At the end of the lecture students should be able to:
- Explain what other sources of evidence are used in addition to statistical inference to draw conclusions about causative relationships
Alderson, M., 1983. An Introduction to Epidemiology, 2nd ed., pp. 320-366. The Macmillan Press Ltd. London.
A brief discussion on inference in epidemiological studies, with many references to discussions of causation and related issues.
Hill, B.H. & I.D. Hill, 1991. Bradford Hill's Principles of Medical Statistics, 12th ed., pp. 270-279. Edward Arnold, London.
Classic discussion of statistical evidence and inference.
Association is not the same as causation
Just because we have established a statistically significant association between two variables does not mean that one causes the other. We need other evidence to make a convincing case.
Bradford Hill's indicators of a causal association
Bradford Hill (ref. above) gave a number of factors which help point towards a causal link between two associated (or correlated) variables. The list below is a slightly condensed version, which contains the essential features.
1. The strength of the relationship
If the risk of disease associated with a particular activity is many times the risk experienced by the 'normal' population we are more likely to accept it as a causal factor.
Is there a dose-response effect? Does increasing the exposure increase the problem? If we remove the cause do we remove the risk?
Has the finding been confirmed by other studies? Ideally, repeatedly observed by different people in different places at different times. Sometimes, however, we may have to make decisions on unique events.
Is the association biologically plausible? This depends on the current state of knowledge.
Does the association conflict with the generally known biology and natural history of the disease? We have to bear in mind the current state of knowledge: we may be uncovering an, as yet, unknown biological mechanism.
Is exposure followed by outcome?
Is the association limited to particular sites or types of disease? The specific association between one exposure and one problem increases our confidence in causality but absence of specificity should not be, necessarily, taken as absence of causation.
Are there other comparable examples of cause and effect relationships?