From: robert.wagner@channel1.com (Robert Wagner)
Date: 18 Apr 94 22:35:00 GMT
Newsgroups: alt.smokers
Subject: Second-hand science
SECOND-HAND SCIENCE:
A STATISTICAL ANALYSIS OF ONE STUDY ON LUNG CANCER AND PASSIVE SMOKING.
Robert Wagner (robert.wagner@channel1.com) 04/01/94
Summary:
A major study, published in JAMA, reached the conclusion "These results
provide support to the body of evidence linking passive smoking to lung
cancer ..". This analysis shows that the study did NOT properly justify
that conclusion. It does so thru a novel method of applying a test for
statistical significance, despite the fact the raw data were not
revealed in the report.
Introduction:
"Active and Passive Smoking and Pathological Indicators of Lung Cancer
Risk in an Autopsy Study", Trichopoulos et al., JAMA (Journal of the
American Medical Association):268:13:1697 (1992) contains two studies:
one on active smoking, which occupies the majority of the report, and
one on passive smoking. Since the Conclusion, as summarized in the
abstract, deals only with passive smoking, one questions the relevance
of the active study. This analysis deals only with the passive study.
The passive study was done on 30 women who had died from causes other
than lung disorders. Roughly 75% died from cardiovascular diseases and
25% were accident victims. Exact numbers were not given in the report,
nor are they particularly significant to this analysis. All the women
were non-smokers and over 35 years old. 17 were married to husbands who
were ever-smokers, 13 were married to husbands who had never smoked
(based on interviews with surrogates, in most cases the husband
himself). Autopsies measured Epithelial, Possibly Precancerous Lesions
(EPPL) in the subjects' lungs and expressed it on a numeric scale ranging
from 0 to 80+, the mean around 41, SE 5.6 (exact numbers not given).
The study divided the subjects into three groups: <40, 40 thru 60, >60.
It ignored the middle group, whose exact size is unknown. Based on other
data in the report (Table 2) it would appear the middle group comprised
at least 25%, thus the number of subjects (N) is approximately 20-22.
The report gives only three numeric facts about the subjects. One is not
useful for determining significance: the difference in EPPL between
passive and non-smokers is 27.88 (5.64 to 50.13) with a two-tailed P of
.02 (at 95% CI). The other two are:
..Number of subjects is less than 30.
..Odds ratio between passive and non-smokers is 6.
Analysis:
What we have is a classically simple four box matrix familar to
statistics students and certainly to epidemiologists:
Exposed Not exposed
Have the disease a b
Do not have the disease c d
Note that numbers are assigned to *classifications*. This is
significant. More familiar statistical analyses are presented with
continuous variables and asked to determine correlation between them.
Lacking evidence to the contrary they assume a normal distribution of
samples as well as the measurement error component: standard error (SE).
Based on that they do regression, T tests and other measures of
variance. None of those can be used on classifications.
Odds ratio is not a proper test for relationship, it is a measure of
effect size. First we must first establish that one variable is
dependent on the other: that exposure 'causes' the disease. Then we can
use odds ratio or relative risk to estimate how many individuals in the
population will be affected. Why? Because in any retrospective study
there is danger of a confounding factor which 'causes' both exposure and
disease. A likely confound in this case is socioeconomic status, which I
will analyze in a companion study. Although odds ratio gives a single
score, multiple tests with different randomly selected subjects would
give a range of scores, presumably with a normal distribution. In order
to establish dependence one must show that the spectrum does not cross
unity within the specified confidence interval (CI).
The test we will use is known as the chi square test for independence.
This is how it is computed, using one hypothetical set of data as an
example.
Exposed Not exposed
Have the disease 6 1
Do not have the disease 6 6
The odds ratio is (a/b) / (c/d), simplified to ad/cb. In this case: 36/6
= 6. The more familiar relative risk (RR) would be (6/12) / (1/7) = 3.5.
To do a chi square test, first we add row and column totals, then use
them to compute the expected number in each cell -- (r * c) / N. For
example cell a is (12 * 7) / 19 = 4.4.
Exposed Not exposed Total
O E O E
Have the disease 6 4.4 1 2.6 7
Do not have the disease 6 7.6 6 4.4 12
Total 12 7 19
Next we compute the difference between observed and expected, square
it, and divide by expected. The sum of the squares of the differences,
in relation to expected, is called x^2 (chi squared).
O E O - E (O - E)^2 / E
a Have, exposed 6 4.4 1.6 .58
b Have, not exposed 1 2.6 1.6- .98
c Do not have, exposed 6 7.6 1.6- .34
d Do not have, not exposed 6 4.4 1.6 .58
Total x^2 = 2.48
Compuationally this can be simplified to:
x2 = (((abs((a*d) - (b*c)))^2)*N) / ((a+b) * (c+d) * (a+c) * (b+d))
Finally we compare x^2 to a threshold obtained from a published table
called "percentage points, chi square distribution". The formula for
computing these values is rather complex and beyond the scope of this
analysis. In this case the value to use is 3.85 (CI=95%, df=1). If x^2
is below that value, we have not demonstrated a statistically
significant relationship between the two variables. In statistical
terminology "we accept the null hypothesis". If x^2 is above that value,
"we reject the null hypothesis" and conclude there is a relationship. In
the example one can see there is no causal connection.
In order to use this analysis on unknown data, I wrote a computer program
which tried all possible integer values in the four boxes, letting N (number
of subjects) range from 4 to 23, identifying those combinations yielding
an odds ratio of 6 (5.9 to 6.1). It found 106 sets meeting this criterion.
*Whatever data the report was based on had to be one of these 106.*
Here is a sample:
a b c d N OR x^2
4 2 4 12 22 6.0 3.275
4 3 2 9 18 6.0 2.921
4 4 1 6 15 6.0 2.142
4 4 2 12 22 6.0 3.275
4 6 1 9 20 6.0 2.400
5 1 5 6 17 6.0 2.298
5 3 3 11 22 6.1 3.709
5 5 1 6 17 6.0 2.298
6 1 1 1 9 6.0 1.148
6 1 2 2 11 6.0 1.638
Result: ALL 106 FAILED THE CHI SQUARE TEST (the null hypothesis was
accepted).
The chi square test is an approximation which is usually satisfactory
when the numbers are large. When the counts are small, as they are
in this case, it is better to use Fisher's Exact Test, described
by Sir Ronald Fisher in 1970. The computationally demanding test
is given by the formula:
P = (a+b)!(c+d)!(a+c)!(b+d)! / (N!a!b!c!d!)
One finds the smallest of the four variables and repeats the calculation
for each value down to zero. The sum of the results is the probability
of the observed deviation from expected. In order to show a
statistically significant dependence between the two variables, i.e. to
reject the null hypothesis, the sum must be less than .05 (CI=95%).
Applying this test to the 106 possible sets where N < 24, all failed
to show even 90% significance. Further, testing ALL possibilities of
N < 30, none of them showed 95% significance.
Conclusion: the data used in the Trichopoulos et al. study shows
no statistically significant relationship between passive smoking and
lung cancer. The study's conclusion is false.
Disclaimer:
I am not associated with the tobacco industry, academia, the medical
profession nor any other group with a bias on tobacco.
---
~ DeLuxe} 1.1 #448sa ~
From: robert.wagner@channel1.com (Robert Wagner)
Date: 18 Apr 94 22:36:00 GMT
Newsgroups: alt.smokers
Subject: Second-hand science
SECOND-HAND SCIENCE:
LUNG CANCER AND PASSIVE SMOKING: AN ARTEFACT OF SOCIOECONOMIC STATUS
Robert Wagner (robert.wagner@channel1.com) 04/01/94
Summary:
A major study, published in JAMA, reached the conclusion "These results
provide support to the body of evidence linking passive smoking to lung
cancer ..". My companion study showed that a direct, causal, relationship
was not statistically significant. This review hypothesizes that both
are related to a confounding factor: socioeconomic status.
Findings:
I first saw the relationship between socioeconomic (SE) status,
environmental tobacco smoke (ETS) and health while reading
Schwartz-Bickenbach et al. The report's authors did not note the
connection.
The study measured cotinine level in infants. Cotinine is a derivative
of nicotine, which is found only in tobacco, thus a reliable measure of
exposure to and absorption of smoke. Findings for newborns was:
- Cotinine in infant -
Mother's status cig/day range median
Smokers < 20 41-430 91
>20 28-588 305
Non-smokers none 0-7 0
(father smokes) passive 0-55 0
body weight (g)
Mother smokes 3165
Father smokes 3365
No smoke 3570
Body length and head circumference were essentially the same for all
categories.
Note that body weight correlates with parents' smoking status. Was lower
weight 'caused' by smoke? No. Where the father smoked and mother did
not, cotinine in the infant is insignificant. We know that cotinine can
cross the placental barrier because smoking mothers' newborns had it, and
in proportion to dosage. In another chart (not shown) they measured
cotinine in mothers' milk and found essentially the same: smoking
mothers had it in proportion to dosage; non-smoking mothers did not have
it, independent of whether fathers (or other household members) smoked.
How, then, did the father's smoking cause low birth weight?
For a likely answer let us now look at some findings quoted by Adler et al.:
Percentage who smoke (US)
Years of education Males Females
<13 41 36
13 -15 30 24
16 25 15 Source:
>16 18 17 Winkleby et al. 1990
Relative risk (RR) of mortality, British civil service
RR
Professional-executive 1.6
Clerical 2.2 Source:
Lowest grade 2.7 Susser et al. 1985
Ratio of observed to expected deaths, US, ages 21-65
Years of education Males Females
16+ .70 .78
13-15 .85 .82
12 .91 .87
9-11 1.03 .91
8 1.07 1.08
5-7 1.13 1.18 Source:
<5 1.17 1.60 Kitagawa & Hauser, 1973
Major psychological depression (Canada)
SE Status Percent
High 1.9
Medium 4.5 Source:
Low 12.4 Murphy et al. 1991
Discussion:
Here we see clear correlations between education/SE status and health,
both physical and mental. We also see a negative correlation with
smoking. *Numerous* other studies have reported similar links between SE
and health. Adler et al. cites about 80. Other studies which we will
look at shortly cite another 50.
Few would argue that education 'causes' better health. Many see society
as a selection process which sorts out individuals who are naturally
'superior' by some measure -- intelligence, social skill, or
psychological well-being -- most likely a combination. In other words,
the same individuals would be as, or nearly as, healthy had they not
been educated or successful.
Others, primarily psychologists such as Adler et al., emphasize a link
with stress. Their thesis is that education and/or higher SE status
reduces psychological stress, depression, hostility and other negative
emotion. They then attempt to show a causal connection between
psychological health and physical health.
Some point to well established links between diet/exercise and health,
then show a correlation between SE status and diet/exercise. I subscribe
to that theory, but implicate other self-care behavior in addition to
diet and exercise. Low SE people make all sorts of bad decisions.
Individually the damages may be too subtle to measure; collectively they
could produce the health effects we see. In the case of cancer, possible
hazards are nutrition, cooking style -- fried food causes cancer
[Erasmus] -- failing to take precautions with chemicals, driving
defective automobiles, occupational exposure to carcinogens and, yes,
smoking.
Low SE people associate with and marry other low SE people. Those
associates smoke more than their high SE counterparts. Statistical tests
find a weak link between ETS and lung cancer. I hypothecate they would
find as strong a link with exposure to bowling, professional wrestling
or supermarket tabloids.
Was there a SE confound in Trichopoulos et al.? The authors say no:
"Adjustment for one or more of the other variables [confounds] ... had
no effect." [1699:col 2:para 2]
"Entering years of schooling and/or occupational groups into
the model did not noticeably affect the regression coefficients, nor was
there a substantial, significant, or suggestive association of EPPL with
any of the occupational groups.." [1699:col 1]
They did not adjust for SE as a covariant because they deemed its effect
insignificant. No regression coefficients are presented to support this,
although they are given for *every one* of the other suspected
confounds: age, gender, urban/rural and smoking habits (which showed no
correlation between EPPL and dosage). Following are EPPL scores by
occupational classification. If there were no correlation between EPPL
and occupation, ideally every classification would read exactly the same
as the overall mean. I added a 'goodness of fit' analysis to test
whether deviations from 50 are significant.
Occupation EPPL SE (std err) (O-E)^2/E
Manual labor 55.18 6.23 .54
Agricultural 54.00 13.15 .32
Employees 42.53 3.79 1.12
Commercial etc 40.15 5.96 1.94
Housekeepers 41.22 5.84 1.54
5.46 vs. 9.49 (df=4, 95%)
SE status
High (last 3) 41.66 1.39
Low (first 2) 54.99 .50
1.89 vs. 3.84 (df=1, 95%)
The analysis says it IS possible for these readings to have come from a
population where every occupation's true mean is the same (had I not
excluded the small (N=5), anomalous (SE=22) Professional class, it would
have failed even this test). But x^2 = 5.46 is certainly not trivial. At
CI=80%, we would have a causal connection. It would be very interesting
to do a regression analysis between EPPL and a continuous variable such
as net worth. Unfortunately the data is not available.
Did the authors simply overlook the possibility of a SE confound?
Let us look at two earlier papers co-written by the report's own authors.
In 1989 Trichopoulos co-authored a paper which concluded that infant
mortality correlates with average income, not with access to doctors nor
hospitals. It goes beyond infant mortality to generalize: "The notion
that SE factors are strong determinants of the health status of the
corresponding populations and that social deprivation causes ill health
is at least a century old; inverse relationships between the SE status
of individuals and their risk of death have been extensively reported in
many studies. ... Other studies ... have shown that ... indices of
medical care ... are also related .. but these associations are usually
weaker, when compared to the former ones." It cites 39 other studies on
the SE confound, which the authors presumably read.
In 1991 Delsedime (a co-author of the JAMA report) co-authored a paper
dealing with survival of cancer patients. It found a correlation between
SE status and survival time. It went beyond cancer to generalize: "The
present report .. suggests the existence of inequalities in health
associated with patient SE status." It is worth noting that F Mollo, the
#2 author of the JAMA report, is mentioned in Acknowledgments. The
report cites 19 other studies on the SE confound.
Since two, or more likely three, of the JAMA authors were keenly aware
of the relationship between SE and health, having *written about its
importance* 3 and 1 years earlier, it seems remarkable that they brushed
it aside in this study. Especially when the numbers show a moderately
strong relationship.
Moreover, these numbers were for the larger (N=206) active study. We
know very little about the demographics of the 30 subjects in the
passive study. It is quite possible there was a bimodal distribution
between the 17 married to smokers and the 13 married to non-smokers.
Bimodal means having two peaks, like a camel. Since all the women had
living husbands, their average age would be less than the full sample.
That they all died before their husband is remarkable. In the full
population this happens only 20% of the time, here we have at least 50%
(30/62). The women were all non-smokers whereas 30% in the full sample
were non-smokers. Only 57% were married to smokers whereas >70% of the
male population (sampled here) were smokers. All of these indications,
taken together, suggest that observations about the larger group are
probably irrelevent to the passive group.
Also, there is a possibility of classification error -- that some of the
non-smokers were actually ex-smokers. It is quite believable that a
grieving husband might lie, forget or not know that his wife had smoked
20 years ago. Is the error not just as likely to apply to unexposed
wives? In this case I believe the errors are biased because low SE
people (smoking husbands) tend to equivocate more frequently than high
SE people. Also because smokers tend to marry other smokers; this is
called concordance bias. I will address these 'subject introduced'
biases in another study.
~References:
Adler et al., "Socioeconomic status and health", American
Psychologist:49:15 (1994)
Ciccone, Mognani, *Delsedime*, et al., "Socioeconomic Status Survival from
Soft Tissue Sarcoma", Amer Jour of Public Health:81:747 (1991)
Erasmus, "Fats that Heal, Fats that Kill", Alive Books, ISBN 0-920470-38-6
Kitagawa & Hauser, "Differential mortality in the United States: A study
in socioeconomic epidemiology", Harvard Univ Press (1973)
Millar & Hunter, "The Relationship Between Socioeconomic Status and
Household Smoking Patterns in Canada.", American journal of health
promotion:AJHP:5:1:36 (1990)
Murphy et al., "Depression and anxiety in relation to social status",
Archives of General Psychiatry:48:223 (1991)
Petridou, Valadia, *Trichopoulos*, et al., "Medical services and
socioeconomic factors: determinants of infant mortality in Greece.",
Hygie:8:20 (1989)
Schwartz-Bickenback et al., "Smoking and Passive Smoking During
Pregnancy and Early Infancy: Effects on Birth Weight, ...", Toxicology
Letters:35:73 (1987)
Susser et al., "Sociology in medicine (3rd ed.)", Oxford Univ Press (1985)
Trichopoulos et al., "Active and Passive Smoking and Pathological
Indicators of Lung Cancer Risk in an Autopsy Study", JAMA (Journal of
the American Medical Association):268:13:1697 (1992)
Winkleby et al., "Social class disparities in risk factors for disease:
Eight-year prevalence patterns by level of education", Preventive
Medicine:19:1 (1990)
---
~ DeLuxe} 1.1 #448sa ~
From: robert.wagner@channel1.com (Robert Wagner)
Date: 18 Apr 94 22:38:00 GMT
Newsgroups: alt.smokers
Subject: Second-hand science
"At a meeting of The International Council for Coordinating Cancer
Research (ICCR) a speaker offered a remarkably candid history of
the activities of the antismoking movement. He said that in the
first phase of the movement the antismoking forces presented
scientific evidence about the health effects of smoking in the
belief that smokers would be rational and stop smoking. Some did,
but many did not, so the antismoking movement adopted a second
strategy: to frighten smokers and make them feel guilty by saying
that they were stupid, short-sighted, and hedonistic. This
increased the number of smokers who stopped smoking, but many
continued to smoke. The third phase involved the "development of
the ETS evidence" ..."where the attention shifted to an attack on
smoke." A fourth strategy has been added, a direct attack on the
"tobacco cartel." The speaker argued that the tobacco companies may
compete for sales and profits but they act like a cartel in their
defense against the attacks of the antismoking movement (Daynard
1991). In these statements the author stated quite openly that the
evidence on the health effects of ETS was consciously "developed"
by the antismoking movement."
Quoted from Bernhardt Lieberman, "What the controversies over the
health effects of exposure to environmental tobacco smoke tell us about
the debates between objectivists and social constructionists.",
available by gopher (veronica) from UManitoba and several other sites.
This is a first-hand report, Lieberman was at the meeting.
We are now in phases 3 and 4.
Lieberman went on to say:
"At the American Lung Association-American Thoracic Society 1991
International Conference a session was held on the topic: Media
Advocacy for Air Pollution and Tobacco control. The session was
designed to teach the attendees how to use the media effectively to
advocate the programs of the association. One speaker advocated the
use of "creative epidemiology," which he left undefined. The term
suggests fraud, and is similar to the half-humorous condemnatory
terms, "creative accounting" and "creative statistics," which
denote the improper manipulation of numbers. ... it is clear that
members of the antismoking movement want to construct beliefs about
the harmfulness of ETS and will use methods many consider
illegitimate, but which they do not consider to be illegitimate, to
construct those beliefs."
---
~ DeLuxe} 1.1 #448sa ~
From: robert.wagner@channel1.com (Robert Wagner)
Date: 18 Apr 94 22:39:00 GMT
Newsgroups: alt.smokers
Subject: Second-hand science
PN Lee writes in "Lung Cancer and Passive Smoking: Association an
Artefact Due to Misclassification of Smoking Habits", Toxicology
Letters, 35:157-162 (1987):
1775 subjects were asked about their current use of tobacco products or
nicotine chewing gum. 1537 provided a sample of saliva for cotinine
analysis. Of 808 who claimed not to be users of such products, 20 (2.5%)
had cotinine values above 30 ng/ml, suggesting their self-reports were
false. In another study, 540 subjects were interviewed on two occasions.
10% [an average between 7.2% to 12.6%, depending on interpretation] of
the subjects claiming on one occasion never to have smoked made
inconsistent statements on the other occasion. A third study showed a
strong tendency for smokers to marry smokers.
In the context of passive smoking, misclassification of current or
ex-smokers as never-smokers, coupled with a tendency for smokers to be
married to smokers ('concordance'), can create an apparent positive
effect of passive smoking when no effect exists. This can be seen in
Table I, where a 5% misclassification rate leads to a spurious excess
risk of lung cancer in relation to spouse smoking of 44%
TABLE 1
Hypothetical Example of Bias Due to Misclassification of 5% of
Smoking Subjects as Non-Smokers.
Assumed Observed Passive
Subject Spouse N RR N RR Effect
NS NS 60 1 60+2=62 1.61 1
NS S 40 1 40+3=43 2.33 1.44
NS Total 100 1 100+5=105 1.90
-End of quote -
Of the *numerous* studies, a finding of RR 1.44 is fairly
representative, actually a little high. The two big ones (with highest
power) both found 1.2, as did the EPA's meta-analysis.
Wouldn't it be ironic if they were actually measuring peoples'
propensity to lie rather than the effect of smoke? Ironic isn't the
right word, make that 'fitting'.
Studies used in the EPA meta-analysis
Original Adjusted by EPA
RR CI (95%) RR CI (90%)
Brownson 1987 1.68 0.39,2.97 1.50 0.48,4.72
Brownson 1992 1.00 0.80,1.20 not used
Buffler 1984 1.00 0.80,1.20 0.68 0.34,1.81
Butler 1988 2.00 n/a 2.01 0.61,6.73
Correa 1983 2.07 0.81,5.26 1.89 0.85,4.14
Fontham 1991 1.28 0.93,1.75 1.28 1.03,1.60
Garfinkel 1985 1.12 0.94,1.60 1.27 0.91,1.79
Garfinkel 1981 1.17 0.85,1.89 1.16 0.89,1.52
Humble 1987 1.78 0.60,5.40 2.00 0.83,4.97 (original CI=90%)
Janerich 1990 0.93 0.55,1.57 0.79 0.52,1.17
Kabat 1984 0.90 0.46,1.76 0.73 0.27,1.89
Stockwell 1992 1.60 0.80,3.00 not used
Wu 1985 1.20 0.50,3.30 1.32 0.59,2.93
Meta-analysis 1.19 1.04,1.35
The most important thing to note here is that all of the CI's cross
unity, most by a wide margin. The EPA managed to overcome that by
combining them and lowering CI from 95% to 90% (the first time they'd
ever done that). Had they not excluded the two studies they did, even
the meta-analysis would have been inconclusive. The Stockwell study was
certainly a Major one and has the appearance of quality. It was
sponsored by the National Cancer Institute, Dept of Health and Human
Services.
As it is, the EPA meta-analysis is just *barely* significant (assuming
one regards RR as a test of significance, which it really isn't).
Most use RR > 1.2 as a rule of thumb to discount unrecognized bias.
Some use 1.5 or even 2.0 because when there IS a connection the ratio
is typically around 6-10 or or more.
In this case, each of the 13 studies failed to show significance on
its own. Even after lowering the CI to 90%, we still have 12/13 failing
to show significance. The solution: combine 11 of them to get
'CIs' which barely squeak over the line (1.04).
Combining studies into a meta-analysis *is* a valid technique, but only
when the studies are similar. Stein denigrated the combination of
heterogeneous studies as a "computational exercise". Fleiss has written
extensively on the criteria for combinability. An excellent in-depth
discussion is in:
Fleiss & Gross, "Meta-analysis in Epidemiology ...", J Clin
Epidemiol:44:2:127-139 (1991).
Stein, "Meta-analysis from one FDA reviewer's perspective", Proc
Biopharmaceutical Section of the American Statistical Assoc:1988:18.
---
~ DeLuxe} 1.1 #448sa ~
From: robert.wagner@channel1.com (Robert Wagner)
Date: 18 Apr 94 23:19:00 GMT
Newsgroups: alt.smokers
Subject: Second-hand science
Let's look at one of the reports the EPA threw out: "Environmental
Tobacco Smoke and Lung Cancer Risk in Nonsmoking Women", Stockwell et
al, J Natl Cancer Inst. 1992;84:18:1417-1421. If finds a 2.4 Odds Ratio
in women who reported > 40 years of exposure in adulthood or > 22 years
in childhood and adolescence. It says "We found no statistically
significant increase in risk associated with exposure to ETS at work or
during social activities." (I believe that one Politically Incorrect
sentence is the reason it was rejected by the EPA). Nor did it find any
correlation in adult exposure less than 40 years. It cites several other
studies including this: "A study of lifetime nonsmoking women in Hong
Kong by Koo et al showed that exposure to ETS was associated with
elevated risk of squamous cell and large-cell carcinomas."
Ok, let's look at "Measurements of Passive Smoking and Estimates of Lung
Cancer Risk Among Non-smoking Chinese Females", Koo et al, Int. J.
Cancer, 1987;39:162-169. It is an interesting read. Contrary to the cite
it concludes "In our analysis of *all* never-smoked cases, the lack of a
dose-response pattern, and an almost consistent drop in RR at the
highest doses of exposure would seem to lead little, or only weak
support for the passive smoking linkage with lung cancer ... the effects
of passive smoking might be so weak that they are easily overshadowed by
other environmental factors such as diet or exposure to inhaled
gaseous/particulate matter from other sources."
Quite a difference from Stockwell's characterization, eh? Typical
findings in the Koo study look like this:
Cig/day by husband: Adjusted RR
0 1.00
1-10 2.33
11-20 1.74
21+ 1.19
By period of life:
No exposure 1.00
Childhood only 2.07
Adulthood only 1.68
Both childhood and adulthood 0.64 !
Mean hours of exposure per day (>25 years)
< 1.5 1.47
> 1.5 1.07
These are not atypical. Other studies have found the same inverse
relationship to dosage.
----------------------------------------------------------------
Let us look at another contradiction. The potentiating effect between
air pollution and smoke wrt lung cancer was first described in a major
paper co-authored by Trichopoulos himself: "A Case-Control Study of Air
Pollution and Tobacco Smoking in Lung Cancer among Women in Athens",
Katsouyanni, Trichopoulos, et al. Preventive Medicine:20:271. It
concluded: "This study indicates that air pollution is probably
unrelated to lung cancer among nonsmokers but .. among smokers air
pollution may have an interactive effect. " It found RR=1.35 for 15
years smoking and 2.23 for 30 years.
That was in 1991. In 1992 Trichopoulos wrote (in the JAMA report) "The
present data cannot substantiate or challenge the notion that air
pollution, of the type and levels present in Athens, is an important
determinant of EPPL .. the results of the present investigation do not
deviate from the "negative" evidence generated by other
pathology-focused studies that have attempted to examine the effect of
general air pollution on lung cancer risk, and *cannot facilitate the
resolution of this important and controversial issue*."
A remarkable tergiversation in only one year's time. :-)
--------------------------------------------------------
I just found this newsgroup. The above is a sample of articles written
for another forum on another net; there's megabytes more.
I'm a high-school dropout who lives in Lubbock, Tx (don't go by origin
line), works with truck stops in the Southwest and thinks communications
is our only hope to stop neo-puritans from destroying the world. More on
that model later. :)
---
~ DeLuxe} 1.1 #448sa ~ Intellectual. Will ponder for food.