerns. Phase 2 arm of 5 treated participants
DeLong (2011) ROBINS-I Critical Serious Low Critical risk. Ecological design; no individual linkage
Tomljenovic and Shaw (2011) ROBINS-I Critical Serious Low Critical risk. Ecological design; no individual linkage
Madsen et al. (2003) ROBINS-I Critical Serious Low Critical risk. Ecological time trend; ascertainment changed over the period
Geier and Geier (2007a) Case series tool Not applicable Serious Unclear Critical risk. No comparator or denominator
Geier and Geier (2007b) ROBINS-I Critical Unclear Unclear Not rated. The primary report was not obtained, and no risk-of-bias rating is assigned on a secondary description
Geier et al. (2013) ROBINS-I Serious Serious Unclear Serious to critical risk. Phase 1 passive surveillance without denominator; phase 2 unreplicated
Singh et al. (2002) ROBINS-I Moderate Moderate Unclear Moderate risk on design; unreplicated for over two decades
Holmes et al. (2003) ROBINS-I Serious Serious Unclear Critical risk. Control selection and unfalsifiable auxiliary hypothesis
Li et al. (2014) SYRCLE Unclear Low Low Unclear risk. Randomisation and blinding not reported; dose 20 times human equivalent
Crépeaux et al. (2017) SYRCLE Unclear Low Unclear Unclear risk. Allocation and blinding not reported
Khadka et al. (2023) SYRCLE Unclear Low Low Unclear risk. Allocation concealment not reported
Oh et al. (2014) SYRCLE Low Low Low Low risk
Two patterns are worth noting. The ecological analyses receive a critical rating irrespective of the direction of their findings, which places the Danish thimerosal-removal trend in the same category as the ecological analyses that report positive associations. The null cohorts are not uniformly low risk: exposure misclassification in claims data and participation shortfalls in case-control recruitment produce serious ratings in two of them. Neither observation changes the overall weight of evidence, and both are recorded so that the basis of the weighting is visible.

3.4. Boundary Condition. Immune Activation and Neurodevelopment in Animal Models

Rodent studies established that strong maternal immune activation can induce offspring neurodevelopmental abnormalities. Administration of polyinosinic:polycytidylic acid triggered stereotyped behaviors and cortical defects in offspring via IL-6 signaling (Smith et al., 2007). Blocking IL-17A in similar models normalized cortical lamination and behaviors (Choi et al., 2016). Microbiota-dependent production of IL-17A has also been implicated (Kim et al., 2017).
In humans, maternal infection and fever have been associated with higher autism risk. A Danish registry cohort of 1,612,342 children found no association between any maternal infection requiring hospitalization during pregnancy and autism spectrum disorder overall (adjusted hazard ratio 1.14; 95% CI 0.96 to 1.34), but reported elevated risk for maternal viral infection in the first trimester (adjusted hazard ratio 2.98; 95% CI 1.24 to 7.15) and bacterial infection in the second trimester (adjusted hazard ratio 1.42; 95% CI 1.08 to 1.87), The analysis made 33 comparisons without adjustment for multiple testing, so the two positive estimates are compatible with chance at the stated alpha (Atladóttir et al., 2010). A Norwegian birth cohort of 95,754 children found that maternal fever in the second trimester was associated with increased autism risk (adjusted odds ratio 1.40; 95% CI 1.09 to 1.79), rising to 3.12 (95% CI 1.28 to 7.63) with three or more fever episodes after 12 weeks of gestation; risk was only minimally reduced among women who took acetaminophen for fever (Hornig et al., 2018). In contrast, maternal influenza or Tdap vaccination showed no increased autism risk in large U.S. studies (Zerbo et al., 2017; Becerra-Culqui, 2018). Certainty is therefore moderate that infection increases risk but high that vaccination does not.

3.5. Boundary Condition. Magnitude Limits From Childhood Vaccine and Autism Epidemiology

Large-scale studies consistently show no link between vaccines and autism. A Danish cohort of 537,303 children reported no increased autism risk after MMR vaccination (Madsen et al., 2002). A U.S. case-control study of 256 children with autism spectrum disorder and 752 matched controls found no association between cumulative vaccine antigen exposure in the first two years of life and autism (DeStefano et al., 2013). A meta-analysis pooling five cohort studies of 1,256,407 children and five case-control studies of 9,920 children reported no relationship (Taylor et al., 2014). A later Danish study of 657,461 children confirmed these findings (Hviid et al., 2019). Thimerosal exposure has been examined in four study populations. A two-phase U.S. study of 124,170 infants at two health maintenance organizations, with reevaluation among 16,717 children at a third, found no significant increased risk of autism or attention-deficit disorder in any analysis, although it reported associations with tics at one site (relative risk 1.89; 95% CI 1.05 to 3.38) and language delay at another (relative risk 1.13; 95% CI 1.01 to 1.27) that did not replicate at the third site; the result is a consistent null for autism accompanied by site-specific signals that did not replicate, which is weaker than a clean null (Verstraeten et al., 2003). A U.K. retrospective cohort of 109,863 children found no evidence that thimerosal exposure via DTP or DT vaccines causes neurodevelopmental disorders, with the possible exception of tics (hazard ratio 1.50 per dose at 4 months; 95% CI 1.02 to 2.20), alongside several statistically significant negative associations with increasing exposure (Andrews et al., 2004). A separate U.K. prospective cohort of more than 14,000 children examined cognitive and behavioural outcomes rather than autism and found no convincing evidence of a deleterious effect, with 1 of 69 adjusted results in the hypothesized direction (Heron & Golding, 2004). Autism prevalence in California continued to rise among children aged 3 to 5 years through March 2007 despite the exclusion of more than trace thimerosal from nearly all childhood vaccines (Schechter & Grether, 2008). Other vaccines show no associations. A Vaccine Safety Datalink cohort of 679,942 children found that whole-cell pertussis vaccine raised the risk of febrile seizures on the day of vaccination (relative risk 5.70; 95% CI 1.98 to 16.42) but was not associated with nonfebrile seizures, and that children with post-vaccination febrile seizures were at no higher risk of subsequent seizures or neurodevelopmental disability (Barlow et al., 2001). A study of 95,727 children, including siblings of autistic children, found no increased autism risk in fully vaccinated children (Jain et al., 2015).
The most recent and comprehensive evidence further supports these conclusions. Andersson et al. (2025) published a nationwide Danish cohort study of 1,224,176 children born between 1997 and 2018, finding no evidence of increased risk for 50 chronic disorders, including autism spectrum disorder (adjusted HR = 0.93) and ADHD (adjusted HR = 0.90), associated with cumulative aluminum exposure from childhood vaccines. At its meeting on 27 November 2025, in a statement published in December 2025, the WHO Global Advisory Committee on Vaccine Safety (GACVS) assessed two new systematic reviews covering 31 primary research studies published from 2010 to August 2025 from multiple countries and reaffirmed that no causal link exists between vaccines and autism spectrum disorder.

3.6. Boundary Condition. Confirmed Adverse Events With Quantifiable Incidence

Several adverse events have been confirmed and quantified through active surveillance systems that carry defined denominators, such as the Vaccine Safety Datalink and claims-based cohorts. The incidence figures below, and the term rare as used in this review, refer to these active-surveillance estimates. Figures that can be derived only from passive spontaneous reports are treated separately as lower bounds and are not described as rare.
Table 2. Quantifiable Adverse Events Following Vaccination.
Table 2. Quantifiable Adverse Events Following Vaccination.
Adverse Event Vaccine Incidence Source
Febrile seizures Measles-containing 25-34 additional per 100,000 children vaccinated Barlow et al., 2001
Intussusception Rotavirus 1.1-1.5 per 100,000 first-dose recipients Patel et al., 2011; Yih et al., 2014
Immune thrombocytopenic purpura MMR 1 per 22,300-25,000 doses Miller et al., 2001; Black et al., 2003
Anaphylaxis All vaccines 1.31 per million doses (95% CI 0.90-1.84) McNeil et al., 2016
Myocarditis (original series; passive reporting rate, not incidence) mRNA COVID-19 52.4-105.9 reports per million 2nd doses (males 12-24) Oster et al., 2022
Myocarditis (2023-2024 formula) mRNA COVID-19 ≈8 per million doses overall; ≈27 per million (males 12-24) FDA, 2025
Measles-containing vaccine is estimated to cause 25 to 34 additional febrile seizures per 100,000 children vaccinated, and whole-cell pertussis vaccine 6 to 9 per 100,000 (Barlow et al., 2001). MMRV increases risk compared with separate MMR and varicella vaccination, with a relative risk of 1.98 (95% CI 1.43 to 2.73) for seizure during days 7 to 10 and an excess of 4.3 febrile seizures per 10,000 doses (95% CI 2.6 to 5.6), equivalent to one additional febrile seizure for every 2,300 MMRV doses given instead of separate vaccines (Klein et al., 2010). Immune thrombocytopenic purpura is estimated at 1 per 22,300 to 25,000 MMR doses within six weeks of vaccination (Miller et al., 2001; Black et al., 2003). Anaphylaxis is rare across vaccines, with 33 confirmed vaccine-triggered cases after 25,173,965 doses in the Vaccine Safety Datalink, a rate of 1.31 per million doses (95% CI 0.90 to 1.84) (McNeil et al., 2016). Intussusception attributable to rotavirus vaccination was estimated at 1.1 to 1.5 excess cases per 100,000 recipients of the first dose of the pentavalent vaccine in U.S. infants (Yih et al., 2014), and at approximately 1 per 51,000 infants in Mexico and 1 per 68,000 in Brazil after the monovalent vaccine, against roughly 80,000 diarrhea hospitalizations and 1,300 deaths prevented each year in those countries (Patel et al., 2011).
Myocarditis following mRNA COVID-19 vaccines remains a confirmed adverse event. Passive surveillance reporting rates after the second dose were highest in adolescent males aged 16 to 17 years (105.9 per million BNT162b2 doses), followed by males aged 12 to 15 years (70.7 per million) and men aged 18 to 24 years (52.4 per million BNT162b2 doses and 56.3 per million mRNA-1273 doses) (Oster et al., 2022). Updated FDA-mandated labeling (June 2025) based on insurance claims data for the 2023-2024 formulations reports approximately 8 per million doses in individuals aged 6 months through 64 years, and approximately 27 per million doses in males aged 12 through 24. A Stanford University study (Cao et al., 2025) identified CXCL10 and IFN-γ as likely molecular drivers of vaccine-associated myocarditis. At IDWeek 2025, Moderna presented data from a phase 4 randomized crossover study of 997 participants aged 12-30 showing no cases of myocarditis or pericarditis with the updated 2024-2025 formulation, though the sample size was acknowledged as potentially too small to detect this rare event (Pfeiffer et al., 2026). A systematic review and meta-analysis found that the risk of myocarditis after COVID-19 infection is approximately 42 times higher than after vaccination (Choi et al., 2025). The comparison sets cases per million infections against cases per million doses and is restricted to adolescents receiving BNT162b2, so it orders the two risks rather than expressing a like-for-like ratio. Most cases of vaccine-associated myocarditis have favorable outcomes, with a French national cohort study reporting lower cardiovascular complications at 18 months compared with conventional myocarditis (Semenzato et al., 2024).
A meta-analysis of six adverse event monitoring systems covering about 23 million vaccinated people found that influenza A (H1N1) 2009 monovalent inactivated vaccines were associated with a small increased risk of Guillain-Barré syndrome (incidence rate ratio 2.35; 95% CI 1.42 to 4.01), corresponding to approximately 1.6 excess cases per million people vaccinated (Salmon et al., 2013). A case-centered analysis of nearly 64 million vaccine doses identified only 7 cases of transverse myelitis and 8 of acute disseminated encephalomyelitis during the primary risk interval and found no association between vaccination and either outcome, concluding that any excess risk of acute disseminated encephalomyelitis is unlikely to exceed 1.16 cases per million doses of Tdap (Baxter et al., 2016).

3.7. Boundary Condition. Adjuvants and Mechanistic Safety

Aluminum adjuvants persist in phagocytes and can migrate in animal models (Crépeaux et al., 2017), yet large human cohort studies show no link to autism or neurodevelopmental disorders (Hviid et al., 2003). Pharmacokinetic modeling indicates that aluminum exposure from vaccines remains within safety limits (Mitkus et al., 2011). The 2025 Andersson et al. study provides the most comprehensive data to date, examining 1,224,176 children and finding no increased risk for any of 50 chronic conditions associated with cumulative aluminum vaccine exposure.
In an integrated analysis of more than 1,700 children and adolescents, MF59-adjuvanted influenza vaccines produced more local and systemic reactions than non-adjuvanted vaccines but no increase in serious adverse events (Black et al., 2010). However, AS03-adjuvanted H1N1 vaccines were linked to narcolepsy in Nordic populations (Partinen et al., 2012). This illustrates that adjuvants can occasionally be associated with rare, subgroup-specific outcomes. Newer adjuvants, such as Matrix-M and CpG motifs, show acceptable safety profiles in early trials, though pediatric exposure remains limited (Lal et al., 2015; Cooper et al., 2005).

4. Discussion

This systematic review synthesized evidence on vaccines and their potential interactions with gut-brain axis biology, encompassing the microbiome, cytokine signaling, barrier integrity, and mitochondrial indices, as well as neurodevelopmental outcomes and adverse events. The findings indicate that vaccines engage immune and inflammatory pathways that intersect with gut-brain axis biology. However, durable alterations or widespread neurodevelopmental harms have not been supported by high-certainty evidence. The evidence further shows that vaccines are associated with specific adverse events that can be quantified through active surveillance, whereas associations with autism or persistent microbiome disruption have not been detected at the population level.
The review also emphasizes the importance of distinguishing between signal detection and incidence estimation. Active surveillance systems allow precise quantification of events such as febrile seizures, intussusception, or myocarditis. In contrast, passive reporting systems such as VAERS or EudraVigilance can detect unusual patterns but cannot provide incidence estimates due to underreporting, stimulated reporting, and lack of denominators (Hazell & Shakir, 2006).
Microbiome-Vaccine Interactions
The most consistent evidence relates to the influence of the microbiome on vaccine immunogenicity. Infant cohorts in South Asia and Africa link early microbiome composition to vaccine responses: higher Bifidobacterium abundance predicted stronger responses in Bangladesh (Huda et al., 2019), and microbiome composition differed between rotavirus vaccine responders and nonresponders in Ghana (Harris et al., 2017). Randomized antibiotic exposure impaired influenza vaccine responses in adults with low pre-existing immunity, though not in those already primed (Hagan et al., 2019). The 2025 prospective study by Ryan et al. extends these findings by reporting reduced antibody titres at 7 months of age following neonatal antibiotic exposure, in a cohort of 191 infants followed to 15 months, with Bifidobacterium abundance at the time of vaccination correlating with those titres six months later. These findings have implications for public health, suggesting that optimizing gut microbial health may enhance vaccine effectiveness.
In contrast, current evidence has not shown that vaccines cause lasting microbiome disruption. A longitudinal adult cohort showed no significant change in gut microbiome composition after COVID-19 vaccination (Boston et al., 2024). Pediatric data are lacking. Overall, the weight of evidence indicates that vaccination interacts with but does not destabilize microbiome ecosystems.
Maternal Immune Activation
Maternal immune activation models provide mechanistic plausibility for immune-neurodevelopmental links. Rodent experiments demonstrate that viral mimetics and bacterial mimetics can trigger interleukin-6 and interleukin-17A pathways that alter cortical development and produce autism-like behaviors (Smith et al., 2007; Choi et al., 2016). However, translation to vaccination contexts requires caution, as the intensity of immune activation in these models exceeds the transient cytokine elevations observed following routine immunization. Human studies associate maternal fever with increased autism risk in one prospective cohort, while a registry analysis of maternal infection found no overall association, with positive estimates confined to two of 33 uncorrected trimester-specific comparisons (Atladóttir et al., 2010; Hornig et al., 2018), yet maternal influenza and Tdap vaccination show no such association (Zerbo et al., 2017; Becerra-Culqui et al., 2018). Thus, while maternal immune activation can influence neurodevelopment, controlled vaccination during pregnancy does not replicate these risks.
Vaccines and Autism
Extensive epidemiological evidence rules out large associations between vaccines and autism spectrum disorder. Large national cohorts and meta-analyses reported no link between MMR vaccination and autism (Madsen et al., 2002; Taylor et al., 2014; Hviid et al., 2019). Thimerosal shows no association with autism in these cohorts, with isolated and non-replicated signals for tics and language delay (Verstraeten et al., 2003; Andrews et al., 2004; Schechter & Grether, 2008). As with the measles, mumps and rubella literature, the exposure measured is not the exposure this review is concerned with. Evidence on other vaccines, including DTaP, hepatitis B, and varicella, is not supplied by the cohorts above, which measured measles, mumps and rubella vaccination, whole-cell pertussis vaccine, or cumulative antigen exposure; no study identified by this review examined those products in relation to neurodevelopmental outcomes.
The evidence base has been further strengthened by developments in 2025. The Andersson et al. nationwide Danish cohort study of over 1.2 million children found no association between cumulative aluminum vaccine exposure and 50 chronic disorders, including autism and ADHD, with the upper bounds of confidence intervals for most outcomes incompatible with even moderate relative increases in risk. The WHO GACVS reaffirmation (assessed 27 November 2025; statement published December 2025), based on 31 primary research studies from multiple countries published since 2010, provides a current international expert consensus that no causal link between vaccines and autism has been established. A prospectively registered systematic review published in 2026 applied the same appraisal instruments used here, namely RoB 2, ROBINS-I, a modified ROBINS-I for case series, and GRADE, to 59 human studies of aluminium-adjuvanted vaccination comprising 37 case series, 11 randomised controlled trials, 9 cohort studies, and 2 ecological studies published to November 2025. Its outcome-level findings are more differentiated than a summary conclusion conveys. For autism, two ecological studies reporting positive correlations were rated at critical risk of bias and very low certainty, while the Danish cohort of 1,224,176 children showed no association and no dose-response at moderate certainty. For asthma, a United States cohort of 326,991 children reported adjusted hazard ratios of 1.19 to 1.26 in children with and without eczema, which lost significance in sensitivity analyses, against a null result in the Danish cohort. Macrophagic myofasciitis rested on 210 reports across case series with no control group in any of them, rated very low certainty. The one outcome with a reproducible association was persistent injection-site nodules and granulomas, uncommon at under one percent, local and self-limited, rated low to moderate certainty. The review declined meta-analysis on grounds of heterogeneity, noted that its conclusions rest on a minority of randomised trials and large cohorts because case series and ecological studies predominate, and observed that the nodule and macrophagic myofasciitis literatures cluster geographically in Scandinavia and France respectively, with limited independent replication (Doyon-Plourde et al., 2026). Two further 2026 papers reach compatible conclusions on aluminium pharmacokinetics and safety, though both are narrative rather than systematic and are weighted accordingly (Nirenberg et al., 2026; Moser & Offit, 2026). Across these studies, estimates for the exposures measured cluster at or slightly below the null, with confidence intervals that exclude the moderate or large effects that early formulations of the hypothesis proposed. They do not exclude small subgroup-specific effects, and they do not address exposures other than those measured.
Interpretation of this evidence base should also acknowledge ongoing scientific and regulatory debate. The Andersson et al. (2025) cohort, although large and widely cited, has drawn methodological commentary. One published critique argues that unvaccinated children were pooled with a low-exposure group, which is not an appropriate contrast for a dose-response analysis; a separate observation is that a majority of reported hazard ratios fall below 1.0, which may indicate residual bias. Both are recorded as raised rather than established. The study has not been corrected or retracted, and is weighted here on its design and precision. In parallel, the regulatory environment shifted during 2025 and 2026. In November 2025 the United States Centers for Disease Control and Prevention revised the wording of its vaccine-and-autism webpage to state that available studies have not definitively ruled out a contribution of infant vaccines to autism (Centers for Disease Control and Prevention, 2025; Stobbe, 2025), and an Advisory Committee on Immunization Practices working group began reviewing aluminum adjuvants (Advisory Committee on Immunization Practices, 2025; Dall, 2025). These developments reflect an evolving policy context rather than new high-quality evidence of harm, and they stand alongside the reaffirmed international scientific consensus that no causal link between vaccines and autism has been established. The present review reports the available evidence and these differing positions without adjudicating between them.
Adverse Events
Vaccines can cause specific adverse events, several of which are confirmed with quantifiable incidence. Febrile seizures, intussusception, immune thrombocytopenic purpura, anaphylaxis, and myocarditis are all established vaccine-associated events with known incidence ranges as detailed in Section 3.6. The myocarditis data have evolved substantially since early reports; updated 2023-2024 formulation data suggest lower rates than the original mRNA COVID-19 vaccine series, and mechanistic understanding has advanced with the identification of CXCL10 and IFN-γ as likely drivers (Cao et al., 2025). The favorable prognosis of most vaccine-associated myocarditis cases, compared with infection-associated or conventional myocarditis, provides important context for risk-benefit communication.
Adjuvants
Aluminum adjuvants remain widely used, and animal studies show persistence and migration of aluminum particles (Crépeaux et al., 2017). Pharmacokinetic modeling demonstrates that cumulative infant exposure from vaccines remains below safety thresholds (Mitkus et al., 2011). The 2025 Andersson et al. study represents the most comprehensive population-level evaluation to date and found no associations with chronic disease outcomes. Squalene-based adjuvants such as MF59 increased local and systemic reactogenicity without increasing serious adverse events in children and adolescents (Black et al., 2010), while the AS03-narcolepsy association in Nordic countries remains an important example of rare, subgroup-specific adverse outcomes (Partinen et al., 2012).
Taken together, the mechanistic and epidemiological findings summarized above have prompted the present authors to consider whether a vaccine-modulated gut-brain axis inflammatory syndrome might serve as an interpretive construct. The construct originates with the authors of this review and has no independent presence in the indexed literature; it is noted here as hypothesis-generating and is not adopted, defined, or treated as an outcome anywhere in this review. No study identified in this review demonstrated such a syndrome, and the available high-certainty evidence does not support the existence of a distinct vaccine-induced inflammatory syndrome affecting the gut-brain axis. The term is used here only to organize biologically plausible pathways for future testing, and it should not be interpreted as a validated diagnosis or as evidence of population-level harm.

Quantitative Bias Analysis

An extended treatment of mortality misattribution and of the number of deaths expected by chance after vaccination from background mortality alone is given in Appendix B, together with the designs that address it. It is placed there because it concerns the interpretation of mortality surveillance rather than the gut-brain axis questions this review examines.
Because most evidence on vaccines and neurodevelopment is observational, the certainty of any near-null finding depends on how strongly unmeasured bias could push estimates toward or below the null. Three sources of directional bias are most relevant. First, the healthy-vaccinee effect and confounding by indication: children with emerging developmental concerns may have vaccination deferred, while healthier children are vaccinated on schedule, so the vaccinated group can appear lower-risk for reasons unrelated to the vaccine. Second, underreporting in passive surveillance: spontaneous-reporting systems such as VAERS and EudraVigilance capture only a fraction of adverse events, so passive-surveillance incidence figures are lower bounds rather than estimates. Third, publication and selective-reporting bias: if studies suggesting an association are less likely to be published, cited, or funded than null studies, the visible literature would be skewed toward the null. The analysis below quantifies the potential impact of each source, without assuming a particular cause, so that readers can judge how much bias would be required to alter the conclusions.
Confounding sensitivity was assessed with the E-value framework (VanderWeele & Ding, 2017). For a representative adjusted estimate near the null (illustratively an adjusted hazard ratio of 0.93), Table 5 reports the bias factor and the strength of unmeasured confounding, expressed as a risk ratio with both vaccination and outcome, that would be required to reconcile the observed value with progressively larger true harmful effects. A true harm as large as an aHR of about 1.10 could be produced by an unmeasured confounder associated with both vaccination and outcome by a risk ratio near 1.65, which is within the plausible range of a healthy-vaccinee mechanism. Reconciling the same data with a true harm of 1.25 or more would require confounding of risk ratio 2.0 or greater acting consistently across studies, which is less plausible but not impossible. The near-null result is therefore robust against attributing large effects to confounding, but it cannot exclude small-to-modest true effects.
Table 5. Unmeasured Confounding Required to Reconcile an Observed Near-Null Estimate (aHR 0.93) with a True Harmful Effect.
Table 5. Unmeasured Confounding Required to Reconcile an Observed Near-Null Estimate (aHR 0.93) with a True Harmful Effect.
Hypothesized True aHR Required Bias Factor Required Confounder Strength (RR)
1.00 1.08 1.36
1.10 1.18 1.65
1.25 1.34 2.02
1.50 1.61 2.61
2.00 2.15 3.72
Underreporting sensitivity is summarized in Table 6. This correction applies to passive-surveillance signals and to any adverse-event incidence derived from spontaneous reports, not to the registry-based association studies, which ascertain outcomes from complete diagnostic records rather than voluntary reports. If the reporting fraction is about 6 percent, the complement of the 94 percent median under-reporting rate found across 37 studies of adverse drug reactions (Hazell & Shakir, 2006), the true incidence of a passively reported event could be roughly seventeen times the observed figure, and higher still where reporting is worse. Active-surveillance systems with defined denominators, such as claims databases and the Vaccine Safety Datalink, are far less affected, so the confirmed adverse-event incidence ranges in Section 3.6 are relatively robust, whereas any event known only from passive reports should be read as a lower bound.

Dissemination and Reporting Bias: A Prespecified Assessment

Concerns have been raised that findings suggesting vaccine-associated harm may be less likely to be published, funded, or cited than null findings, and that the visible literature may therefore be skewed. This report treats that proposition as an empirical question rather than as an assumption, on the principle that a claim which explains a null result and a positive result equally well explains neither. Suppression, if it operates at a material scale, leaves detectable footprints: asymmetry between effect size and study precision, divergence between registered and reported outcomes, systematic differences by funding source, and a gap between the indexed and the unindexed literature. The analyses prespecified in Table 7 are designed to look for those footprints. They are stated with their limits declared, so that a negative result cannot be dismissed as inattention and a positive result cannot be dismissed as opportunism. Because this review was not registered, these analyses were specified during its conduct rather than before it, and the review does not describe them as prespecified. Four of the five are set out as protocols for work not undertaken here, and the status column records which is which.
Table 7. Analyses of Dissemination and Reporting Bias: Scope, Limits, and Status.
Table 7. Analyses of Dissemination and Reporting Bias: Scope, Limits, and Status.
Analysis What It Can Detect What It Cannot Establish Status in This Review
Funnel plot and Egger regression on pooled estimates Small-study effects: asymmetry between effect size and precision Intent; asymmetry has several causes besides suppression, and the tests have low power when studies are few Prespecified; requires the study-level dataset of effect estimates and standard errors
Comparison of published outcomes against pre-registered protocols (ClinicalTrials.gov, WHO ICTRP) Selective outcome reporting: outcomes registered but not reported, or primary outcomes switched after the fact Whether an unreported outcome was suppressed or simply never analyzed Prespecified; requires registry identifiers for each included interventional study
Stratification of effect estimates by funding source and author conflict-of-interest disclosure Sponsorship effects: systematic divergence between industry-funded and independently funded results Causation; funding correlates with other study features such as size and design Prespecified; requires funding and disclosure extraction for each included study
Grey-literature, preprint, dissertation, and non-English search Whether unpublished or less visible findings differ systematically from the indexed literature That no suppressed study exists; only that none was found by these channels Prespecified; requires extension of the search to these sources
Fail-safe N (Orwin) and precision-weighted equivalent The volume of contrary unpublished evidence required to overturn the pooled estimate That such evidence does or does not exist; it quantifies only what would be needed Computed below from an illustrative estimate; no meta-analysis was performed
Four of these five analyses require a study-level dataset that this review has not yet compiled, namely effect estimates with standard errors, registry identifiers, and funding and disclosure fields for each included study. They are therefore reported here as a prespecified protocol rather than as completed results, and the corresponding cells are marked accordingly. This is stated plainly because presenting uncomputed analyses as though they had been performed would itself be a reporting bias, of exactly the kind the section exists to detect.
Regarding the first analysis, funnel-plot asymmetry and its formal tests must be interpreted with care. Asymmetry is not equivalent to publication bias: it can arise from genuine between-study heterogeneity, from differences in study quality or intensity of exposure, from true effect modification by study size, and from chance, and the tests have low power when the number of studies is small or their sizes are similar (Egger et al., 1997; Sterne et al., 2011). A symmetric funnel therefore does not demonstrate that suppression is absent, and an asymmetric one does not demonstrate that it is present. Both directions of overinterpretation are avoided here.
Regarding the second, comparison of published articles against their registered protocols is the most direct available test of selective outcome reporting, and empirical work has demonstrated that such reporting occurs across the medical literature, with outcomes that reach statistical significance more likely to be fully reported (Chan et al., 2004). Applying this to the included interventional studies would quantify, rather than assert, the extent of any outcome switching in this field. Regarding the third, systematic evidence establishes that industry-sponsored studies more often report results favorable to the sponsor than independently funded studies (Lundh et al., 2017), which makes stratification by funding source and disclosure a necessary sensitivity analysis rather than an insinuation. Regarding the fourth, extending the search to grey literature, preprints, dissertations, and non-English sources tests directly whether the indexed literature is unrepresentative, and its failure to find divergent unpublished findings would be informative without being conclusive.
The fifth analysis can be computed now, and it is the one that most directly addresses the scale of suppression that would be required to change the conclusions. No meta-analysis was performed in this review, so no pooled estimate exists. The calculation below therefore takes an adjusted hazard ratio of 0.93 as an illustrative value, chosen because it is close to the estimates reported by the largest individual cohorts, and Table 8 reports two quantities conditional on it. The first is the conventional count-based fail-safe N (Orwin, 1983), which asks how many suppressed studies of a given effect size would be required to shift the pooled estimate to a specified threshold. The second, and the more meaningful in this setting, is the precision-weighted equivalent, which asks what share of the total statistical weight of the published evidence base the suppressed studies would have to carry. The distinction matters because meta-analytic estimates are weighted by precision, and the published cohorts here are exceptionally large, so a count of studies is a misleading measure of the evidence they would have to outweigh.
The results cut in both directions and are reported as such. To move the pooled estimate merely to the null, from 0.93 to 1.00, the suppressed literature would need to carry roughly 18 percent of the statistical weight of the published evidence if each hidden study showed a hazard ratio of 1.50. Against a published base on the order of three million children, that corresponds to an unpublished literature of roughly half a million children studied and never reported. Two qualifications apply to that base. The Danish cohorts overlap substantially, so the count of distinct children is smaller than the sum of the published cohort sizes, and statistical weight depends on event counts rather than on the number of children enrolled. The figure should therefore be read as an order of magnitude rather than an estimate. That is a large quantity of hidden evidence, but it is not an inconceivable one, and this report does not claim otherwise. To move the pooled estimate to a modest harm of 1.10, however, the suppressed literature would need to carry approximately 54 percent of the weight of the entire published evidence base, corresponding on the same assumptions to something on the order of 1.6 million children studied and never reported. To reach 1.25 it would need to exceed the published base outright. A hidden literature of that magnitude, sustained across many countries, funders, and independent research groups over three decades, is a substantially stronger claim than the underreporting documented in passive surveillance, and it is not supported by any evidence identified in this review.
Two boundaries on this analysis are stated explicitly. First, the fail-safe calculation is conditional on its assumptions: on the illustrative estimate used, on the effect size posited in the hidden studies, and on the size of the published base. Different assumptions yield different thresholds, and the table is provided so that readers may substitute their own rather than accept these. Orwin himself noted that the statistic lacks an underlying statistical model and functions as a heuristic device rather than an inferential one, and the calculation here should be read in that spirit (Orwin, 1983). The precision-weighted column is an extension beyond the original statistic rather than part of it, introduced here because the count-based form weights all studies equally while the published cohorts in this literature differ by orders of magnitude in size. Second, and more fundamentally, none of these analyses can establish intent, and the absence of a detectable footprint does not prove that no study was ever suppressed. What they establish is narrower and more defensible: the magnitude of suppression that would be required to overturn each conclusion, and whether the visible literature bears the marks that large-scale suppression would be expected to leave. Reporting that honestly, in both directions, is the only treatment of this question that a reader of any prior conviction can check.
Publication and selective-reporting bias can be screened with funnel-plot asymmetry and Egger's regression test (Egger et al., 1997), but these methods have limited statistical power and can be confounded by genuine between-study heterogeneity: a symmetric funnel does not prove the absence of suppression, and an asymmetric one does not prove its presence. Concerns about publication, reporting, and citation bias in vaccine-safety research have been raised, and the direction and magnitude of any such bias in this field are contested and cannot be resolved from the association estimates alone. Prospective protocol registration, access to full participant-level datasets, and inclusion of grey literature reduce this risk and are recommended for future syntheses.
Table 9 translates these considerations into certainty ratings. Under standard appraisal, the observational vaccine-neurodevelopment evidence is best rated moderate rather than high, because GRADE begins observational evidence at low certainty and the upgrade for plausible confounding does not apply when the finding already sits at the null. Under the stronger assumption that most studies share a directional bias toward the null, these ratings fall further to low or very low. The calibrated conclusion is that the available evidence does not support large vaccine-attributable increases in neurodevelopmental risk, that small-to-modest effects cannot be excluded, and that resolving the residual uncertainty requires designs that difference out shared confounding, including sibling-comparison and other within-family designs, negative-control outcomes, and triangulation across data sources with differing biases.
Table 9. Certainty of Evidence Under Standard Versus Bias-Adjusted Appraisal.
Table 9. Certainty of Evidence Under Standard Versus Bias-Adjusted Appraisal.
Outcome Domain Certainty (Standard Appraisal) Certainty (Assuming Shared Bias Toward Null)
Childhood vaccination and autism spectrum disorder Moderate Low to very low
Maternal vaccination (influenza, Tdap) and autism Moderate Low
Quantified adverse events (active surveillance) Moderate to high Moderate
Passive-surveillance signals Very low Very low (lower bound)
Direction A: microbiome influence on vaccine immunogenicity Moderate Low

Microbiome, Adverse Events, and the Limits of Plausibility

The confirmed vaccine adverse events summarized in Section 3.6 are not, on current evidence, gut-brain axis disorders, and no study has shown that gut microbiome composition determines who develops them. A mechanistic rationale nonetheless exists and is worth stating explicitly as a hypothesis. The microbiome calibrates baseline immune tone, shifting the balance of effector and regulatory T cells and setting the level of tonic inflammatory signaling (Belkaid & Hand, 2014; Lynn et al., 2022). It is required for optimal antibody responses to some vaccines through innate sensing of microbial products, for example TLR5 sensing of flagellin (Oh et al., 2014), and the same innate priming could in principle amplify an inflammatory reaction to an adjuvant or a lipid nanoparticle. Dysbiosis that raises intestinal permeability increases circulating microbial products and low-grade systemic inflammation, which can affect the blood-brain barrier relevant to neurological events, while microbial metabolites such as short-chain fatty acids provide anti-inflammatory regulatory buffering. The most specific channel is molecular mimicry: the established precedent is Campylobacter jejuni, whose lipo-oligosaccharide mimics nerve gangliosides and triggers Guillain-Barre syndrome (Yuki et al., 2004), so a gut community harboring epitopes resembling cardiac, neural, or platelet antigens could in principle seed cross-reactive lymphocytes that a vaccine later expands. Each channel is biologically plausible, and none has been demonstrated for vaccine adverse events in people.
Distinct from the confirmed events is a chronic-symptom cluster reported by a subset of individuals after COVID-19 vaccination, variously termed post-vaccination syndrome or post-acute COVID-19 vaccination syndrome. It is not recognized by health authorities and is poorly characterized. Early immunophenotyping in a small, self-selected sample reported persistent circulating spike antigen in some participants, along with T-cell dysregulation and markers of Epstein-Barr virus reactivation (Bhattacharjee et al., 2025). These signatures are immunologic and antigenic rather than gut-brain axis findings, the samples are small and the causal connection to vaccination remains unestablished, and the investigators describe the condition as probably heterogeneous. A gut-brain contribution to symptom persistence, by analogy with long COVID and myalgic encephalomyelitis, is a reasonable hypothesis to test but has not been demonstrated.
Because these mechanisms are frequently described as biologically plausible, it is worth distinguishing plausibility from causation, since they are very different evidentiary bars (Table 7). Biological plausibility requires only a credible mechanism consistent with known biology, is often supported by analogy or animal data, and can be satisfied for an effect and for its opposite at the same time. Causal inference requires an observed association that survives attempts to explain it away by chance, bias, and confounding, together with correct temporal order, ideally a dose-response gradient, replication across independent designs, and, in the strongest case, a change in the outcome when the exposure is altered (Hill, 1965). Plausibility earns a hypothesis the right to be tested. It does not move it toward being true.
Table 10. Biological Plausibility Versus Causal Inference.
Table 10. Biological Plausibility Versus Causal Inference.
Dimension Biological Plausibility Causal Inference
Question asked Is there a credible mechanism? Does the exposure actually change the outcome?
Evidence that satisfies it Known physiology, analogy, in vitro or animal data Association robust to chance, bias, and confounding; correct temporality; dose-response; replication; experiment
Establishes that the effect exists? No Yes, when the criteria converge
Symmetric (supports an effect and its absence)? Yes No
Status for microbiome and adverse events Met Not established
The Hill viewpoints presuppose an association observed at the individual level whose causal status is then interrogated, and no such association has been measured for the hypothesis considered here. The table is therefore used to specify what would have to be demonstrated rather than to appraise causation, and the status column records how much of that remains undone. Hill offered these as viewpoints for judgment rather than a checklist, and only temporality is strictly necessary. Table 11 records the status of each for this hypothesis. Of the nine, only plausibility and analogy are met, and these are the two that carry the least causal weight and are the easiest to satisfy for a false hypothesis as well as a true one. Every consideration that does the substantive work of causal inference is unmet, and almost entirely because the direct human evidence has not yet been gathered rather than because it has been sought and found absent. This is absence of evidence, not evidence of absence, and it argues neither for nor against the hypothesis. The same caution applies to the mechanistic material summarised earlier in this section. Chains of plausibility, from the ketogenic diet and seizure threshold, from hypocretin neuron projections, or from vagal involvement in cardiac inflammation, satisfy only the plausibility viewpoint, which Table 10 records as symmetric and therefore as capable of supporting an effect and its absence equally. Each link in those chains is untested for the hypothesis at issue, and their accumulation does not convert plausibility into evidence.
A discussion of benefit-risk framing, including why the phrase safe and effective is imprecise as a matter of epidemiology and why a harms-only presentation is its mirror image, is given in Appendix C with the corresponding benefit-risk comparison. It is placed there for the same reason: it concerns how vaccine evidence is communicated rather than the questions this review examines.

Gut-Brain Axis Mechanisms of the Confirmed Adverse Events

The confirmed adverse events summarized in Section 3.6 are ordinarily described as organ-specific entities with distinct mechanisms, and they do not constitute a single syndrome. That description is accurate, but taken alone it understates the extent to which several of these events engage gut-brain axis biology, either in their pathogenesis or in their downstream consequences. Two general principles establish the connection. First, events producing systemic cytokine elevation engage pathways through which circulating cytokines act on the central nervous system, producing sickness behaviour and transiently modulating blood-brain barrier permeability. This is a general property of systemic inflammation rather than a feature specific to vaccination, and it therefore does not discriminate between events that are gut-brain axis disorders and events that merely provoke an inflammatory response. Stated as a criterion it would be satisfied by almost any acute illness, which is the same failure of specificity this review identifies in the ASIA criteria (Ameratunga et al., 2017). It is recorded here as a reason the question is worth asking, not as evidence that the answer is yes. Second, the gut is a principal immune organ and a reservoir of the mimicry antigens that drive several autoimmune events, so it is a plausible upstream source of the cross-reactive immunity these events express. The axis is therefore not a bystander in these events even when the event is named for another organ. This section examines each in turn, with the evidence status of each link stated explicitly, and Table 12 summarizes the mapping.
Guillain-Barre syndrome is the clearest instance, and it is properly understood as a gut-to-nerve axis disease. Its classic form follows infection with the gut pathogen Campylobacter jejuni, whose lipo-oligosaccharide carries carbohydrate structures that mimic the human nerve ganglioside GM1, so that antibodies raised against the organism cross-react with peripheral nerve and produce the paralysis (Yuki et al., 2004). This is an established causal mechanism, not a hypothesis, and it demonstrates that a gut organism can seed cross-reactive immunity that manifests as a neurological syndrome. Whether post-vaccination Guillain-Barre syndrome arises through the same class of molecular-mimicry mechanism is a hypothesis rather than an established finding: no study identified by this review demonstrates cross-reactive binding between a vaccine antigen and a human neural target in cases of post-vaccination Guillain-Barre syndrome, and Table 12 records the mechanism as untested. Mimicry is treated here as the most plausible candidate bridge between gut biology and this adverse event, on the strength of the Campylobacter precedent, not as a demonstrated one. What remains untested is whether variation in gut microbiome composition, including carriage of mimicry-bearing organisms, modifies individual susceptibility to the post-vaccination form.
Narcolepsy following the AS03-adjuvanted H1N1 vaccine is a genuine central nervous system disorder with direct axis relevance. It results from the loss of a small population of hypothalamic neurons that produce the wake-promoting peptide hypocretin, also known as orexin, and is exceptionally tightly associated with the HLA-DQB1*06:02 allele, which indicates an immune-mediated pathogenesis (Liblau et al., 2024; Nohynek et al., 2012). The axis relevance is anatomical as well as immunological: hypocretin neurons project widely to autonomic control structures, including the dorsal motor nucleus of the vagus nerve and the nucleus of the tractus solitarius, which are the central relays of gut-brain communication. Loss of these neurons therefore disrupts a node of the axis itself, and the disorder is accompanied by autonomic and metabolic disturbance in addition to the sleep phenotype. This event is thus not merely a rare harm to be counted; it is an example of vaccine-associated autoimmunity striking a hypothalamic structure with direct gut-brain regulatory function.
Febrile seizures are conventionally attributed to fever alone, which understates the mechanism. Fever is cytokine-driven, and interleukin-1 beta is directly implicated: the hippocampus synthesizes interleukin-1 beta in response to hyperthermia, intracerebral application of the cytokine lowers the threshold for experimental febrile seizures, and animals lacking the interleukin-1 receptor type 1 are resistant to hyperthermia-induced seizures (Dubé et al., 2005). Taken together these results separate the action of interleukin-1 signaling from that of temperature, and the wider role of interleukin-1 beta in epileptogenesis has been reviewed elsewhere (Vezzani & Baram, 2007). This makes febrile seizures a neuroimmune event. The axis connection is that seizure threshold is itself modulated by the gut microbiome: the anti-seizure effect of the ketogenic diet requires an intact microbiota, is abolished in germ-free or antibiotic-treated animals, and can be restored by specific commensal taxa (Olson et al., 2018). A microbiome contribution to individual susceptibility to vaccine-associated febrile seizures is therefore mechanistically coherent, and it is untested.
The remaining events have weaker but non-zero axis relevance. Myocarditis following messenger RNA vaccines is primarily a cardiac inflammatory event driven by interferon gamma and CXCL10 (Cao et al., 2025), but these are systemic mediators that act on the central nervous system, and cardiac inflammation involves autonomic and vagal pathways that form part of the same regulatory network. Anaphylaxis has a neuroimmune and vagal dimension, mediated by mast cells, which are abundant in the gut and are themselves regulated by the microbiome, and by histamine signaling and vagally mediated hypotension. Thrombosis with thrombocytopenia syndrome and immune thrombocytopenic purpura are the least axis-relevant in mechanism, although the former may present with cerebral venous sinus thrombosis and thus with a direct neurological consequence. In none of these cases does the review claim a gut-brain axis cause; the point is narrower, namely that the inflammatory milieu of these events intersects the axis and has not been characterized.
Table 12. Gut-Brain Axis Relevance of the Confirmed Vaccine Adverse Events.
Table 12. Gut-Brain Axis Relevance of the Confirmed Vaccine Adverse Events.
Adverse Event Primary Mechanism Gut-Brain Axis Relevance Evidence Status of the Axis Link
Guillain-Barre syndrome Molecular mimicry; anti-ganglioside antibodies Direct gut-to-nerve axis: the classic trigger is the gut organism Campylobacter jejuni via GM1 mimicry Mechanism established for the infection-triggered form; microbiome modification of susceptibility untested
Narcolepsy (AS03 H1N1) Autoimmune loss of hypothalamic hypocretin (orexin) neurons; HLA-DQB1*06:02 Hypocretin neurons project to the dorsal motor nucleus of the vagus and nucleus tractus solitarius, central relays of the axis Neuronal loss and HLA association established; axis and microbiome contribution untested
Febrile seizures Interleukin-1 beta mediated lowering of seizure threshold, not temperature alone Seizure threshold is microbiome-modulated; the anti-seizure effect of the ketogenic diet requires an intact microbiota Interleukin-1 beta causal role established in animals; microbiome link to vaccine-associated seizures untested
Myocarditis (mRNA) Immune-mediated cardiac inflammation via interferon gamma and CXCL10 Systemic cytokines act on the central nervous system; autonomic and vagal involvement in cardiac inflammation Cardiac mechanism established; axis consequences not characterized
Anaphylaxis IgE and complement activation; mast cell degranulation Gut mast cells are microbiome-regulated; histamine and vagally mediated hypotension Plausible; not studied in the vaccine context
Thrombosis with thrombocytopenia; immune thrombocytopenic purpura Anti-PF4 immunothrombosis; anti-platelet autoimmunity Limited; cerebral venous sinus thrombosis is a neurological consequence rather than an axis mechanism Weak axis relevance; no direct evidence
A separate and wholly unstudied question concerns the downstream consequences of these events for the axis. An adverse event is rarely an isolated insult: it is typically accompanied by acute illness, hospitalization, and in many cases antibiotic exposure, each of which is independently capable of perturbing the microbiome and of producing transient neuroinflammation. Whether individuals who experience a confirmed vaccine adverse event exhibit persistent alterations in microbiome composition, barrier integrity, or cytokine tone in the months that follow is a question that no identified study has addressed. This is stated as an evidence gap rather than as a suspicion of harm, and it is a tractable one: it requires longitudinal follow-up of individuals with confirmed events against matched vaccinated controls without events, with axis endpoints measured directly.
The neutrality of this analysis should be explicit. Nothing in this section constitutes evidence that vaccines cause gut-brain axis injury. What it establishes is that the mechanistic bridges between several confirmed adverse events and gut-brain biology are real, are in some cases well characterized on the infection side, and have not been examined on the vaccination side. Describing these events only as organ-specific entities, without noting their neuroimmune and axis dimensions, would understate the scope of the review as stated in its objectives. Consistent with the Bradford Hill appraisal presented above, these links remain plausible and untested, and they identify precisely where the unfinished science lies.

Appraisal of Frequently Cited Studies Reporting Positive Associations

A number of studies reporting positive associations between vaccines or vaccine components and autism are frequently cited in public discussion. These studies were retrieved by the search and are within the scope of the review, and a synthesis that omitted them without explanation would be open to the charge of selective exclusion. They are therefore included here and appraised using the same instruments applied to every other study in this review, including those whose conclusions point toward the null. Table 15 records the appraisal.
Two of these studies share a single decisive design limitation. DeLong (2011) correlated vaccination uptake with autism and speech impairment prevalence across United States states, and Tomljenovic and Shaw (2011) correlated cumulative aluminum exposure with autism prevalence across countries. Both are ecological analyses, in which exposure and outcome are measured as population averages rather than linked within individuals. Ecological correlation cannot support individual-level causal inference, and it is additionally vulnerable in this setting to ascertainment confounding, because jurisdictions with greater healthcare access tend to have both higher vaccination coverage and higher rates of autism diagnosis. The World Health Organization vaccine safety committee reviewed the aluminum analyses on precisely this ground, noting that ecological studies cannot be used to assert a causal association because they do not link exposure to outcome in individuals. A further point is relevant given the formal causal appraisal presented earlier in this review: Tomljenovic and Shaw report applying the Bradford Hill criteria to their ecological correlations. The Hill viewpoints presuppose an association observed at the individual level whose causal status is then interrogated; applying them to an aggregate correlation does not establish causation but restates the correlation in causal vocabulary.
The remaining studies are limited by design or by subsequent events. Geier and Geier (2007a) is a case series of nine patients with no comparison group and no denominator, a design that cannot support causal inference regardless of the findings reported. Hooker et al. (2014) and the commentary by Chhawchharia and Puliyel (2014) both rest on a reanalysis of a case-control dataset in which the original matched design was not preserved in the reanalysis; the companion reanalysis paper was subsequently retracted. Singh et al. (2002) is a genuine serological study, and it is the most substantive of the group, but its central finding, an unusual measles antibody detected in autistic sera, has not been independently replicated in more than two decades, and non-replication is decisive for a claim of this kind. Two papers by the Tomljenovic and Shaw group on aluminum adjuvants and neurodevelopment have been retracted by their journals, one withdrawn over concerns regarding scientific soundness and one retracted at the request of the editor-in-chief and the authors after gel images in several figures were found to be incorrectly presented, with the original gels no longer available to verify or repeat (Li et al., 2017).
Table 15. Appraisal of Frequently Cited Studies Reporting Positive Associations.
Table 15. Appraisal of Frequently Cited Studies Reporting Positive Associations.
Study Design Reported Finding Principal Limitation and Weight Assigned
DeLong (2011) Ecological (state-level regression) Higher vaccination uptake correlated with higher autism and speech impairment prevalence Ecological fallacy: aggregate correlation cannot link exposure to outcome in individuals; ascertainment confounding by healthcare access. Not weighted for causal inference
Tomljenovic & Shaw (2011) Ecological (cross-country correlation) Aluminum adjuvant exposure correlated with autism prevalence (Pearson r 0.89 to 0.94) Ecological fallacy; Bradford Hill criteria applied to aggregate correlation, which is a category error. Group has two related retractions. Not weighted
Geier & Geier (2007a) Case series (n = 9) Nine consecutive clinic patients with regressive autism and reported markers of mercury exposure. No measured comparison underlies the encephalopathy designation No comparator, no denominator; lowest tier of causal evidence. Substantial author conflicts of interest. Not weighted
Geier & Geier (2007b) Case-control (n = 53 cases) Reported an association between maternal thimerosal-containing Rho(D) immune globulin and autistic disorder Comparison group reportedly drawn from a different population sampled in an earlier decade, which is not a concurrent comparator. This description is taken from a secondary source; the primary report was not obtained, so the limitation is unverified. Not weighted
Hooker et al. (2014) Narrative review of six prior studies No new data. Re-describes six published datasets and characterises their handling as malfeasance Contains no primary analysis to appraise; the companion reanalysis it relies on did not preserve the matched design of the source data and was retracted. Not weighted
Chhawchharia & Puliyel (2014) Commentary Restates the reanalysis claim Not primary research; depends entirely on the reanalysis above. Not weighted
Singh et al. (2002) Serology, case-control (n = 125 vs 92) An unusual measles antibody in 75 of 125 autistic sera (60 percent) and 0 of 92 controls; over 90 percent of those sera also positive for anti-myelin basic protein Most substantive of the group, but the central finding has not been independently replicated in over twenty years. Weighted as an unreplicated finding
The appraisal above applies the same standard used elsewhere in this review. The Andersson et al. (2025) cohort, whose findings point toward the null, was likewise subjected to its published methodological critique rather than accepted uncritically, and the aluminum evidence streams were recorded as discordant in the Navigation Guide integration rather than resolved by preference. Applying an equivalent instrument to studies whose conclusions point in the opposite direction is a requirement of consistency, not an expression of bias. The relevant question for each study is what its design can support, and the answers recorded in Table 15 follow from design and replication status rather than from the direction of the reported result.
The studies whose findings point toward the null are appraised on the same columns, because an appraisal applied to one side of a question and not to the other cannot be checked by a reader who does not already share its conclusions. Two features recur across this literature and bear on how far it reaches. The first is outcome ascertainment: the largest cohorts identify autism from administrative or psychiatric registries without individual chart review, which is what makes cohorts of this size possible and which measures recorded diagnosis rather than diagnosis verified case by case. The second is indirectness in the GRADE sense. These studies test measles, mumps and rubella vaccination, thimerosal exposure, or cumulative antigen count. With the exception of Andersson et al. (2025) they do not test aluminium adjuvants, and none tests the microbiome-mediated pathway that is the subject of this review.
Table 16. Appraisal of Major Studies Reporting Null Associations.
Table 16. Appraisal of Major Studies Reporting Null Associations.
Study Design and Size What It Establishes Limitations Bearing on Interpretation
Madsen et al. (2002) Danish national cohort, 537,303 children Adjusted RR 0.92 (0.68 to 1.24) for autistic disorder after MMR Registry-based outcome ascertainment; validation substudy covered 40 of 316 cases; no data on family history of autism; tests MMR only
Hviid et al. (2019) Danish national cohort, 657,461 children Fully adjusted HR 0.93 (0.85 to 1.02) for autism after MMR No individual medical charts were reviewed. The susceptible-subgroup analysis for children with an autistic sibling rests on 37 exposed cases and is correspondingly imprecise. Tests MMR only
Jain et al. (2015) U.S. claims cohort, 95,727 children with older siblings Adjusted RR 0.91 to 1.10 across ages and dose counts Exposure from payment claims; reported MMR rates 4 to 14 percent below national survey estimates, so some vaccinated children are counted as unvaccinated, biasing toward the null. ASD identified by claims algorithm with 87 percent positive predictive value. Privately insured population
DeStefano et al. (2013) U.S. case-control, 256 cases and 752 controls Adjusted OR 0.999 per 25-unit increase in cumulative vaccine antigen Participation was 48 percent of eligible cases and 32 percent of eligible controls. Cumulative antigen count does not index adjuvant content
Taylor et al. (2014) Meta-analysis, 1,256,407 children in cohorts and 9,920 in case-control studies Pooled OR 0.99 (0.92 to 1.06) for vaccination and autism Two of five cohorts draw on overlapping Danish birth cohorts, identified as a possible duplicate-data problem. Component studies test MMR, thimerosal and mercury only
Andersson et al. (2025) Danish national cohort, 1,224,176 children, 50 outcomes aHR 0.93 for autism and 0.90 for ADHD with cumulative aluminium exposure Aluminium exposure derived from schedule and product records rather than measured. Some sibling analyses diverge from the main model. High national uptake limits the unexposed contrast
Verstraeten et al. (2003) Two-phase U.S. HMO study, 124,170 infants plus 16,717 No increased risk of autism or attention-deficit disorder in any analysis Site-specific associations with tics (RR 1.89) and language delay (RR 1.13) did not replicate across sites. A published erratum corrected tenfold decimal errors in the reported exposure guidelines
Barlow et al. (2001) Vaccine Safety Datalink cohort, 679,942 children Elevated febrile seizure risk on the day of DTP and 8 to 14 days after MMR; no elevation of nonfebrile seizures Case ascertainment differed across the four participating organisations. Vaccination status taken from automated data only, which the report notes may underascertain vaccination and bias relative risks downward. Neurodevelopmental follow-up rests on 273 children with an upper confidence limit of 4.2
These limitations do not reverse the direction of the findings, but their bearing runs in a particular direction and should be stated as such. Non-differential misclassification of exposure or outcome attenuates a true association toward the null. It cannot manufacture a protective estimate from a harmful one, but it is precisely the mechanism by which a small true effect would be rendered undetectable. The null results are therefore strong evidence against associations large enough to survive attenuation, and weaker evidence against small ones. The purpose of tabulating them is narrower. This literature is strong evidence about the exposures it measured, it is imprecise in the susceptible subgroups where the hypothesis would locate any effect, and it is indirect with respect to the mechanism examined here. Its weight against a general association between vaccination and autism is therefore substantially greater than its weight against the specific hypothesis this review sets out, and the two should not be treated as interchangeable. Section 4 returns to this point, because the gap it identifies is not an absence of large studies but an absence of large studies of the relevant exposure.

Funding, Conflicts of Interest, and Reflexive Application of the Risk-of-Bias Framework

Applied to the included literature, the domain identifies disclosures on both sides of the question, and they are recorded here rather than raised selectively. Verstraeten was an employee of the Centers for Disease Control and Prevention during the conduct of the thimerosal study and was employed by GlaxoSmithKline at the time of publication, as stated in the journal erratum, which also corrected tenfold decimal errors in the reported Environmental Protection Agency exposure guidelines (Erratum, 2004; Verstraeten et al., 2003). In Huda et al. (2019), a senior author is a cofounder of and shareholder in a company marketing a Bifidobacterium longum subspecies infantis probiotic, and another author’s institution received trial funding from the same company, which is material because the paper’s conclusion recommends trials of that intervention. The randomized troponin study of mRNA-1273 was funded by Moderna and most of its authors are company employees (Pfeiffer et al., 2026). On the other side, the 93 patients in the ASIA case series had all sought legal consultation before inclusion (Zafrir et al., 2012), and several of the mercury-hypothesis papers appraised in Table 15 originate from authors with documented involvement in vaccine-injury litigation. The narcolepsy literature carries disclosures in the opposite direction: among the authors of the Finnish registry analysis that established the AS03 association, one reported honoraria from GlaxoSmithKline and Pfizer, one was a co-investigator on a GlaxoSmithKline-funded study, one reported consultancies and funding from several pharmaceutical companies, and one was principal investigator of a GlaxoSmithKline-funded study and had received funding from Solvay for an unrelated pandemic vaccine trial; the study itself was funded by the Finnish National Institute for Health and Welfare and the Ministry of Social Affairs and Health, and the authors state that the funders had no role in design, analysis, or the decision to publish (Nohynek et al., 2012). That this disclosure sits alongside a finding unfavourable to a vaccine, and that the review nonetheless treats the finding as sound, is the point: the domain records sponsorship, it does not predict the direction of a result. None of these disclosures determines the weight assigned to a study in this review; weight follows from design, replication status, and the risk-of-bias instruments applied in Tables 15 to 21. The disclosures are reported because the Navigation Guide treats them as a bias domain to be assessed rather than assumed in either direction, and because reporting them only where they favour one conclusion would itself be a form of selective appraisal.

Appraisal of the Mercury and Thimerosal Literature

A second cluster of frequently cited work concerns thimerosal and mercury. It is appraised here on the same basis. The studies in this cluster fall into distinguishable categories, and the reason each carries little or no evidentiary weight differs by category rather than by conclusion.
Three are speculative papers published in Medical Hypotheses, a journal whose stated purpose is to publish untested ideas. For the two published before 2010 the journal operated editorial review rather than external peer review; it introduced peer review in 2010, so that characterisation does not apply to the 2011 paper, which is weighted instead on the ground that a hypothesis paper reports no data. Bernard et al. (2001), Richmand (2011), and Geier and Geier (2005) fall in this class, and the second is titled as a hypothesis on its face. A hypothesis paper cannot serve as evidence for the hypothesis it proposes, and citing it as such is circular. Two others are narrative reviews rather than primary research: Ratajczak (2011) and Blaylock (2008). A further paper, Geier et al. (2013), is primary research and is appraised on its design: its first phase draws incidence comparisons from the Vaccine Adverse Event Reporting System, a passive system with no denominator and known differential reporting, and its second phase is a case-control analysis within the Vaccine Safety Datalink reporting odds ratios of 2.18, 2.11, and 2.11 for organic mercury exposure from Thimerosal-containing hepatitis B vaccine in the first, second, and sixth months of life. The Vaccine Safety Datalink phase is the more informative of the two, and its findings have not been replicated by independent groups using the same database; all investigators on the study disclose involvement in vaccine or biologic litigation. The paper carries no retraction, correction, or expression of concern, and is distinct from a separate 2014 paper by an overlapping author in the same journal that was retracted. A systematic review does not draw evidence from narrative reviews but from the primary studies beneath them, and the immunoexcitotoxicity construct advanced by Blaylock, like the syndrome constructs discussed elsewhere in this review, is a proposed mechanism without an independent primary evidence base.
Two studies in this cluster warrant specific comment because their reported results differ from the use commonly made of them. Walker et al. (2006) challenged cultured lymphocytes from autistic children and their non-autistic siblings with thimerosal and measured no difference between the two groups in the induced transcriptional response. The study therefore reports a null result for the differential-susceptibility hypothesis, and its finding of a heat-shock response is a general cellular observation rather than an autism-specific one. Holmes et al. (2003) found lower mercury concentrations in the first haircuts of autistic children than in controls, which is the opposite of the direction predicted by a mercury-exposure hypothesis. The finding was reconciled with the hypothesis by proposing that autistic children excrete mercury poorly. That auxiliary proposal renders the hypothesis unfalsifiable, since high measured mercury and low measured mercury are then both read as confirmation, and this review treats unfalsifiability as a disqualifying property. The finding has also not been replicated.
Two remaining studies are primary research and are treated accordingly. Geier and Geier (2007b) reported an association between maternal Rho(D) immune globulin containing thimerosal and autistic disorder. On the description available, cases were drawn from the authors’ own clinic and the comparison rested on the frequency of Rh-negativity among women attending the same centres in an earlier decade, which is not a concurrent comparison group. That limitation is taken from a secondary description of the study rather than from the primary report, which was not obtained, and it is marked as unverified in Table 15 for that reason. Li et al. (2014) is the strongest study in this cluster: a transcriptomic analysis in a legitimate toxicology journal reporting neurodevelopmental and behavioral effects in neonatally exposed mice. The administered dose was approximately twenty times that used in human infant immunization. It is therefore admitted to the non-human evidence stream with a downgrade for indirectness, on the same basis as the maternal immune activation models, which likewise demonstrate real effects at stimulus magnitudes exceeding those of routine vaccination. It establishes that high-dose neonatal thimerosal is neurotoxic in mice; it does not establish that schedule-level exposure is neurotoxic in humans.
Table 17. Appraisal of the Mercury and Thimerosal Literature.
Table 17. Appraisal of the Mercury and Thimerosal Literature.
Study Category Reason Weight Is Limited or Withheld
Bernard et al. (2001) Hypothesis paper (Medical Hypotheses) Speculative venue without conventional peer review; proposes the hypothesis rather than testing it. Authors were founders of an advocacy organization. Not weighted
Richmand (2011) Hypothesis paper (Medical Hypotheses) Self-described hypothesis; no primary data. Not weighted
Geier & Geier (2005) Hypothesis paper (Medical Hypotheses) Speculative treatment proposal; no primary data. Not weighted
Geier et al. (2013) Two-phase study (VAERS; VSD case-control) Phase I rests on passive surveillance without a denominator; phase II reports odds ratios of about 2.1 to 2.2 but has not been independently replicated in the same database; all investigators disclose vaccine litigation involvement. Not weighted
Ratajczak (2011) Narrative review Not primary research. Not weighted
Blaylock (2008) Narrative review; coined construct Immunoexcitotoxicity is a proposed mechanism without an independent primary evidence base. Not weighted
Walker et al. (2006) In vitro challenge Thimerosal challenge up-regulated heat shock protein transcripts and zinc up-regulated metallothioneins, with no measured difference between autistic and sibling cells. A null result for differential susceptibility. Not weighted as supportive
Holmes et al. (2003) Case-control (hair mercury) Hair mercury 0.47 ppm in 94 autistic children versus 3.63 ppm in 45 matched controls, that is lower in the exposed-hypothesis group and opposite to the predicted direction; rescued by an auxiliary excretion hypothesis that renders the claim unfalsifiable. Unreplicated. Not weighted
Li et al. (2014) Animal transcriptomics Reported neurodevelopmental delay, reduced social interaction, and altered neuronal and endocrine pathway transcription in treated mice, with the endocrine effect confined to males. Dose approximately twenty times human immunization levels. Admitted to the non-human stream with a downgrade for indirectness
Singh & Jensen (2003) Serology Measles IgG higher in 88 autistic children than in 32 controls and 15 siblings, with an antibody to a 74 kDa measles protein in 43 of 52 autistic children and 0 of 45 controls and siblings. Separate sample from Singh et al. (2002), overlapping authorship; the finding remains unreplicated. Weighted as unreplicated

The Thimerosal Removal Natural Experiment as a Completed Falsification Test

The mercury hypothesis makes a specific and testable prediction. If thimerosal exposure from childhood vaccines caused a substantial share of autism, then removing thimerosal should be followed by a decline in autism incidence in the birth cohorts no longer exposed. This prediction was subjected to a natural experiment. Thimerosal was discontinued from vaccines in Denmark in 1992 and phased out of routine childhood vaccines in the United States beginning in 1999. In the Danish population register, autism incidence did not fall after the discontinuation. It continued to rise, including among children born after thimerosal was removed (Madsen et al., 2003). The equivalent decline did not occur in the United States either.
This is a population-level time-trend comparison, and it shares the design limitations this review applies to the ecological analyses appraised in Table 14: exposure and outcome are measured as population aggregates rather than linked within individuals, and autism ascertainment changed over the same period through successive revisions of diagnostic criteria and registration practice, which could in principle mask a decline. Its evidential value does not rest on design strength but on the specificity of the prediction. The hypothesis in its substantial-cause form predicted a fall in incidence following removal of the exposure; incidence rose. A prediction that fails in the direction opposite to the one specified is informative even from a design that could not have established causation had it succeeded, and it is this asymmetry, rather than the study design, that gives the result weight here. The hypothesis predicted a fall; the observed incidence rose. The prediction therefore failed in the direction opposite to the one specified, and the appropriate conclusion is that thimerosal exposure from the childhood schedule is not a substantial cause of autism at the population level. Consistency requires noting the critique of this evidence as well: the Danish analysis is ecological in design, subject to the same limitations this review applies to ecological studies favoring the opposite conclusion, diagnostic criteria and ascertainment changed over the study period, and a subsequent disclosure indicated that later data were not included in the published figures. These are legitimate qualifications. They do not, however, supply the decline that the hypothesis requires, and the absence of that decline across two national populations is the finding that bears on the hypothesis.

Post-Vaccination Neuroinflammatory and Multi-System Case Series

A distinct body of clinical literature describes neuroinflammatory and multi-system syndromes reported after vaccination. It is treated separately here because it is the evidence most directly relevant to the gut-brain axis question in humans, and because the studies within it differ sharply in what their designs can support. Table 18 records that appraisal.
Zafrir et al. (2012) analyzed 93 patients who developed systemic syndromes after recombinant hepatitis B vaccination, with a mean latency of 43.2 days from the final dose. The clinical profile is the closest published approximation to a gut-brain-immune phenotype identified in this review: neuropsychiatric complaints in 70 percent, systemic fatigue in 42 percent, and explicit gastrointestinal symptoms, including nausea, abdominal pain, diarrhea, and decreased appetite, in 50 percent, with elevated autoantibody titres in 80 percent of those tested. The concurrence of neuropsychiatric and gastrointestinal features in half the cohort is a genuine descriptive finding and is the principal empirical basis in humans for the mechanistic hypotheses advanced in this review.
Three limitations bound what this series can support, and they are structural rather than incidental. The cohort was assembled from an autoimmunity referral setting, so patients entered the series because they had presented with post-vaccination symptoms, which makes the observed association partly circular. There is no denominator and no control group, so the series can characterize a phenotype but cannot estimate frequency or establish causation, and the reported symptom prevalences cannot be compared against background rates in the unvaccinated. Most importantly, the Autoimmune/Inflammatory Syndrome Induced by Adjuvants (ASIA) framework within which the cases were classified is itself a proposed rather than a validated construct: its diagnostic criteria have not been validated, and a formal critique has argued that its first major criterion, exposure to a vaccine adjuvant or an infection, is met by essentially the entire population and therefore lacks specificity (Ameratunga et al., 2017). This review accordingly treats ASIA as hypothesis-generating and not as an established clinical entity.
Khayat-Khoei et al. (2022) described seven patients with MRI-confirmed central nervous system demyelination of the optic nerve, brain, or spinal cord occurring 1 to 21 days after mRNA vaccination. The composition of the series is as informative as its existence: four of the seven were exacerbations of previously known and stable multiple sclerosis, two were new-onset multiple sclerosis, and one was new-onset neuromyelitis optica, and all responded to standard immunotherapy, with most returning to baseline. Read carefully, this is not evidence of a population-level hazard but of a strong immune stimulus precipitating events in individuals who were already predisposed. It therefore supports the individual-susceptibility hypothesis that this review identifies as the principal untested question, and it is cited here on that basis rather than as evidence of incidence.
A narrative review of published neurological events after COVID-19 vaccination (Garg & Paliwal, 2022) catalogues the reported spectrum, including myelitis, acute disseminated encephalomyelitis, and Guillain-Barre syndrome, and invokes molecular mimicry in general terms as a candidate mechanism. It does not quantify how often neuro-autoantibodies are present, and it presents no structural or sequence modelling linking the vaccine-encoded spike protein to myelin oligodendrocyte glycoprotein or aquaporin-4; a single aquaporin-4-positive case is cited from a prior case report. The review is therefore admitted for the descriptive spectrum only. The autoantibody findings are mechanistically important, because they identify a measurable molecular mimicry pathway of the same class as the Campylobacter jejuni mechanism in Guillain-Barre syndrome discussed above. Two cautions are necessary. Sequence-similarity modeling is weak evidence for mimicry on its own, since homology searches will identify shared motifs between almost any two proteins, and confirmation requires demonstrated cross-reactive binding. And a compilation of published case reports carries the heaviest possible reporting bias and has no denominator, so the distribution of cases across vaccine platforms reflects publication and deployment patterns rather than relative risk, and no incidence can be inferred from it.
Martin et al. (2021) supplies the comparative control that the case series cannot. Using the self-controlled case series design, the same method this review endorses in the bias analysis above, it linked 46 verified pediatric cases of acute disseminated encephalomyelitis in Victoria, Australia, to the national immunization register, comparing each child's post-vaccination risk window against that child's own unexposed control period. The relative incidence was 1.041 for the 5 to 28 day window (95 percent CI 0.323 to 3.356) and 0.585 for the 2 to 42 day window (95 percent CI 0.182 to 1.886). Consistency requires that this review accept this result on the same terms it endorses the method: for acute disseminated encephalomyelitis following routine childhood vaccination, no excess above the background rate was detected.
Consistency equally requires stating what this study does not show, and it is here that the review departs from the interpretation offered in some secondary summaries. Only three of the 46 cases were vaccine-proximate. With three exposed events, the estimate is extremely imprecise, and the upper bound of the confidence interval for the narrow risk window reaches 3.356. The study therefore excludes a large increase in risk but cannot exclude a threefold increase, and it should not be described as robust evidence of no effect. It is a null result with wide bounds. The same standard applied elsewhere in this review to ecological studies reporting positive associations must be applied to imprecise studies reporting null ones, and imprecision is not evidence of absence.
The most consequential observation, however, concerns what none of these studies tested. Martin et al. examined acute disseminated encephalomyelitis, a discrete, acute, hospital-codable neurological event with an established case definition. It did not examine the chronic, multi-system, neuro-gastrointestinal symptom cluster that Zafrir et al. describe, which has no validated case definition, is not captured by hospital admission codes, and has never been subjected to a self-controlled or denominator-based design. The negative finding for one outcome therefore does not extend to the other. This is the central gap identified by this review: the outcome that most closely resembles a gut-brain axis phenotype is the one outcome for which no adequately controlled study exists. Establishing a validated case definition and testing it in a self-controlled design against a defined denominator is the study that would resolve this question, and it has not been done.
Table 18. Post-Vaccination Neuroinflammatory and Multi-System Series: What Each Design Can and Cannot Establish.
Table 18. Post-Vaccination Neuroinflammatory and Multi-System Series: What Each Design Can and Cannot Establish.
Study Design and size Denominator or control What it can establish What it cannot establish
Zafrir et al. (2012) Case series, 93 patients, hepatitis B vaccine None; autoimmunity referral setting Descriptive phenotype: co-occurring neuropsychiatric (70%) and gastrointestinal (50%) symptoms with autoantibodies (80%) Causation, frequency, or comparison with background; all 93 patients had sought legal consultation, so selection is both litigation-driven and partly circular; ASIA criteria unvalidated
Khayat-Khoei et al. (2022) Case series, 7 patients, mRNA vaccines None That a strong immune stimulus can precipitate CNS demyelination in predisposed individuals (4 of 7 had known MS); events were treatable Incidence; population-level risk; whether vaccination raises risk above background
Garg & Paliwal (2022) Narrative review of published case reports None The descriptive spectrum of reported neurological events; that molecular mimicry has been proposed in general terms Incidence; relative risk by platform (distribution reflects reporting, not risk); autoantibody frequency, which the review does not report; any spike-to-MOG or spike-to-AQP4 homology claim, which the review does not make
Martin et al. (2021) Self-controlled case series, 46 ADEM cases, national register Yes; each child serves as their own control That no excess of ADEM above background was detected after routine childhood vaccination A precise estimate: only 3 exposed cases; upper confidence bound 3.356, so a threefold increase is not excluded. Says nothing about chronic multi-system outcomes

Interventional Evidence: Microbiome Modulation Around Vaccination

The Bradford Hill appraisal above identified experiment, the demonstration that intervening on the exposure changes the outcome, as the strongest consideration and the most conspicuously absent for the hypothesis that microbiome composition modifies susceptibility to vaccine adverse events. One randomized trial bears directly on it, and the appraisal is revised accordingly.
The IMPACT trial (Wong et al., 2023) randomized 453 adults aged 65 or older or with type 2 diabetes, in a single-centre, double-blind, placebo-controlled design, to three months of a synbiotic formulation of three Bifidobacterium strains with prebiotics (SIM01) or to placebo. Critically for the present question, randomization was performed at the time of COVID-19 vaccination, and all participants received either the inactivated CoronaVac or the mRNA vaccine BNT162b2 within one week of starting the intervention. The pre-specified primary outcome, a composite of adverse health outcomes, occurred in 2.9 percent of the SIM01 group versus 12.6 percent of the placebo group at one month (p = 0.0002), and in 0 percent versus 3.1 percent at three months (p = 0.0252). The authors additionally report that the incidence of possible adverse events attributed to the COVID-19 vaccines was lower in the SIM01 group at one month (2.9 percent versus 9.6 percent) and at three months (0 percent versus 1.2 percent). Metagenomic sequencing confirmed that the intervention colonized as intended, with enrichment of Bifidobacterium adolescentis and other butyrate-producing taxa, and the abundance of B. adolescentis was itself associated with fewer adverse health outcomes.
This is, to the knowledge of this review, the only randomized controlled trial that has modulated the gut microbiome around the time of vaccination and measured adverse outcomes. It constitutes genuine interventional evidence, and the Bradford Hill experiment consideration can no longer be recorded as unmet. Four limitations bound what it establishes, and each is material.
First, the primary outcome is a composite of any new symptom or disease producing a self-reported negative impact on activities of daily living, not vaccine adverse events as such; the vaccine-specific comparison is reported in a single sentence without a p value and appears to be a secondary observation rather than a pre-specified endpoint. Second, and most consequential for interpretation, 61.3 percent of the composite events at one month were gastrointestinal upset, namely bloating, constipation, diarrhoea, abdominal discomfort, and loss of appetite. The intervention is a gut-targeted synbiotic, so a reduction in gastrointestinal symptoms is close to an expected direct effect of the product on the organ system in which most of the outcome events were counted, and it does not by itself demonstrate that the microbiome gates systemic vaccine reactivity. Third, the composite also includes ordinary infections: at three months, 60 percent of the events were infections, including a wound infection and an incident COVID-19 case, which are not vaccine adverse events. Fourth, outcome ascertainment was self-reported and subjective.
Applying the risk-of-bias framework adopted by this review, the conflict-of-interest domain must also be recorded. Two of the senior authors are founding members of the company commercializing the intervention and are named inventors on patent applications covering the tested formulation, and a further author serves as its honorary medical advisor. This does not invalidate the trial, whose randomization, blinding, allocation concealment, and pre-specified endpoints are methodologically sound. It is recorded because this review applies the same disclosure standard to a finding it might welcome as it applies to the contested studies appraised in Table 15 and Table 19. Consistency requires that a conflict of interest be noted whether or not the result is congenial.
The defensible conclusion is narrow but real. Randomized modulation of the gut microbiome around the time of vaccination reduced a composite of adverse health outcomes, and probably reduced mild vaccine-attributable events, in an elderly and diabetic population. This is the first interventional support for the proposition that microbiome state influences post-vaccination reactivity, and it moves the Bradford Hill experiment consideration from unmet to partially met. It does not establish that microbiome composition determines susceptibility to the confirmed serious adverse events of Section 3.6: no myocarditis, Guillain-Barre syndrome, narcolepsy, or anaphylaxis occurred in either arm, and the trial was neither designed nor powered to detect them. An observational companion study from the same group (Ng et al., 2022) reported that gut microbiota composition was associated with both SARS-CoV-2 vaccine immunogenicity and adverse events, which is consistent in direction. The study that would close the remaining gap is a trial of the same design, adequately powered, in a general population, with vaccine adverse events as a pre-specified primary endpoint ascertained objectively rather than by self-report.

5. Limitations

This review has several limitations. Heterogeneity across study designs made direct comparisons difficult. Randomized controlled trials, observational cohorts, mechanistic animal studies, and surveillance data each carry distinct biases. Confounding remains a concern in observational autism studies, though many adjust for socioeconomic and parental factors. Passive surveillance systems suffer from severe underreporting: across 37 studies the median under-reporting rate to spontaneous reporting systems was 94 percent (interquartile range 82 to 98 percent), and it remained 95 percent for more serious or severe reactions (Hazell & Shakir, 2006). This makes incidence estimates impossible from such data. Geographic generalizability is also limited, since most large cohort studies originate in high-income countries. Rare events remain inherently difficult to study, as even large cohorts may produce wide confidence intervals. Publication bias is possible in microbiome and mechanistic studies, where positive findings may be more likely to be reported. Although no language restriction was applied at the search stage, only English-language publications were included, so any relevant evidence published in other languages is absent and the review is exposed to language bias. The search was conducted through PubMed, Google Scholar, and the Walden University Library federated discovery service rather than through separate native-interface searches of each constituent database. Federated discovery does not apply database-specific controlled vocabulary and does not return per-database hit counts, so the search cannot be reproduced database by database and its recall in Embase, Cochrane, Scopus and CINAHL individually is unknown. Google Scholar, which supplied part of the retrieval, ranks results by an undisclosed relevance algorithm, caps accessible results irrespective of the number reported, and does not return a stable result set on repeat searching, so the portion of the search conducted through it is not reproducible either. The executed query strings were not retained, which compounds both limitations. The review is not registered. Registration with PROSPERO was attempted and could not be completed, because a synthesis that integrates human and non-human evidence streams does not fall within its eligible review categories. The consequence is that the prespecification of the eligibility criteria, outcomes and appraisal methods cannot be independently verified, and readers should weigh the reported analyses accordingly. The primary search closed on 5 September 2025. A targeted supplementary search to 27 September 2026 was conducted in four areas selected because the evidence base was known to be active: aluminium adjuvant safety, myocarditis after messenger RNA vaccination, regulatory positions on vaccines and autism, and gut microbiome effects on infant vaccine response. No comparable update was undertaken for Guillain-Barre syndrome, narcolepsy, febrile seizures, or chronic post-vaccination symptom clusters, so evidence published in those areas after September 2025 is not represented here. The supplementary search was not run across the full set of sources used in the primary search.

The Absence of a Countable Outcome

One barrier to resolving the questions this review leaves open is the absence of a countable outcome. The chronic, multi-system neuro-gastrointestinal symptom cluster described in case series (Zafrir et al., 2012) has no validated case definition, is not captured by hospital admission codes, and has not been examined in any self-controlled or denominator-based design. Negative findings for discrete, codable events such as acute disseminated encephalomyelitis (Martin et al., 2021) therefore do not extend to this phenotype. Any adequate definition would need to meet several requirements. It would require new onset of both gastrointestinal and neurological dysfunction in people without prior diagnoses in either domain, persistence beyond the period of ordinary post-vaccination reactogenicity, and objective abnormality on a validated measure rather than symptom report alone. It would exclude conditions and exposures that reproduce the phenotype through established mechanisms, notably prior functional gastrointestinal or neuropathic disease, recent enteric infection, and SARS-CoV-2 infection or other post-acute infection syndromes. It would assess vaccination only after a case has been classified, so that exposure is not written into the definition, and it would include an explicit non-case category so that it can be falsified. Definitions built from individually common symptoms are prone to low specificity (Ameratunga et al., 2017), so specificity should take priority over sensitivity. This review does not propose such a definition and does not assert that the phenotype exists as a distinct entity. A definition meeting these requirements is being developed separately and is not cited here as evidence.

6. Conclusions

The two directions of effect at the vaccine and gut-brain axis interface are supported to very different degrees, and the principal finding of this review is that asymmetry. In the direction running from the microbiome to the vaccine, the evidence is substantial: microbial composition and diversity in early life are associated with the magnitude and persistence of vaccine responses, and the mechanism is supported in animal models. The randomized human evidence is narrower than it is often reported to be: in a two-phase trial of 33 adults, antibody responses were largely unaffected in the first phase, and the reduction in H1N1-specific neutralization and IgG1 appeared only in the second phase, which was restricted to participants with low pre-existing immunity and randomized five of eleven to antibiotics. That result is consistent with the observational cohorts and with the animal work, but it rests on a small treated group and has not been replicated. This direction is rated moderate certainty and is the better-established of the two. In the direction running from the vaccine to the gut-brain axis, which is the direction the causal hypotheses in this field assert, the evidence is thin rather than negative. A single longitudinal adult cohort found compositional stability after vaccination; pediatric data are close to absent; and no study located by this review measured barrier function, mitochondrial indices, or cytokine trajectories following routine immunization in children. The available evidence does not show persistent disruption, and it is not sufficient to exclude it. Maternal fever in the second trimester was associated with increased autism risk in a prospective cohort, while a registry analysis of maternal infection found no overall association, with positive estimates confined to two of 33 uncorrected trimester-specific comparisons. Maternal influenza and Tdap vaccination were not associated with increased autism risk in the cohorts examined. Large-scale epidemiology, including the 2025 Danish cohort of more than 1.2 million children and the WHO GACVS reaffirmation, does not support large vaccine-attributable increases in autism risk for the exposures measured. Those exposures are measles-mumps-rubella vaccination, thimerosal, cumulative antigen count and, in the Danish cohort, cumulative aluminium. No study identified by this review examined DTaP, hepatitis B or varicella in relation to neurodevelopmental outcomes, and none tested the microbiome-mediated pathway that is the subject of this review, so this literature bounds the magnitude of any such effect without testing the mechanism. Under standard appraisal this evidence is of moderate certainty, and the quantitative bias analysis in Section 4 indicates it is robust against large confounding-induced masking of harm but cannot exclude small-to-modest effects. Several adverse events, including febrile seizures, intussusception, immune thrombocytopenic purpura, myocarditis, and anaphylaxis, are causally associated with vaccination and quantifiable in incidence. Other signals remain uncertain due to reliance on passive surveillance.
Adjuvant research shows biologically plausible mechanisms in animals but no reports of large-scale neurodevelopmental effects in humans. The example of AS03 and narcolepsy illustrates that rare, subgroup-specific outcomes can occur and underscores the importance of active surveillance.
No syndrome construct is proposed or adopted here. Future research should focus on longitudinal pediatric cohorts integrating microbiome, immune, and neurodevelopmental data; improved biomarker measurement of barrier and mitochondrial function; enhanced global active surveillance systems; and targeted microbiome interventions to optimize vaccine effectiveness. Transparent communication of both established risks and residual uncertainties is essential for evidence-based vaccine policy.

Supplementary Materials

The supplementary information can be downloaded at the website of this paper posted on Preprints.org.

Funding

Funded by the Bia-Echo Foundation, a private philanthropic foundation supporting work in reproductive equity, healthy ageing, and criminal justice reform. The funder had no role in the selection or appraisal of the evidence, the assignment of certainty ratings, or the drafting of the conclusions.

Conflicts of Interest

This review adopts funding source and conflict of interest as an explicit risk-of-bias domain, following the Navigation Guide methodology and the systematic evidence that industry sponsorship is associated with results favorable to the sponsor (Lundh et al., 2017; Woodruff & Sutton, 2014). Consistency requires that the domain be applied reflexively rather than only to other investigators. The authors therefore disclose that this work was supported by the Bia-Echo Foundation, a private philanthropic foundation that funds work in reproductive equity, healthy ageing, and criminal justice reform. The funder had no role in the selection or appraisal of the evidence, the assignment of certainty ratings, or the drafting of the conclusions, and no role in the separate development of a case definition for the chronic neuro-gastrointestinal phenotype referred to in Section 5. The authors further disclose a non-financial interest of their own: the interpretive construct referred to in Section 4 originates with them, and they are separately developing a case definition for the phenotype it describes. A reader should weigh the sections concerning that phenotype in that light. Neither the construct nor the case definition is treated as evidence at any point in this review, and no certainty rating, appraisal, or conclusion depends on either.

Appendix A. Studies Included in the Synthesis

The 67 studies below constitute the evidence synthesised in this review. Sixty were identified through the primary search to 5 September 2025 and seven through the targeted supplementary search to 27 September 2026. Full bibliographic details for each appear in the reference list. Publications cited in the report for context but excluded from the synthesis under the eligibility criteria are not listed here and are identified in Section 3.1.
The fourth column records the instrument applicable to each design: ROBINS-I for non-randomised studies, RoB 2 for randomised trials, SYRCLE for animal experiments, an adapted tool for case series, and an equivalent instrument for systematic reviews and meta-analyses. Studies for which no instrument applies, namely passive-surveillance analyses and in vitro work, are identified as such. Domain-level ratings for a subset of these studies are reported in Table 4.
No Study Design Appraisal instrument
1 Andersson et al. (2025) Cohort (national registry) ROBINS-I
2 Andrews et al. (2004) Cohort (retrospective) ROBINS-I
3 Ardura-Garcia et al. (2024) Systematic review AMSTAR-2 or equivalent
4 Barlow et al. (2001) Cohort (VSD) ROBINS-I
5 Baxter et al. (2016) Case-centred analysis ROBINS-I
6 Becerra-Culqui et al. (2018) Cohort ROBINS-I
7 Bhattacharjee et al. (2025) Preprint, cross-sectional ROBINS-I
8 Black, C. et al. (2003) Cohort (GPRD) ROBINS-I
9 Black, S. et al. (2010) Pooled trial analysis RoB 2
10 Borgognone et al. (2022) Animal experiment SYRCLE
11 Boston et al. (2024) Cohort (longitudinal) ROBINS-I
12 Cao et al. (2025) Animal and in vitro SYRCLE
13 Choi, Y. et al. (2025) Meta-analysis AMSTAR-2 or equivalent
14 Cooper et al. (2005) RCT RoB 2
15 Crepeaux et al. (2017) Animal experiment SYRCLE
16 DeLong (2011) Ecological ROBINS-I
17 DeStefano et al. (2013) Case-control ROBINS-I
18 Doyon-Plourde et al. (2026) Systematic review AMSTAR-2 or equivalent
19 Geier and Geier (2007a) Case series Case series tool
20 Geier and Geier (2007b) Case-control ROBINS-I
21 Geier et al. (2013) Two-phase (VAERS; VSD) ROBINS-I
22 Hagan et al. (2019) RCT (two phase) RoB 2
23 Harris et al. (2017) Nested case-control ROBINS-I
24 Heron and Golding (2004) Cohort (prospective) ROBINS-I
25 Holmes et al. (2003) Case-control ROBINS-I
26 Huda et al. (2019) Cohort (prospective) ROBINS-I
27 Hviid et al. (2003) Cohort (national registry) ROBINS-I
28 Hviid et al. (2019) Cohort (national registry) ROBINS-I
29 Jain et al. (2015) Cohort (claims) ROBINS-I
30 Khadka et al. (2023) Animal experiment SYRCLE
31 Khayat-Khoei et al. (2022) Case series Case series tool
32 Klein et al. (2010) Cohort (VSD) ROBINS-I
33 Lal et al. (2015) RCT RoB 2
34 Lei et al. (2017) Meta-analysis AMSTAR-2 or equivalent
35 Li, X. et al. (2014) Animal experiment SYRCLE
36 Madsen et al. (2002) Cohort (national registry) ROBINS-I
37 Madsen et al. (2003) Ecological time trend ROBINS-I
38 Martin et al. (2021) Self-controlled case series ROBINS-I
39 McNeil et al. (2016) Cohort (VSD) ROBINS-I
40 Miller et al. (2001) Cohort ROBINS-I
41 Mosca et al. (2008) Human mechanistic ROBINS-I
42 Narvaez-Miranda et al. (2026) Cohort (longitudinal) ROBINS-I
43 Ng et al. (2022) Cohort ROBINS-I
44 Nohynek et al. (2012) Cohort (registry) ROBINS-I
45 Oh et al. (2014) Animal experiment SYRCLE
46 Oster et al. (2022) Passive surveillance No instrument applicable
47 Partinen et al. (2012) Cohort (registry) ROBINS-I
48 Patel et al. (2011) Case-control ROBINS-I
49 Pfeiffer et al. (2026) RCT RoB 2
50 Ryan et al. (2025) Cohort and gnotobiotic ROBINS-I and SYRCLE
51 Salmon et al. (2013) Meta-analysis (SCRI) AMSTAR-2 or equivalent
52 Schechter and Grether (2008) Ecological time trend ROBINS-I
53 Semenzato et al. (2024) Cohort (national) ROBINS-I
54 Singh and Jensen (2003) Case-control (serology) ROBINS-I
55 Singh et al. (2002) Case-control (serology) ROBINS-I
56 Taylor et al. (2014) Meta-analysis AMSTAR-2 or equivalent
57 Tomljenovic and Shaw (2011) Ecological ROBINS-I
58 Verstraeten et al. (2003) Cohort (two phase, HMO) ROBINS-I
59 Vestergaard et al. (2004) Cohort (national registry) ROBINS-I
60 Walker et al. (2006) In vitro (human cells) No instrument applicable
61 Wong et al. (2023) RCT RoB 2
62 Xu et al. (2021) Cohort (multi-site) ROBINS-I
63 Xu et al. (2024) Self-controlled case series ROBINS-I
64 Yih et al. (2014) Self-controlled / cohort ROBINS-I
65 Zafrir et al. (2012) Case series Case series tool
66 Zerbo et al. (2017) Cohort ROBINS-I
67 Zimmermann and Curtis (2018) Systematic review AMSTAR-2 or equivalent

Appendix B. Mortality Misattribution and Expected Background Deaths

A further threat, distinct from underreporting, is outcome misclassification, and in particular the attribution of cause of death. When a person dies in the days after vaccination while also carrying an infection such as COVID-19 or another serious condition, the death may be recorded under that condition rather than examined as a possible vaccine event. If this occurs systematically, cause-specific surveillance would undercount any vaccine-attributable deaths. At the same time, temporal proximity alone is not evidence of causation, because deaths occur continuously in every population and a predictable number will always fall by chance in the days after any common exposure. Both statements are true together, and both can be addressed with the same objective tools.
The scale of coincidental clustering can be estimated directly from background mortality. Table B1 shows the number of deaths expected within one day, one week, and one month of vaccination among one million vaccinated people, based on ordinary age-specific mortality and assuming the vaccine has no effect at all. Among older adults, hundreds of deaths per million within a single day, and thousands within a month, are expected purely by chance. Because hundreds of millions of older adults were vaccinated worldwide, tens of thousands of deaths shortly after vaccination were statistically inevitable regardless of any causal effect. A death on the day after vaccination is therefore expected background unless the observed count exceeds this expectation. The analytic task is not to count deaths that follow vaccination, but to compare the observed number with the expected number.
Table B1. Deaths Expected by Chance After Vaccination from Background Mortality Alone (per Million Vaccinated, Assuming No Vaccine Effect).
Table B1. Deaths Expected by Chance After Vaccination from Background Mortality Alone (per Million Vaccinated, Assuming No Vaccine Effect).
Illustrative age band Approx. annual mortality Expected deaths per million within 1 day within 7 days within 30 days
18 to 39 0.1% 3 19 82
40 to 64 0.6% 16 115 493
65 to 74 2% 55 384 1,644
75 to 84 5% 137 959 4,110
85 and older 14% 384 2,685 11,507
Two designs make this comparison while remaining robust to the misattribution concern. Observed-versus-expected analysis compares deaths in a defined post-vaccination window with the number predicted from background rates. The self-controlled case series design uses each person as their own control, comparing mortality in a risk window after vaccination with mortality in a control window in the same individual, which removes all fixed between-person confounding. Critically, when these designs are applied to all-cause mortality they do not depend on how any death was labeled, so cause-of-death misclassification cannot conceal an excess: a true wave of vaccine-attributable deaths misattributed to COVID-19 would still raise all-cause mortality in the risk window above expectation. Large applications of these designs, including a Vaccine Safety Datalink self-controlled case series across eight sites (Xu et al., 2024) and multi-organization cohort analyses (Xu et al., 2021), found no increase in all-cause or non-COVID-19 mortality after COVID-19 vaccination, and in several strata mortality was lower among the vaccinated, a pattern consistent with the healthy-vaccinee effect.
That these systems can detect real fatal harms is shown by the fact that they did. Thrombosis with thrombocytopenia syndrome after adenoviral-vector vaccines, and myocarditis after mRNA vaccines, were identified and quantified through active surveillance, and regulatory restrictions on adenoviral-vector vaccines followed in several countries. Reports of a small increase in cardiac-related mortality in self-controlled analyses, more pronounced in males, would be consistent with the recognised myocarditis signal; no such analysis was among the studies included here. These pooled estimates are heterogeneous and some included studies have been contested. Two limitations remain and are stated plainly. The healthy-vaccinee effect biases all-cause comparisons toward showing vaccination as protective, so a small true excess could be partly masked, and the observed-versus-expected approach depends on the accuracy of the background rates used. Neither limitation is removed by counting deaths that happen to follow vaccination, which is why the observed-versus-expected framework, rather than raw temporal association, is the objective standard.

Appendix C. Benefit-Risk Framing and the Limits of Safe and Effective

Benefit-Risk Framing and the Limits of Safe and Effective

A recurring feature of public communication about vaccines, and about medical interventions generally, is the phrase safe and effective. As a matter of epidemiology the phrase is imprecise, because no intervention is exempt from the possibility of harm, and effectiveness is never absolute (Jefferson & Heneghan, 2023). Safety is not a property an intervention possesses on its own. It exists only in comparison: an intervention can be safer or less safe than a specified alternative, where the alternative may be a different intervention or no intervention at all. For a vaccine, the relevant alternative is the disease it prevents, so the honest unit of evaluation is a benefit-risk comparison rather than a safety label.
This distinction matters for the integrity of the present review, which throughout quantifies vaccine adverse events with care but has, until this point, quantified only one arm of the comparison. Presenting the absolute risk of a vaccine-associated harm without the absolute risk of the corresponding disease outcome is itself a form of selective presentation, and it is the mirror image of the safe and effective slogan: the slogan omits the harms, and a harms-only table omits the benefits. A complete account requires both arms and an explicit comparator. Table C1 provides this for the vaccines with quantified adverse events in Section 3.6, drawing the disease-side figures from public health surveillance.
The comparison in Table C1 is not a like-for-like ratio, and it should not be read as one, because the two arms are expressed in different denominators: adverse events are counted per dose administered, while disease complications are counted per case of natural infection. Translating between them requires knowing the probability of infection in the absence of vaccination, which varies with coverage, contact patterns, and time. The table is therefore an ordering of magnitudes rather than a single decisive number, and its purpose is to make the comparator visible rather than to compute a universal verdict. Even at the level of magnitudes the ordering is informative, though the margin varies by pairing. For measles, death occurs in roughly 1 to 3 per 1,000 infections, that is 1,000 to 3,000 per million, against immune thrombocytopenic purpura at about 40 per million doses, a difference of roughly 25 to 75 times. Encephalitis at roughly 1 per 1,000 infections against febrile seizures at 250 to 340 per million doses is a difference of only three to four times, and the two outcomes are not of comparable severity. The comparison is informative in ordering and should not be summarised as a single ratio.
Table C1. Benefit-Risk Comparison for Vaccines With Quantified Adverse Events.
Table C1. Benefit-Risk Comparison for Vaccines With Quantified Adverse Events.
Vaccine Quantified adverse event (per dose) Corresponding disease outcome (per infection, unless noted) Comparator and notes
Measles-containing (MMR) Febrile seizures 25 to 34 per 100,000; immune thrombocytopenic purpura 1 per 22,300 to 25,000 Encephalitis approx. 1 per 1,000 cases; death 1 to 3 per 1,000 cases; pneumonia approx. 1 per 20 cases; hospitalization approx. 1 per 5 unvaccinated cases Per dose versus per infection; disease-side figures from CDC measles surveillance
Rotavirus Intussusception 1.1 to 1.5 excess cases per 100,000 first-dose recipients Severe gastroenteritis and dehydration; before vaccination a leading cause of infant diarrheal hospitalization and, globally, of infant diarrheal death Per dose versus per unvaccinated child; wild rotavirus is itself associated with intussusception
mRNA COVID-19 Myocarditis approx. 8 per million doses overall; higher in young males Myocarditis after SARS-CoV-2 infection is several times higher than after vaccination; infection also carries age-dependent hospitalization and death Per dose versus per infection; see Section 3.6 for the infection-versus-vaccination comparison
The practical lesson, and the one most relevant to trust, is that candor about harms is not in tension with an accurate account of benefit; it is a precondition for it. When an intervention is presented as simply safe and a harm later becomes visible, confidence is undermined, and the erosion is greatest when earlier doubts were dismissed rather than addressed (Jefferson & Heneghan, 2023). The evidence-based alternative is to state the benefit-risk balance explicitly, including the harms that are real and the ones that are not yet explained, and to leave the informed judgment to the reader. This review adopts that standard, and Table C1 is its concrete expression for the events it has quantified.

References

  1. Advisory Committee on Immunization Practices. Childhood and Adolescent Immunization Schedule Workgroup: Terms of reference. Centers for Disease Control and Prevention. 2025. Available online: https://www.cdc.gov/acip/downloads/child-schedule-tor-508.pdf.
  2. Ameratunga, R.; Gillis, D.; Gold, M.; Linneberg, A.; Elwood, J. M. Evidence refuting the existence of autoimmune/autoinflammatory syndrome induced by adjuvants (ASIA). J. Allergy Clin. Immunol. Pract. 2017, 5(6), 1551–1555. [Google Scholar] [CrossRef] [PubMed]
  3. Andersson, N. W.; Bech Svalgaard, I.; Hoffmann, S. S.; Hviid, A. Aluminum-adsorbed vaccines and chronic diseases in childhood: A nationwide cohort study. Ann. Intern. Med. 2025, 178(10), 1369–1377. [Google Scholar] [CrossRef] [PubMed]
  4. Andrews, N.; Miller, E.; Grant, A.; Stowe, J.; Osborne, V.; Taylor, B. Thimerosal exposure in infants and developmental disorders: A retrospective cohort study in the United Kingdom does not support a causal association. Pediatrics 2004, 114(3), 584–591. [Google Scholar] [CrossRef] [PubMed]
  5. Ardura-Garcia, C.; Curtis, N.; Zimmermann, P. Systematic review of the impact of intestinal microbiota on vaccine responses. npj Vaccines 2024, 9, 254. [Google Scholar] [CrossRef] [PubMed]
  6. Atladóttir, H. Ó.; Thorsen, P.; Østergaard, L.; Schendel, D. E.; Lemcke, S.; Abdallah, M.; Parner, E. T. Maternal infection requiring hospitalization during pregnancy and autism spectrum disorders. J. Autism Dev. Disord. 2010, 40(12), 1423–1430. [Google Scholar] [CrossRef] [PubMed]
  7. Barlow, W. E.; Davis, R. L.; Glasser, J. W.; Rhodes, P. H.; Thompson, R. S.; Mullooly, J. P.; et al. The risk of seizures after receipt of whole-cell pertussis or measles, mumps, and rubella vaccine. N. Engl. J. Med. 2001, 345(9), 656–661. [Google Scholar] [CrossRef] [PubMed]
  8. Baxter, R.; Lewis, E.; Goddard, K.; Fireman, B.; Bakshi, N.; DeStefano, F.; et al. Acute demyelinating events following vaccines: A case-centered analysis. Clin. Infect. Dis. 2016, 63(11), 1456–1462. [Google Scholar] [CrossRef] [PubMed]
  9. Becerra-Culqui, T. A.; Getahun, D.; Chiu, V.; Sy, L. S.; Tseng, H. F. Prenatal tetanus, diphtheria, acellular pertussis vaccination and autism spectrum disorder. Pediatrics 2018, 142(3), e20180120. [Google Scholar] [CrossRef] [PubMed]
  10. Belkaid, Y.; Hand, T. W. Role of the microbiota in immunity and inflammation. Cell 2014, 157(1), 121–141. [Google Scholar] [CrossRef] [PubMed]
  11. Bernard, S.; Enayati, A.; Redwood, L.; Roger, H.; Binstock, T. Autism: A novel form of mercury poisoning. Med. Hypotheses 2001, 56(4), 462–471. [Google Scholar] [CrossRef] [PubMed]
  12. Bhattacharjee, B.; Lu, P.; Monteiro, V. S.; Tabachnikova, A.; Wang, K.; Hooper, W. B.; et al. Immunological and antigenic signatures associated with chronic illnesses after COVID-19 vaccination. medRxiv 2025. [Google Scholar] [CrossRef]
  13. Black, C.; Kaye, J. A.; Jick, H. MMR vaccine and idiopathic thrombocytopaenic purpura. Br. J. Clin. Pharmacol. 2003, 55(1), 107–111. [Google Scholar] [CrossRef] [PubMed]
  14. Black, S.; Della Cioppa, G.; Malfroot, A.; Nacci, P.; Nicolay, U.; et al. Safety of MF59-adjuvanted versus non-adjuvanted influenza vaccines in children and adolescents: An integrated analysis. Vaccine 2010, 28, 7331–7336. [Google Scholar] [CrossRef] [PubMed]
  15. Blaylock, R. L. A possible central mechanism in autism spectrum disorders, part 1. Altern. Ther. Health Med. 2008, 14(6), 46–53. Available online: https://pubmed.ncbi.nlm.nih.gov/19043938/. [PubMed]
  16. Bonhoeffer, J.; Kohl, K.; Chen, R.; Duclos, P.; Heijbel, H.; Heininger, U.; Jefferson, T.; Loupi, E. The Brighton Collaboration: Addressing the need for standardized case definitions of adverse events following immunization (AEFI). Vaccine 2002, 21(3-4), 298–302. [Google Scholar] [CrossRef] [PubMed]
  17. Borgognone, A.; Elizalde-Torrent, A.; Casadellà, M.; Romero, L.; Escribà, T.; Parera, M.; Català-Moll, F.; Noguera-Julian, M.; Brander, C.; Olvera, A.; Paredes, R. Vaccination with an HIV T-cell immunogen induces alterations in the mouse gut microbiota. npj Biofilms Microbiomes 2022, 8, 104. [Google Scholar] [CrossRef] [PubMed]
  18. Boston, R. H.; Guan, R.; Kalmar, L.; Beier, S.; Horner, E. C.; Beristain-Covarrubias, N.; et al. Gut microbiome remains stable following COVID-19 vaccination in healthy and immuno-compromised individuals. Life Sci. Alliance 2024, 7(4), e202302529. [Google Scholar] [CrossRef] [PubMed]
  19. Cao, X.; Manhas, A.; Chen, Y.-I.; Caudal, A.; Mondejar-Parreno, G.; Zhu, W.; Liu, W.; Kong, X.; Zeng, W.; Liu, L.; Zhao, S. R.; Jahng, J. W. S.; Utz, P. J.; Nadeau, K. C.; Nishiga, M.; Wu, J. C. Inhibition of CXCL10 and IFN-gamma ameliorates myocarditis in preclinical models of SARS-CoV-2 mRNA vaccination. Sci. Transl. Med. 2025, 17. [Google Scholar] [CrossRef] [PubMed]
  20. Carabotti, M.; Scirocco, A.; Maselli, M. A.; Severi, C. The gut-brain axis: Interactions between enteric microbiota, central and enteric nervous systems. Ann. Gastroenterol. 2015, 28(2), 203–209. [Google Scholar] [PubMed]
  21. Centers for Disease Control and Prevention. Autism and vaccines. U.S. Department of Health and Human Services, 19 November 2025. Available online: https://www.cdc.gov/vaccine-safety/about/autism.html.
  22. Chan, A. W.; Hróbjartsson, A.; Haahr, M. T.; Gøtzsche, P. C.; Altman, D. G. Empirical evidence for selective reporting of outcomes in randomized trials: Comparison of protocols to published articles. JAMA 2004, 291(20), 2457–2465. [Google Scholar] [CrossRef] [PubMed]
  23. Chhawchharia, R.; Puliyel, J. M. Commentary: Controversies surrounding mercury in vaccines: Autism denial as impediment to universal immunisation. Indian J. Med. Ethics 2014, 11(4). [Google Scholar] [CrossRef] [PubMed]
  24. Choi, G. B.; Yim, Y. S.; Wong, H.; Kim, S.; Kim, H.; Kim, S. V.; Hoeffer, C. A.; Littman, D. R.; Huh, J. R. The maternal interleukin-17a pathway in mice promotes autism-like phenotypes in offspring. Science 2016, 351(6276), 933–939. [Google Scholar] [CrossRef] [PubMed]
  25. Choi, Y.; Lee, J. S.; Choe, Y. J.; Lee, H.; Yoon, Y.; Shin, S. H.; et al. Myocarditis and pericarditis are temporally associated with BNT162b2 COVID-19 vaccine in adolescents: A systematic review and meta-analysis. Pediatr. Cardiol. 2025, 46(8), 2193–2206. [Google Scholar] [CrossRef] [PubMed]
  26. Cooper, C. L.; Davis, H. L.; Angel, J. B.; Morris, M. L.; Elfer, S. M.; Seguin, I.; Krieg, A. M.; Cameron, D. W. CPG 7909 adjuvant improves hepatitis B virus vaccine seroprotection in antiretroviral-treated HIV-infected adults. AIDS 2005, 19(14), 1473–1479. [Google Scholar] [CrossRef] [PubMed]
  27. Crépeaux, G.; Eidi, H.; David, M.-O.; Baba-Amer, Y.; Tzavara, E.; Giros, B.; Authier, F.-J.; Exley, C.; Shaw, C. A.; Cadusseau, J.; Gherardi, R. K. Non-linear dose-response of aluminium hydroxide adjuvant particles: Selective low dose neurotoxicity. Toxicology 2017, 375, 48–57. [Google Scholar] [CrossRef] [PubMed]
  28. Cryan, J. F.; Dinan, T. G. Mind-altering microorganisms. Nat. Rev. Neurosci. 2012, 13(10), 701–712. [Google Scholar] [CrossRef] [PubMed]
  29. Dall, C. CDC advisers create new work group to review childhood vaccines. CIDRAP News. October 2025. Available online: https://www.cidrap.umn.edu/childhood-vaccines/cdc-advisers-create-new-work-group-review-childhood-vaccines.
  30. DeLong, G. A positive association found between autism prevalence and childhood vaccination uptake across the U.S. population. J. Toxicol. Environ. Health Part A 2011, 74(14), 903–916. [Google Scholar] [CrossRef] [PubMed]
  31. DeStefano, F.; Price, C. S.; Weintraub, E. S. Increasing exposure to antibody-stimulating proteins and polysaccharides in vaccines is not associated with risk of autism. J. Pediatr. 2013, 163(2), 561–567. [Google Scholar] [CrossRef] [PubMed]
  32. Doyon-Plourde, P.; Chong, J.; Abrams, E. M.; Pless, R.; Young, K.; Tunis, M.; Zafack, J. Aluminium adjuvants in vaccines and potential health effects: Systematic review. BMJ 2026, 393, e088921. [Google Scholar] [CrossRef] [PubMed]
  33. Dubé, C.; Vezzani, A.; Behrens, M.; Bartfai, T.; Baram, T. Z. Interleukin-1beta contributes to the generation of experimental febrile seizures. Ann. Neurol. 2005, 57(1), 152–155. [Google Scholar] [CrossRef] [PubMed]
  34. Drossman, D. A. Functional gastrointestinal disorders: History, pathophysiology, clinical features, and Rome IV. Gastroenterology 2016, 150(6), 1262–1279. [Google Scholar] [CrossRef] [PubMed]
  35. Egger, M.; Davey Smith, G.; Schneider, M.; Minder, C. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997, 315(7109), 629–634. [Google Scholar] [CrossRef] [PubMed]
  36. Fasano, A. Intestinal permeability and its regulation by zonulin. Clin. Gastroenterol. Hepatol. 2012, 10(10), 1096–1100. [Google Scholar] [CrossRef] [PubMed]
  37. FDA. FDA approves required updated warning in labeling of mRNA COVID-19 vaccines regarding myocarditis and pericarditis following vaccination [FDA Safety Communication]. 25 June 2025. Available online: https://www.fda.gov/vaccines-blood-biologics/safety-availability-biologics/fda-approves-required-updated-warning-labeling-mrna-covid-19-vaccines-regarding-myocarditis-and.
  38. Fiorentino, M.; Sapone, A.; Senger, S.; Camhi, S. S.; Kadzielski, S. M.; Buie, T. M.; Kelly, D. L.; Cascella, N.; Fasano, A. Blood-brain barrier and intestinal epithelial barrier alterations in autism spectrum disorders. Mol. Autism 2016, 7, 49. [Google Scholar] [CrossRef] [PubMed]
  39. Furness, J. B. The enteric nervous system and neurogastroenterology. Nat. Rev. Gastroenterol. Hepatol. 2012, 9(5), 286–294. [Google Scholar] [CrossRef] [PubMed]
  40. Garg, R. K.; Paliwal, V. K. Spectrum of neurological complications following COVID-19 vaccination. Neurol. Sci. 2022, 43(1), 3–40. [Google Scholar] [CrossRef] [PubMed]
  41. Geier, D. A.; Geier, M. R. The potential importance of steroids in the treatment of autistic spectrum disorders and other disorders involving mercury toxicity. Med. Hypotheses 2005, 64(5), 946–954. [Google Scholar] [CrossRef] [PubMed]
  42. Geier, D. A.; Geier, M. R. A case series of children with apparent mercury toxic encephalopathies manifesting with clinical symptoms of regressive autistic disorders. J. Toxicol. Environ. Health Part A 2007a, 70(10), 837–851. [Google Scholar] [CrossRef] [PubMed]
  43. Geier, D. A.; Geier, M. R. A prospective study of thimerosal-containing Rho(D)-immune globulin administration as a risk factor for autistic disorders. J. Matern.-Fetal Neonatal Med. 2007b, 20(5), 385–390. [Google Scholar] [CrossRef] [PubMed]
  44. Geier, D. A.; Hooker, B. S.; Kern, J. K.; King, P. G.; Sykes, L. K.; Geier, M. R. A two-phase study evaluating the relationship between Thimerosal-containing vaccine administration and the risk for an autism spectrum disorder diagnosis in the United States. Transl. Neurodegener. 2013, 2, 25. [Google Scholar] [CrossRef] [PubMed]
  45. Guyatt, G. H.; Oxman, A. D.; Vist, G. E.; Kunz, R.; Falck-Ytter, Y.; Alonso-Coello, P.; Schünemann, H. J. GRADE: An emerging consensus on rating quality of evidence and strength of recommendations. BMJ 2008, 336(7650), 924–926. [Google Scholar] [CrossRef] [PubMed]
  46. Hagan, T.; Cortese, M.; Rouphael, N.; Boudreau, C.; Linde, C.; Maddur, M. S.; Das, J.; Wang, H.; Guthmiller, J.; Zheng, N.-Y.; Huang, M.; Uphadhyay, A. A.; Gardinassi, L.; Petitdemange, C.; McCullough, M. P.; Johnson, S. J.; Gill, K.; Cervasi, B.; Zou, J.; Pulendran, B. Antibiotics-driven gut microbiome perturbation alters immunity to vaccines in humans. Cell 2019, 178(6), 1313–1328.e13. [Google Scholar] [CrossRef] [PubMed]
  47. Harris, V. C.; Armah, G.; Fuentes, S.; Korpela, K. E.; Parashar, U.; Victor, J. C.; et al. Significant correlation between the infant gut microbiome and rotavirus vaccine response in rural Ghana. J. Infect. Dis. 2017, 215(1), 34–41. [Google Scholar] [CrossRef] [PubMed]
  48. Hazell, L.; Shakir, S. A. W. Under-reporting of adverse drug reactions. Drug Saf. 2006, 29(5), 385–396. [Google Scholar] [CrossRef] [PubMed]
  49. Heron, J.; Golding, J. Thimerosal exposure in infants and developmental disorders: A prospective cohort study in the United Kingdom does not support a causal association. Pediatrics 2004, 114(3), 577–583. [Google Scholar] [CrossRef] [PubMed]
  50. Hill, A. B. The environment and disease: Association or causation? Proc. R. Soc. Med. 1965, 58(5), 295–300. Available online: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1898525/. [CrossRef] [PubMed]
  51. Holmes, A. S.; Blaxill, M. F.; Haley, B. E. Reduced levels of mercury in first baby haircuts of autistic children. Int. J. Toxicol. 2003, 22(4), 277–285. [Google Scholar] [CrossRef]
  52. Hooijmans, C. R.; Rovers, M. M.; de Vries, R. B. M.; Leenaars, M.; Ritskes-Hoitinga, M.; Langendam, M. W. SYRCLE’s risk of bias tool for animal studies. BMC Med. Res. Methodol. 2014, 14, 43. [Google Scholar] [CrossRef] [PubMed]
  53. Hooker, B.; Kern, J.; Geier, D.; Haley, B.; Sykes, L.; King, P.; Geier, M. Methodological issues and evidence of malfeasance in research purporting to show thimerosal in vaccines is safe. BioMed Res. Int. 2014, 2014, 247218. [Google Scholar] [CrossRef] [PubMed]
  54. Hornig, M.; Bresnahan, M. A.; Che, X.; Schultz, A. F.; Ukaigwe, J. E.; Eddy, M. L.; et al. Prenatal fever and autism risk. Mol. Psychiatry 2018, 23(3), 759–766. [Google Scholar] [CrossRef] [PubMed]
  55. Huda, M. N.; Ahmad, S. M.; Alam, M. J.; Khanam, A.; Kalanetra, K. M.; Taft, D. H.; Raqib, R.; Underwood, M. A.; Mills, D. A.; Stephensen, C. B. Bifidobacterium abundance in early infancy and vaccine response at 2 years of age. Pediatrics 2019, 143(2), e20181489. [Google Scholar] [CrossRef] [PubMed]
  56. Hviid, A.; Stellfeld, M.; Wohlfahrt, J.; Melbye, M. Association between thimerosal-containing vaccine and autism. JAMA 2003, 290(13), 1763–1766. [Google Scholar] [CrossRef] [PubMed]
  57. Hviid, A.; Hansen, J. V.; Frisch, M.; Melbye, M. Measles, mumps, rubella vaccination and autism: A nationwide cohort study. Ann. Intern. Med. 2019, 170(8), 513–520. [Google Scholar] [PubMed]
  58. Jain, A.; Marshall, J.; Buikema, A.; Bancroft, T.; Kelly, J. P.; Newschaffer, C. J. Autism occurrence by MMR vaccine status among US children with older siblings with and without autism. JAMA 2015, 313(15), 1534–1540. [Google Scholar] [CrossRef] [PubMed]
  59. Jefferson, T.; Heneghan, C. Safe and effective: Health interventions can’t stake such a claim: So what should we say? Trust the Evidence [Substack]. 12 June 2023. Available online: https://trusttheevidence.substack.com/p/safe-and-effective.
  60. Khadka, S.; Omura, S.; Sato, F.; Tsunoda, I. Adjuvant injections altered the ileal and fecal microbiota differently with changes in immunoglobulin isotypes and antimycobacterial antibody responses. Int. J. Mol. Sci. 2023, 24(3), 2818. [Google Scholar] [CrossRef] [PubMed]
  61. Khayat-Khoei, M.; Bhattacharyya, S.; Katz, J.; Harrison, D.; Tauhid, S.; Bruso, P.; Houtchens, M. K.; Edwards, K. R.; Bakshi, R. COVID-19 mRNA vaccination leading to CNS inflammation: A case series. J. Neurol. 2022, 269(3), 1093–1106. [Google Scholar] [CrossRef] [PubMed]
  62. Kim, M.; Qie, Y.; Park, J.; Kim, C. H. Gut microbial metabolites fuel host antibody responses. Cell Host Microbe 2016, 20(2), 202–214. [Google Scholar] [CrossRef] [PubMed]
  63. Kim, S.; Kim, H.; Yim, Y. S.; Ha, S.; Atarashi, K.; Tan, T. G.; Longman, R. S.; Honda, K.; Littman, D. R.; Choi, G. B.; Huh, J. R. Maternal gut bacteria promote neurodevelopmental abnormalities in mouse offspring. Nature 2017, 549(7673), 528–532. [Google Scholar] [CrossRef] [PubMed]
  64. Klein, N. P.; Fireman, B.; Yih, W. K.; Lewis, E.; Kulldorff, M.; Ray, P.; et al. Measles-mumps-rubella-varicella combination vaccine and the risk of febrile seizures. Pediatrics 2010, 126(1), e1–e8. [Google Scholar] [CrossRef] [PubMed]
  65. Koh, A.; De Vadder, F.; Kovatcheva-Datchary, P.; Bäckhed, F. From dietary fiber to host physiology: Short-chain fatty acids as key bacterial metabolites. Cell 2016, 165(6), 1332–1345. [Google Scholar] [CrossRef] [PubMed]
  66. Lal, H.; Cunningham, A. L.; Godeaux, O.; Chlibek, R.; Diez-Domingo, J.; Hwang, S.-J.; Levin, M. J.; McElhaney, J. E.; Poder, A.; Puig-Barberà, J.; Vesikari, T.; Watanabe, D.; Weckx, L.; Zahaf, T.; Heineman, T. C.; the ZOE-50 Study Group. Efficacy of an adjuvanted herpes zoster subunit vaccine in older adults. N. Engl. J. Med. 2015, 372(22), 2087–2096. [Google Scholar] [CrossRef] [PubMed]
  67. Lauria, G.; Hsieh, S. T.; Johansson, O.; Kennedy, W. R.; Leger, J. M.; Mellgren, S. I.; et al. European Federation of Neurological Societies/Peripheral Nerve Society guideline on the use of skin biopsy in the diagnosis of small fiber neuropathy. Eur. J. Neurol. 2010, 17(7), 903–912. [Google Scholar] [CrossRef] [PubMed]
  68. Lei, W.-T.; Shih, P.-C.; Liu, S.-J.; Lin, C.-Y.; Yeh, T.-L. Effect of probiotics and prebiotics on immune response to influenza vaccination in adults: A systematic review and meta-analysis of randomized controlled trials. Nutrients 2017, 9(11), 1175. [Google Scholar] [CrossRef] [PubMed]
  69. Li, D.; Tomljenovic, L.; Li, Y.; Shaw, C. A. Retracted: Subcutaneous injections of aluminum at vaccine adjuvant levels activate innate immune genes in mouse brain that are homologous with biomarkers of autism. J. Inorg. Biochem. Retracted. 2017, 177, 39–54. [Google Scholar] [CrossRef] [PubMed]
  70. Li, X.; Qu, F.; Xie, W.; Wang, F.; Liu, H.; Song, S.; Chen, T.; Zhang, Y.; Zhu, S.; Wang, Y.; Guo, C.; Tang, T.-S. Transcriptomic analyses of neurotoxic effects in mouse brain after intermittent neonatal administration of thimerosal. Toxicol. Sci. 2014, 139(2), 452–465. [Google Scholar] [CrossRef] [PubMed]
  71. Liblau, R. S.; Latorre, D.; Kornum, B. R.; Dauvilliers, Y.; Mignot, E. J. The immunopathogenesis of narcolepsy type 1. Nat. Rev. Immunol. 2024, 24, 33–48. [Google Scholar] [CrossRef] [PubMed]
  72. Lindblad, E. B. Aluminum compounds for use in vaccines. Immunol. Cell Biol. 2004, 82(5), 497–505. [Google Scholar] [CrossRef] [PubMed]
  73. Lundh, A.; Lexchin, J.; Mintzes, B.; Schroll, J. B.; Bero, L. Industry sponsorship and research outcome. Cochrane Database Syst. Rev. 2017, 2, MR000033. [Google Scholar] [CrossRef] [PubMed]
  74. Lynn, D. J.; Benson, S. C.; Lynn, M. A.; Pulendran, B. Modulation of immune responses to vaccination by the microbiota: Implications and potential mechanisms. Nat. Rev. Immunol. 2022, 22(1), 33–46. [Google Scholar] [CrossRef] [PubMed]
  75. Madsen, K. M.; Hviid, A.; Vestergaard, M.; Schendel, D.; Wohlfahrt, J.; Thorsen, P.; et al. A population-based study of measles, mumps, and rubella vaccination and autism. N. Engl. J. Med. 2002, 347(19), 1477–1482. [Google Scholar] [CrossRef] [PubMed]
  76. Madsen, K. M.; Lauritsen, M. B.; Pedersen, C. B.; Thorsen, P.; Plesner, A. M.; Andersen, P. H.; Mortensen, P. B. Thimerosal and the occurrence of autism: Negative ecological evidence from Danish population-based data. Pediatrics 2003, 112(3), 604–606. [Google Scholar] [CrossRef] [PubMed]
  77. Martin, T. J.; Fahey, M.; Easton, M.; Clothier, H. J.; Samuel, R.; Crawford, N. W.; Buttery, J. P. Acute disseminated encephalomyelitis and routine childhood vaccinations: A self-controlled case series. Hum. Vaccines Immunother. 2021, 17(8), 2578–2585. [Google Scholar] [CrossRef] [PubMed]
  78. Mayer, E. A.; Knight, R.; Mazmanian, S. K.; Cryan, J. F.; Tillisch, K. Gut microbes and the brain: Paradigm shift in neuroscience. J. Neurosci. 2014, 34(46), 15490–15496. [Google Scholar] [CrossRef] [PubMed]
  79. McNeil, M. M.; Weintraub, E. S.; Duffy, J.; Sukumaran, L.; Jacobsen, S. J.; Klein, N. P.; et al. Risk of anaphylaxis after vaccination in children and adults. J. Allergy Clin. Immunol. 2016, 137(3), 868–878. [Google Scholar] [CrossRef] [PubMed]
  80. Miller, E.; Waight, P.; Farrington, C. P.; Andrews, N.; Stowe, J.; Taylor, B. Idiopathic thrombocytopenic purpura and MMR vaccine. Arch. Dis. Child. 2001, 84(3), 227–229. [Google Scholar] [CrossRef] [PubMed]
  81. Mitkus, R. J.; King, D. B.; Hess, M. A.; Forshee, R. A.; Walderhaug, M. O. Updated aluminum pharmacokinetics following infant exposures through diet and vaccination. Vaccine 2011, 29(51), 9538–9543. [Google Scholar] [CrossRef] [PubMed]
  82. Moser, C. A.; Offit, P. A. Aluminum exposure from vaccines and diet. JAMA 2026, 335(11), 939–942. [Google Scholar] [CrossRef] [PubMed]
  83. Mosca, F.; Tritto, E.; Muzzi, A.; Monaci, E.; Bagnoli, F.; Iavarone, C.; O’Hagan, D.; Rappuoli, R.; De Gregorio, E. Molecular and cellular signatures of human vaccine adjuvants. Proc. Natl. Acad. Sci. 2008, 105(30), 10501–10506. [Google Scholar] [CrossRef] [PubMed]
  84. Narváez-Miranda, J.; Sohn, M. B.; Velasquez-Portocarrero, D.; Gill, A. L.; Beblavy, R.; Castro-Melendez, D.; Ejiofor, K.; Qiu, X.; Laniewski, N.; Groff, B.; Brunner, J.; Ras, M.; Leger, A.; Macomber, A.; Caddy, S. L.; Jiang, B.; O’Connor, T.; Gill, S. R.; Scheible, K. Early-life gut microbiome composition and rotavirus vaccine-induced IgA responses in U.S. infants: A longitudinal cohort study. eBioMedicine 2026, 129, 106360. [Google Scholar] [CrossRef] [PubMed]
  85. Ng, S. C.; Peng, Y.; Zhang, L.; Mok, C. K. P.; Zhao, S.; Li, A.; et al. Gut microbiota composition is associated with SARS-CoV-2 vaccine immunogenicity and adverse events. Gut 2022, 71(6), 1106–1116. [Google Scholar] [CrossRef] [PubMed]
  86. Nirenberg, E.; Maldonado, Y. A.; Hoffman, S. A. The role and safety of aluminum adjuvants in childhood vaccines. Pediatrics 2026, 157(3), e2025074874. [Google Scholar] [CrossRef] [PubMed]
  87. Nohynek, H.; Jokinen, J.; Partinen, M.; Vaarala, O.; Kirjavainen, T.; Sundman, J.; Himanen, S.-L.; Hublin, C.; Julkunen, I.; Olsén, P.; Saarenpää-Heikkilä, O.; Kilpi, T. AS03 adjuvanted AH1N1 vaccine associated with an abrupt increase in the incidence of childhood narcolepsy in Finland. PLoS ONE 2012, 7(3), e33536. [Google Scholar] [CrossRef] [PubMed]
  88. Oh, J. Z.; Ravindran, R.; Chassaing, B.; Carvalho, F. A.; Maddur, M. S.; Bower, M.; Hakimpour, P.; Gill, K. P.; Nakaya, H. I.; Yarovinsky, F.; Sartor, R. B.; Gewirtz, A. T.; Pulendran, B. TLR5-mediated sensing of gut microbiota is necessary for antibody responses to seasonal influenza vaccination. Immunity 2014, 41(3), 478–492. [Google Scholar] [CrossRef] [PubMed]
  89. Olson, C. A.; Vuong, H. E.; Yano, J. M.; Liang, Q. Y.; Nusbaum, D. J.; Hsiao, E. Y. The gut microbiota mediates the anti-seizure effects of the ketogenic diet. Cell 2018, 173(7), 1728–1741. [Google Scholar] [CrossRef] [PubMed]
  90. Orwin, R. G. A fail-safe N for effect size in meta-analysis. J. Educ. Stat. 1983, 8(2), 157–159. [Google Scholar] [CrossRef] [PubMed]
  91. Oster, M. E.; Shay, D. K.; Su, J. R.; Gee, J.; Creech, C. B.; Broder, K. R.; et al. Myocarditis cases reported after mRNA-based COVID-19 vaccination in the US from December 2020 to August 2021. JAMA 2022, 327(4), 331–340. [Google Scholar] [CrossRef] [PubMed]
  92. Page, M. J.; McKenzie, J. E.; Bossuyt, P. M.; Boutron, I.; Hoffmann, T. C.; Mulrow, C. D.; Shamseer, L.; Tetzlaff, J. M.; Akl, E. A.; Brennan, S. E.; Chou, R.; Glanville, J.; Grimshaw, J. M.; Hróbjartsson, A.; Lalu, M. M.; Li, T.; Loder, E. W.; Mayo-Wilson, E.; McDonald, S.; Moher, D. The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. Syst. Rev. 2021, 10(1), 89. [Google Scholar] [CrossRef] [PubMed]
  93. Partinen, M.; Saarenpää-Heikkilä, O.; Ilveskoski, I.; Hublin, C.; Linna, M.; Olsén, P.; Nokelainen, P.; Alén, R.; Wallden, T.; Espo, M.; Rusanen, H.; Olme, J.; Sätilä, H.; Arikka, H.; Kaipainen, P.; Julkunen, I.; Kirjavainen, T. Increased incidence and clinical picture of childhood narcolepsy following the 2009 H1N1 pandemic vaccination campaign in Finland. PLoS ONE 2012, 7(3), e33723. [Google Scholar] [CrossRef] [PubMed]
  94. Patel, M. M.; Richardson López-Collada, V.; Bulhões, M. M.; De Oliveira, L. H.; Bautista Márquez, A.; Flannery, B.; et al. Intussusception risk and health benefits of rotavirus vaccination in Mexico and Brazil. N. Engl. J. Med. 2011, 364(24), 2283–2292. [Google Scholar] [CrossRef] [PubMed]
  95. Pfeiffer, S.; Rhee, M.; Striplin, S.; Urdaneta, V.; Straus, W.; Wang, X.; Girard, B.; Priddy, F.; Chalkias, S. Evaluation of cardiac troponin levels in individuals aged 12-30 years following mRNA-1273 vaccination: Results from a randomized placebo-controlled trial. Open Forum Infect. Dis. 2026, 13(4), ofag139. [Google Scholar] [CrossRef] [PubMed]
  96. Plotkin, S. A. History of vaccination. PNAS 2014, 111(34), 12283–12287. [Google Scholar] [CrossRef] [PubMed]
  97. Praharaj, I.; John, S. M.; Bandyopadhyay, R.; Kang, G. Probiotics, antibiotics and the immune responses to vaccines. Philos. Trans. R. Soc. B 2015, 370(1671), 20140144. [Google Scholar] [CrossRef] [PubMed]
  98. Ramadan, Y. N.; Alqifari, S. F.; Alshehri, K.; Alhowiti, A.; Alrasheed, T.; Aljohani, F.; Alghamdi, A.; Hetta, H. F.; Mirghani, H. Microbiome gut-brain-axis: Impact on brain development and mental health. Mol. Neurobiol. 2025, 62(8), 10813–10833. [Google Scholar] [CrossRef] [PubMed]
  99. Ratajczak, H. V. Theoretical aspects of autism: Causes, a review. J. Immunotoxicol. 2011, 8(1), 68–79. [Google Scholar] [CrossRef] [PubMed]
  100. Richmand, B. J. Hypothesis: Conjugate vaccines may predispose children to autism spectrum disorders. Med. Hypotheses 2011, 77(6), 940–947. [Google Scholar] [CrossRef] [PubMed]
  101. Ryan, F. J.; Clarke, M.; Lynn, M. A.; Benson, S. C.; McAlister, S.; Giles, L. C.; et al. Bifidobacteria support optimal infant vaccine responses. Nature 2025, 641, 456–464. [Google Scholar] [CrossRef] [PubMed]
  102. Salmon, D. A.; Proschan, M.; Forshee, R.; Gargiullo, P.; Bleser, W.; Burwen, D. R.; et al. Association between Guillain-Barré syndrome and influenza A (H1N1) 2009 monovalent inactivated vaccines in the USA: A meta-analysis. Lancet 2013, 381(9876), 1461–1468. [Google Scholar] [CrossRef] [PubMed]
  103. Sanna, S.; van Zuydam, N. R.; Mahajan, A.; Kurilshikov, A.; Vich Vila, A.; Võsa, U.; Mujagic, Z.; Masclee, A. A. M.; Jonkers, D. M. A. E.; Oosting, M.; Joosten, L. A. B.; Netea, M. G.; Franke, L.; Zhernakova, A.; Fu, J.; Wijmenga, C.; McCarthy, M. I. Causal relationships among the gut microbiome, short-chain fatty acids and metabolic diseases. Nat. Genet. 2019, 51(4), 600–605. [Google Scholar] [CrossRef] [PubMed]
  104. Schechter, R.; Grether, J. K. Continuing increases in autism reported to California’s developmental services system: Mercury in retrograde. Arch. General. Psychiatry 2008, 65(1), 19–24. [Google Scholar]
  105. Sejvar, J. J.; Kohl, K. S.; Gidudu, J.; Amato, A.; Bakshi, N.; Baxter, R.; et al. Guillain-Barre syndrome and Fisher syndrome: Case definitions and guidelines for collection, analysis, and presentation of immunization safety data. Vaccine 2011, 29(3), 599–612. [Google Scholar] [CrossRef] [PubMed]
  106. Semenzato, L.; Le Vu, S.; Botton, J.; Bertrand, M.; Jabagi, M.-J.; Drouin, J.; et al. Long-term prognosis of patients with myocarditis attributed to COVID-19 mRNA vaccination, SARS-CoV-2 infection, or conventional etiologies. JAMA 2024, 332(16), 1367–1377. [Google Scholar] [CrossRef] [PubMed]
  107. Sheldon, R. S.; Grubb, B. P.; Olshansky, B.; Shen, W. K.; Calkins, H.; Brignole, M.; et al. 2015 Heart Rhythm Society expert consensus statement on the diagnosis and treatment of postural tachycardia syndrome, inappropriate sinus tachycardia, and vasovagal syncope. Heart Rhythm 2015, 12(6), e41–e63. [Google Scholar] [CrossRef] [PubMed]
  108. Singh, V. K.; Jensen, R. L. Elevated levels of measles antibodies in children with autism. Pediatr. Neurol. 2003, 28(4), 292–294. [Google Scholar] [CrossRef] [PubMed]
  109. Singh, V. K.; Lin, S. X.; Newell, E.; Nelson, C. Abnormal measles-mumps-rubella antibodies and CNS autoimmunity in children with autism. J. Biomed. Sci. 2002, 9(4), 359–364. [Google Scholar] [CrossRef] [PubMed]
  110. Smith, S. E. P.; Li, J.; Garbett, K.; Mirnics, K.; Patterson, P. H. Maternal immune activation alters fetal brain development through interleukin-6. J. Neurosci. 2007, 27(40), 10695–10702. [Google Scholar] [CrossRef] [PubMed]
  111. Sterne, J. A. C.; Hernán, M. A.; Reeves, B. C.; Savović, J.; Berkman, N. D.; Viswanathan, M.; Henry, D.; Altman, D. G.; Ansari, M. T.; Boutron, I.; Carpenter, J. R.; Chan, A.-W.; Churchill, R.; Deeks, J. J.; Hróbjartsson, A.; Kirkham, J.; Jüni, P.; Loke, Y. K.; Pigott, T. D.; Higgins, J. P. T. ROBINS-I: A tool for assessing risk of bias in non-randomised studies of interventions. BMJ 2016, 355, i4919. [Google Scholar] [CrossRef] [PubMed]
  112. Sterne, J. A. C.; Savović, J.; Page, M. J.; Elbers, R. G.; Blencowe, N. S.; Boutron, I.; Cates, C. J.; Cheng, H.-Y.; Corbett, M. S.; Eldridge, S. M.; Emberson, J. R.; Hernán, M. A.; Hopewell, S.; Hróbjartsson, A.; Junqueira, D. R.; Jüni, P.; Kirkham, J. J.; Lasserson, T.; Li, T.; Higgins, J. P. T. RoB 2: A revised tool for assessing risk of bias in randomised trials. BMJ 2019, 366, l4898. [Google Scholar] [CrossRef] [PubMed]
  113. Sterne, J. A. C.; Sutton, A. J.; Ioannidis, J. P. A.; Terrin, N.; Jones, D. R.; Lau, J.; et al. Recommendations for examining and interpreting funnel plot asymmetry in meta-analyses of randomised controlled trials. BMJ 2011, 343, d4002. [Google Scholar] [CrossRef] [PubMed]
  114. Stobbe, M. vaccine safety. In webpage changed to contradict scientific conclusion that vaccines don’t cause autism; Associated Press, via PBS NewsHour, 20 November 2025; Available online: https://www.pbs.org/newshour/health/cdc-vaccine-safety-webpage-changed-to-contradict-scientific-conclusion-that-vaccines-dont-cause-autism.
  115. Taylor, L. E.; Swerdfeger, A. L.; Eslick, G. D. Vaccines are not associated with autism: An evidence-based meta-analysis. Vaccine 2014, 32(29), 3623–3629. [Google Scholar] [CrossRef] [PubMed]
  116. Tomljenovic, L.; Shaw, C. A. Do aluminum vaccine adjuvants contribute to the rising prevalence of autism? J. Inorg. Biochem. 2011, 105(11), 1489–1499. [Google Scholar] [CrossRef] [PubMed]
  117. VanderWeele, T. J.; Ding, P. Sensitivity analysis in observational research: Introducing the E-value. Ann. Intern. Med. 2017, 167(4), 268–274. [Google Scholar] [CrossRef] [PubMed]
  118. Erratum. Erratum: Safety of thimerosal-containing vaccines: A two-phased study of computerized health maintenance organization databases. Pediatrics 2004, 113(1), 184. [Google Scholar] [CrossRef]
  119. Verstraeten, T.; Davis, R. L.; DeStefano, F.; Lieu, T. A.; Rhodes, P. H.; Black, S. B.; et al. Safety of thimerosal-containing vaccines: A two-phased study of computerized health maintenance organization databases. Pediatrics 2003, 112(5), 1039–1048. [Google Scholar] [CrossRef]
  120. Vestergaard, M.; Hviid, A.; Madsen, K. M.; Wohlfahrt, J.; Thorsen, P.; Schendel, D.; et al. MMR vaccination and febrile seizures: Evaluation of susceptible subgroups and long-term prognosis. JAMA 2004, 292(3), 351–357. [Google Scholar] [PubMed]
  121. Vezzani, A.; Baram, T. Z. New roles for interleukin-1 beta in the mechanisms of epilepsy. Epilepsy Curr. 2007, 7(2), 45–50. [Google Scholar] [CrossRef] [PubMed]
  122. Walker, S. J.; Segal, J.; Aschner, M. Cultured lymphocytes from autistic children and non-autistic siblings up-regulate heat shock protein RNA in response to thimerosal challenge. NeuroToxicology 2006, 27(5), 685–692. [Google Scholar] [CrossRef] [PubMed]
  123. WHO GACVS. Statement of the WHO Global Advisory Committee on Vaccine Safety (GACVS) on vaccines and autism; World Health Organization, 11 December 2025; Available online: https://www.who.int/philippines/news/detail-global/11-12-2025-statement-gacvs-vaccines-autism.
  124. Wong, M. C. S.; Zhang, L.; Ching, J. Y. L.; Mak, J. W. Y.; Huang, J.; Wang, S.; et al. Effects of gut microbiome modulation on reducing adverse health outcomes among elderly and diabetes patients during the COVID-19 pandemic: A randomised, double-blind, placebo-controlled trial (IMPACT study). Nutrients 2023, 15(8), 1982. [Google Scholar] [CrossRef] [PubMed]
  125. Woodruff, T. J.; Sutton, P. The Navigation Guide systematic review methodology: A rigorous and transparent method for translating environmental health science into better health outcomes. Environ. Health Perspect. 2014, 122(10), 1007–1014. [Google Scholar] [CrossRef] [PubMed]
  126. Xu, S.; Huang, R.; Sy, L. S.; Glenn, S. C.; Ryan, D. S.; Morrissette, K.; et al. COVID-19 vaccination and non-COVID-19 mortality risk: Seven integrated health care organizations, United States, December 14, 2020 to July 31, 2021. MMWR Morb. Mortal. Wkly. Rep. 2021, 70(43), 1520–1524. [Google Scholar] [CrossRef] [PubMed]
  127. Xu, S.; Sy, L. S.; Hong, V.; Farrington, P.; Glenn, S. C.; Ryan, D. S.; et al. Mortality risk after COVID-19 vaccination: A self-controlled case series study. Vaccine 2024, 42(7), 1731–1737. [Google Scholar] [CrossRef] [PubMed]
  128. Yano, J. M.; Yu, K.; Donaldson, G. P.; Shastri, G. G.; Ann, P.; Ma, L.; Nagler, C. R.; Ismagilov, R. F.; Mazmanian, S. K.; Hsiao, E. Y. Indigenous bacteria from the gut microbiota regulate host serotonin biosynthesis. Cell 2015, 161(2), 264–276. [Google Scholar] [CrossRef] [PubMed]
  129. Yih, W. K.; Lieu, T. A.; Kulldorff, M.; Martin, D.; McMahill-Walraven, C. N.; Platt, R.; et al. Intussusception risk after rotavirus vaccination in U.S. infants. N. Engl. J. Med. 2014, 370(6), 503–512. [Google Scholar] [CrossRef] [PubMed]
  130. Yuki, N.; Susuki, K.; Koga, M.; Nishimoto, Y.; Odaka, M.; Hirata, K.; et al. Carbohydrate mimicry between human ganglioside GM1 and Campylobacter jejuni lipooligosaccharide causes Guillain-Barre syndrome. Proc. Natl. Acad. Sci. 2004, 101(31), 11404–11409. [Google Scholar] [CrossRef] [PubMed]
  131. Zafrir, Y.; Agmon-Levin, N.; Paz, Z.; Shilton, T.; Shoenfeld, Y. Autoimmunity following hepatitis B vaccine as part of the spectrum of autoimmune (auto-inflammatory) syndrome induced by adjuvants (ASIA): Analysis of 93 cases. Lupus 2012, 21(2), 146–152. [Google Scholar] [CrossRef] [PubMed]
  132. Zerbo, O.; Qian, Y.; Yoshida, C.; Fireman, B. H.; Klein, N. P.; Croen, L. A. Association between influenza infection and vaccination during pregnancy and risk of autism spectrum disorder. JAMA Pediatr. 2017, 171(1), e163609. [Google Scholar] [CrossRef] [PubMed]
  133. Zimmermann, P.; Curtis, N. The influence of probiotics on vaccine responses: A systematic review. Vaccine 2018, 36(2), 207–213. [Google Scholar] [CrossRef] [PubMed]
Table 6. Sensitivity of Passive-Surveillance Incidence to Underreporting.
Table 6. Sensitivity of Passive-Surveillance Incidence to Underreporting.
Assumed Reporting Fraction Multiplier on Observed Rate Illustrative: 2 Per Million Observed Implies
50% 2x 4 per million
10% 10x 20 per million
6% (Hazell & Shakir median) 16.7x 33 per million
5% 20x 40 per million
1% 100x 200 per million
Table 8. Fail-Safe N and Precision-Weighted Suppressed Evidence Required to Shift a Pooled Estimate of aHR 0.93.
Table 8. Fail-Safe N and Precision-Weighted Suppressed Evidence Required to Shift a Pooled Estimate of aHR 0.93.
Threshold the Pooled Estimate Must Reach Effect in Each Suppressed Study Count-Based Fail-Safe N (Equal Weights) Required Weight of Suppressed Evidence, as a Share of the Entire Published Evidence Base
aHR 1.00 aHR 1.25 9 studies 0.33 (33 percent)
aHR 1.00 aHR 1.50 5 studies 0.18 (18 percent)
aHR 1.00 aHR 2.00 3 studies 0.10 (10 percent)
aHR 1.10 aHR 1.25 35 studies 1.31 (131 percent)
aHR 1.10 aHR 1.50 15 studies 0.54 (54 percent)
aHR 1.10 aHR 2.00 8 studies 0.28 (28 percent)
aHR 1.25 aHR 1.50 43 studies 1.62 (162 percent)
aHR 1.25 aHR 2.00 17 studies 0.63 (63 percent)
Table 11. Bradford Hill Considerations Applied to the Microbiome-to-Adverse-Event Hypothesis.
Table 11. Bradford Hill Considerations Applied to the Microbiome-to-Adverse-Event Hypothesis.
Consideration What It Asks Status Note
Strength Size of the association Undefined No association has been measured in people
Consistency Recurs across studies and settings Unmet No direct studies exist to be consistent
Specificity One exposure, one effect Low Proposed mechanisms are broad; the weakest criterion
Temporality Cause precedes effect Untested (necessary) Needs pre-vaccination sampling; reverse causation is the main threat
Biological gradient Dose-response Untested No graded exposure-to-risk data
Plausibility A credible mechanism exists Met (weak) Immune set-point, mimicry, barrier tone; cheap and symmetric
Coherence Fits known biology Partial Only as a secondary modifier of dominant genetic and demographic risk
Experiment Intervening changes the outcome Partially met One RCT (SIM01) randomized at vaccination reduced a composite of adverse outcomes; GI-dominated, self-reported, patent-holding authors; not powered for the serious events
Analogy Similar established relationships Met (weak) Microbiome shapes immunogenicity; Campylobacter mimicry in GBS
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.