Focus on Alternative and Complementary Therapies
www.pharmpress.com/fact
Focus Alternat Complement Ther©2005 Pharmaceutical Press
Focus Altern Complement Ther 2004; 9: 1–2
Some CAM enthusiasts are less than convinced that scientists’ ‘obsession’ with causality is reasonable. They claim that reductionistic concepts of this type are not applicable to CAM and believe that, in CAM, things are so subtle and complex that the ‘naivety’ of linear thinking is simply not applicable. I beg to differ and will use an example from the recent CAM literature to explain why.
Risberg et al. recently conducted a longitudinal study in which they initially recorded the CAM usage of Norwegian cancer patients.1 Subsequently this population was followed up, with the surprising result that, on average, those who used CAM lived less long than those who did not use CAM. In principle, such a finding is compatible with two dramatically different interpretations: (i) CAM use causes early death or (ii) other factors associated with CAM use cause early death. To deny that causation is irrelevant in CAM would be equivalent to saying that it is irrelevant which of the two interpretations is correct. Such a view is obviously untenable, but how can we establish causality? How can we find out which of the two interpretations of Risberg et al.’s finding is correct?
For many years six criteria have been known in science that can be used to assess the likelihood of causality of associations:
In the case of the Risberg study, one might argue that some forms of CAM are burdened with risks that could hasten death; such risks have repeatedly been described.2 One could also assume that, in some cases, CAM use entails giving up life-saving conventional therapies that in turn could be responsible for early death.3 Thus the biological plausibility of CAM being a causal factor does theoretically exist.
Specificity of association, in the above example, means that only CAM use, and not some other associated factor, is linked with early deaths. The researchers should thus look at their data carefully to exclude the influence of any other variable of potential importance (see below). Only if that is possible and done thoroughly can we be sure about specificity. The obvious problem is that any data set is limited, and one can never draw conclusions about factors that have not been assessed in the first place.
Consistency of the observations is equally important; it is the reason why scientists insist on independent replication with monotonous regularity. If an observed phenomenon (e.g. CAM users die earlier) is real (i.e. due to a causal relationship), it will show up over and over again. In other words, if researchers in the US or elsewhere repeated the Risberg study, they would observe similar phenomena. In the absence of consistency of an association, the observed phenomenon might well be an artefact of some unknown confounding factor.
It is obvious that the cause can never take place after the effect; temporal correctness of the association is thus an essential precondition for causality. In a famous Italian randomised study, patients’ charts were submitted to ‘energy healing’ years after the patients had been dismissed from hospital.4 Patients whose charts received the intervention turned out to be associated with shorter hospital stays than those whose charts were left untouched. This study is an excellent example of the importance of temporal correctness; this factor alone can demonstrate the absurdity of a result. In the case of the Risberg study, CAM use was obviously before death. Thus, based on this criterion, causality is a possibility.
The strength of an association describes the size and the robustness of the difference between two groups. In the Risberg study, 79% of the CAM users and 65% of the non-users died during the 8-year follow-up period. This effect is significant but it is difficult to estimate the strength of the association using these data alone.
Dose–response relationships are good indicators of causality. Did patients who used CAM once a month live longer than those who used CAM once a week, and did that group live on average longer than those patients who used CAM on a daily basis? Similarly, one could ask whether patients who used 10 forms of CAM died earlier than those who employed five forms of CAM, etc.
This discussion shows that causality matters – not just in an academic sense but in a real and practical way. It also demonstrates that causality can be assessed systematically and that such an assessment can be a prerequisite of understanding what is going on. In many instances, this scrutiny will open up new relevant questions that need further research. In Risberg’s study, for instance, one would want to conduct a careful analysis according to disease severity. It is possible (even likely) that CAM users were more severely diseased than non-users (they could, for instance, have tried CAM out of desperation with the severity of their cancer). If that were true, the association between CAM use and early death would have failed the specificity test – thus it would probably not be a cause–effect relationship. Consider this and then try to argue against the importance of assessing causality in CAM!
