Misconceptions about the healthy vaccinee bias

Having posted numerous articles about the healthy vaccinee effect (bias), I sometimes encountered misconceptions, which I will address here.

1. The healthy vaccinee bias is a type of selection bias

No. It is a type of confounding bias. To understand why, we need to understand how biases are classified in causal diagrams. Older methods of classification blur the picture. Like any methodological area, you need to invest time in learning the topic. You can start here, and if you want to read more, try here (“On the definition of a confounder.”) Neither is light reading. Methodology is a very complicated business.

2. There are different types of healthy vaccinee effect

No. As I wrote, the healthy vaccinee bias is confounding bias, which arises from variables called confounders. A confounder is a shared cause of the exposure (vaccinated or not) and the outcome (dead or alive). That’s the causal structure of confounding bias.

Under the null (no effect of vaccination), we omit the arrow from E to D.

Any variable, C, that meets this causal structure is a confounder, whatever it is. The name of the cause (C) of vaccination status (E) does not matter. It could be the fact that a patient in a hospice is not vaccinated or that a homeless person does not care about vaccines. As far as confounding is concerned, it makes no difference whether the explanation is the mindset of a physician or the mindset of another person. The only methodological distinction is between time-stable confounders (e.g., sex chromosomes) and time-varying confounders (e.g., smoking status).

Here is a generic example of a time-varying confounder (C), which may take different values over time. The sequence of short arrows arises from an axiom of continuity.

3. There is short-term and time-varying healthy vaccinee effect versus long-term and time-invariant healthy vaccinee effect.

No. As I explained above, there is no methodological difference between a terminally ill patient who is not vaccinated (“short”) and a homeless person who is not vaccinated (“long”). They translate into indistinguishable confounding paths in a generic causal diagram. Moreover, time-dependent effects may change continuously, and any dichotomy on the time axis is usually artificial. Our minds like to simplify complex reality and to convert continuous traits or phenomena into “simple” categories.

There are data to show that the healthy vaccinee bias is not stable over time, but is gradually attenuated. Again, no dichotomy between “short” and “long.” It is a continuum.

I explained elsewhere:

“Given the shorter life expectancy of the unvaccinated cohort, the most vulnerable members of that cohort died earlier. The remaining people gradually made up a somewhat ‘healthier’ surviving cohort, thereby narrowing the non-Covid mortality gap between the unvaccinated and the vaccinated.”

The causal structure that corresponds to that text is called competing risks (without censored observations!). The magnitude of the confounding bias is attenuated because a negative confounding path is operating. Homelessness might cause non-Covid death, and death from another cause “prevents” later death from Covid. We don’t die twice. Try to draw the causal diagram, adding + and- over the arrows. It is called a “sign diagram.”

4. The effect of Covid vaccines on non-Covid death is not fully a pseudo-effect. They may have a true (beneficial) effect on death from other causes. (And let’s forget about vaccine-related fatalities.)

I like this one. It is an excellent illustration of a well-known idea in the philosophy of science: every observation is compatible with at least two theories. (And that’s one explanation of why no theory is ever logically proven.)

I have a better example, which I used to teach: We observe someone being shot and dropping dead. Was he killed by the shot? Perhaps not. Perhaps he had a heart attack just before the bullet hit, and he would have died anyway. Do you want to counter by an autopsy report? Perhaps the report was falsified. You get the point. The only question is how far you are willing to go on the trail to endless absurdity in order to reject (or save) a theory.

If you want to claim that flu vaccines and Covid vaccines protect against stroke, design a study, get positive results, and start prescribing them to prevent stroke. Potential effects are general theories. Competing theories compete by using solid, direct data.

That vaccinated people have “healthier characteristics” has solid empirical support, and therefore the healthy vaccinee effect is real. The claim that not all of it is bias rests on shaky empirical grounds for the time being.

5. We take care of the healthy vaccinee effect by relying on available methods for adjustment.

No. They do not capture the full spectrum of health-related variables, as shown repeatedly for flu vaccines and Covid vaccines. Sometimes, measured variables don’t even make a difference. There was a study from Israel that showed exceptional effectiveness, where “adjusted” estimates and “crude” estimates were almost identical.

6. The bias correction factor using non-Covid deaths should be replaced by bias factors from a set of various types of “negative control” outcomes.

No. Where is the evidence that it is better? How exactly will the correction be done? Which set of outcomes should be chosen, and why? It is irresponsible to make claims like this, if truth seeking is the goal. Competing theories on the preferred methodology should be explicit, rationalized, and testable. Not vague handwaving.

7. We should first adjust for measured variables and take care of the remaining healthy vaccinee bias.

That’s a serious suggestion, but it is not necessarily correct. The topic has not been studied, and I have reasons to believe that that approach might be inferior to my single-step approach, which Høeg also used implicitly. I have yet to read arguments to the contrary. Try to read here to find out how complicated the issue is before telling others what is “right”.

The last one is long.

8. Proponents of Covid vaccines have applied the same level of scrutiny to an endless number of “great effectiveness” studies and to much fewer studies and posts that criticized “great effectiveness” studies. Moreover, it has been easy to publish the latter in scientific journals. There was no publication bias, certainly not in the case of the healthy vaccinee effect.

True or false?