Anti-tobacco means never having to say
weak associations are sorry

October 15, 1999


The American Council on Science and Health has teed off on the Cato Institute report "Lies, Damned Lies, & 400,000 Smoking-Related Deaths."

The Cato report, authored by Robert Levy and Rosalind Marimont, critiqued the oft-repeated claim that smoking kills 400,000 every year. Levy and Marimont's report rankled the anti-smoking establishment so much that the CDC actually attempted to defend itself. Usually, the anti-smoking industry ignores criticism, lest it give its opponents a platform.

But, alas, the ACSH report, "A Critical Assessment of 'Lies, Damned Lies, & 400,000 Smoking-Related Deaths' by Robert Levy and Rosalind Marimont Published in Regulation,* Fall 1998," is itself in need of some critical assessment.

The ACSH report makes four key points (Note: The ACSH text highlighted below does not necessarily present the full ACSH point of view. Please refer to the full text of the ACSH report.)

"1. Levy and Marimont discount over one-third of the estimated 400,000 annual deaths caused by smoking with the erroneous claim that 'small' increases in the risk of disease or death marked by relative risks less than 2.0 are 'statistically insignificant,' and "insufficiently reliable to conclude that a particular agent (e.g., tobacco) caused a particular disease."

Relative risks less than 2.0 - and even less than 3.0 - are often referred to as "weak statistical associations." A weak statistical association is inherently unreliable because it shows too slight a difference between the study groups being compared. This is problematic because the data used in epidemiologic studies is not so precise that it is capable of making such fine distinctions. In contrast, the relative risk for heavy smoking and lung cancer is on the order of 10 to 20 and no one disputes that association.

The classic quote on this issue comes from renown epidemiologist Sir Austin Bradford Hill. In his famous speech defining the criteria by which epidemiologic studies should be evaluated, Hill noted the weak association between smoking and heart disease - i.e., a relative risk on the order 2.0 - was not very credible because a variety of uncontrollable confounding factors could easily explain the slight difference in heart disease rates between smokers and nonsmokers.

The ACSH comment, "Given the pervasiveness of a risk factor, such as smoking, and the prevalence of some of the diseases it causes, small relative risks can, and do, represent serious threats to public health," is simply wrong. A relative risk only can measure whether a statistical association exists. It does not measure "risk," on an individual or population basis. The statistical measure "relative risk" is derived by comparing two groups of study subjects. The interpretation of a relative risk statistic is limited by what is being compared - groups that tend to be small and not representative of the general population.

Even accepting ACSH's erroneous interpretation of relative risk, I challenge ACSH to produce the "small relative risks" that "represent serious threats to public health."

"2. Levy and Marimont argue that the American Cancer Society's Cancer Prevention Survey (CPS) a widely used data set for the calculation of public health statistics is unrepresentative of the general population and is therefore 'the wrong sample [to use] as a standard of comparison' when estimating smoking-related deaths in the US."

Who cares whether the CPS is representative? It's not very good data, representative or not. The data was collected by thousands of unprofessional volunteers who queried friends, relatives and neighbors about their health and lifestyle habits. (So tell me Aunt Mary, how much do you smoke and drink every day?)

In addition to uncontrolled data collection, the data was never verified - not even a sample - for accuracy. A weak association derived from the CPS data set is, ipso facto junk.

"3. The authors state that the Centers for Disease Control and Prevention (CDC) fails "to control for obvious confounding variables" in its calculation of smoking-related deaths. They argue that after accounting for other factors that may contribute to deaths among smokers, the CDC's estimate should be greatly reduced."

It is well-known that smokers tend to have unhealthier lifestyles than nonsmokers. Smokers tend to drink more, eat worse diets, exercise less, and have lower socio-economic status - all of these factors are significantly associated with premature mortality. It is difficult, if not impossible, to separate the effects of smoking from all relevant competing risk factors and their interactions -- no matter how much statistical massaging is conducted.

"4. Finally, Levy and Marimont purport that the impact of smoking-related mortality is overstated, particularly with respect to children, given that the majority of smoking-related deaths occur late in life."

ACSH says "In fact, it has been estimated that over one-half of all smoking-related deaths occur between ages 35 and 69, which translates into an average loss of roughly 23 years of life."

This factoid is based on the definition of "smoking-related death." Once you get past deaths from lung cancer, emphysema and perhaps a few other diseases, "smoking-related death" becomes quite uncertain. Many deaths attributed to smoking depend on weak statistical associations.

Finally, consider this abstract from a recent study in the American Journal of Epidemiology about smoking and lifespan:

The effect of smoking and physical activity on active and disabled life expectancy was estimated using data from the Established Populations for Epidemiologic Studies of the Elderly (EPESE). Population-based samples of persons aged > or = 65 years from the East Boston, Massachusetts, New Haven, Connecticut, and Iowa sites of the EPESE were assessed at baseline between 1981 and 1983 and followed for mortality and disability over six annual follow-ups. A total of 8,604 persons without disability at baseline were classified as "ever" or "never" smokers and doing "low," "moderate," or "high" level physical activity. Active and disabled life expectancies were estimated using a Markov chain model. Compared with smokers, men and women nonsmokers survived 1.6-3.9 and 1.6-3.6 years longer, respectively, depending on level of physical activity. When smokers were disabled and close to death, most nonsmokers were still nondisabled. Physical activity, from low to moderate to high, was significantly associated with more years of life expectancy in both smokers (9.5, 10.5, 12.9 years in men and 11.1, 12.6, 15.3 years in women at age 65) and nonsmokers (11.0, 14.4, 16.2 years in men and 12.7, 16.2, 18.4 years in women at age 65). Higher physical activity was associated with fewer years of disability prior to death. These findings provide strong and explicit evidence that refraining from smoking and doing regular physical activity predict a long and healthy life. [Source: Am J Epidemiol 1999 Apr 1;149(7):645-53.]

Somewhat different than what ACSH says, huh?

Conclusion

ACSH, like any group that fights against junk science, is often and wrongly criticized by the junk science crowd as a being a "tool" of industry. But instead of simply advocating sound science and criticizing junk science, ACSH tries to compensate -- it tries to buy credibility with the junk science crowd -- by being as anti-tobacco as possible. The ACSH mindset is that small amounts of chemicals are safe, unless they're in tobacco smoke.

ACSH's shrill anti-tobacco-ism hasn't bought credibility with the junk science crowd. I doubt it ever will. If ACSH wants to advocate against smoking, that's fine. But resorting to junk science, for whatever reason, will only hurt the organization's credibility with its allies in the anti-junk science movement.


Comments on this posting?

Click here to post a public comment on the Trash Talk Bulletin Board.

Click here to send a private comment to the Junkman.
1