Commentary: Teaching causal thinking (Commentary on ‘The association between tobacco and coronary heart disease’)
Lorann F. Stallones · International Journal of Epidemiology · 2015
In a time of electronic access to published papers and computer programs capable of conducting meta-analyses and systematic reviews, it is difficult to remember the work that was involved in conducting a comprehensive literature review of an area of research. Obtaining all relevant papers from libraries other than the one where you worked could take weeks. Hand-searching through citation references, the list of cited papers in published papers and compiling a list of those took intense concentration and meticulous attention to details contained in each paper. The report prepared by Dr RA Stallones, who was my father, is just such an accomplishment.1 When he was working on something like this he travelled everywhere with a rather large briefcase so he could carry the papers, his notepads and his graph paper. The topic of interest, the association between tobacco smoke and coronary heart disease, was a significant issue at that time but the link was not firmly established, reducing the motivation for the public health community to implement prevention strategies. Using what are now well established criteria to evaluate whether the relationship between tobacco smoke and coronary heart disease represented a causal link, he evaluated the literature available at that time, doing by hand what could not be done by computer, in order to draw conclusions with regard to the causal nature of the association. He also presented alternative explanations which would account for the findings of the studies, other than a causal relationship, including unknown sources of bias resulting in spurious associations and what we would call publication bias resulting in suppression of findings contrary to prevailing doctrine. The paper is much like a lesson in the use of epidemiological findings to establish a causal association by compiling all of the evidence, arraying it in a logical manner and computing risk ratios as appropriate. The thoughtful application of what have become to be known Hill’s postulates,2 which were included in the introduction to Smoking and Health,3 to a specific question of causality makes his paper a tool for teaching epidemiology in an applied manner rather than in an abstract way. The refinements of his use of standard epidemiology (person, place, time) in the paper represents an exercise that provides us with an example of why each of those can be used to frame a logical conclusion with regard to a causal association. For example, the fact that males (person) smoked more and also had higher rates of coronary health disease provided a clue to explore the association. He discussed the fact that the finding of an association was found in different countries (place). He addressed the issue of changing smoking patterns coinciding with changing patterns of coronary heart disease (time). In the third section of the paper, he refines his discussion of the association by specifying the association as related to myocardial infarction. Expanding on the logical discussion of a spurious association, he suggests that assessing all the studies in different populations, at different times and with different methods, makes the argument that there is some hidden but consistent bias unlikely. For those of us who teach epidemiology to students, we often find them invoking some hidden bias as an explanation for a research finding and it would serve us all well to have them go back and try to explain how that might operate across a wide range of studies. Dr. Stallones comments on the lack of a reasonable pathogenetic mechanism to support a direct causal association. This is another excellent reminder to the teachers that sometimes having an underlying mechanism is important but, more importantly, when there is no plausible biological mechanism we may have to wait for our colleagues in the basic sciences to point the way. Epidemiological studies can serve to identify gaps in our understanding of pathogenesis at times. His careful review of risk factors for coronary heart disease, which may be directly or indirectly explaining the association with smoking, provides another opportunity to teach our students how to carefully consider alternative explanations for associations we find in epidemiological studies. Finally, his conclusion that social action to reduce tobacco use was needed in light of lack of clear scientific evidence of a causal association is a reflection the dilemma of many of the public actions that are proposed. Balancing individual choice and preference with the potential for great benefit to the population is always a paradox that public health professionals face. As John Locke wrote: ‘For where is the man that has incontestable evidence of the truth of all that he holds, or of the falsehood of all he condemns … ?’4 Epidemiological knowledge is always changing, and critical review of the current state of understanding of causal relationships using logic and assessing alternative explanations is something we need to teach students to do and continue to do throughout their professional careers. This paper provides one template for doing that careful assessment. Conflict of interest: None declared.