5: Causation models (Bradford Hill criteria)

Section 1: Foundations - Causation Models (Bradford Hill Criteria)

Establishing a causal relationship, rather than mere association, is a core challenge in epidemiology. While definitive proof often requires experimental designs (like RCTs), much epidemiological evidence comes from observational studies. Sir Austin Bradford Hill, in 1965, proposed a set of nine criteria to help assess whether an observed association might plausibly reflect a causal effect. These are not rigid rules but a framework for evaluating evidence.

1. Strength of Association: Strong associations (e.g., large relative risks) are less likely to be explained solely by bias or confounding. A very strong association (like smoking and lung cancer) provides more compelling evidence than a weak one. However, weak associations can still be causal (e.g., passive smoking and lung cancer).
2. Consistency: Repeatedly observing the association in different populations, using different study designs, and by different investigators strengthens the case for causality. If multiple studies across diverse settings find a similar link, it's less likely to be due to chance or local bias.
3. Specificity: This refers to a cause leading to a single specific effect, or an effect resulting from a single specific cause. While appealing, it's the weakest criterion. Many exposures cause multiple diseases (e.g., smoking), and many diseases have multiple causes. Its presence can be supportive, but its absence doesn't rule out causation.
4. Temporality: This is the only essential criterion. The cause must precede the effect in time. Demonstrating exposure occurs before disease onset is fundamental. Cohort studies are generally better at establishing temporality than case-control studies.
5. Biological Gradient (Dose-Response): Evidence of increasing risk with increasing levels of exposure (e.g., higher doses, longer duration) supports a causal interpretation. Observing a trend where heavier smokers have higher lung cancer rates than lighter smokers strengthens the argument.
6. Plausibility: A biologically plausible mechanism, based on current scientific knowledge, makes the association more credible. However, lack of known mechanism doesn't disprove causation, especially if the evidence is otherwise strong – science may simply be incomplete.
7. Coherence: The causal interpretation should not conflict with the generally known facts of the natural history and biology of the disease. It should fit reasonably well with existing knowledge about disease processes and population patterns.
8. Experiment: Evidence from experimental settings (e.g., RCTs, animal studies, or natural experiments like removing an exposure) can provide strong support. If removing the exposure reduces disease incidence, causality is strongly suggested.
9. Analogy: If an association is analogous to another established causal relationship, it lends some support. For example, if one drug in a class causes birth defects, it raises suspicion about similar drugs.

The Bradford Hill criteria guide epidemiologists in weighing the totality of evidence. They emphasize the need to consider multiple aspects of an association beyond statistical significance. No single criterion is necessary or sufficient, but collectively, they help distinguish causal relationships from non-causal associations, informing public health action and clinical practice.