Since my original review, other analysts have independently found some of the same numerical problems. A deeper look also raises additional questions about healthcare use, vaccination-status misclassification, practice effects, temporality and multiple testing.
Update: October 7, 2026
When I first went through Comparative Health Outcomes of Vaccinated versus Unvaccinated Children: A Retrospective Cohort Analysis, I came away with a fairly simple conclusion:
There may be signals in this dataset worth investigating.
But the analysis presented in the paper does not establish that vaccines caused vaccinated children to be less healthy.
Since then, I’ve gone back through the study again and looked for responses, corrections, and independent analyses.
Something important has happened.
Other reviewers have now independently identified several of the same problems I raised, including the bizarre Table 4 denominator that somehow contains more vaccinated children than the study says existed.
And the deeper I look, the less this resembles a simple debate over whether the paper has been peer-reviewed.
The bigger issue is whether the published numbers, exposure definitions, and analysis are internally reliable enough to support the claims being made from them.
The paper remains on Zenodo as Version v1, published and last modified September 26. As of October 7, I could not find a revised version correcting the tables or a public response from the authors explaining the discrepancies discussed below. Zenodo
First, the basic design
The researchers analyzed electronic medical records from eight U.S. pediatric practices encompassing 6,239 children. The Zenodo abstract describes vaccination status as at least one vaccine before 13 months versus “zero vaccines ever.” Nine diagnostic categories were examined using ICD codes, with overall, sex-stratified and vaccine-dose analyses. Zenodo
Children’s Health Defense then promoted the findings as evidence that vaccinated children had substantially higher odds — often described as higher “risk” — of several conditions, including recurrent respiratory disease, ear infections and “strep.” Children’s Health Defense
Those associations are reported in the paper. What the study does not establish is that vaccination caused them. And there are some much more basic problems to settle first.
Table 4 still doesn’t add up
This was the first thing that really stopped me when I originally reviewed the paper.
The study says its vaccinated cohort contains:
2,414 children.
But the respiratory-disease line in Table 4 contains:
1,369 vaccinated children with disease
plus
1,324 vaccinated children without disease.
That equals:
2,693 vaccinated children.
That is 279 more vaccinated children than supposedly exist in the cohort.
Since my article was published, this discrepancy has now been independently identified by at least two other analyses.
Vaxopedia/StopAntiVaxPropaganda raised the same 2,693-versus-2,414 discrepancy on October 2. Tech ARP independently reported it again on October 6. Stop Anti-Vax Propaganda
This matters because this isn’t an argument over vaccine policy, confounding or causal philosophy. It’s arithmetic.
And my original observation goes a step further.
When the male and female respiratory counts in Tables 5 and 6 are combined, they do total the stated 2,414 vaccinated children.
Those counts are:
523 + 484 = 1,007 respiratory cases
720 + 687 = 1,407 noncases
Total = 2,414
But those numbers produce an odds ratio of approximately 1.19, rather than the 1.6 reported in Table 4.
Maybe there is an innocent table-construction error. Maybe two analytic populations got mixed together. Maybe a row was copied incorrectly.
But somebody needs to explain it. Until then, one of the paper’s most heavily promoted findings has an unresolved denominator problem.
Hooker has a long history with vaccinated-vs.-unvaccinated claims
Brian Hooker is not a neutral newcomer to this research question. For years, he has argued that vaccinated children have worse health outcomes than unvaccinated children, and he has both published and promoted studies designed to support that conclusion.
In 2020, Hooker and Neil Miller published a vaccinated-versus-unvaccinated analysis in SAGE Open Medicine using records from three pediatric practices. The paper reported higher odds of developmental delay, asthma, ear infections, and gastrointestinal disorders among vaccinated children. Children’s Health Defense promoted the study under the headline that unvaccinated children had “better health outcomes.” ResearchGate
Hooker and Miller followed that with another paper in 2021 examining vaccinated and unvaccinated children while adjusting for breastfeeding and type of birth. That study again reported poorer outcomes among vaccinated children. ResearchGate
Hooker has also spent years promoting Anthony Mawson’s 2017 vaccinated-versus-unvaccinated homeschool survey and similar studies. Children’s Health Defense maintains a page prepared by Hooker compiling multiple “vaxxed vs. unvaxxed” studies, including Mawson, Hooker and Miller, and work by James Lyons-Weiler and Paul Thomas. CHD Pennsylvania
In 2023, Hooker co-authored Vax-Unvax: Let the Science Speak with Robert F. Kennedy Jr., a book built around more than 70 comparisons of vaccinated and unvaccinated groups. Hooker has continued presenting these studies publicly as evidence that unvaccinated children are healthier. Children’s Health Defense
That history does not invalidate the new Yengst et al. study. A researcher’s prior views do not determine whether the data are correct.
But it does make methodological rigor especially important.
Several of the studies Hooker has previously promoted have themselves been heavily criticized for selection bias, inadequate control of confounding, self-reported outcomes, nonrepresentative populations, or other methodological problems. Science Feedback’s 2023 review of Hooker’s vaccinated-versus-unvaccinated claims concluded that the collection presented a one-sided picture and noted that some of the prominently cited studies had been retracted or contained major methodological limitations. Science Feedback
That makes the current paper’s problems especially relevant. Once again we have a highly selected population, inadequate adjustment for healthcare use and confounding, unclear exposure classification, irregular dose-response patterns and, this time, an apparently impossible denominator in a central results table.
The point isn’t that Hooker’s history automatically makes Yengst et al. wrong.
The point is that this is a longstanding advocacy position of one of the paper’s authors, so the appropriate response is not to give the new study less scrutiny. It is to demand enough transparency and reproducibility that the analysis can stand independently of the authors’ prior beliefs.
And right now, some of its most basic numbers still don’t.
Healthcare-use bias isn’t hypothetical
The paper relies heavily on diagnoses appearing in electronic medical records.
For several outcomes, one diagnosis wasn’t enough. Children had to accumulate at least three diagnoses during twelve months. That makes healthcare utilization critically important.
The more often a child sees a pediatrician, the more opportunities that child has to receive a vaccine. The same additional visits create more opportunities for a physician to document an ear infection, rash, respiratory illness, sore throat or fever.
Conversely, a child whose parents use pediatric care less frequently may experience similar minor illnesses without ever generating three medical-record diagnoses.
The authors acknowledge they don’t have complete healthcare-utilization information. And we now have very good reason not to treat that as a hypothetical concern.
A large 2013 JAMA Pediatrics study followed more than 323,000 children and found that undervaccinated children had significantly fewer outpatient visits than age-appropriately vaccinated children. Children undervaccinated specifically because of parental choice also had fewer outpatient and emergency-department encounters. PubMed
That is almost exactly the sort of systematic difference that can distort an electronic-record comparison like this one.
You cannot simply equate fewer diagnoses in the chart with less disease in the child when one group interacts differently with the healthcare system.
There is another EHR problem: are the exposure groups themselves accurate?
This deserves more attention than I gave it originally. The paper wants to distinguish children who received vaccines from children who received none.
But those classifications are being constructed from medical records across eight practices.
How was “zero vaccines ever” established?
Were state immunization registries checked? Were outside pediatricians captured? Health-department vaccination clinics? Pharmacies? Other health systems? Did families change providers?
This is not an obscure concern.
Researchers specifically studying the feasibility of vaccinated-versus-undervaccinated safety studies have examined vaccination-status misclassification in electronic health records.
In a 2017 Vaccine study involving more than 361,000 children, researchers used chart review and parent surveys to validate EHR vaccination classifications. They also found that parents of undervaccinated children were more likely to report using alternative medical providers. PubMed
A subsequent study explicitly examined both confounding and misclassification bias in childhood-immunization-schedule research and concluded that measurement and control of disease risk factors need careful consideration. PubMed
That doesn’t prove Yengst et al. misclassified anybody. It means the paper needs to demonstrate how it ruled that problem out.
And the “unvaccinated” definition is still confusing
The Zenodo abstract says the comparison is:
at least one vaccine before 13 months versus zero vaccines ever. Zenodo
But the manuscript also discusses 469 children whose first vaccination occurred after 13 months and indicates that they were retained in the primary analysis as unvaccinated before 13 months.
Those definitions don’t cleanly match.
Other reviewers have now noticed the same problem. Stop Anti Vax Propaganda
If those 469 children are in the reference group, then this isn’t simply “vaccinated versus never vaccinated.”
If they’re excluded from the never-vaccinated group, the totals and language need to make that clear.
Exposure classification is the foundation of a study like this. It should not require readers to reverse-engineer what happened to hundreds of children.
The sample itself is extraordinarily unusual
The paper’s comparison group contains 3,825 of 6,239 children - about 61% of the entire sample.
Because of the ambiguity surrounding those 469 late-vaccinated children, I would not automatically say all 3,825 remained completely unvaccinated. If the 469 are removed, there would still be approximately 3,356, or about 54% of the entire cohort, left in that category.
Either way, this is an extraordinarily unusual pediatric population.
For context, the large EHR validation study mentioned above found that only 1.3% of more than 361,000 children had received no vaccines by 24 months. PubMed
Those aren’t perfectly comparable definitions or populations, so don’t turn that into “61% versus 1.3%” as though it were an apples-to-apples prevalence comparison.
But the contrast tells us something important:
These eight practices are highly selected. That makes representativeness and confounding even more important.
Tech ARP has now independently raised the same concern about how unusual the study population is compared with ordinary U.S. pediatric populations. Tech ARP
Eight practices introduce another confounder
There is an additional issue I would add to my original critique: practice-level effects.
The children came from eight different pediatric practices.
Suppose one practice has a relatively high vaccination rate and also codes respiratory illnesses aggressively.
Suppose another practice serves many vaccine-refusing families and is less likely to bring children in for minor respiratory illnesses.
Pooling those practices together can create an apparent relationship between vaccination and diagnosis even if the relationship is much smaller — or absent, within individual practices.
This is why site, clinic, or provider can matter in multi-practice observational datasets.
The Vaccine Safety Datalink’s methodological work on studying the childhood schedule has emphasized that these observational comparisons are inherently complex and require careful attention to confounding and bias. PubMed
I would want to see the Yengst data stratified by practice, and preferably analyzed with practice incorporated directly into the statistical model.
Which came first? The disease or the vaccine?
The original article already raised temporality, but the problem is even more fundamental than “the authors didn’t provide enough dates.”
The exposure is determined by vaccination through 13 months. Yet some diagnoses apparently occurred extremely early in life.
If a child develops a respiratory condition at three weeks of age and receives a first vaccine at four months, that child may eventually be classified as vaccinated.
But the disease at three weeks obviously cannot have been caused by a vaccine received months later.
That means a valid causal analysis has to make sure that outcomes counted against the exposed group actually occurred after exposure.
Ideally, vaccination would be treated as a time-varying exposure or the analysis would define appropriate index dates and begin outcome observation afterward.
The authors themselves acknowledge that their analysis did not establish the timing of particular vaccinations relative to diagnoses.
So the problem isn’t simply that chronology wasn’t reported. The design can potentially assign pre-exposure disease to the eventual exposed group.
That seriously weakens causal interpretation.
“Strep throat” still isn’t really a clean strep-throat endpoint
Children’s Health Defense has emphasized the large association with recurrent “strep.”
But the paper’s diagnostic definition isn’t confined to streptococcal pharyngitis.
Yes, J02.0 represents streptococcal pharyngitis.
But the composite also includes much broader diagnostic families. Some cover nonspecific upper-respiratory infections, croup and other infections. The coding list even extends into categories for gonococcal and chlamydial disease.
Official ICD documentation distinguishes these diagnoses rather than treating them as synonymous with streptococcal throat infection. Centers for Medicare & Medicaid Services
StopAntiVaxPropaganda independently noticed this problem after my article appeared. Stop Anti Vax Propaganda
So if the strongest headline is going to be that vaccinated children had dramatically more recurrent strep throat, the endpoint should actually measure recurrent strep throat.
As written, it is a much broader diagnostic bucket.
And R19 is still not a respiratory disease
The paper’s respiratory category reportedly includes R19.
R19 is not a respiratory-disease family. It belongs to signs and symptoms involving the digestive system and abdomen.
Again, there may be an innocent explanation. Perhaps R19 was supposed to be R09. Perhaps it’s only a manuscript typo and wasn’t used in the code.
But that distinction matters.
If it’s merely a typo, correct the table. If R19 actually entered the analysis, rerun the analysis.
Other reviewers independently flagged this after my original article as well. Stop Anti Vax Propaganda
The list of nine outcomes doesn’t even stay consistent
The Methods and diagnostic-code table include nausea/vomiting.
The main outcome table instead contains sinusitis.
Yet no corresponding sinusitis definition appears where you would expect it.
Any single mistake like that could be mundane.
But we’re no longer dealing with one typo.
We now have an impossible vaccinated denominator, conflicting exposure definitions, an unusual “strep” composite, a gastrointestinal code in a respiratory outcome, and an outcome appearing in the result table that doesn’t match the outcome-definition table.
At some point “probably a typo” stops being an adequate response.
The paper needs a corrected version.
The “dose response” remains much less impressive than advertised
The dose-response argument sounds powerful until you actually look at the sequence of odds ratios.
For the paper’s strep endpoint: 1.5 → 2.2 → 3.4 → 2.7
That’s an increase followed by a decrease.
For respiratory disease: 1.2 → 1.6 → 2.0 → 0.50
At the highest exposure category, the association completely reverses.
For developmental delay: 0.82 → 0.66 → 0.74 → 1.4
The middle categories point toward lower odds before the highest group flips toward higher odds. Calling all of that a straightforward harmful dose-response relationship is difficult to justify.
And the interpretive asymmetry remains a problem.
When a vaccinated group appears healthier, the paper invokes possible selection effects or healthy-vaccinee bias.
When a vaccinated group appears less healthy, biological vaccine harm receives much more attention.
You can’t use confounding selectively.
If bias can create a protective association, it can also create a harmful one.
Multiple comparisons deserve more than a footnote
The paper examines nine outcomes overall, sex-specific analyses and four vaccine-exposure categories.
That produces dozens of statistical tests.
A rough count is about 63 comparisons if all nine outcomes are considered across the overall, sex-stratified and four dose-category analyses.
The authors state that they did not correct for multiple comparisons. That does not make every significant finding disappear.
But it matters particularly when highlighting one subgroup result because its p-value crossed 0.05 after examining many different combinations.
Take the much-publicized developmental-delay finding in the 10+ category: the Zenodo abstract reports OR 1.4, p = 0.0031. Zenodo
For illustration only, if all 63 comparisons were treated as one statistical family, a simple Bonferroni threshold would be approximately 0.0008.
Under that deliberately conservative approach, p = 0.0031 would not qualify.
I’m not arguing that Bonferroni across exactly 63 tests is necessarily the correct solution. Researchers can reasonably define multiplicity families differently, and less conservative procedures exist.
The point is that “p = 0.0031” does not exist in isolation when it was selected from dozens of analyses.
CHD still confuses odds with risk
This criticism also survives intact. The study reports odds ratios.
CHD and Hooker repeatedly translate them into language such as “60% more likely” or “double the risk.” Children’s Health Defense
For uncommon outcomes, odds ratios and risk ratios can be similar.
For common outcomes they can differ substantially, with odds ratios exaggerating the apparent relative change when interpreted as risk ratios. This is a longstanding statistical issue. BMJ
Using Table 4 as printed - notwithstanding its denominator problem - the respiratory proportions are roughly 50.8% versus 39.6%.
That’s about a 28% relative difference in observed proportions, not “60% higher risk.”
The 60% number comes from treating an OR of 1.6 as though OR and relative risk were interchangeable.
They aren’t.
CHD’s own dose-category reporting remains wrong
This wasn’t something critics imposed on the paper. It’s a discrepancy between the paper and CHD’s own coverage.
CHD reported the developmental-delay increase in the 6–9 dose group.
But Table 7 gives that group an OR of approximately 0.74.
The 1.4 result belongs to the 10+ category.
The same type of error occurs with rhinitis: the OR of 2.3 belongs to the 10+ group rather than the 6–9 group.
When an advocacy organization is using a paper produced by its own staff to make a public case, getting the exposure categories right seems like the minimum requirement.
What has happened since my original article?
This is worth emphasizing.
The criticism has not gone away under scrutiny. It has begun to be independently reproduced.
On October 2, Vaxopedia/StopAntiVaxPropaganda independently identified the oversized Table 4 vaccinated denominator, the inconsistent definition of “unvaccinated,” the broad strep coding, the misplaced R19 category and the lack of healthcare-utilization control. That article also linked back to my original analysis. Stop Anti Vax Propaganda
On October 6, Tech ARP independently identified the same 2,693 versus 2,414 problem and highlighted healthcare ascertainment and the unusually selected eight-practice sample. Tech ARP
Meanwhile, supportive promotion has continued. A post published October 3 presents the paper as evidence that earlier or greater childhood vaccination is not beneficial. Opinyuns
Steve Kirsch has also added the Yengst paper to a compilation whose headline describes the included work as being “all published in the peer-reviewed literature,” even though the Yengst paper is currently a Version v1 Zenodo upload, not a peer-reviewed journal publication. Kirsch Substack
And as of October 7, the Zenodo record still shows Version v1 with no visible corrected version. Zenodo
What would actually strengthen this study?
None of this requires dismissing the dataset. If anything, a cohort containing a large population of minimally vaccinated or unvaccinated children could be scientifically useful.
But usefulness depends on doing the analysis properly.
I would like to see the authors reconcile every Table 4 denominator; explicitly account for the 469 children first vaccinated after 13 months; document how vaccination status was validated outside the originating practices; report healthcare-visit frequency; stratify or adjust for practice; control important demographic, birth, and family confounders; use validated diagnostic definitions; make vaccine exposure time-varying so diagnoses cannot precede the exposure supposedly causing them; address multiplicity; publish analytic code; and provide enough de-identified data for independent reproduction.
Then rerun the study.
If the associations survive all of that, my assessment changes.
That would be genuinely interesting evidence.
The bottom line
I began with the position that this study should not be dismissed merely because it came from Children’s Health Defense or because it hasn’t been peer-reviewed.
I still hold that position. But scrutiny works both ways.
The deeper problem is that some of the paper’s basic numbers do not reconcile, its exposure groups are ambiguous, its disease definitions contain questionable coding choices, healthcare utilization isn’t controlled, important confounders aren’t adjusted, temporal order isn’t reliably established, and dozens of comparisons are interpreted without correction for multiplicity.
Since I first pointed these issues out, other reviewers have independently found several of the same things.
And the central Table 4 question remains unanswered:
How does a study with 2,414 vaccinated children perform an analysis containing 2,693 vaccinated children?
Before anyone tells me this paper proves vaccinated children are less healthy, answer that. Then show that the diagnosis happened after the vaccine.
Then show that the result survives healthcare-use adjustment, practice effects, confounding, corrected outcome definitions and a reproducible reanalysis.
That isn’t moving the goalposts. That’s the goalpost.
And if a result is real, it should survive getting the numbers right.
Resources
Yengst T, Ray H, Jablonowski K, Hooker BS. Comparative Health Outcomes of Vaccinated versus Unvaccinated Children: A Retrospective Cohort Analysis. Zenodo, September 26, 2026.
Primary source for the study discussed in this article.
Read the Zenodo paperMichael Nevradakis. “Vaccinated Kids Have Significantly Higher Odds of Developing Multiple Chronic Health Conditions.” Children’s Health Defense, September 29, 2026.
CHD’s coverage and interpretation of the study. Children’s Health Defense
Read the CHD articleVaxopedia / StopAntiVaxPropaganda. “Let’s Talk About Vaccinated vs Unvaccinated Children Again.” October 2, 2026.
Independent critique identifying the Table 4 denominator discrepancy, inconsistent vaccination definitions, ICD-code problems and healthcare-utilization concerns. Stop Anti-Vax Propaganda
Read the analysisAdrian Wong. “Fact Check: Children’s Health Defense’s Vaccinated vs Unvaccinated Children Study.” Tech ARP, October 6, 2026.
Independent review that also identifies the 2,693-versus-2,414 vaccinated-child discrepancy and medical-record ascertainment problems. Tech ARP
Read the Tech ARP fact-checkGlanz JM et al. A population-based cohort study of undervaccination in 8 managed care organizations across the United States. JAMA Pediatrics. 2013;167(3):274–281. DOI: 10.1001/jamapediatrics.2013.502.
Found different healthcare-utilization patterns among undervaccinated and age-appropriately vaccinated children. PubMed
PubMed recordDaley MF et al. Assessing misclassification of vaccination status: Implications for studies of the safety of the childhood immunization schedule. Vaccine. 2017;35(15):1873–1878. DOI: 10.1016/j.vaccine.2017.02.058.
Examines vaccination-status misclassification in electronic medical records and outside-provider use among undervaccinated families. PubMed
PubMed record
Free full text at PubMed CentralNewcomer SR et al. Assessing Potential Confounding and Misclassification Bias When Studying the Safety of the Childhood Immunization Schedule. Academic Pediatrics. 2018.
Shows why confounding and accurate vaccination classification must be explicitly addressed in schedule-level observational studies. PubMed
PubMed recordGlanz JM et al. White Paper on studying the safety of the childhood immunization schedule in the Vaccine Safety Datalink. Vaccine. 2016;34 Suppl 1:A1–A29. DOI: 10.1016/j.vaccine.2015.10.082.
Discusses suitable designs, exposure definitions, confounding and bias when comparing different childhood vaccination patterns. PubMed
PubMed recordInstitute of Medicine. The Childhood Immunization Schedule and Safety: Stakeholder Concerns, Scientific Evidence, and Future Studies. 2013.
Major National Academies review of how whole-schedule vaccine-safety questions can be investigated. National Academies Publications
National Academies report pageDavies HTO, Crombie IK, Tavakoli M. “When can odds ratios mislead?” BMJ. 1998;316:989. DOI: 10.1136/bmj.316.7136.989.
Explains why odds ratios should not automatically be interpreted as relative risks, particularly when outcomes are common. BMJ
Read the BMJ articleGrant RL. “Converting an odds ratio to a range of plausible relative risks for better communication of research findings.” BMJ. 2014;348:f7450. DOI: 10.1136/bmj.f7450.
Further discussion of the distinction between odds ratios and relative risks. BMJ
Read the BMJ articleCenters for Medicare & Medicaid Services. ICD-10-CM Codes.
Official ICD-10 coding resources useful for checking the diagnostic categories used in the Yengst paper. Centers for Medicare & Medicaid Services
CMS ICD-10-CM resourcesU.S. Food and Drug Administration. “Common Ingredients in FDA-Approved Vaccines.”
Official reference for vaccine ingredients, including aluminum-containing adjuvants. fda.gov
FDA vaccine-ingredient informationHooker BS, Miller NZ. “Analysis of health outcomes in vaccinated and unvaccinated children: Developmental delays, asthma, ear infections and gastrointestinal disorders.” SAGE Open Medicine. 2020;8. DOI: 10.1177/2050312120925344.
Hooker and Miller’s earlier vaccinated-versus-unvaccinated study using records from three U.S. medical practices. PubMed
Read the paper on PubMedSAGE. “Corrigendum to ‘Analysis of health outcomes in vaccinated and unvaccinated children.’” 2026.
The correction added a previously omitted acknowledgement that the Pediatric Health Outcomes Initiative referred the physicians from the three practices supplying the EMR data and that one physician was a PHOI board member. The authors stated PHOI did not handle the data, design or financing. Sage Journals
Read the SAGE corrigendumSAGE. “Publisher’s note to ‘Analysis of health outcomes in vaccinated and unvaccinated children.’” August 28, 2026.
SAGE says reader concerns led to an Expression of Concern. One concern was addressed by the corrigendum; another concerned possible reuse of data in later articles without disclosure. SAGE said it lacked enough information to resolve the latter and called its investigation inconclusive. The publisher’s note and corrigendum replaced the earlier Expression of Concern. Sage Journals
Read the SAGE publisher’s noteHooker BS, Miller NZ. “Health effects in vaccinated versus unvaccinated children, with covariates for breastfeeding status and type of birth.” Journal of Translational Science. 2021;7. DOI: 10.15761/JTS.1000459.
A second Hooker-Miller vaccinated-versus-unvaccinated analysis, this time using survey data associated with three medical practices and models incorporating breastfeeding and birth type. OAText
Read the 2021 paperKennedy RF Jr., Hooker BS. Vax-Unvax: Let the Science Speak. Skyhorse Publishing, 2023.
Hooker co-authored a full book centered on comparisons between vaccinated and unvaccinated populations, demonstrating that this research question and argument long predate Yengst et al. The publisher describes the book as drawing on more than 100 studies involving vaccinated-versus-unvaccinated comparisons. Skyhorse Publishing
Publisher’s page for Vax-UnvaxScience Feedback. “Claim by Brian Hooker that unvaccinated children are healthier compared to vaccinated children relies on flawed studies.” July 28, 2023.
An independent review of studies Hooker publicly cited in support of the claim that unvaccinated children are healthier, discussing selection bias, confounding and other methodological limitations. Science Feedback
Read the Science Feedback review