The Healthy User Effect and Who Buys Supplements

This is written to be used rather than admired. Each section below is a decision about the healthy user effect in observational research, and each one has a default.

What the effect is

People who take up a health behaviour differ systematically from those who do not, in ways that independently affect their health. Supplement users on average exercise more, smoke less, drink less, weigh less and attend screening more reliably.

They also tend to have more education and higher income, both of which predict better health outcomes on their own. A study finding that supplement users are healthier has found something true and has not found that the supplement did it. The effect is one of the most consistent findings in observational epidemiology and it is not a fringe objection.

Why adjustment does not fix it

Researchers adjust statistically for measured differences, which requires having measured them and measured them accurately. Smoking recorded as yes or no misses how much, for how long and how long since stopping, so adjustment removes only part of it. Factors nobody recorded, from sleep to social connection to unmeasured illness, cannot be adjusted for at all.

Where the claim is carefully worded, residual confounding is the standard term for what remains, and it usually points in the same direction as the original bias. Adjusted results are better than unadjusted ones and are not equivalent to a randomised comparison.

The reverse version

People who feel unwell also change behaviour, which can create an association pointing the other way. Someone with an undiagnosed illness may lose weight, drink less alcohol and stop exercising, making those behaviours look harmful.

This reverse causation explains several counterintuitive findings that resurface periodically in the popular press. Excluding the first few years of follow-up is one way researchers test for it, and good studies report that analysis. A finding that disappears when early events are excluded was probably illness causing the behaviour rather than the reverse.

Where it has misled everyone

Several nutrients showed strong protective associations in observational studies and then failed in large randomised trials. That pattern has repeated often enough to be predictable, and it is the strongest practical argument for waiting for trials. In a few cases the trials found harm at high doses, which observational data had given no reason to expect.

Where the claim is carefully worded, the lesson is not that observational research is worthless but that it generates hypotheses rather than settling them.

Anyone citing observational data as proof of a supplement benefit is ignoring a well-documented history.

Why randomisation solves it

Assigning people at random distributes both known and unknown characteristics roughly evenly between groups. It balances the factors nobody thought to measure, which is the specific thing adjustment cannot do. This is why the randomised trial occupies the position it does, despite being expensive, slow and often impractical.

Follow the mechanism and trials also have limits, including short duration, selected participants and dropout, and none of these is the confounding problem. Where randomisation is impossible, techniques exist to strengthen observational inference, and none of them fully replaces it.

Spotting it in a headline

Look for the word associated, which is the honest signal that no causal comparison was made. Ask whether the behaviour in question is one that health-conscious people tend to adopt, which covers most wellness products.

Ask what was adjusted for and whether the obvious confounders appear in that list. Ask whether a randomised trial of the same question exists, since it usually does and usually shows less. The reflex worth building is to picture the two groups of people rather than the two treatments.

The takeaway

The vitamin takers were already different people. Adjustment removes what was measured, and randomisation removes what nobody thought to measure.

An extraordinary mechanism needs better evidence than testimonials, and usually has less.