Mechanism, Directness, Dose & Generalization

A mechanism can be plausible without producing the claimed real-world effect.

An observed effect can also be real even before its mechanism is fully understood.

Good evidence evaluation keeps these questions separate.

Mechanistic evidence asks how an effect could occur

Mechanistic evidence can include:

• biochemical pathways;
• physical interactions;
• receptor binding;
• device physics;
• toxicological processes;
• physiological models;
• intermediate biological effects.

This evidence helps answer whether a proposed causal pathway is plausible.

It does not automatically answer how large the real-world outcome is.

Direct outcome evidence asks whether the claimed effect occurs

If the claim is:

Exposure X causes disease Y in humans.

then evidence of a molecular interaction is indirect.

Direct evidence would observe the relevant health outcome in the relevant population, under relevant exposure conditions.

Mechanistic and outcome evidence can support each other, but they answer different questions.

Dose is often decisive

A substance can produce a biological effect at one dose and no detectable meaningful effect at another.

The same principle applies to:

• radiation;
• chemicals;
• nutrients;
• drugs;
• electromagnetic exposure;
• temperature;
• mechanical stress.

Evidence at very high experimental exposure cannot automatically be generalized to ordinary exposure.

That is why “it has an effect” is incomplete without:

at what dose?

Exposure route matters too

Inhaling a substance can differ from ingesting it.

Injecting a compound can differ from eating it.

Skin exposure can differ from systemic exposure.

Experimental delivery methods can therefore create effects that do not map directly onto real-world exposure.

Surrogate outcomes require validation

A biomarker or intermediate endpoint may be easier to measure than the outcome people actually care about.

But changing a surrogate does not guarantee changing the final outcome.

Blood markers, imaging findings and laboratory endpoints can be useful only to the extent that their relationship with meaningful outcomes is established for the question at hand.

This is one reason GRADE treats indirect outcomes as a potential reason to lower certainty.

Generalization is an empirical question

A study can be internally valid and still apply poorly to a new setting.

Differences may include:

• age;
• disease severity;
• genetics;
• environment;
• dose;
• treatment implementation;
• follow-up time;
• healthcare context.

External validity asks whether the result transfers beyond the study conditions.

Animal and cell evidence have real value

Cell and animal models can reveal mechanisms, identify hazards and guide experiments.

The mistake is not using them.

The mistake is skipping the translational steps and presenting a preclinical effect as though it established a human clinical outcome.

The evidential claim should match the evidential level.

Converging direct and mechanistic evidence is strongest

A causal interpretation is especially persuasive when:

• the real-world effect is observed;
• the dose-response pattern is coherent;
• the mechanism is plausible;
• timing fits;
• different study designs converge.

No single component is always required, but agreement across them reduces alternative explanations.

The Claim by Claim translation check

Whenever evidence looks impressive, we ask:

Does this study test the same population, exposure, dose, outcome and timescale as the viral claim?

If not, the gap has to be stated rather than silently crossed.