What we actually mean when we talk about "vaccine safety" — and why it's not a binary question
The first trap begins with the term "safety" itself. In everyday thinking, this is an absolute category: either a substance is safe or it isn't. In medicine, safety is a risk-benefit ratio, measured in specific populations under specific conditions. More details in the section Pseudo-drugs and Counterfeits.
When we say a vaccine is "safe," we mean: its benefits significantly outweigh the risks of side effects for the vast majority of recipients. This doesn't mean an absence of risks — it means their acceptability in the context of protection against disease.
Vaccine safety is not a binary question, but a probabilistic statement about the balance of benefits and harms in a specific population.
🔬 Defining adverse events: why "after" doesn't mean "because of"
An adverse event in clinical practice is any undesirable medical occurrence that happens after vaccine administration, regardless of causal relationship. This is a critically important distinction.
If a million people receive a vaccine within a month, statistically thousands of events will occur among them: heart attacks, strokes, exacerbations of chronic conditions, accidents. Most would have happened regardless of vaccination — this is the background rate of events in the population.
- Background rate
- The natural frequency of medical events in a population, unrelated to any intervention. Without accounting for it, any temporal coincidence appears to be causation.
- Signal
- A statistically significant excess of observed event frequency over the background rate. Requires additional analysis to confirm causality.
- Noise
- Random coincidences and background events that are mistakenly interpreted as vaccine adverse effects.
The task of monitoring systems is to separate signal from noise, to identify events whose frequency statistically significantly exceeds the background rate. This requires comparative data, control groups, and statistical analysis, not simply counting all events after vaccination (S002).
🧾 Three levels of evidence: from correlation to causation
Epidemiology distinguishes three levels of relationship between exposure and outcome, each requiring increasingly rigorous evidence.
| Level | Definition | Example | Sufficient for conclusion? |
|---|---|---|---|
| Correlation | Statistical association: two events occur together more often than by chance | Headaches are reported more frequently after vaccination | No — requires control for confounders |
| Association | Correlation that persists after controlling for known confounding factors | The relationship remains after accounting for age, sex, comorbidities | No — requires biological mechanism and RCTs |
| Causation | Proven cause-and-effect relationship through a set of criteria: temporal sequence, biological plausibility, dose-response, reproducibility, RCT data | Vaccine causes event through known mechanism, confirmed in different populations | Yes — this is grounds for action |
Most disputes about vaccine safety arise from conflating these levels. People observe a correlation (an event occurred after vaccination) and immediately interpret it as causation, bypassing all intermediate stages of proof.
⚙️ Framework for analysis: from clinical trials to post-marketing surveillance
The vaccine safety assessment system includes several sequential stages, each detecting adverse effects of different rarity.
- Preclinical studies — detect gross toxic effects in animal models
- Phase I (dozens of participants) — basic safety and immunogenicity
- Phase II (hundreds of participants) — expanded safety assessment, optimal doses
- Phase III (thousands to tens of thousands of participants) — randomized controlled trials, detect adverse effects with frequency of approximately 1:1,000 and higher (S002)
- Post-marketing surveillance — continuous monitoring after approval, when millions receive the vaccine, detects rare effects (1:100,000 and less frequent)
Even the largest clinical trials cannot detect very rare adverse effects. It was precisely during post-marketing surveillance that rare cases of thrombosis with thrombocytopenia after vector COVID-19 vaccines were identified, for example (S001).
Absence of a signal in clinical trials doesn't mean absence of rare adverse effects — it means the sample size was insufficient to detect them.
Seven Arguments That Sound Convincing — and Why They Require Careful Examination, Not Immediate Acceptance
Before examining the evidence, it's necessary to honestly present the strongest arguments of vaccination skeptics. This is not a straw man, but a steel man — the most convincing version of the opposing position. Only then can we conduct an honest analysis. For more details, see the section Extreme Diets and Miracle Cures.
⚠️ Argument One: "Clinical Trials Are Too Short to Detect Long-Term Effects"
Critics point out that most vaccine clinical trials last months, not years, and therefore cannot detect side effects that manifest years after vaccination. This is particularly relevant for new platforms such as mRNA vaccines. Indeed, the median follow-up in the ChAdOx1 nCoV-19 study was 3.4 months at the time of interim analysis (S002).
The question sounds logical: how can we be certain about the absence of effects that appear five or ten years later? But there's a trap here: most vaccine side effects manifest within the first 6–8 weeks, not years. This is a biological fact, not an assumption.
⚠️ Argument Two: "Adverse Event Reporting Systems Are Incomplete and Underestimate True Frequency"
Passive surveillance systems, such as VAERS in the US or Yellow Card in the UK, depend on voluntary reports from healthcare workers and patients. Studies show that a significant proportion of adverse events go unreported — a phenomenon known as underreporting.
This is true. But an incorrect conclusion is drawn from this fact: if the system captures only 10–20% of actual events, then the true frequency of side effects is supposedly 5–10 times higher. In reality, underreporting concerns mild, transient reactions (injection site pain, fever), not serious complications. Serious events are reported much more completely.
⚠️ Argument Three: "Manufacturer Conflicts of Interest Distort Safety Data"
Vaccine clinical trials are funded and often conducted by the manufacturers themselves, who have a direct financial interest in positive results. The history of the pharmaceutical industry includes cases of concealing unfavorable data, manipulating study designs, and selective publication of results.
- Conflict of Interest Is a Real Problem
- Manufacturers are indeed interested in positive results. But this doesn't mean the data are automatically falsified. Regulatory agencies (FDA, EMA) independently verify primary data, and results are published in peer-reviewed journals where they're analyzed by competing researchers.
- Where the Real Vulnerability Lies
- The vulnerability is not in concealing serious side effects (these are detected quickly), but in insufficient study of rare complications requiring large samples and long-term follow-up. This is a problem of scale, not honesty.
⚠️ Argument Four: "Individual Variability Makes Population Data Irrelevant"
Even if a vaccine is safe for 99.9% of the population, this doesn't guarantee safety for a specific individual with their unique genetic profile, medical history, and current health status. Pharmacogenetics shows that the same drug can be safe for some people and toxic for others.
This is true in principle, but leads to a paradox: if we reject population data as irrelevant, then we have no way to make a decision about vaccination at all. Individual risk can only be assessed through population studies stratified by risk groups. This isn't perfect, but it's the only tool we have.
⚠️ Argument Five: "Accelerated COVID-19 Vaccine Development Meant Cutting Corners on Safety"
Vaccines that typically take 10–15 years to develop were created and approved in 10–12 months. Skeptics claim that such acceleration was only possible by skipping or shortening critical safety testing stages. Emergency Use Authorization implies less stringent evidence requirements than full approval.
Here it's important to distinguish: acceleration occurred not by shortening trial phases, but by parallelizing stages, increasing funding, and simplifying bureaucracy. Phases I, II, and III were conducted fully, with complete control groups (S002). What actually shortened was the time between trial completion and approval, not the trials themselves.
⚠️ Argument Six: "Historical Examples of Vaccines Withdrawn Due to Safety Issues"
The history of vaccinology includes cases where vaccines were withdrawn after serious side effects were identified following mass use. The rotavirus vaccine RotaShield was withdrawn in 1999 due to its association with intussusception. The 1976 swine flu vaccine was linked to Guillain-Barré syndrome.
These examples demonstrate not the failure of the safety system, but its success. Surveillance systems identified problems and vaccines were withdrawn. RotaShield was withdrawn after 15 cases of intussusception per 1.5 million doses — meaning the system worked. This is not an argument against monitoring, but an argument in its favor.
⚠️ Argument Seven: "Pressure on Physicians and Censorship of Critical Opinions Suppress Open Discussion"
Healthcare workers who express doubts about vaccine safety face professional ostracism, accusations of spreading misinformation, and even threats to their licenses. Social networks and media platforms remove content critical of vaccines.
Honesty is needed here: yes, there is asymmetry in public discussion. But we need to distinguish two phenomena. First — moderation of clearly false information (for example, claims about microchips in vaccines). Second — suppression of legitimate scientific questions about rare side effects or long-term data. The first is justified, the second is not. The problem is that the boundary between them is often blurred.
What Randomized Controlled Trials Show — And Why They Remain the Gold Standard Despite All Limitations
Randomized controlled trials (RCTs) are the gold standard of evidence-based medicine. Their design minimizes systematic errors and allows establishing causal relationships with a high degree of confidence. More details in the section Myths About Psychosomatics.
📊 ChAdOx1 nCoV-19 Study Design: Four Countries, 11,636 Participants, Double-Blinding
The interim analysis of the ChAdOx1 nCoV-19 vaccine (AstraZeneca) included data from four RCTs in Brazil, South Africa, and the United Kingdom (S002). 11,636 participants were randomized: 5,807 received the vaccine, the rest received control vaccines (meningococcal MenACWY or saline).
Double-blinding eliminates placebo and nocebo effects. Neither participants nor researchers assessing outcomes knew who received the experimental vaccine. This prevents bias in recording adverse effects — if a physician knows about the vaccine, they may monitor symptoms more closely, artificially inflating their frequency in the experimental group.
Double-blinding is not a formality, but a mechanism that eliminates cognitive traps of the observer and participant at the design level.
🧪 Safety Profile: Local and Systemic Reactions Compared to Control
Local reactions (pain, redness, swelling at the injection site) were more common in the vaccine group — this is expected for any vaccine that elicits an immune response. Systemic reactions (fatigue, headache, myalgia, fever) also predominated in the vaccine group, especially after the first dose (S002).
Most reactions were mild or moderate and resolved within a few days. Critically: the frequency of serious adverse events (SAE — events requiring hospitalization or life-threatening) was comparable between vaccine and control groups.
Of 168 serious adverse events, only three were deemed possibly related to the vaccine, and all resolved without consequences. This means the vaccine did not increase the risk of serious medical events compared to background frequency.
🧾 Causality Assessment Methodology: WHO Algorithm and Bradford Hill Criteria
When an adverse event is recorded, its causal relationship with the vaccine is assessed using standardized algorithms. WHO developed an approach considering: temporal relationship (event within a biologically plausible period after vaccination), alternative explanations (other diseases, medications), biological plausibility of mechanism, data on similar cases.
- Bradford Hill Criteria (population level)
- Strength of association — how much the risk is increased; consistency — is it reproduced in different studies; specificity — is it linked to a specific outcome; temporal sequence — does exposure precede outcome.
- Biological gradient and mechanism
- Dose-response relationship; is there an understandable mechanism of action; is it consistent with other knowledge (coherence).
- Experimental evidence and analogy
- Does risk change when exposure changes; are there similar causal relationships in other contexts.
No single criterion is absolutely necessary or sufficient. Their totality distinguishes true causality from random correlation.
📌 Post-Marketing Data: Detecting Rare Events Across Millions of Doses
Clinical trials, even large ones, have limited statistical power for rare adverse events. An event with a frequency of 1:100,000 may not occur in a study with 20,000 participants. Post-marketing surveillance is critically important precisely for this.
Rare cases of thrombosis with thrombocytopenia (TTS) after vector COVID-19 vaccines (frequency ~1:100,000) were identified precisely through post-marketing monitoring (S001). This is not a failure of the safety system, but its success: clinical trials detect common and moderately rare effects, post-marketing surveillance detects very rare ones.
After TTS was identified, recommendations for diagnosis and treatment were developed, and application guidelines were adjusted for different age groups considering risk-benefit ratios. The system works.
🔎 Active Surveillance vs. Passive: VSD and Sentinel Systems
Passive systems (VAERS) have known limitations: incomplete reporting, absence of control groups, inability to establish event frequency. To overcome these limitations, active surveillance systems were created.
| System | Data Source | Capabilities |
|---|---|---|
| Vaccine Safety Datalink (VSD) | Medical records of millions of people from large healthcare organizations (USA) | Population studies with control groups, precise calculation of event frequency (S006) |
| Sentinel FDA | Insurance company data and electronic medical records | Real-time monitoring, rapid safety signal detection, formal epidemiological studies |
The VSD system conducted several studies of the human papillomavirus (HPV) vaccine that found no increased risk of autoimmune diseases, despite reports in passive systems (S001). This demonstrates how active surveillance with control groups can distinguish true signals from passive reporting artifacts.
Active surveillance systems can rapidly conduct formal epidemiological studies to assess causality. When a suspected adverse event arises, researchers compare event frequency in vaccinated and unvaccinated groups, controlling for confounders (age, sex, comorbidities), and apply causality criteria.
The link to analysis of vaccine and autism myths shows how active surveillance debunks false associations that appear convincing in passive systems.
Why Our Brains Are So Bad at Assessing Causality — and Which Cognitive Traps Make Vaccine Myths So Persistent
Even with quality data available, people systematically make errors in assessing cause-and-effect relationships. This isn't a matter of intelligence or education — these are fundamental features of how human cognition works. Learn more in the Mental Errors section.
🧩 The Post Hoc Ergo Propter Hoc Fallacy: "After" Is Automatically Interpreted as "Because Of"
The Latin phrase "post hoc ergo propter hoc" means "after this, therefore because of this." This is one of the most common logical fallacies. If event B occurred after event A, our brain automatically assumes that A caused B.
This heuristic made evolutionary sense: in our ancestors' environment, quickly establishing cause-and-effect relationships (even false ones) was safer than slow analysis. Better to mistakenly avoid a safe plant than to eat a poisonous one. But in the modern world, where we encounter millions of events, this heuristic produces massive numbers of false positives.
If a million people get vaccinated within a month, among them there will statistically be hundreds of heart attacks, strokes, cancer diagnoses — simply because these events occur at a certain frequency in any population. Temporal proximity to vaccination doesn't make the vaccine the cause of these events.
🕳️ Base Rate Neglect: We Don't Account for How Often Events Occur Without Vaccination
The base rate is the frequency of an event in a population without exposure. If we want to assess whether a vaccine increases the risk of event X, we need to compare the frequency of X among the vaccinated with the frequency of X among the unvaccinated.
People systematically ignore the base rate, focusing only on the absolute number of cases among the vaccinated. If in a country with a population of 10 million people there are 50,000 heart attacks annually (base rate 0.5%), and during the year 5 million people get vaccinated, among whom approximately 2,000 heart attacks will occur within a month after vaccination, this doesn't mean the vaccine caused these heart attacks. This is the expected number of events that would have occurred in this group regardless of vaccination.
- Determine the base rate of the event in the target population (without vaccination)
- Calculate the expected number of cases among the vaccinated for the same period
- Compare the observed number with the expected number
- If observed ≈ expected, there is no causal relationship
- If observed > expected, conduct additional analysis with a control group
🧬 The Availability Effect: Vivid Stories Overshadow Statistics
The availability heuristic causes us to assess the probability of an event by the ease with which examples come to mind. A dramatic story about a person who developed a serious illness after vaccination is emotionally vivid and easily memorable.
Statistical data about millions of people who had no side effects are abstract and don't evoke an emotional response. Social media amplifies this effect by creating clusters of similar stories. If you read ten stories about side effects in a row, your brain will assess the risk as very high, even if those ten cases are all that occurred among millions of vaccinated people.
One patient with a rare side effect whose story spreads on social media has a greater impact on risk perception than safety data (S002) obtained from millions of people.
🎯 Confirmation Bias: We Search for Evidence of Our Hypothesis, Not Its Refutation
Confirmation bias is the tendency to seek, interpret, and remember information that confirms our existing beliefs. If you're already convinced that vaccines are dangerous, you'll actively collect stories about side effects and ignore safety data.
This isn't malicious intent — it's an automatic process. Our brain conserves resources by prioritizing information that's consistent with our existing model of the world. Reevaluating contradictory information requires cognitive effort that the brain prefers to avoid.
Vaccine safety monitoring systems (S004), (S006) are specifically designed to overcome this bias: they collect data actively (not relying on voluntary reports), use control groups, and apply statistical methods that don't allow confirmation bias to distort results.
🔄 Pattern Illusion: The Brain Sees Connections Where None Exist
The human brain is a pattern-seeking machine. This ability helped our ancestors survive, but it also causes us to see patterns in random data. If you're looking for a connection between vaccination and autism, you'll find it, even if it doesn't exist.
Classic example: Wakefield's study found a correlation between the MMR vaccine and autism, but it was later revealed to be a fabrication. However, the myth remained alive because it fit an existing pattern: the vaccine is administered at a certain age, autism is often diagnosed at the same age, so the connection seems obvious.
| Cognitive Trap | How It Works | How to Overcome It |
|---|---|---|
| Post hoc ergo propter hoc | Temporal sequence is interpreted as causality | Compare event frequency among vaccinated and unvaccinated |
| Base rate neglect | Absolute number of cases seems high without context | Always request the base rate and expected number |
| Availability effect | Vivid stories seem more probable | Rely on systematic data, not anecdotes |
| Confirmation bias | We search for evidence of our hypothesis | Actively seek disconfirming evidence |
| Pattern illusion | Brain sees connections in random data | Use statistical tests, not visual analysis |
⚙️ Why Vaccine Myths Are So Persistent: Synergy of Traps
None of these traps work in isolation. They reinforce each other, creating a self-sustaining belief system. A person hears a story about a side effect (availability effect), assumes the vaccine caused it (post hoc), doesn't check the base rate (base rate neglect), searches for additional confirming stories (confirmation bias), and sees a pattern that confirms their belief (pattern illusion).
Active safety monitoring systems (S001), (S005) work precisely because they bypass these traps. They collect data systematically, use control groups, apply statistical methods, and publish results regardless of whether they confirm or refute the hypothesis of vaccine harm.
Cognitive traps aren't a sign of stupidity. They're a sign that we're using evolutionarily ancient thinking systems to solve modern problems that require statistical thinking.
Understanding these mechanisms is the first step toward overcoming them. When you hear a story about a side effect, you can ask: what's the base rate of this event? Was it compared with a control group? Is this one case or a systematic pattern? These questions don't deny the reality of side effects — they simply require evidence that goes beyond cognitive traps.
For a deeper understanding of how vaccine myths spread and become entrenched, see the analysis of the vaccine-autism myth and the guide to evaluating sources of vaccine information.
