Adverse event after vaccine: judging causality

A 16-year-old adolescent arrives at the emergency department with chest pain three days after a vaccine. In the same week, a 14-month-old girl has a febrile seizure eight days after the MMRV vaccine, and a 60-year-old man is admitted with ascending weakness four weeks after the flu vaccine. All three families ask the same question: was it the vaccine?

The answer requires reasoning similar to a differential diagnosis. The sequence in time is the starting point, not the proof. Many events happen after a vaccine simply because they happen all the time, and some are indeed caused by it. Telling one from the other depends on defining the event well, knowing the expected frequency without the vaccine, and knowing how the data were collected.

This text shows how the World Health Organization organizes this assessment, which pitfalls appear along the way, and how population studies confirmed or ruled out famous associations.

Key points

• The WHO assesses the causality of an event after vaccination in four steps: eligibility, checklist, algorithm, and classification (Bellavite, 2020).
• In active surveillance, most events are mild and expected, such as fever and local pain, and a portion are coincidental (Mathew, 2026).
• Spontaneous reporting systems capture only part of the cases and have no denominator, so they serve to generate signals, not to measure risk (Miller, 2020; Zauche, 2026).
• Comparing observed cases with those expected from the background rate showed the excess of myocarditis after mRNA vaccines in young men (Naveed, 2022).
• Adjustment for confounders undid an apparent association between the 2009 pandemic vaccine and Guillain-Barré syndrome (Dieleman, 2011).
• A real and small risk must be weighed against the benefit, as in the case of intussusception after the rotavirus vaccine (Clark, 2019).

Summary

• What an adverse event after vaccination is
• Why the sequence in time misleads
• The WHO algorithm in four steps
• From reporting to signal: how the systems work
• Observed versus expected and other designs
• When the risk is real: weighing against the benefit
• How to manage the case in practice
• How NEXO helps train this reasoning
• Frequently asked questions
• Take-home messages

What an adverse event after vaccination is

An adverse event following immunization is any unwanted medical occurrence that happens after a vaccine, without this implying, by definition, a causal relationship with it. The definition is purposely broad. It allows recording everything that happens and only then deciding what is related to the vaccine (Bellavite, 2020).

The classification used by the WHO divides events with a causal relationship into four types. There are reactions linked to the product itself, such as fever and local pain. There are reactions due to a quality defect in the batch. There are reactions due to immunization error, such as administration by the wrong route or inadequate storage. And there are reactions linked to anxiety around the administration, such as fainting and hyperventilation (Mathew, 2026).

In addition to these, there are coincidental events, which would have happened anyway, and indeterminate ones, in which the evidence does not allow a conclusion. A final category gathers cases without enough information for any assessment (Bellavite, 2020).

What active surveillance shows

An active surveillance study at a tertiary hospital in Mysuru, India, followed children by telephone after 21,303 administrations of routine vaccines. About 10 in 100 children had some event, and severe ones, with hospitalization, were 1.27 per 1,000 administrations (Mathew, 2026). Fever was the most common event, followed by pain at the site (Mathew, 2026).

In the causality assessment, about 91 in 100 events were classified as reactions linked to the product, almost 5 in 100 as coincidental, about 3 as immunization error, and fewer than 1 as indeterminate. There was one death after the pentavalent vaccine, considered coincidental after investigation (Mathew, 2026).

Why the topic matters for the clinician

The clinician is usually the first to see the event and hear the family's question. The answer influences that child's next vaccine, the family's trust and, when the case is reported, surveillance across the whole country. Answering "it has nothing to do with it" without assessing is as risky as confirming the cause without criteria.

Why the sequence in time misleads

The reasoning "after this, therefore because of this" is one of the oldest shortcuts of human thought. It is among the classic heuristics of clinical practice and gains strength when the event is serious and the family looks for an explanation.

The background rate

Every day, children have febrile seizures, adolescents have viral myocarditis and adults develop Guillain-Barré syndrome, with or without vaccine. When millions of people are vaccinated in a few weeks, some of these events will fall, by pure chance, in the days after administration. The right question is whether more cases appear than expected.

This is the same problem as the base rate in diagnosis: without knowing the background frequency, any isolated number looks alarming. Intussusception illustrates the point well. The background incidence in infants is around 74 per 100,000 children in the first year of life, with huge variation between regions, from 9 to 328 (Yen, 2016). Any assessment of the rotavirus vaccine must start from that number.

The probability before the event

Causal reasoning resembles pretest probability. An event that already has a proven association with the vaccine, within the expected time interval, starts with a high probability of causal relation. An event without biological plausibility and without a population signal starts with low probability, and the sequence in time, on its own, changes that estimate little.

The retrospective view

After the event happens, the link with the vaccine seems evident, and any detail that reinforces it gains weight. Retrospective bias leads to reconstructing the story backwards. The antidote is to follow a fixed script, decided before looking at the outcome.

The WHO algorithm in four steps

The 2018 WHO manual organizes the individual assessment of causality into four steps. The first checks whether the case is eligible: there must be a clear diagnosis, preferably according to a standardized case definition, and identification of the vaccine. Without a diagnosis, there is nothing to assess (Bellavite, 2020).

The second step is a checklist. It asks whether there is strong evidence of another cause, whether there is a known association between the vaccine and the event, whether the time interval is compatible, whether there was an immunization error or quality problem and whether there is evidence against the causal relation. The third step takes these answers to an algorithm, and the fourth classifies the case (Bellavite, 2020).

The four final categories

At the end, the case is classified as consistent with a causal relation, inconsistent (when another cause explains the event), indeterminate (when the evidence is insufficient or conflicting) or unclassifiable (when data are missing) (Bellavite, 2020). The Indian study used subcategories: A1 to A4 for the types of causal reaction, B1 and B2 for the indeterminate ones and C for the coincident ones (Mathew, 2026).

Until 2013, WHO used a different scale, with the categories certain, probable, possible, unlikely and unclassifiable (Puliyel, 2018). The change sought to make the process more reproducible, but it also drew criticism.

Criticism of the algorithm

Puliyel and Naik argue that the revised algorithm only recognizes as causal the reactions already described in epidemiological studies. Thus, a new event, detected after registration, tends to fall into the coincident or indeterminate categories. The authors cite deaths of children with congenital heart disease after the pentavalent vaccine, in which the presence of another factor would lead the algorithm to rule out the vaccine as the cause, although WHO has come to recommend precautions in such cases (Puliyel, 2018).

Bellavite points to a related problem: many vaccine reactions result from inflammatory and immune responses in which genetic and environmental factors interact with the vaccine. By looking for "the" cause, the algorithm tends to classify as inconsistent an event in which the vaccine was one cause among several. The author proposes excluding the vaccine only when the other cause is independent of it (Bellavite, 2020).

These critiques are a reminder that the algorithm is a standardization tool, not a definitive answer. An indeterminate case remains useful information, especially when other similar cases start to appear.

The case definition comes before the cause

The first step of the algorithm depends on a well-defined diagnosis. For this, the Brighton Collaboration publishes standardized definitions for events of interest, with levels of diagnostic certainty. They allow a case of myocarditis in one country to be compared with a case in another.

Applying these definitions to real data is not simple. The European consortium VAC4EU created a process to validate 16 safety outcomes of covid vaccines, 13 of them with a Brighton definition. In 747 assessments of fictitious cases made by 33 assessors, overall agreement was moderate (kappa of 0.55), and it was practically nil for thrombosis with thrombocytopenia syndrome (Dehghan Tarazjani, 2026).

The message for the clinician is direct. If trained specialists disagree about the diagnosis of a rare event, the record in the medical chart needs to be detailed: symptoms, dates, tests and what was ruled out. The same care appears in conditions that mimic others, such as syncope and functional seizures, in which fainting from anxiety shortly after administration can be mistaken for a convulsion.

An example of active surveillance with a case definition

In the Puglia region of Italy, an active surveillance program followed 2,540 children vaccinated with the combined measles, mumps, rubella and varicella vaccine in their second year of life, with 2,149 completing follow-up. Of the 109 serious events, 82 were classified as consistent with a causal relationship by the WHO algorithm, and all resolved. The rate of convulsion consistent with the vaccine was 0.05 per 100 children followed, lower than in studies that did not perform causality assessment (Stefanizzi, 2019).

From reporting to signal: how the systems work

Assessing an individual case is rarely enough to prove or rule out a causal relationship. That depends on population data, which come from systems with different objectives.

Spontaneous reporting

VAERS, in the United States, receives reports from anyone: professionals, manufacturers, patients and family members. It is fast and sensitive to unexpected events, but it suffers from underreporting and does not know the number of vaccinated people, so it does not allow rates or relative risks to be calculated (Zauche, 2026).

How much VAERS captures depends on the event. Compared with rates from an active system, sensitivity for anaphylaxis ranged from 13 to 76 per 100, depending on the vaccine, and for Guillain-Barré, from 12 to 64. Earlier studies had estimated 47 per 100 for intussusception after rotavirus vaccine and 68 per 100 for paralytic polio associated with oral vaccine (Miller, 2020). Serious and well-known events are reported more; mild or unexpected events, less.

Active systems with a denominator

The Vaccine Safety Datalink, created in 1990, brings together clinical and vaccination records from 13 integrated health organizations and covers more than 15 million people per year. Because it knows vaccinated and unvaccinated people, it allows rates to be compared and signals to be confirmed or ruled out (Zauche, 2026).

V-safe, created during the pandemic, collected responses from more than 10 million people by text message about reactions after covid vaccines. The limitation is the lack of an unvaccinated group, which forces comparison with estimated background rates. A network of specialists, CISA, supports the assessment of complex individual cases (Zauche, 2026).

Reporting bias

Reporting changes with public attention. When an event gains media prominence, reports rise, even without a change in real risk. Therefore, an increase in reports is a signal to investigate, not an estimate of risk. The table below summarizes the role of each source.

Data source	What it does well	Main limitation	Use in reasoning
Individual case report	Describes the event in detail	Mixes cause and coincidence	Generates a hypothesis
Spontaneous reporting (VAERS)	Detects rare and unexpected events quickly	Underreporting, no denominator, sensitive to media	Generates a signal
Active surveillance (V-safe, follow-up cohorts)	Measures frequent reactions more completely	Often no unvaccinated group	Estimates the frequency of expected reactions
Databases with a denominator (VSD)	Compares observed and expected rates	Depends on the quality of the clinical record	Tests the signal
Analytical studies (case-control, self-controlled case series)	Adjust for confounding and estimate relative risk	Require a case definition and timing	Confirms or rules out the association

Observed versus expected and other designs

The studies that test a signal use designs with strengths and weaknesses, the same ones that appear in the critical reading of articles.

Observed versus expected: the case of myocarditis

In British Columbia, Canada, a population study compared cases of myocarditis after mRNA vaccines with the number expected from historical rates. In the seven days after administration, the rate was 0.97 per 100,000 doses, about 15 times the expected rate (Naveed, 2022). At 21 days, the rate was 1.37 per 100,000, about 7 times the expected rate (Naveed, 2022). The highest ratio, close to 148, appeared in men aged 18 to 29 after the second dose of the mRNA-1273 vaccine (Naveed, 2022).

A living review of 46 studies confirmed the pattern. Incidence was higher in boys aged 12 to 17, between 50 and 139 per million, and most cases in adolescents and adults occurred in young men, 2 to 4 days after the second dose. Most were hospitalized for a short time, 2 to 4 days (Pillay, 2022). Here, all the criteria aligned: a population signal, a short and consistent time window, a defined risk group and biological plausibility.

Adjusting for confounding: Guillain-Barré and the pandemic vaccine

A multinational case-control study in Europe assessed Guillain-Barré syndrome after the 2009 pandemic influenza vaccines. The crude estimate suggested an almost threefold higher risk (odds ratio of 2.8) (Dieleman, 2011). After adjustment for influenza-like illness or recent respiratory infection and for the seasonal vaccine, the association disappeared (adjusted odds ratio of 1.0, with an interval of 0.3 to 2.7) (Dieleman, 2011).

The explanation is that respiratory infection itself is a known cause of Guillain-Barré, and people with recent infections also sought the vaccine more often. The example shows why adjustment for confounders is an essential part of the reasoning, one of the biases every student should recognize.

The self-controlled case series

The self-controlled case series compares, in each person who had the event, the frequency within a risk window after the vaccine with the frequency outside it. Because each person is their own control, fixed factors such as genetics and social condition cancel out. The method has power almost as good as a cohort when coverage is high and the risk window is short, but it does not provide absolute risk on its own (Stowe, 2020).

With these methods, Stowe and colleagues reviewed three neurological associations. The suspicion that the hepatitis B vaccine would cause multiple sclerosis was not confirmed. For Guillain-Barré, there is a small increase in risk after influenza vaccines, much smaller than the risk after the infection itself. For narcolepsy, there was strong evidence of an association with an adjuvanted pandemic vaccine used in 2009 and 2010: in Finland, the risk was 13 times higher in children and young people aged 4 to 19 (Stowe, 2020).

The historical reference is the 1976 swine flu program in the United States, associated with about 1 excess Guillain-Barré case per 100,000 vaccinated people and suspended because of it (Stowe, 2020).

When the risk is real: weighing it against the benefit

Some associations are confirmed, and the question becomes another one: does the benefit outweigh the risk? Intussusception after the rotavirus vaccine is the most studied example.

The current vaccines showed no risk in the large pre-licensure trials, each with more than 60,000 infants, but post-marketing surveillance detected a small increase in some countries, especially right after the first dose, on the order of 1 to 6 excess intussusceptions per 100,000 vaccinated infants (Yen, 2016). The risk was much lower than that of the first rotavirus vaccine, withdrawn from the market in the United States in 1999 because of about 1 excess intussusception per 10,000 vaccinated infants (Yen, 2016).

A model for 135 low- and middle-income countries estimated that a three-dose schedule together with the DTP vaccine would prevent about 74,000 deaths from rotavirus gastroenteritis and cause about 201 excess deaths from intussusception. That gives a ratio of 369 deaths prevented for each death caused (Clark, 2019). To communicate this kind of number, natural frequencies are usually clearer than ratios.

What this changes for the individual patient

For the teenager with myocarditis after the vaccine, the probable causal link changes management: investigation, treatment, avoidance of exertion and, later, a conversation about the next vaccines. For the girl with a febrile seizure eight days after the MMRV vaccine, the interval needs to be compared with the window described for fever after MMRV, and guidance on the next dose should take this into account. For the man with Guillain-Barré, investigation of a recent infection is part of the assessment.

How to manage the case in practice

The reasoning when facing an event after vaccination can follow a simple order, which works as a strategy of decision under uncertainty.

First, the diagnosis

Before the cause comes the diagnosis. What exactly happened, when did it start, which tests confirm it and which differential diagnoses were considered? A vasovagal reaction right after the injection, for example, needs to be separated from anaphylaxis, as discussed in the differential diagnosis in anaphylaxis.

Then, the checklist questions

Is there another strong cause? Is there a known association with this vaccine? Does the interval fit the described risk window? Was there an error in administration, storage or batch? Is there anything that argues against a causal link? Recording the answers in the medical record helps whoever assesses the case later (Bellavite, 2020).

Report and communicate

Reporting is the professional's responsibility, even when the causal link seems unlikely. Reporting feeds the systems that detect new signals, and underreporting is the main limitation of these systems (Miller, 2020).

For the family, it helps to explain what is known, what is still being assessed and what changes for the next vaccines, checking understanding with teach-back. A confirmed causal link rarely rules out all future vaccines; it usually changes the choice of product, the injection site or the observation time.

Calibrate confidence

It is easy to be too certain in both directions: to state that the vaccine caused the event or that it had nothing to do with it. Calibrating confidence helps use the right words: probable, possible, unlikely, indeterminate.

How NEXO helps train this reasoning

In NEXO, the case of the day is fictional, with a real clinical presentation and up to six progressive clues, and you propose the diagnosis by searching for the ICD-10 code. Separating what came before from what caused it is exactly the exercise of reading the clues in the right order without sticking to the first one.

The limits on using the game are described on the NEXO economy page. If you research the teaching of clinical reasoning, the page for researchers explains how to get in touch. Professionals who write or review cases find information under curation and creators, and institutions can learn about the proposal on the partnerships page. To practice searching for the code, see how the ICD-10 code is built.

Frequently asked questions

Is every event after a vaccine caused by it?

The definition of an adverse event following immunization includes any occurrence after the vaccine, with or without a causal link. Some events are coincidental, and causality assessment serves precisely to separate them (Mathew, 2026).

How does the WHO assess causality?

In four steps: eligibility with a clear diagnosis, a checklist, an algorithm and a final classification as consistent, inconsistent, indeterminate or unclassifiable (Bellavite, 2020).

Do VAERS data show the risk of a vaccine?

VAERS data serve to generate signals. The system receives reports from anyone, captures only part of the cases and does not know the number of vaccinated people (Zauche, 2026; Miller, 2020).

What is an observed versus expected analysis?

It is the comparison between the cases that appear after the vaccine and those that would be expected from the background rate in the same period. It showed the excess of myocarditis after mRNA vaccines in young men (Naveed, 2022).

Does the flu vaccine cause Guillain-Barré?

There is a small increase in risk after influenza vaccines, much smaller than the risk caused by the infection itself (Stowe, 2020). For the 2009 pandemic vaccine, adjustment for recent infection undid the apparent association (Dieleman, 2011).

Does the WHO algorithm have limitations?

Critics point out that it tends to classify new or multifactorial events as coincidental or indeterminate (Puliyel, 2018; Bellavite, 2020).

Does a confirmed risk mean the vaccine should be suspended?

It depends on the balance between risk and benefit. For the rotavirus vaccine, the estimate was 369 deaths prevented for each excess death from intussusception (Clark, 2019).

Take-home messages

• The sequence in time is the starting point of the assessment, not proof of cause.
• The background rate of the event is essential to know whether there are more cases than expected.
• The WHO algorithm follows four steps and begins with a well-defined diagnosis.
• Spontaneous reporting generates signals; databases with a denominator and analytical studies test them.
• Adjusting for confounders can undo apparent associations, as with Guillain-Barré and the pandemic vaccine.
• Real risks exist, such as myocarditis after mRNA in young men, and they must be weighed against the benefit.
• Reporting, recording in detail and communicating with calibration are part of the care.

Play the case of the day on NEXO.

References

1. Bellavite P. Causality assessment of adverse events following immunization: the problem of multifactorial pathology. F1000Res. 2020. PMID 32269767. DOI 10.12688/f1000research.22600.2.
2. Puliyel J, Naik P. Revised World Health Organization (WHO)'s causality assessment of adverse events following immunization–a critique. F1000Res. 2018. PMID 30026925. DOI 10.12688/f1000research.13694.2.
3. Mathew M, Sanathan SR, Ravi MD. Understanding adverse events following immunisation: evidence of vaccine safety and coincidental events from 3 years of surveillance. BMJ Paediatr Open. 2026. PMID 41871912. DOI 10.1136/bmjpo-2025-003938.
4. Miller ER, McNeil MM, Moro PL, Duffy J, et al. The reporting sensitivity of the Vaccine Adverse Event Reporting System (VAERS) for anaphylaxis and for Guillain-Barré syndrome. Vaccine. 2020. PMID 33039207. DOI 10.1016/j.vaccine.2020.09.072.
5. Zauche LH, Miller ER, Kim SS, Jazwa A, et al. Monitoring Vaccine Safety: United States Vaccine Safety Surveillance Systems. J Pediatr Health Care. 2026. PMID 41478707. DOI 10.1016/j.pedhc.2025.07.016.
6. Naveed Z, Li J, Spencer M, Wilton J, et al. Observed versus expected rates of myocarditis after SARS-CoV-2 vaccination: a population-based cohort study. CMAJ. 2022. PMID 36410749. DOI 10.1503/cmaj.220676.
7. Dieleman J, Romio S, Johansen K, Weibel D, et al. Guillain-Barré syndrome and adjuvanted pandemic influenza A (H1N1) 2009 vaccine: multinational case-control study in Europe. BMJ. 2011. PMID 21750072. DOI 10.1136/bmj.d3908.
8. Clark A, Tate J, Parashar U, Jit M, et al. Mortality reduction benefits and intussusception risks of rotavirus vaccination in 135 low-income and middle-income countries: a modelling analysis of current and alternative schedules. Lancet Glob Health. 2019. PMID 31607466. DOI 10.1016/S2214-109X(19)30412-7.
9. Yen C, Healy K, Tate JE, Parashar UD, et al. Rotavirus vaccination and intussusception – Science, surveillance, and safety: A review of evidence and recommendations for future research priorities in low and middle income countries. Hum Vaccin Immunother. 2016. PMID 27322835. DOI 10.1080/21645515.2016.1197452.
10. Stowe J, Andrews N, Miller E. Do Vaccines Trigger Neurological Diseases? Epidemiological Evaluation of Vaccination and Neurological Diseases Using Examples of Multiple Sclerosis, Guillain–Barré Syndrome and Narcolepsy. CNS Drugs. 2020. PMID 31576507. DOI 10.1007/s40263-019-00670-y.
11. Stefanizzi P, Stella P, Ancona D, Malcangi KN, et al. Adverse Events Following Measles-Mumps-Rubella-Varicella Vaccination and the Case of Seizures: A Post Marketing Active Surveillance in Puglia Italian Region, 2017–2018. Vaccines (Basel). 2019. PMID 31591347. DOI 10.3390/vaccines7040140.
12. Pillay J, Gaudet L, Wingert A, Bialy L, et al. Incidence, risk factors, natural history, and hypothesised mechanisms of myocarditis and pericarditis following covid-19 vaccination: living evidence syntheses and review. BMJ. 2022. PMID 35830976. DOI 10.1136/bmj-2021-069445.
13. Dehghan Tarazjani A, Weibel D, Aurelius T, Zwiers LC, et al. Harmonizing Case Validation for Vaccine Safety in Real-World Data: The VAC4EU Experience. Clin Epidemiol. 2026. PMID 41890771. DOI 10.2147/CLEP.S558974.