During the COVID era, I was accused of being anti-vaccine.
I wasn’t then, and I’m not now.
I questioned the scientific methods used to support claims about a product being promoted to a nation. I questioned whether the outcomes measured justified the benefits being communicated. Above all, I questioned the explanation of immune function that appeared to underpin much of the public discussion.
Those questions came from a longstanding interest in immunology.
One person who applied the anti-vaccine label retreated from it remarkably quickly when we discussed my actual position. But the exchange illustrated the problem: the accusation arrived before the understanding.
A label can make a complicated argument easier to dismiss. It does very little to answer it.
My position was—and remains—that a medical intervention should demonstrate the benefit being claimed for it. Its methods, assumptions and limitations should be open to examination. That expectation applies to vaccines, medicines, dietary interventions and my own proposals.
What does “effective” actually mean?
When vaccination is recommended, the practical questions concern what happens to people.
Will it reduce their likelihood of developing a particular disease? Will it reduce severe illness, disability or death? How long does protection last? What are the adverse effects? How does that balance change with age, previous exposure and existing health?
“Effective” needs an outcome attached to it.
Effective against laboratory-confirmed infection? Symptomatic disease? Hospitalisation associated with a particular pathogen? Death from that disease? Illness from all causes?
These are related questions, but they are not interchangeable.
A vaccine is generally given to prevent future disease. Its clinical value depends on what happens after vaccination compared with what would have happened without it. The choice of outcome therefore matters enormously.
My concern was that a precisely defined research finding could become a much broader promise as it moved into public discussion.
A result concerning one laboratory-confirmed disease could sound like an equivalent reduction in illness generally. A result concerning pathogen-associated hospitalisation could be heard as the same reduction in hospitalisation overall.
The narrower finding may be valuable. The broader claim still needs evidence.
The comparison group can already be ill
One method worth understanding is the test-negative case-control design.
In its conventional form, researchers recruit people who meet a clinical definition—for example, people seeking medical attention for respiratory symptoms.
They test for the pathogen under investigation. Those testing positive become the cases. Those testing negative become the controls. Researchers then compare vaccination histories between those groups.
The controls are therefore not necessarily healthy people. They may be substantially ill, but they did not test positive for the pathogen defining a case.
There is a rationale for this approach. Selecting people who sought care for similar symptoms can reduce some differences in healthcare-seeking behaviour. But the interpretation depends on assumptions about recruitment, testing and the relationship between vaccination and the illnesses represented in the control group. Systematic review of test-negative methodology
The important question is what that comparison allows us to conclude.
Does it estimate protection against a specified disease? Does it establish the intervention’s effect on the overall burden of illness? Does it capture adverse outcomes that fall outside the case definition?
Those questions should remain visible when the effectiveness percentage is communicated.
The polio example
The distinction is particularly striking in research involving paralysis.
A study published in 2012 used records covering 46,977 children with acute flaccid paralysis in Pakistan and Afghanistan.
Children with wild poliovirus identified in stool were classified as poliomyelitis cases. Children with two adequate stool samples containing neither wild-type nor vaccine-related poliovirus were classified as having non-polio acute flaccid paralysis and could serve as controls.
Both groups had experienced paralysis.
Other viruses could still be present in children classified as non-polio cases. The paper reported other enteroviruses in approximately 30% of non-polio paralysis stool samples in two regions, citing unpublished surveillance observations. That finding did not establish what caused each child’s paralysis.
Children with vaccine-derived poliovirus were excluded from the analysis. O’Reilly and colleagues, 2012
The clinical problem and the laboratory classification were separate parts of the study.
The effectiveness analysis concerned paralysis associated with the specified wild poliovirus. It did not independently establish the vaccine’s complete effect on paralysis from every cause.
A Nigerian study published in 2008 also excluded children whose stools contained vaccine virus. That exclusion was disclosed; those children were not simply counted as healthy controls. Nevertheless, an analysis excluding them could not, on its own, provide a complete assessment of all potentially vaccine-associated paralysis. Finding vaccine virus in stool would not, by itself, establish that it caused the paralysis either. Jenkins and colleagues, 2008
That leaves a legitimate question:
What evidence establishes protection against the specified disease, and what additional evidence establishes the overall balance of clinical benefits and harms?
A population recommendation needs both.
Could the measurement improve without an equivalent clinical benefit?
This is the methodological concern I wanted discussed.
Suppose an intervention changed the probability of detecting a pathogen without producing a corresponding improvement in the clinical condition. A test-defined outcome could then overstate the clinical benefit.
Investigating that possibility would require evidence about sampling, timing, pathogen detection and the course of illness. It would require looking beyond the classification alone.
But a possibility must not be presented as a demonstrated explanation.
The existence of paralysed test-negative controls does not establish that vaccination merely suppressed a stool-test result while leaving the same poliovirus disease untouched. Nor does it demonstrate that vaccination replaced one cause of paralysis with another.
Those are hypotheses requiring their own evidence.
The criticism I can support is that a disease-specific effectiveness estimate is not a complete account of an intervention’s clinical consequences.
That remains a substantial criticism when the estimate is used to support a broader public claim.
The percentage and the promise
Consider a simplified example.
Imagine two otherwise comparable groups of 10,000 people followed for the same period. Assume each affected person experiences one illness episode.
In one group, there are 100 episodes caused by the target pathogen and 900 caused by other conditions. In the other, there are 50 target-pathogen episodes and the same 900 other episodes.
Target-pathogen illness has fallen by 50%. Total illness episodes have fallen from 1,000 to 950—a reduction of 5%.
This is an invented example to illustrate the distinction, not a finding from the polio research.
The benefit is real: 50 episodes have been prevented. But describing the outcome simply as “50% less illness” would misrepresent what happened.
Severity also matters. Preventing a potentially fatal illness is not equivalent to preventing a mild cold. Counting all episodes together can obscure an important benefit, just as an overly broad headline can exaggerate one.
I want to know what changed, how much it changed and what remained outside the analysis.
The design is not automatically invalid
A serious critique has to acknowledge what the method can do.
Case-control studies deliberately select people according to their outcomes. We cannot infer the population’s overall illness rate by noticing that everyone recruited into the study was ill. Sick controls do not, by themselves, make a study invalid.
There is also evidence that test-negative methods can estimate protection meaningfully. A 2013 analysis applied them to randomised influenza vaccine and RSV antibody trial data and obtained estimates closely matching those from the full trials. De Serres and colleagues, 2013
A 2025 analysis of five randomised COVID-19 vaccine trials likewise found agreement under conditions without the confounding and selection problems that complicate observational research. Andrews and colleagues, 2025
That evidence supports using the method appropriately.
It does not make every application reliable, remove the need to investigate adverse effects or justify translating a narrowly defined result into an unrestricted claim about health.
My objection concerns the assumptions and the interpretation, particularly when the confidence of the public message exceeds the scope of the study.
Why “trust the experts” was insufficient
I value expertise. Much of my thinking has developed through reading people who have spent their careers investigating questions I could not answer alone.
Expertise should help explain why an outcome was selected, why a comparison is appropriate and what would change the conclusion.
An instruction to trust cannot perform those tasks.
When an intervention becomes part of national policy, asking about absolute benefits, relative benefits, adverse effects and differences between groups is reasonable.
Commercial involvement does not automatically invalidate research. Institutional endorsement does not place it beyond examination.
I wanted the reasoning made explicit. I wanted uncertainty acknowledged without it being treated as a threat to the entire recommendation.
That is how trust becomes justified.
My deeper questions concerned immune function
The research methods were only part of my concern.
I wanted to understand why apparently comparable exposures produced markedly different outcomes. Why did one person experience little disturbance while another became seriously ill? What determined whether an immune response was protective, inadequate or itself damaging?
A simplified account of a foreign invader meeting a defensive immune system did not adequately answer those questions for me.
Modern immunology is much richer than that account. Researchers investigate tissue damage, immune regulation, metabolism, prior exposure and individual variation. It would be inaccurate to suggest that everyone in the field subscribed to one simplistic model.
My frustration was with how little of that complexity survived in public discussion.
A scientific theory is an explanatory framework. Calling it a theory does not make it a guess. But no framework becomes complete because it is familiar.
I was interested in what a model explained, where it struggled and whether other perspectives could help us understand more.
Cunliffe’s morphostasis and Matzinger’s danger model
Jamie Cunliffe’s work on morphostasis appealed to me because it considered immune activity within the broader maintenance of tissue organisation and integrity.
It encouraged me to ask what the organism was trying to preserve or restore, and how immune activity contributed to that process. Cunliffe, Morphostasis and immunity
Polly Matzinger’s danger model challenged the adequacy of foreignness alone as an explanation of immune activation. It emphasised signals associated with damage and distress, directing attention towards the condition of the tissues in which an encounter occurs. Matzinger, The danger model: a renewed sense of self
These approaches helped shape my thinking.
They do not establish that pathogens are irrelevant or that vaccination cannot confer protection. They offer ways to investigate how the organism responds to a challenge and why that response has different consequences in different circumstances.
My proposed proportional danger theory
I developed these ideas into what I called proportional danger theory, introduced in my earlier writing on immune theory and developed further in my Proportional Danger Theory working paper on EPK Research.
My proposal considers a new challenge in relation to the person’s existing physiological condition and the resources available to respond and recover.
The context I wanted to investigate included other infections or viral exposures, inflammatory activity, nutritional inadequacy, underlying disease and the demands already being placed on tissue maintenance and repair.
A new exposure arrives in a person with an existing biological history.
My suggestion was that the significance of the challenge depends partly on that context: the disturbances already present and the capacity available to accommodate and resolve them.
The objective is an appropriate response followed by recovery. Greater immune activation is not automatically a better outcome.
This remains a proposed framework, not an established explanation of individual disease outcomes. The working paper is a conceptual preprint, not a peer-reviewed validation study. It sets out evidence, limitations and tests that could challenge the hypothesis.
It should face the same scrutiny I ask of other explanations.
Context also does not operate in one simple direction. Prior exposure can provide protection. Interactions between infections vary. A vigorous response can sometimes contribute to injury. We cannot simply add up exposures and assume that more always means worse.
Why the individual matters
A controlled SARS-CoV-2 human challenge study reported in 2024 identified abortive, transient and sustained infection patterns, alongside differences in local and systemic responses.
Even under controlled exposure conditions, participants did not follow an identical biological course. That supports investigating individual differences; it does not establish my particular explanation for them. Lindeboom and colleagues, 2024
The distinction between resistance and disease tolerance is also relevant.
Resistance concerns controlling a pathogen. Tolerance concerns limiting the harm associated with infection without necessarily reducing the pathogen burden. Medzhitov, Schneider and Soares, 2012
That distinction matters when deciding what an intervention should achieve and how its effects should be measured.
The person’s functioning, symptoms and recovery matter alongside the pathogen measurement.
None of this justifies blaming people who become seriously ill. Healthy people can suffer severe infections. Diet and underlying health are parts of a complicated picture, not guarantees of protection or measures of personal virtue.
The discussion I wanted to have
A vaccine can demonstrate a clinical benefit while our understanding of immune function remains incomplete. Equally, a plausible account of how an intervention should work cannot replace evidence that it improves the outcomes being claimed.
We need both clinical evidence and explanations that remain open to testing.
I wanted to know precisely what vaccination had been shown to prevent, the size of that benefit, which people the finding applied to and what risks remained. I wanted the limitations of the method to remain visible when the results were communicated.
I also wanted room to explore competing explanations of immune function without that interest being mistaken for blanket opposition to vaccination.
Those questions leave room for vaccination. They leave room for disagreement about a particular product, schedule, population or recommendation.
The anti-vaccine label erased those distinctions before the discussion began.
I was not anti-vaccine then. I am not anti-vaccine now.
I remain interested in immunology, uncomfortable with certainty that outruns the evidence, and willing to ask questions—including questions about my own ideas.

