Estimating the Numbers of Smoking-Related Deaths   

Letter from CATO to JAMA Vol. 284 No. 18, 8nov00

To the Editor:  
            In debunking the notion that smoking causes more than 400,000 deaths each year, Ms Marimont and I¹ were critical of studies that failed to control for confounding variables like alcohol consumption, exercise, diet, occupation, and income. In response, Dr Thun and colleagues² purport to have controlled for the first 4 of those variables plus race and education. Thun et al conclude that "federal estimates of deaths caused by smoking are not substantially altered by adjustment for behavioral and demographic factors."

That conclusion is entirely unjustified. Indeed, Thun et al acknowledge the major defect of their research: "Smokers and lifelong nonsmokers are more similar with respect to socioeconomic and educational status in our study than in the contemporary United States." Then they inexplicably gloss over that defect as if it were inconsequential, when, in fact, it hopelessly compromises the validity of their results.

The study by Thun et al is based on the Cancer Prevention Study II (CPS-II) database, which consists of approximately 1 million of the American Cancer Society's members and friends. The individuals who participated were more affluent than average, overwhelmingly white, and college graduates who generally did not have hazardous jobs. Each of those characteristics tends to reduce the death rate of the CPS-II sample so that this population has an average life expectancy that is substantially longer than that of the US population. Astonishingly, even smokers in the CPS-II sample have a lower death rate than the national average for smokers and nonsmokers combined.

The bias is significant: in a 1989 report, the US surgeon general estimated that smoking caused 335,600 deaths. When Sterling et al³ recalculated those numbers, replacing the distorted CPS-II sample with a more representative baseline from surveys conducted by the National Center for Health Statistics, they found that the number of smoking-related deaths declined to 203,200. Accordingly, the surgeon general's report overstated the death count by 65% by choosing the wrong standard of comparison.

Moreover, the study by Thun et al does not control directly for the single confounding variable that nearly every researcher regards as critical: income.⁴ Because the CPS-II database excludes low-income persons, Thun et al could not have controlled for income even if they had attempted to do so. Thus, the authors disregard 2 elementary principles of statistics: first, conclusions for a whole population cannot be drawn from a nonrepresentative sample. Second, the effect of a variable (like education or race) cannot be evaluated by using a database in which the variable is essentially constant.

Robert A. Levy, PhD, JD
Cato Institute
Washington, DC
 1. Levy RA, Marimont RB. Lies, damned lies, & 400,000 smoking-related deaths. Regulation. 1998;21:24-29.

2. Thun MJ, Apicella LF, Henley SJ. Smoking vs other risk factors as the cause of smoking-attributable deaths. JAMA. 2000;284:706-712. ABSTRACT  |  FULL TEXT  |  PDF  |  MEDLINE

3. Sterling TD, Rosenbaum WL, Weinkam JJ. Risk attribution and tobacco-related deaths. Am J Epidemiol. 1993;138:128-139. MEDLINE

4. Goode E. For good health, it helps to be rich and important. New York Times. June 1, 1999:F1.

In Reply: The main points of our article are that cigarette smoking kills an extraordinary and unacceptable number of people in the United States each year and that these deaths cannot be attributed to other characteristics of smokers rather than smoking itself. Dr Levy prefers the estimate of 203,200 proposed by studies funded by the tobacco industry¹ rather than the approximately 418,000 estimated by the Centers for Disease Control and Prevention.² Either number of deaths is unacceptable. The issues raised by Levy must be debated, however, because his views³ have been publicized widely in the lay press and submitted as court evidence by the tobacco industry to illustrate how it has been wronged by the health care community.⁴

Levy argues that relative risk (RR) estimates derived from the American Cancer Society's prospective study should not be generalized to the average US population because the CPS-II cohort is healthier and more thoroughly middle class than the US general population. Certainly poverty does not protect smokers from the harmful effects of tobacco. Levy incorrectly claims that the massive CPS-II has too little variation in educational and socioeconomic status to examine this issue. Table 1 indicates that the CPS-II includes more than 50,000 men and women within each level of educational attainment. For lung cancer, the estimates of RR and the percentage of deaths among current smokers that are attributable to smoking (attributable percentage [AP]) are essentially constant across all levels of education. The corresponding estimates for coronary heart disease and stroke are slightly lower among persons with less than a high school education than those with higher educational levels. However, this subgroup comprises only 13.8% of the CPS-II cohort and 16.6% of the US population aged 25 years and older; small variations in this subgroup have little effect on the overall US estimates. Other cohort studies have observed similar RR and AP estimates associated with cigarette smoking in the general US population, although these studies are relatively small and have wider confidence intervals.

Levy also errs on 3 other points. First, the lack of direct information on income in our study does not invalidate the use of education and occupation as proxy measures for socioeconomic status. Numerous studies indicate that data on education and occupation capture the socioeconomic factors that affect health.⁵ Second, Levy characterizes the lower estimates of smoking-attributable deaths as being less "biased" than the federal estimates, whereas, in reality, they are merely incorrect. The estimate by Sterling et al¹ is based on faulty methods⁶ and excludes a large number of cardiovascular deaths caused by smoking. Third, Levy and Marimont propose that all diseases for which the RR associated with current smoking is below 2.0 should be excluded from attributable-risk calculations.³ This exclusion is unjustified, since it deftly reclassifies all deaths from ischemic heart disease, stroke, and cancers of the pancreas and bladder as having nothing to do with smoking, despite extensive evidence to the contrary.

Michael J. Thun, MD, MS
Louis F. Apicella, MSPH
S. Jane Henley, MSPH
American Cancer Society
Atlanta, Ga 

1. Sterling TD, Rosenbaum WL, Weinkam JJ. Risk attribution and tobacco-related deaths. Am J Epidemiol. 1993;138:128-139. MEDLINE

2. Cigarette smoking-attributable mortality and years of potential life lostUnited States, 1990. MMWR Morb Mortal Wkly Rep. 1993;42:645-649. MEDLINE

3. Levy RA, Marimont RB. Lies, damned lies, & 400,000 smoking-related deaths. Regulation. 1998;21:24-29.

4. Levy RA, Marimont RB. Our "dammed lies" spark another exchange. Regulation. 2000;23:2-11.

5. Berkman L, ed, Kawachi I, ed. Social Epidemiology. New York, NY: Oxford University Press; 2000:13-35.

6. Malarcher A, Schulman J, Epstein L, et al. Methodological issues in estimating smoking-attributable mortality in the United States. Am J Epidemiol. 2000;152:573-584. MEDLINE
 
Letters Information   
Guidelines for Letters
 
Letters Section Editors: Stephen J. Lurie, MD, PhD, Senior Editor; Phil B. Fontanarosa, MD, Executive Deputy Editor.

If you have come to this page from an outside location click here to get back to mindfully.org