The Cancer Journal - Volume 6, Number 1 (January-February 1993)
Does smoking cause suicide?
The graph below depicts a linear association between the number of cigarettes that someone smokes and his risk of committing suicide (1). Based on the chi-square test the trend is highly significant: p<0.0001. 361,662 men aged 35-57 were followed up for 12 years, and the graph is based on the analysis of death rates per 10,000 person-years of risk. Now pause for a while, examine the figure and try to make up your mind. Does smoking cause suicide? What is the exact meaning of the graph and the significance estimate of p<0.0001? Is it correct or false? Or perhaps it hides a fallacy?
In the introduction to their paper the authors cite other similar associations which led to the following conclusions: physical activity is a risk factor for diabetes. Taking aspirin protects against colon cancer. Smoking could increase the risk of cervical cancer, or predispose to coronary heart disease. Are these conclusions justified? or do they "need substantial improvement to produce trustworthy scientific evidence"? (2). Epidemiology applies two criteria in order to conclude that an association is a causal relationship: 1) the risk has to increase with the intensity of the agent; 2) the association has to be "independent" of other risk factors. Both requirements were met by the present association and yet the authors did not like it.
They tried to sidestep the difficulty by showing a similar association between smoking and homicide, concluding that in spite of the epidemiological rigor, the conclusion that the two associations are causal is not justified. How then to distinguish between genuine and spurious causal associations? Since epidemiology fails to distinguish between the two, the authors postulated a new criterion, "plausibility", concluding that these associations are implausible and add: "The lack of plausible mechanisms to explain smoking/suicide and smoking/homicide relationships is, however, the exception rather than the rule in epidemiology". But: "If epidemiology can throw up a spurious independent dose-response relation in smoking/suicide, how many more such relations, for other exposures and other diseases, are equally spurious?".
You are witnessing a piece of information with dubious value. An association without meaning; and yet many such associations were described in the literature. The significance of this association is impressive, p<0.0001, yet meaningless. The graph represents disinformation since it is neither true nor false. Statistically the association is correct and the p-value justified, but in reality it is meaningless.
In their important book "Follies and Fallacies in Medicine" (3), Skrabanek and McCormick state that the relationship between two events A and B may be of four kinds:
1. A causes B (cause)
2. B causes A (consequence)
3. A and B share a common cause C (collateral)
4. A and B are associated by chance (coincidence).
The same applies also for the association between smoking and cancer (4,5): 1) smoking may initiate cancer, 2) cancer may trigger an urge to smoke, 3) the two may never interact and the observed association results from a third process that was not considered in the study, 4) the association is coincidental.
The issue cannot be resolved by epidemiological arguments. Only medicine has the conceptual tools to resolve it. Instead of admitting the inadequacy of their method, epidemiologists adjust, stratify (6-8) and now propose the doubtful criterion of plausibility. This correlation fallacy has far-reaching implications since it underlies clinical trials. And since clinical trials attempt to direct therapy, this fallacy breeds iatrogenesis. Modern epidemiology is utterly confused and should therefore reconsider its basic premises and objectives
1. Smith GD, Phillips AN, Neaton JD. Smoking as "independent" risk factor for suicide: illustration of an artifact from observational epidemiology? Lancet 340, 709-712, 1992.
2. Feinstein AR. Scientific standards in epidemiologic studies of the menace of daily life. Science 242, 1257-1263, 1988.
3. Skrabanek P, McCormick J. Follies and Fallacies in Medicine. Prometheus Books Buffalo NY, p. 28, 1990.
4. Zajicek G. Epidemiology spreads disinformation. Cancer J 5, 240, 1992.
5. Zajicek G. To smoke or not to smoke? Cancer J 5, 70, 1992.
6. Zajicek G. Progress against cancer: are we winning the war? Cancer J 3, 2, 1990.
7. Zajicek G. Cancer wars. Cancer J 4, 4-5, 1991.
8. Zajicek G. Meta-analysis and chaos. Cancer J 4, 152-153, 1991.