robert wagner

A recent (5/27/04) report from the US Surgeon General says:

" Assuming that one-third of adult
current smokers and 10 percent of adult former smokers
die from smoking-related diseases, and that current
smoking patterns continue without a marked
increase in cessation, an estimated 25 million persons
(adults and children) alive in 1995 will die prematurely
from smoking-related illnesses (CDC 1996a);"
(page 885, chapter 7)

Dividing 25M by the US population of 280M gives 9%. Multiplying that
by 2M annual deaths gives 179,000 from smoking-related illnesses.
Notice it doesn't say caused by smoking. Many diseases are
statistically 'related' to smoking but not caused by it. For instance,
leukemia.

The same chapter of the same report says:

"Smoking caused an estimated total of 263,600
deaths in males and 176,500 deaths in females (total
440,200) in the United States each year from 1995–1999
(Table 7.3)."
(page 858, chapter 7)

How did the body count jump from 179K 'related to smoking' to 440K
'caused by smoking'?

The source of 440K is CDC/SAM/CPS-II, shown in table 7.3, plus three
additional categories:

Male Female
Cancer 105.7 53.9
Heart 87.6 55.0
Lungs 53.7 44.3
Infant .6 .4
Total CDC 247.6 153.6 401.2
Fire 1.0
ETS cancer 3.0
ETS heart 35.0
Total 440.2

This is how SAM-II works, using acute myeloid leukemia (AML) as an
example. Start with total deaths from the disease:
Male Female
AML deaths 3,200 2,700

Divide them into smoking status, using table 7.2. Since years and
death ages are not given in the report, I'll estimate using 1992 age
45-64 (30.6 MCS, 40.5 MFS, 26.8 FCS, 23.8 FFS)

M-current M-former F-current
F-former
AML deaths 979 1,296 724
643

Go to Table 7-1.1 to get the relative risk of smokers (relative to
never-smokers).
M-current M-former F-current
F-former
Acute myeloid leukemia 1.9 1.3 1.1
1.4

Calculate the number of deaths in each column caused by smoking as
D - (D/rr) which can be simplified ((rr - 1) * D) / rr (easier to do
on a calculator)

M-current M-former F-current
F-former
Smoking caused 464 299 66
184
Total by gender 763 250
Round to nearest 100 800 300
Shown on table 7.3 800 300

What's wrong with that? They didn't count all deaths from AML as
caused by smoking; they used relative risk to calculate how many of
them were caused by smoking.

The report TELLS you what's wrong with that .. on page 880 where it
says, "Numerous cohort studies provide RR estimates for
smoking-related diseases and mortality." Relative risks from cohort
studies are useful for estimating effect size when cause and effect
(etiology) can be determined by other means; cohort studies are no
good for 'proving' cause and effect. There is too much danger of bias
and unrecognized confounders.

What other means? Best are experimental studies that demonstrate the
mechanism of etiology. Microbiologists using electron microscopes show
before and after pictures of the molecules affected. Second best are
case-control studies, an epidemiologic (statistical) technique much
better than cohort studies.

Relative risks less than 3.0 are usually considered statistically
insignificant, or at least require more intense scrutany than a cohort
study. Yet MOST diseases on the Surgeon General's report are less than
3, including all but one in the heart category.

The 1992 EPA report on ETS deleted the whole chapter on cardiovascular
effects, found in preliminary versions, for that very reason -- there
was some correlation but it was statistically insignificant. The
Surgeon General reinstated it to inflate the danger from ETS by
1,200%.


CDC has turned the process upside-down by using an estimation
technique to 'prove' causation.

Why is a male smoker NINE TIMES as likely to get AML 'from smoking' as
a female? If AML were caused by a chemical in smoke, the relative risk
for males and females would be about the same. And why do former
female smokers have four times the increased risk of current smokers?
Such disparities strongly suggest a confounding factor is involved.

Here's my hypothesis. Known risk factors for AML are radiation,
chemotherapy and exposure to benzene, which is abundant in gasoline
fumes. Suppose tobacco smoke doesn't cause AML directly, but weakens
the body's defenses against it. Persons exposed to gasoline AND
tobacco smoke will get AML more than average. If either is absent,
there will be no increased incidence.

When a smoking couple pulls up to the gas pump, it's usually the man
who pumps gas. Women who pump gas usually do so because they don't
have a partner. My conjecture will be supported if it can be shown
that smoking women are four times a likely to have a partner who pumps
her gas vs. former smoking women. It would also be supported if it
were found there's no increased AML in men living in states that don't
allow self-service gasoline, such as New Jersey.


http://www.surgeongeneral.gov/library/smokingconsequences/