A non-technical Discussion of
"Risk Attribution and Tobacco Related Deaths"

There are now three methods to estimate the number of deaths due to tobacco use, one used by the OTA, one used by the Surgeon General, and a new method used by WHO. All three methods derive their basic data from the two large cohort studies by the American Cancer Society of smoking in relation to the death rates of a million men and women, better known now as Cancer Prevention Survey I (CPS-I) and Cancer Prevention Survey II (CPS-II).

All three methods assume that the death rates observed in CPS-I and CPS-II are representative of the death rates of the entire U.S. population. Unfortunately that assumption simply is not true. The American Cancer Society study populations for both CPS I and CPS II consist of mostly affluent individuals, mostly white, Protestant, urban, married college graduates, etc. who moreover are generally not exposed to hazardous occupations, As a consequence, the members of The American Cancer Society and their friends and acquaintances who volunteered for the American Cancer Society studies (i.e. CPS-I and -II) enjoy a much lower death rate than does the entire U.S. population.

Because all three methods base their calculations in one way or another on the low mortality rates of the American Cancer Society's studies, an apparent large number of excess deaths appears to exist. But this apparent number of lives lost is an illusion.

We do the same calculations as were done by the OTA, by the Surgeon General, by the WHO, and by others, but use data representative of the U.S. population instead of the American Cancer Society. Our data comes from large surveys of the entire U.S. population and of its deaths, conducted by The National Center for Health Statistics (NCHS), specifically the National Mortality Followback and the National Health Interview surveys.

With respect to the Surgeon General's estimate of the number of excess deaths that occurred in 1985 due to tobacco use, our major findings are summarized in Table 5, (page 9). This table gives the number of excess deaths (1) estimated by the 1989 Surgeon General's report based on data from CPS-II, data that are not representative of the U.S. population, and (2) estimated by our calculations based on a representative sample of the U.S. population. The lower part of the table gives the number of excess deaths (in thousands). The Surgeon General's report estimates 229,900 male excess deaths and 105,700 female excess deaths, for a total of 335,600 excess deaths for the year 1985. Using data from a representative sample of the United States instead of CPS, the same calculations yield 135,100 excess male and 68,100 excess female deaths, for a total of 203,200 deaths. The calculation above assumes that the only differences between smokers and nonsmokers is that smokers smoke whereas nonsmokers do not. This is not true; smokers as a whole are quite different from nonsmokers in a number of characteristics which affect their mortality. For example, smokers tend to drink more alcohol, have less income, work in more hazardous occupations, and eat fewer green vegetables than do nonsmokers. Therefore, when we compare smokers to nonsmokers we are also to some extent comparing, for example, people with lower incomes to people with higher incomes. By ignoring differences in income between smokers and non- smokers, some deaths that are due to income-related factors are counted as due to smoking.

When we adjust the calculations above to properly account for differences in income and alcohol consumption between smokers and nonsmokers, the total number of excess deaths is estimated to be 150,000, or 44% of the Surgeon General's estimate. This number would be reduced further still if we had additional data about occupational and nutritional habits of persons who took part in the NCHS surveys.

It would be fallacious to conclude from our findings that a true number of deaths due to tobacco use would emerge after adjustments were made for all relevant causes. First, to adjust for some highly correlated factors would be practically impossible. (We discuss the difficulties met with in any attempt, for instance, to adjust for occupation or nutrition.) Second, it should not be assumed that the numbers remaining after ad,adjusting for major confounding factors necessarily would be of excess death. They could just as easily be of excess lives saved.

The major flaw in the OTA's method is dramatically demonstrated in our Table 7 (page 11). Very much simplified, they first calculate the estimated number of deaths in the U.S. population if everyone should die with the death rate of never-smokers in the American Cancer Society Study (i.e. in CPS). Next, that number is subtracted from the actual number of observed deaths. The difference is then ascribed to smoking. However, as pointed out above, the CPS samples have much lower death rates than the entire U.S. population. In fact, even smokers in the CPS samples have lower death rates than the entire U.S. population.

As a consequence, if one uses the OTA's method but compares the number of deaths that would have occurred had everyone smoked (and died with the mortality of smokers in the American Cancer Society's sample population) to the number of deaths in the U.S. population, one ends up with saving 277,621 lives. Unfortunately this startling discovery is not a medical miracle. It is sheer nonsense to claim that 277,000 deaths would be averted if everyone would smoke. 277,000 lives would be saved only if the U.S. population would die with the death rate of the smokers in the affluent CPS sample.

Similarly, it is sheer nonsense to claim that 340,000 deaths would be averted if the U.S. population would not smoke. That many deaths would be averted only if the U.S. population would die with the mortality rate of the affluent never-smokers in the CPS study samples.

Peto, Lopez and Borehan of WHO assume that the death rates obtained from CPS-II hold in fact for all developed countries. In Table 6 we compare relative risks for a number of diseases obtained from CPS-II by Peto and from the U.S. population from the National Center of Health Statistics surveys. With one single exception, risks computed for a representative sample of the U.S. population are consistently smaller than risks computed from CPS-II. Thus, Peto's assumption is simply not justified. Death rates based on the CPS-II population do not even estimate the much more appropriate risks obtained from the U.S. study sample, from which after all, subjects in CPS-I and CPS-II were selected.

Besides these major flaws there are number of other problems with these calculations.

For instance, the WHO's calculation model assumes that every population is a mixture of current smokers and never-smokers, but does not contain any former smokers. In fact, this model cannot accommodate former smokers at all.

The exercise of calculating attributable risks and excess deaths derived from them attempts to go beyond what can validly be inferred from relative risks. We may all agree that smoking should be relegated to an activity practiced by consenting adults. However, the fact that smoking appears to have few redeeming qualities does not justify the use of methods of risk calculations which so clearly are questionable.