What the Vaccine Adverse Event Reporting System is, what it isn't, and how to interpret its data correctly
In 1996 the Vaccine Adverse Events Reporting System (VAERS) is officially set up in the USA as a passive surveillance system co-managed by the CDC and FDA that collects reports of adverse events occurring after vaccination. The Vaccine Safety Datalink (VSD) is also set up as a collaborative project between CDC's Immunization Safety Office and nine healthcare organizations to monitor vaccine safety.
The Vaccine Adverse Event Reporting System (VAERS) is one of the most frequently cited — and most frequently misunderstood — sources of vaccine safety data.
Understanding what VAERS is designed to do, and what it cannot tell us, is essential for interpreting any data drawn from it correctly.
VAERS is designed to detect potential safety signals — new or unexpected patterns of adverse events that warrant further investigation. It is not designed to prove causation.
Reports are submitted voluntarily by patients, caregivers, healthcare providers, and manufacturers. Unlike active surveillance, VAERS does not seek out cases — it receives them.
All VAERS data is publicly available and searchable at vaers.hhs.gov. This transparency is a deliberate feature enabling independent researchers to analyze trends.
Healthcare providers and vaccine manufacturers are required by law to report certain adverse events to VAERS. Patients and caregivers may also submit reports voluntarily.
Anyone can submit a VAERS report — patients, parents, caregivers, healthcare providers, and vaccine manufacturers. Healthcare providers are legally required to report:
Epidemiologists look for unexpected patterns or clusters of events that occur more frequently than background rates would predict. This triggers formal investigation, not automatic conclusions.
VAERS data generates hypotheses tested using active surveillance systems like the Vaccine Safety Datalink (VSD) and the BEST System, which can establish causation.
Tracking changes in report rates over time can reveal emerging signals — for example, an unusual increase in a specific event type following a new vaccine rollout.
Peer-reviewed researchers use VAERS as one data source among many, always with appropriate methodological controls and caveats about its limitations.
VAERS captures only a fraction of actual adverse events. How large that fraction is depends on the type and severity of the event. For serious events such as anaphylaxis and Guillain-Barré syndrome, published studies estimate that VAERS captures between 12% and 76% of cases, depending on the vaccine and comparison data source (Shimabukuro et al., 2020). For mild events such as injection-site soreness or low-grade fever, estimates fall below 1% (Rosenthal & Chen, 1995; Lazarus et al., 2010).
A frequently cited claim that 'fewer than 1% of adverse events are reported' draws on multiple sources that each measured something different — drugs vs. vaccines, serious vs. mild events, and different surveillance systems. These distinctions are important and are examined in detail on the VAERS Reporting Efficacy page. This means:
The question of how completely VAERS captures adverse events is central to interpreting its data correctly. VaccinationFacts.com has produced a dedicated reference page examining this topic in depth — including published sensitivity estimates by event type, the three distinct sources behind the commonly cited '1%' figure, documented signal-detection successes, and how VAERS findings are verified through stronger follow-up systems such as the Vaccine Safety Datalink (VSD).
The sections above explain what VAERS is designed to do, what it cannot do, who can submit reports, and why reporting completeness varies. But a critical question remains: when someone reports an adverse event to VAERS, how do scientists determine whether the vaccine actually caused it?
This is the question of causality — and it is the single most important concept for interpreting VAERS data correctly. Without understanding how causality is assessed, it is easy to mistake a temporal association (an event that happened after vaccination) for a causal relationship (an event caused by vaccination). That distinction is at the center of every major debate about vaccine safety data.
The sections that follow explain the frameworks, study designs, institutional systems, and legal standards used to evaluate whether a vaccine caused a specific adverse event — and where those frameworks are themselves debated.
Causality — also called causal inference — means establishing that one event directly caused another. In vaccine safety, this is the difference between two very different statements:
"Event X happened after Vaccine Y" — a temporal association
"Vaccine Y caused Event X" — a causal relationship
The first statement is an observation. The second is a conclusion that requires evidence beyond timing alone.
The assumption that because one event follows another, the first must have caused the second is a reasoning error known in formal logic as post hoc ergo propter hoc — Latin for "after this, therefore because of this." The Institute of Medicine identified this fallacy directly in its 1994 report on childhood vaccine adverse events, noting that "the mere fact that B follows A does not mean that A caused B" (IOM, Adverse Events Associated with Childhood Vaccines, 1994).
The reason timing alone is unreliable is that adverse health events happen every day in every population, regardless of vaccination status. Epidemiologists call these ordinary expected events "background rates." When millions of people receive a vaccine over a short period, some will experience heart attacks, strokes, new diagnoses, and deaths in the days following vaccination — events that would have occurred regardless.
Researchers at the CDC calculated what this looks like in practice. Among 10 million vaccinated people, purely by coincidence from background rates: approximately 237 would die from any cause within one day of vaccination, approximately 1,656 within seven days, and approximately 9,933 within 42 days — all from causes entirely unrelated to the vaccine (Abara et al., Journal of Infectious Diseases, 2022).
This does not mean that vaccines never cause harm. It means that the observation of an adverse event after vaccination is a starting point for investigation, not proof that the vaccine was the cause.
The University of Minnesota's Center for Infectious Disease Research and Policy (CIDRAP) offers a widely used teaching example: autistic symptoms typically become apparent in the second year of life, around the same time children receive several recommended vaccines. Children also learn to walk during the second year of life. No one concludes that walking causes autism. The temporal coincidence between vaccination schedules and the natural developmental window for autism recognition illustrates why scientists require evidence beyond timing before concluding causation. CIDRAP presents this analogy not to minimize vaccine concerns, but as a teaching heuristic to illustrate the mathematical certainty of coincidental timing in large populations (CIDRAP, 2024).
When evaluating any claim that a vaccine caused harm, three questions are useful starting points:
No single question settles the matter, but together they distinguish strong causal evidence from temporal coincidence.
In 1965, British epidemiologist Sir Austin Bradford Hill delivered an address to the Royal Society of Medicine titled "The Environment and Disease: Association or Causation?" In it, he proposed nine viewpoints for evaluating whether an observed statistical association between an exposure and a disease should be interpreted as causal (Hill, Proceedings of the Royal Society of Medicine, 1965).
Hill deliberately called these "viewpoints," not "criteria," and stated explicitly: "None of my nine viewpoints can bring indisputable evidence for or against the cause-and-effect hypothesis and none can be required as a sine qua non." Despite this, they have become one of the most widely cited frameworks in epidemiology and are routinely applied in vaccine safety assessment.
| Viewpoint | What It Asks | Vaccine Example |
|---|---|---|
| Strength of Association | How large is the measured effect? Larger effects are harder to explain by bias alone. | The relative risk of intussusception within 3–7 days of RotaShield vaccination was approximately 37-fold above background — a strong association. |
| Consistency | Has the association been observed repeatedly by different researchers, in different populations? | Myocarditis following mRNA COVID-19 vaccines was identified independently in the U.S., Israel, and multiple Nordic countries. |
| Specificity | Is the exposure associated with a particular outcome rather than many unrelated outcomes? | Hill himself called this the weakest viewpoint, since one cause can have multiple effects. |
| Temporality | Did the exposure precede the outcome? | This is the only viewpoint universally considered essential. The vaccine must have been administered before the adverse event occurred. |
| Biological Gradient | Does a dose-response relationship exist? | Risk differences between first and second mRNA vaccine doses for myocarditis serve as a functional equivalent of dose-response. |
| Plausibility | Is there a biologically plausible mechanism? | Hill cautioned that plausibility "depends upon the biological knowledge of the day" and should not be used to dismiss unfamiliar associations. |
| Coherence | Does the causal interpretation fit with what is known about the disease's natural history? | The causal link between OPV and paralytic polio was coherent with known mechanisms of viral reversion in the intestine. |
| Experiment | Does removing or modifying the exposure change the outcome? | After RotaShield was withdrawn in 1999, intussusception rates associated with rotavirus vaccination declined — supporting causation. |
| Analogy | Are similar agents already known to cause similar effects? | The occurrence of thrombotic events with the Janssen COVID-19 vaccine was analogous to heparin-induced thrombocytopenia, a known immune-mediated clotting disorder. |
The Bradford Hill viewpoints remain influential more than 60 years after their introduction. They are used by the WHO Global Advisory Committee on Vaccine Safety, referenced in the IOM/NASEM causality framework, and cited in peer-reviewed vaccine safety studies worldwide.
However, they are one framework among several for reasoning about causation. Modern causal inference methods have developed alongside the Bradford Hill viewpoints. These include directed acyclic graphs (DAGs), which visually map assumed causal relationships; sufficient-component cause models, which describe how multiple factors can combine to produce an outcome; and the GRADE system, which rates the certainty of evidence (Shimonovich et al., European Journal of Epidemiology, 2021).
The viewpoints are best understood as a structured way to organize evidence and think carefully about causation — not as a checklist where a fixed number of boxes must be checked before causation can be accepted or rejected.
Phillips and Goodman, writing in Epidemiologic Perspectives & Innovations (2004), observed that Hill's paper "is almost exclusively cited as the source of the 'Bradford-Hill criteria' for inferring causation, despite Hill's explicit statement that cause-effect decisions cannot be based on a set of rules." They argued that the viewpoints are sometimes applied as threshold requirements — demanding that multiple viewpoints be satisfied before any causal inference is permitted — in a way Hill did not intend.
Epidemiologists Kenneth Rothman and Sander Greenland have argued that the strength-of-association viewpoint can be misleading. A small measured effect does not necessarily mean there is no causal relationship — confounding, measurement error, and study design can all reduce the apparent size of a real effect. Conversely, large associations can sometimes arise from bias rather than causation. Rothman and Greenland's position, articulated across multiple editions of their textbook Modern Epidemiology and in subsequent commentary, is that no single viewpoint — including strength — should be treated as dispositive.
Shimonovich et al. (European Journal of Epidemiology, 2021) systematically compared the Bradford Hill viewpoints against three modern causal inference approaches and found "enduring importance" for four viewpoints (strength, temporality, plausibility, and experiment) but "limited utility" for coherence and analogy specifically. They concluded that the viewpoints remain useful but benefit from being supplemented with more formal methods.
Fedak et al. (Emerging Themes in Epidemiology, 2015) reached a similar conclusion, stating that the viewpoints "should not be used as a heuristic for assessing causation in a vacuum; rather they should be viewed as a list of possible considerations."
A 2025 preprint applied Kuhnian ideas about how scientific frameworks shape interpretation to argue that the methodological principle "correlation does not imply causation" — while essential as a safeguard — was applied so restrictively during the COVID-19 pandemic that it hindered legitimate investigation of serious adverse events with temporal proximity to vaccination. The authors argued that established methods can sometimes make it harder to recognize new patterns while evidence is still emerging. This paper is a preprint and has not been peer-reviewed, but it represents a perspective that has appeared in academic commentary on pandemic-era pharmacovigilance (Preprints.org, 2025).
The academic debate over Bradford Hill does not mean the viewpoints are useless. It means they are tools that require judgment in application. The strongest causal conclusions draw on multiple viewpoints, multiple studies, and multiple methods — and the weakest rely on any single viewpoint or any single study alone.
If Bradford Hill helps organize causal reasoning, study design helps determine how much weight the evidence deserves.
Different types of studies provide different levels of evidence for causal claims. This hierarchy is widely recognized in evidence-based medicine and is used by the National Academies, the WHO, and regulatory agencies when evaluating vaccine safety.
Not all evidence is equally useful for answering the question "Did this vaccine cause this event?" A case report describing one patient's experience after vaccination is different from a study comparing outcomes in 500,000 vaccinated and unvaccinated people. The difference lies in how well the study design controls for other explanations.
| Study Design | What It Does | Main Strength | Main Limitation |
|---|---|---|---|
| Systematic Reviews & Meta-Analyses | Pool results from multiple studies to assess overall evidence | Largest evidence base; reduces random variation | Quality depends on the underlying studies |
| Randomized Controlled Trials (RCTs) | Randomly assign participants to receive vaccine or placebo | Controls for both known and unknown confounders | Cannot detect events rarer than ~1 per 10,000 doses; ethical constraints post-licensure |
| Cohort Studies | Follow vaccinated and unvaccinated groups over time | Can calculate incidence rates and relative risk | Subject to "healthy vaccinee bias" — healthier people are more likely to be vaccinated |
| Self-Controlled Case Series (SCCS) | Compare adverse event rates during risk windows after vaccination against control periods within the same individual | Each person serves as their own control, eliminating all time-invariant confounders | Only works for acute, transient outcomes with definable risk windows |
| Case-Control Studies | Compare vaccination histories of people who experienced the event to matched controls who did not | Efficient for studying rare outcomes | Vulnerable to selection bias and recall bias |
| Ecological Studies | Compare population-level vaccination rates against population-level disease rates | Can identify broad patterns | Cannot establish individual-level causation (ecological fallacy) |
| Case Reports / Case Series | Describe individual patient experiences | Can generate hypotheses and identify novel events | Cannot establish causation; no comparison group; no denominator |
Pre-licensure vaccine trials — studies conducted before a vaccine is approved for general use — typically enroll tens of thousands of participants. They are large enough to detect common adverse events, but not large enough to detect very rare events. A trial enrolling 60,000 participants cannot reliably detect an adverse event occurring at a rate below roughly 1 per 10,000 doses. The Janssen COVID-19 vaccine's thrombosis with thrombocytopenia syndrome (TTS), which occurred at approximately 3.8 per million doses, could not have been detected in any pre-licensure trial of feasible size. This is why post-licensure surveillance systems exist.
While pre-licensure trials are limited by size regarding rare events, their randomized design remains one of the most effective methods for isolating a vaccine's effect from confounding factors — a strength that observational post-licensure studies cannot fully replicate.
Before causality can be assessed, researchers must agree on what constitutes the adverse event being studied. The Brighton Collaboration, an international network of more than 500 experts from 57 countries established in 2000, develops standardized case definitions for adverse events following immunization. These definitions specify diagnostic criteria at multiple levels of certainty — for example, defining how much clinical evidence is needed to classify a case of anaphylaxis with high, medium, or lower certainty.
Without standardized definitions, studies conducted in different countries or clinical settings may classify the same clinical presentation differently, making it impossible to compare results or pool data for causality assessment (Kohl et al., Advances in Patient Safety, 2005; Brighton Collaboration).
Two major institutional frameworks dominate how vaccine-adverse event causality is formally evaluated: the U.S. National Academies framework and the World Health Organization's AEFI causality assessment algorithm.
The Institute of Medicine (now the National Academies of Sciences, Engineering, and Medicine) developed the most systematic framework used in the United States for evaluating whether vaccines cause specific adverse events. The framework uses four causality categories, refined across reports from 1991 to 2024.
| Category | What It Means |
|---|---|
| "Evidence establishes a causal relationship" | The totality of evidence suggests vaccination can cause this harm, and further research is unlikely to change the conclusion. |
| "Evidence favors acceptance of a causal relationship" | Evidence suggests vaccination might cause this harm, but meaningful uncertainty remains and future studies could lead to a different conclusion. |
| "Evidence is inadequate to accept or reject a causal relationship" | Available evidence is too limited, biased, imprecise, or inconsistent to draw meaningful conclusions either way. This also applies when no relevant studies exist. |
| "Evidence favors rejection of a causal relationship" | Evidence suggests vaccination does not cause this harm, but meaningful uncertainty remains. |
These categories describe how strong the evidence is, not whether an event feels medically serious.
An important design feature: no category exists for definitively establishing that a vaccine does not cause an adverse event. The 2012 IOM committee explained that "it is virtually impossible to prove the absence of a relationship with the same certainty that is possible in establishing the presence of one" (Adverse Effects of Vaccines: Evidence and Causality, 2012).
Most hypothesized associations lacked sufficient evidence to draw conclusions in either direction. The 2012 IOM report examined 158 vaccine-adverse event pairs across eight vaccines. Of these, 14 were classified as "convincingly supports" a causal relationship (including anaphylaxis following multiple vaccines and febrile seizures after MMR), 4 as "favors acceptance," 5 as "favors rejection" (including MMR and autism), and 135 — the vast majority — as "inadequate to accept or reject" (IOM, 2012).
The 2024 NASEM report, focusing specifically on COVID-19 vaccines, reached 85 distinct conclusions across four platform types, supplementing the broader 2012 review of routine childhood and adult vaccines. It established a causal relationship between mRNA vaccines and myocarditis, found evidence favoring acceptance of causal relationships between the Janssen vaccine and both Guillain-Barré syndrome and thrombosis with thrombocytopenia syndrome, and favored rejection for 12 vaccine-event pairs including mRNA vaccines and myocardial infarction and female infertility (NASEM, 2024).
The World Health Organization developed a separate causality assessment system for evaluating individual adverse events following immunization (AEFI). The revised WHO algorithm, endorsed by the Global Advisory Committee on Vaccine Safety (GACVS) in 2012 and updated in 2019, classifies individual AEFI cases into four categories: Consistent with causal association, Inconsistent with causal association, Indeterminate, or Unclassifiable (WHO, Causality Assessment of AEFI: User Manual, 2019).
Unlike the IOM/NASEM framework, which evaluates bodies of evidence across populations, the WHO algorithm is designed for individual case assessment — determining whether a specific patient's adverse event is consistent with a causal association to the vaccine they received.
The primary strength of the WHO AEFI algorithm is its role as a global standardized tool. By providing a common framework for 194 member states, it allows for the aggregation of safety data across diverse populations that would otherwise remain siloed and difficult to compare.
Both frameworks have drawn peer-reviewed criticism.
Puliyel and Naik (F1000Research, 2018) argued that the revised WHO AEFI classification creates a structural paradox: only events already acknowledged in prior epidemiological studies can be classified as vaccine-related. Events observed during post-marketing surveillance that were not seen in Phase 3 trials are categorized as "coincidental" or "unclassifiable." The authors also noted that the WHO definition of causal association requires "no other factor intervening in the process," which effectively excludes cases where a vaccine contributes to harm alongside pre-existing conditions — even when WHO itself advises clinical precautions for those conditions.
Bellavite et al. (PMC, 2024) extended this critique to COVID-19 vaccines specifically, arguing that the WHO algorithm's requirement that an adverse event type be identified in large-scale studies before a causal association can be acknowledged creates circular reasoning for novel vaccines, where post-marketing data is the primary safety evidence. In a separate paper, Bellavite (F1000Research, 2020) argued that most vaccine adverse reactions involve multifactorial pathology — excessive or biased inflammatory responses interacting with genetic susceptibility and pre-existing conditions — and that the WHO algorithm's treatment of "other causes" as exclusion criteria rather than interacting factors systematically underestimates vaccine contribution to some adverse events.
These critiques do not mean formal causality frameworks are unnecessary. They show why structured systems are valuable, while also highlighting areas where some researchers believe the frameworks can miss new, complex, or interacting causes.
Scientific causation and legal causation answer different questions, for different purposes, under different standards of proof. Understanding this distinction matters because vaccine injury compensation decisions are sometimes cited as evidence of scientific causation — or vice versa — when the two systems are designed to operate independently.
The Vaccine Injury Compensation Program (VICP), established by the National Childhood Vaccine Injury Act of 1986 and operational since October 1988, uses a legal standard for causation that is intentionally lower than the scientific standard. Congress designed it this way after vaccine injury lawsuits in the 1980s threatened the vaccine supply.
For injuries listed on the Vaccine Injury Table — a legal list of covered injuries and the time periods in which they must occur after vaccination — causation is legally presumed. If a petitioner received a covered vaccine and sustained a listed injury within the specified period, the government bears the burden of proving an alternative cause.
For injuries not listed on the Table — often called "off-Table" claims — which now account for approximately 90% of all VICP claims (Grey, Harvard Journal on Legislation, 2011), petitioners must prove causation under the Althen test, established in Althen v. Secretary of Health and Human Services, 418 F.3d 1274 (Fed. Cir. 2005). The three prongs require:
The standard of proof is preponderance of the evidence — a legal term meaning "more likely than not," or greater than 50% probability. Petitioners are not required to provide epidemiological studies or published medical literature; medical expert opinion is sufficient. This intentional lowering of the evidentiary bar ensures that individuals are not denied compensation simply because a published scientific study has not yet addressed their specific injury. As the Federal Circuit held: requiring identification and proof of specific biological mechanisms "would be inconsistent with the purpose and nature of the vaccine compensation program" (Knudsen v. Secretary of HHS, 1994).
| Dimension | VICP Legal Standard | Scientific Standard |
|---|---|---|
| Threshold | Greater than 50% probability ("more likely than not") | A result unlikely to be due to chance (statistical significance, typically p < 0.05), replicated across studies |
| Evidence required | Medical expert opinion sufficient | Controlled epidemiological studies expected |
| Focus | Individual case ("Did this vaccine cause this patient's injury?") | Population-level patterns ("Does this vaccine increase risk of this event?") |
| Close calls | Resolved in favor of the petitioner | Remain classified as uncertain ("inadequate to accept or reject") |
| Epidemiological data | Not required | Central to assessment |
Approximately 60% of all VICP compensation has come from negotiated settlements where HHS has not concluded that the vaccine caused the injury (Grey, Harvard Journal on Legislation, 2011). Since 1988, the program has compensated over 12,500 petitions totaling approximately $5.5 billion, funded by a $0.75 excise tax per vaccine dose (HRSA, VICP Data & Statistics, 2025).
This gap between legal and scientific causation is by design, not a system failure. The two systems answer different questions and can therefore reach different kinds of conclusions without necessarily being in conflict. A VICP compensation award does not mean science has established that the vaccine caused the injury. Equally, the absence of scientific proof does not mean the legal system should deny compensation to an individual whose case meets the legal standard.
The following cases illustrate how the causality assessment process works in practice — from initial signal to final determination and policy action. They are selected to show the system producing different outcomes in different directions.
| Vaccine / Event | Initial Signal | Study Evidence | IOM/NASEM Category | Policy Outcome |
|---|---|---|---|---|
| OPV / Paralytic polio | Clinical case identification; vaccine-strain virus isolated | Genetic sequencing confirmed >99% homology with OPV strain; ~1 per 2.4M doses | "Evidence establishes a causal relationship" (IOM, 1991) | ACIP adopted IPV-only schedule in 2000 |
| RotaShield / Intussusception | VAERS reports within months of 1998 licensure | ~37-fold increased risk within 3–7 days; ~1 per 10,000 infants | "Evidence establishes a causal relationship" | ACIP withdrew recommendation October 1999; product withdrawn |
| MMR / Autism | 1998 Lancet case series (later retracted) | Madsen et al. (2002, 537,303 children, adj RR 0.92); Hviid et al. (2019, 657,461 children) | "Evidence favors rejection" (IOM, 2004; reaffirmed 2012) | Universal MMR recommendation maintained; paper retracted 2010 |
| mRNA COVID-19 / Myocarditis | VAERS reports and Israeli surveillance, April 2021 | Multiple systems in U.S., Israel, Nordic countries confirmed elevated risk in younger males after dose 2 | "Evidence establishes a causal relationship" (NASEM, 2024) | ACIP updated guidance; benefits outweighed risk; most cases mild |
| Janssen / TTS | VAERS reports within days of initial doses | Estimated 3.83 per million doses; mechanism analogous to HIT | "Evidence favors acceptance" (NASEM, 2024) | CDC/FDA paused; restricted authorization May 2022; discontinued 2023 |
| Hep B / Multiple sclerosis | Case reports and media coverage, 1990s (France) | Large-scale cohort and nested case-control studies found no increased risk | "Evidence favors rejection" (IOM, 2002) | Recommendation maintained; French suspension reversed after review |
| 1976 Swine Flu / GBS | Surveillance data showing excess GBS cases | Estimated ~1 additional case per 100,000 vaccinated | "Evidence establishes a causal relationship" | Vaccination program suspended after 43 million doses |
These cases show how causality assessment can lead to different conclusions depending on the strength and type of evidence available. The framework has confirmed genuine vaccine-caused injuries (OPV paralysis, RotaShield intussusception, mRNA myocarditis), identified and acted on rare but serious risks (J&J TTS, 1976 swine flu GBS), and found no causal association where evidence was extensively studied (MMR and autism, Hepatitis B and MS). No single case should be used to characterize the framework as either too permissive or too restrictive.
The frameworks described on this page are widely used, but they are not beyond question. Academic debate over how causality should be assessed — and whether existing systems do it well enough — is ongoing.
Several threads run through this debate:
Whether frameworks are applied too rigidly. Some epidemiologists argue that Bradford Hill's viewpoints are treated as strict requirements rather than flexible considerations, potentially blocking legitimate causal inferences when evidence is limited but suggestive (Phillips & Goodman, 2004; Rothman & Greenland).
Whether institutional algorithms create structural barriers. Peer-reviewed critiques have argued that the WHO AEFI algorithm can make it structurally difficult to classify novel or multifactorial adverse events as vaccine-related, particularly when the algorithm requires prior epidemiological confirmation before acknowledging a causal association (Puliyel & Naik, 2018; Bellavite et al., 2024; Bellavite, 2020).
Whether the 'correlation does not imply causation' principle has been applied too broadly. A 2025 preprint argued that during the COVID-19 pandemic, this principle was sometimes used to dismiss temporal signals before adequate investigation, rather than as a starting point for deeper study. Drawing on Kuhnian ideas about how scientific frameworks shape interpretation, the authors argued that established methods can sometimes make it harder to recognize new patterns while evidence is still emerging.
Whether the gap between legal and scientific standards creates confusion. The VICP's lower evidentiary threshold means that compensation awards are sometimes cited as proof of scientific causation, and scientific uncertainty is sometimes cited as evidence that no injury occurred. Neither interpretation is accurate, but the coexistence of two different causation standards creates space for misunderstanding.
These debates appear in peer-reviewed journals, formal methodological critiques, and professional discussions about how causality frameworks should be applied. They reflect an ongoing effort to improve how causality is assessed — not to eliminate the frameworks themselves.
Causality assessment in vaccine safety is a layered process, not a single test.
The Bradford Hill viewpoints provide a conceptual foundation for organizing causal evidence, but they are one framework among several and are best applied with judgment rather than as a rigid checklist. The National Academies and WHO have developed structured systems for evaluating evidence, but those systems have drawn peer-reviewed criticism for how they handle novel, rare, or multifactorial adverse events. Study design determines how much weight evidence carries, and different designs answer different causal questions. Legal and scientific causation operate under different standards, by design, for different purposes.
Historical cases show that vaccine safety frameworks and evidence review have identified real vaccine-caused injuries, prompted action when risks were confirmed, and also found no association where evidence was extensively studied. No single framework, study, or report settles every question.
Different frameworks can lead researchers to weigh the same evidence differently, especially when evidence is limited, evolving, or indirect. The strongest causal conclusions draw on multiple methods, multiple studies, and multiple independent populations. The weakest rely on any single source alone — whether that source is a VAERS report, a single study, or an institutional verdict.
Hill AB. The Environment and Disease: Association or Causation? Proc Royal Soc Med. 1965;58:295–300. PMC →
Original source of the nine viewpoints for evaluating causal associations.
Institute of Medicine. Adverse Events Associated with Childhood Vaccines. National Academies Press. 1994. NCBI Books →
Defined post hoc ergo propter hoc fallacy in vaccine context; foundational IOM vaccine safety report.
Institute of Medicine. Adverse Effects of Vaccines: Evidence and Causality. National Academies Press. 2012. NAP →
Most comprehensive vaccine safety review: 158 pairs evaluated across 4 causality categories.
NASEM. Evidence Review of the Adverse Effects of COVID-19 Vaccination. National Academies Press. 2024. NAP →
85 COVID-19 vaccine conclusions; established mRNA-myocarditis causal relationship.
WHO. Causality Assessment of AEFI: User Manual (2nd ed.). 2019. WHO →
Global standard for individual-case AEFI causality assessment.
Abara WE, et al. Expected Rates of Select Adverse Events After Immunization. J Infect Dis. 2022;225(9):1569–1578. OUP →
Calculated expected background rates to contextualize post-vaccination adverse events.
Phillips CV, Goodman KJ. The Missed Lessons of Sir Austin Bradford Hill. Epidemiologic Perspectives & Innovations. 2004;1:3. Springer →
Argued Hill's viewpoints are misapplied as rigid criteria rather than flexible considerations.
Shimonovich M, et al. Assessing Causality in Epidemiology: Revisiting Bradford Hill. Eur J Epidemiol. 2021;36:873–887. PMC →
Mapped Hill's viewpoints against modern causal inference methods.
Puliyel J, Naik P. Revised WHO's Causality Assessment of AEFI — A Critique. F1000Research. 2018;7:243. PMC →
Argued WHO AEFI classification creates structural paradox excluding novel and multifactorial adverse events.
Bellavite P, et al. WHO Algorithm for Causality Assessment of AEFI: Pitfalls and Suggestions. PMC. 2024. PMC →
Extended WHO critique to COVID-19 vaccines; flagged circular reasoning for novel platforms.
Bellavite P. Causality Assessment of AEFI: The Problem of Multifactorial Pathology. F1000Research. 2020. PubMed →
Argued WHO algorithm underestimates vaccine contribution when other factors interact.
Kohl KS, et al. The Brighton Collaboration. Advances in Patient Safety. 2005. NCBI Books →
International network standardizing AEFI definitions across 57 countries.
Shimabukuro TT, et al. Safety Monitoring in VAERS. Vaccine. 2015;33(36):4398–4405. PubMed →
Comprehensive overview of VAERS methodology, strengths, and limitations.
CDC. Chapter 21: Surveillance for Adverse Events Following Immunization Using VAERS. Manual for Surveillance of VPDs. 2024. CDC →
CDC operational guide documenting VAERS strengths and limitations.
FDA. VAERS Questions and Answers. 2024. FDA →
FDA guidance stating VAERS reports generally cannot determine causation.
McNeil MM, et al. The Vaccine Safety Datalink. Vaccine. 2014;32(42):5390–5398. PMC →
Comprehensive overview of VSD methodology and achievements.
Grey B. The Plague of Causation in the NCVIA. Harvard J on Legislation. 2011;48:2. Harvard JOL →
Analysis of VICP legal causation standards; ~90% of claims now off-Table.
Althen v. Secretary of HHS, 418 F.3d 1274 (Fed. Cir. 2005). FindLaw →
Established the three-prong Althen test for off-Table VICP causation.
HRSA. VICP Data & Statistics. 2025. HRSA PDF →
Compensation statistics: 12,500+ petitions, ~$5.5B total since 1988.
Preprint (2025). Beyond Correlation and Causation. Preprints.org. Preprints.org →
Kuhnian paradigm critique; not peer-reviewed.
How completely VAERS captures adverse events, what the sensitivity studies found, and why the commonly cited "1%" figure requires careful context.
Why a VAERS report cannot establish that a vaccine caused an adverse event.
Full framework: Bradford Hill viewpoints, IOM/NASEM system, WHO AEFI algorithm, study designs, and legal causation.