Hill's Criteria:
Sir Austin Bradford Hill was an English epidemiologist who investigated disease and injury in the workplace. He pioneered randomised clinical trials. Together with Richard Doll, he demonstrated the connection between cigarette smoking and lung cancer — which had been investigated as early as the 1920s by German researchers.
Hill noted that the goal of causal assessment is to determine if there is “any other way of explaining the set of facts before us." His 9 criteria for causation can be stated as follows"
1. Strength of Association: A strong association between exposure to a supposed toxin and the outcome is more likely to be causal. This means that a large effect size or a substantial increase in risk should be observed. (So look for a large correlation. The closer the correlation is to 1.0 the higher the effect size. In the Swanson et al study, for example, correlations of R=0.989 represent a very large effect size. R.A.)
2. Consistency: The association should be consistently observed in different studies, populations, and settings. (This is important in study design. In the Swanson study, very large data sets were collected but across locales — so covered a broad range of populations and settings. R.A.)
3. Specificity: The exposure should be specifically linked to the outcome, rather than being associated with multiple outcomes. (Investigators often compare the outcomes of exposed and unexposed groups)
4. Temporal precedence: The exposure should precede the development of the outcome in time. In other words, the cause should come before the effect. (This is pretty self explanatory. Did a person break his leg and as a consequence fall down the stairs, or did he fall down the stairs and hence break his leg.)
5. Dose-Response Relationship: There should be a dose-response relationship, meaning that increasing levels of exposure are associated with a corresponding increase in the risk of the outcome. (More recent experts have noted that dose-response can be less useful, as confounding variables can intervene. When dealing with issues at the level of genomes, very low exposure levels can none-the-less cause damage that is then replicated by the processes of DNA replication. R.A.)
6. Plausibility: The proposed causal relationship should be biologically plausible based on existing knowledge and understanding of the mechanisms involved. (This, to me, is one of the most important elements of establishing causality. With glyphosate, for example, the mechanisms can be very different from disease to disease, ranging from microbial, though chelations issues, to interfernce with RNA replication. R.A.)
7. Coherence: The causal relationship should be consistent with what is known about the natural history of the disease and other established facts.
8. Experimentation: Experimental evidence, such as intervention studies or randomized controlled trials, can provide stronger evidence for causality.
9. Analogy: Similarities with other established causal relationships can strengthen the argument for a new causal relationship.
So if faced with the old "but correlation does not prove causation" argument, one needs to be armed with a true understanding of mechanism, and how study design can impact the result.