# Vaccine and autism evidence review

Prepared for Getting It Right. Evidence and publication status checked 7 October 2026. This is a structured editorial review, not a registered systematic review or personalized medical advice.

## Editorial decision

Use the requested headline **“Vaccines don't cause autism.”** Under it, explain its scientific scope: large human studies and evidence reviews do not support childhood vaccination as a cause of autism. Evidence is particularly extensive for MMR and thiomersal; aluminium and cumulative antigen exposure have separate evidence bases. The headline is not a mathematical guarantee about every conceivable future vaccine, every rare genetic subgroup, or every possible adverse effect.

The best case is cumulative evidence, not the retraction of one paper or a vote among authorities. Lead with what researchers compared, show the uncertainty, and let readers inspect the sources. Avoid “all studies agree,” “every vaccine has been tested in a placebo-controlled autism trial,” or “scientists know every cause of autism.” Those are unnecessary and inaccurate standards.

## A concise public page draft

### Vaccines don't cause autism

Researchers have looked for this connection in large groups of children, including children with an autistic sibling. The strongest studies have not found that MMR vaccination increases autism diagnoses. In a Danish study following more than 650,000 children, the autism rate after adjustment was similar in children who had and had not received MMR. [Hviid 2019](https://doi.org/10.7326/M18-2101)

Researchers have also studied different questions separately: the preservative thiomersal, aluminium used in some vaccines, and children's exposure to vaccine antigens. These are different ingredients and hypotheses; MMR does not contain thiomersal or an aluminium adjuvant. Current reviews do not support a causal link to autism. [WHO review](https://www.who.int/publications/i/item/9789240123533), [aluminium review](https://www.bmj.com/content/393/bmj-2025-088921)

A child may show developmental changes around the same age as routine vaccinations. That timing can be worrying and deserves attention. Timing alone cannot tell us the cause. Researchers look for higher rates, repeated patterns, and explanations that hold up when comparable children are studied.

Autism is a neurodevelopmental condition. Autistic people's communication, sensory experiences, abilities and support needs vary widely. They deserve respect and appropriate support. The evidence does not support blaming a parent's decision to vaccinate for a child's autism. [WHO autism overview](https://www.who.int/news-room/fact-sheets/detail/autism-spectrum-disorders)

Vaccines can have side effects. This page addresses autism; known vaccine risks belong on their own clearly linked pages.

## What counts as convincing evidence

Our editorial appraisal weights accurate exposure records, standardized or validated outcome ascertainment, comparable controls, adjustment for age/calendar time and healthcare use, adequate follow-up, precision, replication in different settings, preregistration where available, and sensitivity analyses. It does not count publications as votes. Statistical significance alone does not establish causation; a confidence interval crossing one does not prove equivalence.

The relevant question is whether a defined vaccination exposure changes the probability of an autism diagnosis compared with a credible alternative. A population autism prevalence number is not the expected “unvaccinated baseline.” Hazard ratios, odds ratios and cumulative absolute risks are different measures. Do not multiply a current national prevalence by a historical study hazard ratio to invent a personalized excess risk.

## Evidence by question

### 1 MMR and autism

**Hviid et al., 2019, Annals of Internal Medicine.** DOI 10.7326/M18-2101. [Journal record](https://doi.org/10.7326/M18-2101); [full original article PDF](https://nmd.bg/wp-content/uploads/2019/03/MMR-vaccine-and-ASD-201904160-m182101.pdf).

657,461 Danish children born 1999–2010; 6,517 autism diagnoses over 5.03 million person-years. Adjusted HR 0.93 (95% CI 0.85–1.02). Vaccination was time-varying; Cox models used attained age and accounted for birth year, sex, other vaccines, sibling history and perinatal risk factors. Subgroup and post-vaccination timing analyses did not show a consistent increase. Individual charts were not reviewed; residual confounding, diagnostic delay, exclusion of specified syndromes and sparse subgroups limit generalization. “MMR-unvaccinated” does not mean never receiving any vaccine. Funding: Novo Nordisk Foundation and Danish Ministry of Health; authors reported no funder role. Full main methods/results/limitations and indexed author correspondence inspected. Protocol/code/data were not publicly supplied with the article. Passage locators: Statistical Analysis; Results; Figure 3; Discussion. This is strong observational evidence, not a randomized experiment.

**Madsen et al., 2002, NEJM.** DOI 10.1056/NEJMoa021134. [Original](https://www.nejm.org/doi/full/10.1056/NEJMoa021134); [PubMed](https://pubmed.ncbi.nlm.nih.gov/12421889/).

537,303 Danish children born 1991–1998; adjusted RR 0.92 (0.68–1.24) for autistic disorder, 0.83 (0.65–1.07) for other spectrum diagnoses. Different birth cohort from Hviid 2019, but overlapping investigators and registry infrastructure. Indexed original methods/results inspected; not a new independent country. Age adjustment is essential because vaccination and diagnosis both change sharply with age; crude rates can compare unlike person-time. Precision is lower than the later cohort. Avoid claiming either study proves a protective effect against autism.

**Jain et al., JAMA 2015, corrected January 2016.** [Full current article](https://jamanetwork.com/journals/jama/fullarticle/2275444/).

95,727 US privately insured children with older siblings, including 1,929 with an autistic older sibling. For the higher-risk group, adjusted RR for two MMR doses versus none at age five was 0.56 (0.30–1.04); for those without an affected sibling, 1.09 (0.76–1.54). No increased association was found. Claims-based ascertainment, vaccine avoidance after developmental concerns, limited generalizability and subgroup precision matter. Do not interpret 0.56 as autism prevention. Full methods, Table 2, limitations, funding and correction statement inspected. NIMH funded; several authors worked for Lewin/Optum, disclosed in the paper. The correction redefined a variable and updated data; use the corrected online version. This is evidence about familial susceptibility, not proof covering every rare medical vulnerability.

**Di Pietrantonj et al., Cochrane 2021.** DOI 10.1002/14651858.CD004407.pub5. [Full review record](https://pmc.ncbi.nlm.nih.gov/articles/PMC8607336/).

Review covers MMR/MMRV effectiveness and multiple harms. Its approximately 23 million total participants must not be advertised as 23 million in an autism analysis. Autism evidence comes from a much smaller subset. The review supports no MMR–autism association while identifying real adverse effects such as febrile seizures. Indexed full review methods were inspected in the earlier dossier; not all underlying primary articles or supplements were reaudited here. Useful synthesis, not an independent dataset.

**Taylor, Swerdfeger and Eslick, Vaccine 2014.** DOI 10.1016/j.vaccine.2014.04.085. [PubMed abstract](https://pubmed.ncbi.nlm.nih.gov/24814559/).

Five cohorts, 1,256,407 children; five case-control studies, 9,920 children. Cohort pooled OR for vaccination and autism 0.99 (0.92–1.06); MMR 0.84 (0.70–1.01). Separate estimates address thiomersal. Two reviewers extracted data; searches through April 2014. Abstract verified; full statistical appendices not inspected. Large pooled totals do not eliminate confounding or dataset overlap. This review predates Jain 2015 and Hviid 2019 and should be paired with later evidence.

**Gidengil et al., Vaccine/AHRQ update 2021.** [PubMed](https://pubmed.ncbi.nlm.nih.gov/34049735/).

Literature search through November 2020; high strength of evidence for no increased autism risk after MMR. It distinguishes specific vaccine–outcome questions and recognizes genuine rare adverse events. Abstract inspected; full AHRQ report not independently audited here. Its MMR conclusion cannot be copied into every individual infant-vaccine row as though equally extensive direct evidence exists.

### 2 Thiomersal or thimerosal

This is an ethylmercury-containing preservative used in some formulations, not an aluminium adjuvant. Its risk cannot be inferred from the name “mercury” alone; chemical form, dose, route and clearance matter. Human exposure–outcome studies are more decisive than a claim that an ingredient sounds dangerous or a test-tube effect is possible.

**Hviid et al., JAMA 2003.** DOI 10.1001/jama.290.13.1763. [Full article](https://jamanetwork.com/journals/jama/fullarticle/197365).

467,450 Danish children; compared thiomersal-containing and thiomersal-free formulations of the same vaccine. RR 0.85 (0.60–1.20) for autism and 1.12 (0.88–1.43) for other spectrum diagnoses; no increasing dose-response. This ingredient-specific comparison is more informative than an uncontrolled vaccinated-versus-never-vaccinated contrast. Calendar changes, ascertainment and residual confounding remain possible. Full article and exposure table inspected. Some population overlap with the older Danish MMR cohort means it is not independent at the child level; exposure question differs.

**Price et al., Pediatrics 2010.** DOI 10.1542/peds.2010-0309. [Journal](https://publications.aap.org/pediatrics/article/126/4/656/65633/Prenatal-and-Infant-Exposure-to-Thimerosal-From).

256 ASD cases and 752 controls in three US managed-care organizations; prenatal and infant exposure reconstructed from records/interviews, with standardized diagnosis. No increased ASD association. Participation and residual confounding limit inference; inverse associations should not be sold as benefit. Publisher entry and WHO study appraisal inspected; full original article not independently audited in this pass. The same study sample underlies the antigen analysis below, so do not add their participant totals as independent replication.

### 3 Cumulative antigen exposure and the schedule

**DeStefano, Price and Weintraub, Journal of Pediatrics 2013.** DOI 10.1016/j.jpeds.2013.02.001. [PubMed](https://pubmed.ncbi.nlm.nih.gov/23545349/).

256 ASD cases/752 matched controls; recorded vaccine antigen content summed through 3 months, 7 months and 2 years. Adjusted OR per 25-unit increase at age two was 0.999 (0.998–1.001); no increased association for ASD with regression. Diagnoses were validated in person. Abstract with detailed methods inspected. Counts of antibody-stimulating proteins/polysaccharides are a proxy for one immunological hypothesis, not a universal measure of biological burden. It does not separately isolate every preservative, adjuvant, product or timing interaction, nor compare all modern schedules. Modest sample, participation bias and proxy limitations deserve explicit mention.

### 4 Aluminium adjuvants

**Andersson et al., Annals 2025, corrected supplement July 17.** DOI 10.7326/ANNALS-25-00997. [Journal](https://doi.org/10.7326/ANNALS-25-00997); [original main-text PDF](https://autismsciencefoundation.org/wp-content/uploads/2025/07/andersson-et-al-2025-aluminum-adsorbed-vaccines-and-chronic-diseases-in-childhood.pdf); [correction record](https://pubmed.ncbi.nlm.nih.gov/40674587/).

1,224,176 Danish children; estimated vaccine aluminium exposure before age two, primary diagnostic follow-up ages two–five, sensitivity follow-up to eight. Overall ASD HR per additional milligram was 0.93 (0.89–0.97). It principally tests dose variation; only 1.2% had zero aluminium-adjuvanted exposure. Main PDF methods/limitations inspected, including birth-year adjustment and dose categories. Important limits: registry outcomes, calendar confounding, delayed diagnosis, restricted exposure range and selection at age two. No specific funding reported. A July 17 correction replaced the supplement; that corrected file was inaccessible here. **Do not claim a complete supplementary-table audit or no increase in every subgroup.** The main article and newer review continue to support no overall ASD increase. Details of the contrary subgroup critique appear below.

**Doyon-Plourde et al., BMJ May 2026.** DOI 10.1136/bmj-2025-088921. [Original](https://www.bmj.com/content/393/bmj-2025-088921); [PMC](https://pmc.ncbi.nlm.nih.gov/articles/PMC13147496/).

Systematic review through November 27, 2025: six databases plus trial registries; registered protocol; independent screening/extraction checks, RoB2/ROBINS-I and GRADE. Fifty-nine studies across all outcomes, but **only three concerned ASD**: two ecological studies and the Danish cohort. ASD certainty rated moderate. It supports no causal association; persistent local nodules/granulomas are recognized reactions. Do not imply the included randomized trials directly tested long-term autism. Indexed full methods and ASD-specific results inspected; publisher PDF/direct PMC open failed. Public Health Agency of Canada authors. The Danish cohort appears within this review; counting both as independent replication would be wrong. The review predates the August 2026 critique below.

### 5 Timing and a proposed gut mechanism

**Honda, Shimizu and Rutter 2005.** [Original](https://acamh.onlinelibrary.wiley.com/doi/10.1111/j.1469-7610.2005.01425.x); [PubMed](https://pubmed.ncbi.nlm.nih.gov/15877763/).

Yokohama birth cohorts included 31,426 children; ASD diagnoses continued after MMR withdrawal. Useful natural-policy triangulation against MMR as a major population driver, not a randomized experiment or proof about all other vaccines. Diagnostic/service changes can affect time trends. Indexed original results and abstract inspected; complete methods/supplement not audited.

**Hornig et al., PLOS ONE 2008.** DOI 10.1371/journal.pone.0003140. [Full paper](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0003140).

Bowel tissue from 25 children with autism and gastrointestinal problems and 13 GI controls tested in three laboratories blinded to diagnosis, including a laboratory involved in earlier positive findings. Results did not support the proposed persistent-measles-virus/gut/autism connection. This is a direct test of one mechanism, with a small selected clinical sample, not proof of all possible mechanisms. Full methods/results inspected.

## Contrary evidence and difficult questions

### The original Wakefield report

The [1998 paper's record](https://pubmed.ncbi.nlm.nih.gov/9500320/) and [2010 retraction record](https://pubmed.ncbi.nlm.nih.gov/20137807/) identify the original and its withdrawal. [Full retraction notice](https://briandeer.com/solved/lancet-retraction-2010.pdf) inspected: it says claims that children were consecutively referred and investigations had ethics approval were incorrect, following the GMC findings. Full retraction was online February 2, 2010; journal issue February 6. The original was a 12-child uncontrolled case series, incapable of estimating a vaccine-caused autism risk. The later [BMJ investigation](https://www.bmj.com/content/342/bmj.c5347) documented discrepancies in histories. The causal conclusion does not depend on attacking Wakefield: later comparative evidence stands on its own. Do not say his small paper ever demonstrated causation.

### “A hepatitis B study found threefold risk”

[Gallagher/Goodman 2010](https://pubmed.ncbi.nlm.nih.gov/21058170/) was **retracted May 21, 2026**. [Publisher notice](https://www.tandfonline.com/doi/full/10.1080/15287394.2026.2673183) and [PubMed retraction record](https://pubmed.ncbi.nlm.nih.gov/42166532/) verified. The notice identifies a small selected sample with 31 autism cases, inadequate analysis/confounder adjustment, a cross-sectional design and unsupported causal conclusion. Authors disagree with retraction. Indexed full notice inspected; direct publisher open failed. Report the reasons, not merely the badge “retracted.” The live CDC page still invokes this study's reported association without acknowledging that status.

### “CDC hid an MMR subgroup effect”

[Hooker 2014 retraction notice](https://link.springer.com/article/10.1186/2047-9158-3-22) states concerns about methods/statistical validity and undeclared interests affecting peer review. Full notice inspected. Allegations about research conduct require their own documentary audit; they do not establish vaccine causation. The US sibling cohort and Danish subgroup studies test susceptibility more directly. Do not say every conceivable subgroup is fully excluded, or use “not statistically significant” as proof of exact absence.

### “Vaccinated versus unvaccinated surveys find more autism”

[Mawson et al. 2017, original PDF](https://www.oatext.com/pdf/JTS-3-186.pdf) inspected. Anonymous maternal survey of 666 homeschooled children, 261 unvaccinated; it reported ASD association. Convenience selection, parent-reported diagnoses/exposures, healthcare-use differences, many comparisons and nonrandom vaccination decisions prevent a reliable causal estimate. Its own Table 6 shows different medical-service use, relevant to diagnosis opportunity. Explain these vulnerabilities rather than claiming studies comparing vaccinated and unvaccinated children do not exist. No claim about its present retraction status is made here.

### “The Danish aluminium supplement contradicts its headline”

[Jablonowski and Hooker, August 2026 critique](https://pmc.ncbi.nlm.nih.gov/articles/PMC13567156/), [PubMed record](https://pubmed.ncbi.nlm.nih.gov/42730380/), is a reanalysis/commentary on the same cohort, not a new population. Indexed full argument inspected. It highlights a later-birth-cohort Asperger estimate of 1.67 (1.01–2.77), categorical risk differences and inverse associations that could reflect bias. It recounts an author response concerning inadequate exposure overlap across birth years, but the original response and corrected supplement were not independently retrieved.

Editorial assessment: these are specific methodological questions worth resolving. Repeated graphs of the same diagnoses are not independent replications; multiple subgroup tests, historical diagnostic categories and positivity/calendar overlap matter. An isolated nominally significant subgroup does not establish broad aluminium-caused autism. Nor should all inverse findings be treated as genuine protection. **Publication gate:** inspect the corrected supplement and original response before writing a definitive point-by-point rebuttal. Do not claim “every analysis was null.” Do not equate historical Asperger diagnosis automatically with a current support-needs level.

### “Mechanisms show it is possible”

Biological plausibility generates testable hypotheses. Cell or animal effects cannot determine whether real vaccine exposures cause human ASD without appropriate dose, route, developmental timing and clinical corroboration. Neuroinflammation is nonspecific. Known neurological reactions or a rare metabolic disorder are not interchangeable with an autism diagnosis. Avoid a confident assertion that all autistic people share one mechanism, or that unknown mechanisms cancel repeated human comparisons.

## What the latest official statements do and do not establish

[Live CDC Autism and Vaccines page](https://www.cdc.gov/vaccine-safety/about/autism.html), dated July 22, 2026, inspected directly. It argues that the blanket no-autism claim is not evidence-based, cites gaps for individual infant vaccines, criticizes observational MMR evidence and flags aluminium subgroup findings. It also acknowledges strong review evidence of no MMR association. A website's wording is not new experimental evidence.

The [National Academies DTaP chapter](https://www.ncbi.nlm.nih.gov/books/NBK190028/) genuinely classified direct DTaP–autism evidence as inadequate to accept or reject causation. Indexed chapter inspected. That gap must not be rewritten as a demonstrated association. Conversely, MMR findings alone cannot establish identical certainty for every schedule/product interaction. This distinction is more honest than saying either “nothing was studied” or “every combination was conclusively tested.”

[WHO's September 2026 publication](https://www.who.int/publications/i/item/9789240123533) reaffirms no causal vaccine–autism link. Its search ends August 2025. Final download failed; the [December 2025 technical proof](https://cdn.who.int/media/docs/default-source/medicines/pharmacovigilance/rev2_vaccines-autism-and-asd-evidence-review-2010-2025_15-dec.pdf?sfvrsn=c31b62bb_1) was inspected. It assessed 31 primary studies and five meta-analyses using multiple reviewers and qualitative bias appraisal. It included positive studies rather than excluding them by conclusion. The positive thiomersal reports clustered in one research group and were judged high-bias/very-low-strength, with problems including control selection, coding and exposure definition. This is a review's appraisal, not independent reanalysis of every dataset. Do not count the proof, final report and WHO statement as three independent reviews.

## Timing, diagnosis and respect

Autistic characteristics may be recognized gradually, and some children experience loss of skills. Recognition, symptom onset and first diagnostic code are distinct dates. A recorded diagnosis years after vaccination does not date the biological onset. Parents' observations can be accurate while the causal interpretation remains uncertain. Developmental concerns deserve assessment regardless of vaccination history.

Potential biases run both directions: early concerns can delay further vaccination; families with greater healthcare access may have more complete vaccine records and earlier diagnoses; birth cohorts differ in awareness and criteria. Adjustment is necessary but imperfect. Age adjustment is not automatically manipulation. Low point estimates should not be described as proof vaccination prevents autism.

The [WHO autism overview](https://www.who.int/news-room/fact-sheets/detail/autism-spectrum-disorders) describes diverse neurodevelopmental characteristics, variable support needs, and contributions from genetic and environmental factors. Do not reduce autism to injury, poisoning or parental error. Increased recorded prevalence alone identifies neither a cause nor the exact contribution of diagnostic change. This review does not attempt a complete autism-etiology assessment.

## Source independence and access audit

- Danish 2002 and 2019 MMR birth cohorts do not overlap, but investigators/infrastructure overlap. The 2003 thiomersal cohort overlaps calendar cohorts with the older MMR work. The 2025 aluminium cohort overlaps later Danish children. Different hypotheses remain informative; participant totals cannot simply be added.
- Price 2010 and DeStefano 2013 use the same US case-control sample. Jain's US claims cohort, Yokohama policy change and Hornig's laboratory study contribute distinct settings/designs.
- Reviews frequently reuse these primary studies. WHO and BMJ reviews are syntheses, not additional million-child cohorts. The aluminium ASD evidence is less independently replicated than MMR evidence.
- Public-health affiliations, industry/foundation relationships, funders and inaccessible raw data should be disclosed. Neither affiliation nor an undeclared-interest allegation substitutes for checking design and results. No blanket “all studies independent of pharmaceutical money” claim is supported.
- Fully inspected main texts: Hviid 2019 PDF, Jain corrected webpage, Hviid 2003 webpage, Andersson original main PDF, Hornig, WHO technical proof, Mawson PDF, Wakefield and Hooker notices. Indexed full methods/results: BMJ aluminium review and current aluminium critique. Abstract-only/detailed summaries: Taylor, Gidengil, Price and antigen study; limited original indexed text for Madsen/Honda. Final WHO PDF, corrected aluminium supplement and original author response remain access gaps.

## Search log and inclusion approach

Search date 7 October 2026; general web search with publisher, PubMed/PMC, WHO, CDC and National Academies verification. Relevant positive-result papers were deliberately sought. No language-complete database export, duplicate independent screening or pooled reanalysis was performed, so this memo must not be called an exhaustive systematic review.

Executed query families:
1. “Hviid 2019 MMR autism 657461 full text”, “M18-2101 sibling hazard methods”, “Madsen 2002 population based MMR autism 537303”.
2. “JAMA autism MMR siblings 2015 95727”, then full Table 2, funding and correction review.
3. “Taylor Swerdfeger Eslick 2014 vaccines autism meta analysis 1256407”, “Safety vaccines updated systematic review 2021”.
4. “Price 2010 prenatal infant thimerosal autism 256 752”, “Hviid 2003 thimerosal 467450”, “DeStefano 2013 proteins polysaccharides autism”.
5. “Aluminum Adsorbed Vaccines Chronic Diseases Childhood 2025 correction”, “ANNALS-25-03233”, “aluminium adjuvants systematic review 2026”, “Asperger 1.67”, “Andersson positivity assumption response”.
6. “Wakefield retraction ileal lymphoid 2010”, “Hooker retraction 2014”, “Gallagher Goodman HepB autism”, verified resulting 2026 retraction.
7. “Mawson pilot comparative vaccinated unvaccinated 2017”, “Honda MMR withdrawal autism”, “Hornig blinded PCR autism enteropathy”.
8. Current WHO vaccine/autism review, live CDC autism page, WHO autism overview and National Academies specific DTaP evidence.

Inclusion priority: direct human autism outcomes, defined exposure/comparator, large well-controlled cohorts, systematic reviews with explicit bias methods; mechanistic studies for specified hypotheses; original contrary claims and correction/retraction records. Animal/cell findings and broad neurodevelopmental composites were not treated as direct autism-risk estimates. News and social posts served only as discovery leads.

## Linked deeper questions and publication checks

Recommended links below the short page:
- What did the large MMR studies compare?
- What about children with an autistic sibling?
- Are thiomersal, aluminium and vaccine antigens the same question?
- Why might developmental changes be noticed after a vaccine?
- What do positive studies claim, and how were they checked?
- Why do CDC and WHO currently use different wording?
- What evidence would change this conclusion?
- What are the real risks and benefits of the relevant vaccine?

Before launch: medically qualified review; verify final WHO version and corrected aluminium supplement/response; update publication-status checks; preserve country, age, exposure and endpoint labels. A credible replicated association with appropriate comparators, clinically verified outcomes, coherent timing/dose effects and robustness to confounding would warrant reconsideration. A single ecological trend, passive report count or a change of official wording does not meet that standard. Keep dated corrections public. The principal MMR conclusion need not wait for resolution of every aluminium secondary-analysis dispute, but the ingredient page must acknowledge its narrower evidence base.