Bradford Hill Criteria for Causation

The nine considerations

1. Strength — large effect sizes harder to explain by confounding.
2. Consistency — replication across populations, settings, studies.
3. Specificity — single cause, single effect (weak criterion; modern diseases are multi-factorial).
4. Temporality — cause must precede effect (the only strictly required criterion).
5. Biological gradient — dose-response relationship.
6. Plausibility — coherent mechanism (depends on state of knowledge).
7. Coherence — fits with existing biological knowledge.
8. Experiment — RCT confirmation or natural experiment.
9. Analogy — similar agents producing similar effects.

How to use them

Hill explicitly stated these are not a checklist. Causation can be inferred without satisfying all; satisfying all does not prove causation. They are a structured way to reason from observational data.

Where they matter in longevity

• Saturated fat → CVD: meets most criteria (strength, consistency, temporality, gradient, plausibility, experiment via intervention trials, coherence).
• Coffee → lower mortality: meets consistency, dose-response, plausibility; lacks RCT confirmation.
• Supplements: many observational signals fail to satisfy temporality (reverse causation) or specificity (confounded by overall health-conscious behaviour).

Limitations

• Doesn’t formalise confounding adjustment.
• "Plausibility" depends on current understanding (can be misleading).
• Modern tools (Mendelian randomization, target-trial emulation) complement Hill’s reasoning.

Related entries

Mendelian randomization, Confounding, Survivorship bias.