A recent Surgeon's General report on smoking adds new cause-and-effect links between smoking and several diseases including cataracts, pneumonia, acute myeloid leukemia, abdominal aortic aneurysm, stomach cancer, pancreatic cancer, cervical cancer, kidney cancer, and periodontitis. Drawing a cause-and-effect link requires great care. A supplementary report, How Do We Conclude That Smoking is a Cause of Disease, states that
For the Surgeon General to conclude that smoking is proven to cause a particular disease, there must be enough scientific evidence that smoking either increases the overall number of cases of the disease or makes the disease occur earlier than it otherwise would. The reports use a number of criteria to guide their judgment:
- Do multiple high-quality studies show a consistent association between smoking and disease?
- Are the measured effects large enough and statistically strong?
- Does the evidence show that smoking occurs before the disease occurs (a temporal association)?
- Is the relationship between smoking and disease coherent or plausible in terms of known scientific principles, biologic mechanisms, and observed patterns of disease?
- Is there a dose-response relationship between smoking and disease?
- Is the risk of disease reduced after quitting smoking?
These are six of the nine criteria offered by Sir Austin Bradford Hill in 1965. The criteria mentioned above are
- Consistency (have the results have been replicated by different researchers and under different conditions)
- Strength (is the risk so large that we can easily rule out other factors)
- Temporality (did the exposure precede the disease)
- Plausibility (is there a credible scientific mechanism that can explain the association)
- Biological gradient (are increasing exposures associated with increasing risks of disease)
- Experimental evidence (does a physical intervention show results consistent with the association)
and the remaining criteria are
- Coherence (is the association consistent with the natural history of the disease)
- Specificity (is the exposure associated with a very specific disease as opposed to a wide range of diseases)
- Analogy (is there a similar result that we can draw a relationship to)
These criteria are very commonly used, although Rothman and Greenland criticize them in their classic textbook, Modern Epidemiology, Second Edition (see pages 24-28). B. Burt Gerstman devotes a full chapter to Hill's criteria in his book, Epidemiology Kept Simple (chapter 12). Edward Tufte liked Hill's criteria so much that he reproduced the original 1965 article by Hill on his web site.
I'd like to talk a bit more about specificity, because that is an important, but neglected criteria.
I am generally skeptical of very general medical claims. For example, Linus Pauling has claimed in an interview with the Academy of Achievement that large doses of vitamin (particularly Vitamin C) can cure conditions from a simple cold through cardiovascular disease, and advanced cancer. It's possible that Vitamin C can cure all of these conditions, but it's hard to identify a mechanism that would explain all this. It is easy, however, to identify a common source of bias in many of the observational studies: people who take the time and trouble to take vitamin supplements also tend to have other good self-care habits. The actual research record on Vitamin C is quite complex, and perhaps it is a bit simplistic on my part to cite this as an example. In general, though, you should be cautious about research claims that drawing sweeping rather than specific conclusions.
There are notable exceptions to this rule about specificity. First, aspirin is a drug that is effective for inflammation, headaches, fever. It is useful as a primary prevention against heart attack and it can reduce the risk of death if taken at the first signs of a heart attack. Aspirin may even reduce the risk of certain types of cancer, including a recent report in JAMA suggesting that regular consumption of aspirin may reduce the risk of breast cancer [Medline].
Also, ironically, the reports about cigarette smoking violate the rule about specificity. It turns out that smoking increases the risks of a broad range of cancers, as well as arthritis, stroke, osteoporosis, peptic ulcers, reproductive disorders, cataracts, thyroid conditions, and baldness. The Surgeon General's report cited above is the best resource about the wide range of diseases linked with smoking, but other sources about this are:
Like all of Hill's criteria, specificity is only a general indication. If an association is specific, that's one piece of evidence in support of that finding, but specificity alone does not guarantee a causal relationship. Neither does lack of specificity refute a causal relationship. You have to look at specificity along with all the other factors before concluding anything.
Additional links on Hill and causation
You can find an earlier version of this page on my original website.