Causal association strength

The subject of what evidence is necessary to conclude that an exposure is causally associated with disease has received much discussion over the years. In 1964, the seminal report to the Surgeon General on Smoking and Health [United States Department of Health, Education, and Welfare (DHEW 1964)] published criteia by which to evaluate whether an exposure was causally related to disease. These criteria were consistency, strength of association, specificity of the association, temporal relationship of the association, and coherence of the association. 

Cohort studies measure the strength of an association between an exposure and outcome of interest or disease by comparing the incidence of disease in the exposed population to the incidence of disease in the unexposed population. Absolute measures of association calculate the difference of disease between the exposed and unexposed groups. Excess or attributable risk is the number of extra cases of disease that the exposure is responsible for, assuming that the relationship between exposure and disease is causal. 

Strength of evidence involves the enumeration of tumours in human and animal studies and determination of their level of statistical significance. Sufficient human evidence demonstrates causality between human exposure and the development of cancer, whereas sufficient evidence in animals shows a causal relationship between the agent and an increased incidence of tumours. Limited evidence in humans is demonstrated by a positive association between exposure and cancer, but a causal relationship cannot be stated. Limited evidence in animals is provided when data suggest a carcinogenic effect, but are less than sufficient. The terms sufficient and limited are used here as they have been defined by the International Agency for Research on Cancer (lARC) and are outlined in 3.6.5.3.I. 

Criteria similar to those listed by Adams (2003) are used to establish causality and are derivatives of Koch s postulates and Hume s criteria. The list includes (1) strength of association, (2) consistency of association, (3) specificity of association, (4) time order or temporality, (5) biological gradient over space and time, (6) experimental evidence available, and (7) biological plausibility. In many instances, especially at a regional scale and over long periods of time, meeting the requirements for each of these criteria can be difficult. 

Certain basic criteria in epidemiology (Susser 1986) have been widely accepted and appear to be equally applicable to ecoepidemiology (Fox 1991). It should, however, be clearly appreciated that these techniques are not designed to establish unambiguous causal relationships, but rather to indicate which criteria may most usefully be used to provide a balanced evaluation in support—or in contradiction—of a given hypothesis it should be noted that the relatively loose term association is consistently used. These five criteria are (1) consistency, (2) strength, (3) specificity, (4) temporal relationship, and (5) coherence. 

Strength Associations which are not robust do not necessarily disprove a causal relationship. 

In the context of medical causation. Sir Austin Bradford Hill, suggested in 1965 that to imply causation from the observation of association we should consider its (1) strength, (2) consistency, (3) specificity - the restriction to specific conditions, (4) temporality - the order of events, (5) dose-response relationship, (6) theoretical plausibility, and (7) coherence - the consistency with other related phenomena. Although these guidelines were presented in the context of medicine and epidemiology, it would be very usefiil to keep these necessary conditions for causality in mind when evaluating crash causation on the basis of statistical associations. 

---