In pharmacology, the criteria for causation are quantitative, objective, and numerous: an agonist must reproduce the action of a putative neurotransmitter; the action of neurotransmitter and agonist must be competitively shifted to the right with the same Schild coefficient by an antagonist, with statistical confidence quantitated with P values. For more complex hypotheses, reproducibility, predictive value, and parsimony are the classic criteria of the Scientific Method. However, stepping out of the laboratory, the scientist needs to find common language with non-scientists, which usually means more philosophy and less quantitation.

The so-called “Bradford-Hill Criteria” have been elevated to the status of "Gold Standard” in determining Causation. The reliance on these criteria and especially flawed application of the criteria lead to erroneous conclusions (Phillips and Goodman, 2004). Hill himself (1965) used the term “considerations” rather than “criteria”, and raised numerous cautions concerning their use.

The first consideration, Strength of Association, is a particularly insidious one. The doubling of risk is sometimes considered to be an indication that some factor is causal, and is often used in legal proceedings to determine if a particular cause is “more likely than not”, and even sometimes construed to purport that a cause is established with “reasonable medical certainty”. It is analogous to the doubling of signal over noise to be considered detectable, a rule of thumb for engineers. However, statistically speaking and based on real world experience, the magnitude of an effect is independent of its statistical significance and true causal role. In one case, a ratio* of 1.4 could be highly significant and indicate causation, whereas in another case, a ratio of 5 could be fortuitous and meaningless.

The application of the Bradford Hill Considerations to the adverse reactions putatively attributed to thimerosal by the Institute of Medicine (IOM, 2001) is an example of use of the guidelines. Perhaps it is a telling outcome, however, that no definitive conclusions were reached concerning the ability of thimerosal to cause autism primarily because of failure to satisfy the last (and most subjective) criterion, Biological Plausibility. The Institute of Medicine did however recommend further investigation, and adverse reactions including autism to thimerosal continue to be controversial today.

The Bradford Hill consideration of Dose-Response (#2) should at a minimum be sharpened to take into consideration Exposure - defined as Dose x Duratioin - as the most likely correlate of Adverse Reactions.

Gerneral Causation and Specific Causation.
Specific Causation is the domain of the physician whereas General Causation is the domain of the scientist. When a patient deals with a patient, N=1 and there is no chance uisually to replicate the experiment. Statistics do not apply. Healing is indeed an art and a physician's opinion is influenced by diverse and subjective factors.

*ratio: Risk Ratio or Signal-to-Noise Ratio.

1. Strength of Association – The strength of an association refers to the magnitude of the measure of effect of an exposure, usually the relative risk or odds ratio, in a study comparing an exposed and an unexposed group. The larger the magnitude of the effect the less likely any observed effect is due to chance, bias or confounding. In general, in observational studies, relative risks of 2 or less are considered to be evidence of a weak association, because the likelihood the effect is due to chance, bias or confounding is greater than if the effect is larger).

2. Dose-Response Relation – The existence of a dose-response relation strengthens an inference that an association is causal. A dose-response relation is defined as an increased strength of association with increased magnitude of exposure.

3. Temporally Correct Association – Exposure must precede the event by at least the duration of disease induction. This consideration may be limited by the fact that knowledge of the pathogenesis and natural history of an adverse event may be insufficient.

4. Consistency of Association – This consideration requires that an association be found regularly in a variety of studies, using different study populations and study methods.

5. Specificity of Association – Uniqueness of an association between an exposure and an outcome provides a stronger justification for a causal interpretation than when the association is nonspecific. However, perfect specificity between an exposure and an effect cannot be expected in all cases because of the multifactorial etiology of many disorders.

6. Biological Plausibility – The existence of a possible mechanism of action that fits existing biologic or medical knowledge is thought to increase the likelihood that an association is causal.

Hill AB, The environment and disease: association or causation?
Proc R Soc Med 1965, 58:295-300.

IOM (Institute of Medicine). Thimerosal-containing vaccines and neurodevelopmental disorders. Washington DC: National Academy Press; 2001

Phillips CV and Goodman KJ, The missed lessons of Sir Austin Bradford Hill. Epidemiol Perspect Innov. 2004, 1:3