Anatomy of a Failure: Why This Latest Vaccine-Autism Paper is Dead Wrong

A methodological autopsy of flawed research that contradicts decades of rigorous evidence

This month, a paper published in "Science, Public Health Policy and the Law" (a WordPress blog, not a scientific journal) claimed that vaccinated children in Florida's Medicaid program had significantly higher rates of neurodevelopmental disorders, including autism. This study by Mawson and Jacob, funded by the National Vaccine Information Center (an anti-vaccine advocacy organization), has become emblematic of how flawed methodology and biased analysis can be used to promote vaccine hesitancy. Despite being a repeat of previously retracted work and contradicting decades of rigorous research showing no link between vaccines and autism, this paper is circulating widely in parent groups, and has gone totally viral on social media…

So, we teamed up with the amazing Dr. Bertha Hidalgo— epidemiologist, and fellow data lover— to do a deep dive on this paper. Let’s discuss…

---

STUDY CLAIM

Claims that vaccinated children in Florida Medicaid program (1999-2011) had significantly higher rates of neurodevelopmental disorders (NDDs) including autism, and that more vaccination visits correlated with higher autism risk.

Note: This conflicts with decades of rigorous analysis of millions of data points that show absolutely zero link between vaccines and autism.

STUDY DESIGN

• Observational study using Florida Medicaid billing claims data

• Population: 47,155 children enrolled in Florida Medicaid from birth to age 9

• Compared "vaccinated" (defined by presence of vaccine billing codes) vs "unvaccinated" (absence of codes)

• Used billing codes to identify NDDs and vaccination status

CRITICAL ISSUES

1. Publication & Funding Red Flags

• Published on WordPress blog, not peer-reviewed journal

• Funded by the National Vaccine Information Center (NVIC), a known anti-vaccine organization

• "Peer-reviewed" by Peter McCullough, known for promoting medical misinformation

• Conflicts with extensive body of rigorous research showing no vaccine-autism link

The study's fundamental flaws begin with its publication process. Legitimate scientific research undergoes rigorous peer review, where independent experts evaluate the methodology, analysis, and conclusions before publication in established journals. This work instead appeared on a WordPress blog that is not a peer-reviewed scientific journal, with review conducted by Peter McCullough, who has promoted medical misinformation. The lead author, Anthony Mawson, has a documented history of retracted vaccine research. Combined with funding from an anti-vaccine advocacy organization (NVIC), these factors raise serious concerns about the study's scientific validity and reliability.

The study's legitimacy is fundamentally compromised by its publication venue (a WordPress blog) and funding source (NVIC, an anti-vaccine organization). Lead author Anthony Mawson has a pattern of conducting methodologically flawed vaccine research, with multiple retractions. This study was "peer-reviewed" by Peter McCullough, a known promoter of medical misinformation. Thus, it bypassed legitimate scientific scrutiny while contradicting extensive peer-reviewed research showing no vaccine-autism link.

2. Data Collection & Classification Problems

The study's foundation rests on problematic data collection and classification methods. It relies solely on billing codes designed for payment, not research. Their 'unvaccinated' classification simply means no vaccine billing codes appear in Medicaid records - missing vaccines given through other programs (including Vaccines for Children), providers, or states. With no verification of vaccination status and no validation of diagnostic codes, this classification error fundamentally undermines their analysis.

A separate methodological flaw is their failure to account for basic confounding variables. For example, in the classic epidemiology example, ice cream sales correlate with drowning rates due to summer weather driving both. The authors' data shows a classic healthcare utilization bias: children with more medical visits are naturally more likely to have documented vaccines and developmental diagnoses. The authors fail to address this fundamental confounding factor.

Most critically, the authors' data claims contain a fundamental contradiction: they assert having 'individual-level data' through DEVEXI to track children across multiple visits using unique identifiers, while simultaneously admitting they lack individual-level data on specific vaccines. If they truly had individual-level Medicaid data with unique identifiers, they should be able to track specific vaccines per child. This discrepancy suggests either a significant misunderstanding of their data capabilities or concerning transparency issues about their methodology. Without clarity on these fundamental data problems, their findings cannot be validated or understood, adding another layer of doubt to a study already compromised by methodological flaws.

3. Selection Bias & Confounding

The study's population selection introduces profound limitations. By examining only Florida Medicaid recipients, the researchers restricted their analysis to a narrow, unrepresentative slice of children - specifically those from low-income families who maintained continuous Medicaid coverage for nine years. This selection immediately raises questions about generalizability.

Even more concerning is their complete failure to account for critical confounding variables. A proper analysis would need to consider family history, parental age, socioeconomic factors, environmental exposures, and crucially, changes in how autism was diagnosed and documented over their 12-year study period. Without controlling for these factors, any perceived associations become essentially meaningless.

The researchers' puzzling decision to focus on 9-year-olds creates multiple methodological problems. While they claim this age was chosen to align with CDC's use of 8-year-olds for autism prevalence monitoring, this choice introduces severe biases. Autism is typically diagnosed between ages 2-4, meaning their approach misses early diagnoses entirely. They also create a 'survival bias' by excluding children who left Medicaid before age 9, artificially extending the time gap between vaccination and diagnosis, and ignoring how diagnostic practices evolved during their study period.

A fundamental epidemiological issue pervades their analysis: healthcare utilization bias. Children who receive vaccinations typically have more frequent medical visits, naturally increasing their likelihood of receiving developmental screenings and subsequent diagnoses. Conversely, unvaccinated children often have less healthcare engagement, potentially leading to underdiagnosis. The researchers' failure to address this basic confounding factor is particularly troubling.

This design appears to enable multiple forms of cherry-picking: they can ignore earlier diagnoses, create artificial associations through extended time windows, and selectively analyze data that supports their predetermined conclusions. Most tellingly, they offer no biological mechanism or explanation for how vaccines might be linked to autism - a crucial missing piece that highlights the speculative nature of their claims.

The culmination of these methodological choices - the restricted population, uncontrolled confounders, arbitrary age cutoff, and unaddressed biases - renders any conclusions from this study essentially meaningless. Each of these issues alone would raise serious concerns; together, they reveal a study design that appears engineered to find associations rather than understand causation.

4. Statistical Analysis Flaws

The study's statistical analysis fails at multiple levels. Let's examine their data presentation first: while their basic prevalence calculations appear mathematically correct (2.6% overall autism prevalence representing 1,226 of 47,155 children, and 4.4% among preterm births representing 220 of 5,009 children), these descriptive statistics only scratch the surface of what's needed for a proper analysis of their claims.

Most concerning is their treatment of temporal data. Despite having access to twelve years of data (1999-2011), the researchers failed to leverage this longitudinal richness. Their analysis ignores crucial changes in diagnostic practices that occurred during this period, making no attempt to account for how autism diagnosis patterns evolved over time. Even basic temporal analysis could reveal whether apparent associations held steady or were artifacts of changing diagnostic practices.

Their statistical methodology appears frozen in time - treating data from 1999 the same as data from 2011, despite dramatic changes in autism diagnosis and documentation practices. They rely on crude odds ratios and problematic metrics like 'vaccination visits' without proper adjustments or sensitivity analyses. These basic measures fail to capture the complexity of the relationship between healthcare encounters and diagnoses.

A robust analysis would need to:

• Consider the timing of vaccinations relative to diagnoses

• Examine trends in diagnosis rates over time

• Address temporal confounding factors

• Employ proper statistical adjustments

• Include sensitivity analyses

• Account for the nested structure of their longitudinal data

The absence of these fundamental statistical considerations renders their conclusions invalid. Their approach demonstrates either a concerning lack of statistical sophistication or a deliberate choice to avoid more rigorous methods that might fail to support their predetermined conclusions.

5. Causal Inference Problems

The study's causal inference approach demonstrates fundamental misunderstandings of epidemiological principles. Beyond making inappropriate causal claims from observational data, their analysis appears to have basic directional errors - potentially even flipping their variables in SAS, casting doubt on their reported odds ratios. More fundamentally, they ignore critical issues of temporality and reverse causality. Children with developmental concerns often have different healthcare utilization patterns, and families may adjust their vaccination decisions based on developmental observations - factors completely unaddressed in their analysis. The glaring absence of a proper control group and their failure to establish temporal sequences between vaccinations and diagnoses makes any causal claims impossible to substantiate.

6. Data Quality & Transparency

The data quality and transparency issues are equally concerning. The authors claim to track individual-level data while simultaneously admitting they cannot track specific vaccines - a contradiction that undermines their entire analytical approach. This becomes particularly problematic in their Table 10 analysis, where they make claims about vaccination patterns despite acknowledging they cannot track individual data. Beyond these internal inconsistencies, the study provides no access to underlying data, insufficient methodological details, and no possibility for external validation. The inability to verify vaccination status accuracy or reproduce their analysis violates basic principles of scientific transparency.

7. Research Design Issues

The research design itself appears engineered to find associations rather than test hypotheses objectively. Without proper control groups, randomization (where feasible), or adjustments for diagnostic changes over time, the design lacks basic elements necessary for valid inference. The poor definition of study periods and inadequate follow-up procedures suggest a concerning pattern: methodological choices that systematically bias toward finding associations without regard for scientific rigor.

Final thoughts…

This study demonstrates significant methodological problems: publication without standard peer review, funding from advocacy organizations with clear agendas, fundamental flaws in design and analysis, and conclusions that contradict extensive scientific evidence. The scientific consensus remains clear and unambiguous - vaccines do not cause autism.

The scientific community is frankly exhausted from repeatedly addressing this thoroughly debunked claim. Despite decades of rigorous research examining millions of children and consistently finding no link between vaccines and autism, we continue to see methodologically flawed studies attempting to resurrect this discredited hypothesis. The time and resources spent repeatedly debunking these claims could be better invested in advancing our understanding of autism's actual causes and developing more effective supports for autistic individuals. We must prioritize evidence-based research that genuinely serves public health and the autism community.

Stay Curious,

Unbiased Science (and Dr. Bertha Hidalgo)