We had asked the author to shorten an earlier longer version of this
editorial, which he had submitted to us. We are posting the earlier
version (unedited) as a rapid response.
Rajendra Kale
Editorials editor, BMJ
---
(Competing interests: None declared by the author)
Passive smoking: will we ever know the truth?
George Davey Smith
Department of Social Medicine
Canynge Hall
Whiteladies Road
Bristol
BS8 2PR
In 1928 Schönherr presented a lung cancer case-series from Chemnitz
and discussed the contribution of smoking to the disease[1]. He noted that
none of the small number of cases among women occurred in smokers, but
concluded that these cancers could have been caused by the inhalation of
their husband's smoke. Thus passive smoking was identified as a potential
cause of lung cancer at the same time as active smoking was first formally
studied as a putative cause. While the issue of personal smoking and lung
cancer is now resolved, the impact of environmental tobacco smoke (ETS)
remains highly controversial[2] [3]. Partly this is because of technical
issues related to measurement error, confounding and the identification of
small risks in epidemiological studies, as Enstrom and Kabat discuss in
their paper in this week's BMJ. However it also reflects the large
investment by the tobacco industry in keeping this issue from closure,
through the funding of research, review articles, commentaries and
methodological critiques from "scientists" who, unsurprisingly, generally
seem to reach the conclusion that the evidence regarding any causal effect
of ETS is limited, and that the profits of their paymasters (and thus the
source of their rewards) should not be interfered with.
The study by Enstrom and Kabat has considerable relevance for the debate
on the health effects of passive smoking. Given the small risks associated
with ETS exposure, meta-analysis has played an important role in
demonstrating an apparent adverse effect on lung cancer, coronary heart
disease (CHD) and chronic obstructive pulmonary disease (COPD). The
consistent elevated risks associated with ETS in these meta-analyses are,
however, potentially influenced by publication bias, with small negative
studies being less likely to get published (and thus into the public
domain) than positive studies[4]. A controversial issue in this regard
relates to a tobacco industry funded analysis of the American Cancer
Society's first Cancer Prevention Study (ACS I)[5]. This has not generally
been included in published meta-analyses, although it would contribute by
far the largest number of events - and thus statistical power - to such an
analysis. The main argument advanced by the meta-analysts who have not
included the study in their reviews is that the published ACS I analysis
was not presented in a format that allowed for the combination of
equivalent effect estimates across studies. Enstrom and Kabat have
analysed the California sub-sample of the ACS I, with considerable
additional follow-up, and have presented the data in a format that allows
inclusion in future meta-analyses. They also present data that give a
handle on the degree to which misclassification of ETS exposure may dilute
the association with mortality. They interpret their findings as null,
although, inevitably, statistical uncertainty remains. Indeed they may
over-emphasise the negative nature of their findings. With respect to
chronic obstructive pulmonary disease (COPD) - plausibly related to ETS
exposure - the estimates based on the most accurately classified exposure
groups give relative risks of 1.80 in men and 1.57 in women. These are
said to be non-significant, but combining them - and there is no good
evidence that ETS exposure has a different effect for men and women -
gives a relative risk of 1.65 (95% confidence intervals 1.0-2.73). A
substantial increased risk of COPD could result from ETS exposure.
Despite this, it is certain that this paper will be hailed as
demonstrating that the detrimental effect of passive smoking has been
overstated, and controversy will continue. What are the issues?
Confounding is clearly important, with ETS exposed individuals in many
situations being likely to display adverse profiles in relation to socio-
economic position and health-related behaviors. The ACS I was established
in 1959, when smoking was much less related to such factors than it is
currently within the US, as is evident from Enstrom and Kabat paper. It
could be argued that this is why smaller - if any - ETS-associated risks
are seen in the first compared to the second American Cancer Society study
(ACS II)[6]. In ACS II, with participants recruited in 1982, women exposed
to ETS had less education than those unexposed, as opposed to the lack of
any such gradient in ACS I. Similarly amongst men in the 1982 cohort there
was little education gradient, whilst amongst men in the 1959 cohort the
exposed group had more education than the unexposed group. These figures
reflect a changing social gradient in smoking amongst men and women over
time. Socio-economic confounding in ACS II would lead to over-estimation
of the effect of ETS, whereas there is relatively little confounding in
ACS I, and what confounding there is could lead to under-estimation of the
ETS effects. The findings of the two studies are, in some respects, in
line with this - in ACS II exposure to ETS was associated with increased
risk of coronary heart disease mortality, while this is not seen in ACS I.
Misclassification is a key issue in studies of passive smoking. It is not
being married to a smoker - the indicator of ETS exposure utilized in the
Enstron and Kabat paper - that leads to disease, rather it is the
inhalation of ETS. As an indicator of ETS exposure the smoking status of
spouses is a highly approximate measure. This will lead to the ETS-
associated risk being under-estimated. Conversely misclassification of
confounders can lead to statistical adjustment failing to fully account
for confounding, leaving apparently "independent" elevated risks that are
residually confounded[7]. Methods of statistically correcting for
misclassification in both the exposure of interest and in confounders
exist, but they are highly dependant upon the validity of assessments of
measurement imprecision7 , which renders them a highly unsatisfactory
method for resolving such issues. In the passive smoking field the tobacco
industry has eagerly discussed measurement error that would lead to the
effect of passive smoking being over estimated, and rely on the work of
its consultants in this regard[8], while ignoring misclassification that
would lead to underestimation of the strength of the ETS-disease
association2.
A second approach to evaluating the risks of passive smoking is to assess
the exposure to known carcinogens produced by ETS. Tobacco industry
consultants have repeatedly suggested that levels of such exposures are to
low to be of concern, with even a heavily exposed passive smoker inhaling
much less than the equivalent of one cigarette a day2. However the amount
of exposure to the 4,000+ compounds within cigarette smoke differs between
passive and active smokers, since sidestream and mainstream smoke have
different compositions. Tobacco-specific nitrosamine 4-(methylnitrosamino)
-1-(3-pyridyl)-1-butanone (NNKs) is metabolized and metabolites excreted
in urine, and levels in non-smoking women married to smokers are about 6%
of those of their spouses (and 8 times higher in women not married to
smokers)[9]. Given the strength of relationship between active smoking
and, say, lung cancer, exposure to 6% of the dose that is received by an
active smoker could easily produce the level of risk associated with
passive smoking[10]. However the exact factors in cigarette smoke
responsible for its detrimental health consequences are not fully
understood and thus these forms of calculation are highly approximate.
Given the considerable problems with measurement imprecision, confounding
and the small predicted excess risks associated with passive smoking,
conventional observational epidemiology is somewhat limited in addressing
this issue. Randomised controlled trials of exposure to ETS will clearly
not be carried out, but it may be possible to improve understanding
through a form of natural experiment: "Mendelian Randomisation"[11].
Genetic polymorphisms that are associated with poor detoxification of
tobacco smoke carcinogens have been identified. The distribution of these
polymorphisms in the population will not be related to the same
behavioural in socioeconomic confounders that ETS exposure is currently
associated with. Amongst people unexposed to such carcinogens there is no
reason to believe that the polymorphisms would be related to lung cancer
risk. However amongst the ETS-exposed a decrease in the ability to
detoxify the carcinogens should be related to increased lung cancer risk,
if ETS exposure is responsible for increased lung cancer risk. Indeed one
study suggested that a null (non-functional) variant for one such
detoxification enzyme, GSTM1, was associated with an increased risk of
lung cancer in non-smoking women exposed to ETS, but not in non-exposed
non-smoking women[12]. However a later study failed to confirm this
finding[13], reflecting one limitation of Mendelian Randomisation, which
is that large sample sizes are required to produce robust results. However
this is a promising strategy if we really want to know whether passive
smoking increases the risk of various diseases.
Enstrom and Kabat's study demonstrates yet again the huge elevated risks
for lung cancer, COPD and - to a lesser extent - coronary heart disease,
seen among active smokers. Why, then, is there interest in passive
smoking? It is obvious that smokers have large potential health gains from
quitting. Several commercially sensitive reasons particular to passive
smoking exist. First, legal action against tobacco companies for
compensation with respect to health damaging consequences of active
smoking are opposed on the grounds that smokers should have known of the
risks. The same cannot be said of those involuntarily exposed to other
people's smoke. Second, adverse health consequences of passive smoke
encourage smoking bans in work places and other public spaces, which would
lead to reduced cigarette sales. Both strictures imposed on smoking and
the effectiveness of health promotion may increase if it is shown that
those that have no say in their exposure to tobacco smoke are being made
sick. The financial interests in this regard make it unlikely that the
debate about passive smoking and health will go away.
REFERENCES
1 Schnönherr E. Beitrag zur Statistik und Klinik der Lungentumoren. Z
Krebsforsch 1928;27:436-50.
2 Davey Smith G, Phillips AN. Passive smoking and health: should we
believe Philip Morris's "experts"? BMJ 1996;313:929-33.
3 Bayer R, Colgrove J. Science, politics, and ideology in the campaign
against environmental tobacco smoke. Am J Public Health 2002;92:949-954.
4 Vandenbroucke JP. Passive smoking and lung cancer: a publication bias?
BMJ 1988;296:391-392.
5 LeVois ME, Layard MW. Publication bias in the environmental tobacco
smoke/coronary heart disease epidemiologic literature. Regulatory
Toxicology and Pharmacology 1995;21:184-191.
6 Steenland K, Thun M, Lally C, Heath C. Environmental tobacco Smoke and
Coronary Heart Disease in the American cancer Society CPS-II Cohort.
Circulation 1996;94:622-628.
7 Phillips AN, Davey Smith G. How independent are independent effects?
Relative risk estimation when correlated exposures are measured
imprecisely. J Clin Epidemiol 1991;44:1223-31.
8 Lee PN, Forey VA. Misclassification of smoking habits as a source of
bias in the study of environmental tobacco smoke and lung cancer.
Statistics in Medicine 1996;15:591-605.
9 Andersen KE, Carmella SG, Bliss RL, Murphy L. Metabolites of a tobacco-
specific Lung carcinogen in nonsmoking women exposed to environmental
tobacco smoking. Journal of the National Cancer Institute 2001;93(5):378-
381)
10 Taylor R, Cumming R, Woodward A, Black M. Passive smoking and lung
cancer: a cumulative meta-analysis. Australian and New Zealand Journal of
Public Health 2001;25(3):203-211.
11 Davey Smith G, Ebrahim S. 'Mendelian randomization': can genetic
epidemiology contribute to understanding environmental determinants of
disease? Int J Epidemiology 2003;32:1-22.
12 Bennett WP, Alavanja MCR, Blomeke B, Vähäkangas KH, Castrén K, Welsh
JA, Bowman ED, Khan MA, Flieder DB, Harris CC. Environmental tobacco
smoke, genetic susceptibility, and risk of lung cancer in never-smoking
women. Journal of the National Cancer Institute 1999;91(23):2009-2014.
13 Malats N, Camus-Radon AM, Nyberg F, Ahrens W, Constantinescu V, Mukeria
A, Benhamou S, Batura-gabryel H, bruske-Hohfeld I, Simonato L, Menezes A,
Lea S, Lang M, Boffeta P. Lung cancer risk in nonsmokers and GSTM1 and
GSTT1 genetic polymorphism. Cancer Epidemiology, Biomarkers &
Prevention 2000;9:827833.
Rapid Response:
Earlier longer version of this editorial
We had asked the author to shorten an earlier longer version of this
editorial, which he had submitted to us. We are posting the earlier
version (unedited) as a rapid response.
Rajendra Kale
Editorials editor, BMJ
---
(Competing interests: None declared by the author)
Passive smoking: will we ever know the truth?
George Davey Smith
Department of Social Medicine
Canynge Hall
Whiteladies Road
Bristol
BS8 2PR
In 1928 Schönherr presented a lung cancer case-series from Chemnitz
and discussed the contribution of smoking to the disease[1]. He noted that
none of the small number of cases among women occurred in smokers, but
concluded that these cancers could have been caused by the inhalation of
their husband's smoke. Thus passive smoking was identified as a potential
cause of lung cancer at the same time as active smoking was first formally
studied as a putative cause. While the issue of personal smoking and lung
cancer is now resolved, the impact of environmental tobacco smoke (ETS)
remains highly controversial[2] [3]. Partly this is because of technical
issues related to measurement error, confounding and the identification of
small risks in epidemiological studies, as Enstrom and Kabat discuss in
their paper in this week's BMJ. However it also reflects the large
investment by the tobacco industry in keeping this issue from closure,
through the funding of research, review articles, commentaries and
methodological critiques from "scientists" who, unsurprisingly, generally
seem to reach the conclusion that the evidence regarding any causal effect
of ETS is limited, and that the profits of their paymasters (and thus the
source of their rewards) should not be interfered with.
The study by Enstrom and Kabat has considerable relevance for the debate
on the health effects of passive smoking. Given the small risks associated
with ETS exposure, meta-analysis has played an important role in
demonstrating an apparent adverse effect on lung cancer, coronary heart
disease (CHD) and chronic obstructive pulmonary disease (COPD). The
consistent elevated risks associated with ETS in these meta-analyses are,
however, potentially influenced by publication bias, with small negative
studies being less likely to get published (and thus into the public
domain) than positive studies[4]. A controversial issue in this regard
relates to a tobacco industry funded analysis of the American Cancer
Society's first Cancer Prevention Study (ACS I)[5]. This has not generally
been included in published meta-analyses, although it would contribute by
far the largest number of events - and thus statistical power - to such an
analysis. The main argument advanced by the meta-analysts who have not
included the study in their reviews is that the published ACS I analysis
was not presented in a format that allowed for the combination of
equivalent effect estimates across studies. Enstrom and Kabat have
analysed the California sub-sample of the ACS I, with considerable
additional follow-up, and have presented the data in a format that allows
inclusion in future meta-analyses. They also present data that give a
handle on the degree to which misclassification of ETS exposure may dilute
the association with mortality. They interpret their findings as null,
although, inevitably, statistical uncertainty remains. Indeed they may
over-emphasise the negative nature of their findings. With respect to
chronic obstructive pulmonary disease (COPD) - plausibly related to ETS
exposure - the estimates based on the most accurately classified exposure
groups give relative risks of 1.80 in men and 1.57 in women. These are
said to be non-significant, but combining them - and there is no good
evidence that ETS exposure has a different effect for men and women -
gives a relative risk of 1.65 (95% confidence intervals 1.0-2.73). A
substantial increased risk of COPD could result from ETS exposure.
Despite this, it is certain that this paper will be hailed as
demonstrating that the detrimental effect of passive smoking has been
overstated, and controversy will continue. What are the issues?
Confounding is clearly important, with ETS exposed individuals in many
situations being likely to display adverse profiles in relation to socio-
economic position and health-related behaviors. The ACS I was established
in 1959, when smoking was much less related to such factors than it is
currently within the US, as is evident from Enstrom and Kabat paper. It
could be argued that this is why smaller - if any - ETS-associated risks
are seen in the first compared to the second American Cancer Society study
(ACS II)[6]. In ACS II, with participants recruited in 1982, women exposed
to ETS had less education than those unexposed, as opposed to the lack of
any such gradient in ACS I. Similarly amongst men in the 1982 cohort there
was little education gradient, whilst amongst men in the 1959 cohort the
exposed group had more education than the unexposed group. These figures
reflect a changing social gradient in smoking amongst men and women over
time. Socio-economic confounding in ACS II would lead to over-estimation
of the effect of ETS, whereas there is relatively little confounding in
ACS I, and what confounding there is could lead to under-estimation of the
ETS effects. The findings of the two studies are, in some respects, in
line with this - in ACS II exposure to ETS was associated with increased
risk of coronary heart disease mortality, while this is not seen in ACS I.
Misclassification is a key issue in studies of passive smoking. It is not
being married to a smoker - the indicator of ETS exposure utilized in the
Enstron and Kabat paper - that leads to disease, rather it is the
inhalation of ETS. As an indicator of ETS exposure the smoking status of
spouses is a highly approximate measure. This will lead to the ETS-
associated risk being under-estimated. Conversely misclassification of
confounders can lead to statistical adjustment failing to fully account
for confounding, leaving apparently "independent" elevated risks that are
residually confounded[7]. Methods of statistically correcting for
misclassification in both the exposure of interest and in confounders
exist, but they are highly dependant upon the validity of assessments of
measurement imprecision7 , which renders them a highly unsatisfactory
method for resolving such issues. In the passive smoking field the tobacco
industry has eagerly discussed measurement error that would lead to the
effect of passive smoking being over estimated, and rely on the work of
its consultants in this regard[8], while ignoring misclassification that
would lead to underestimation of the strength of the ETS-disease
association2.
A second approach to evaluating the risks of passive smoking is to assess
the exposure to known carcinogens produced by ETS. Tobacco industry
consultants have repeatedly suggested that levels of such exposures are to
low to be of concern, with even a heavily exposed passive smoker inhaling
much less than the equivalent of one cigarette a day2. However the amount
of exposure to the 4,000+ compounds within cigarette smoke differs between
passive and active smokers, since sidestream and mainstream smoke have
different compositions. Tobacco-specific nitrosamine 4-(methylnitrosamino)
-1-(3-pyridyl)-1-butanone (NNKs) is metabolized and metabolites excreted
in urine, and levels in non-smoking women married to smokers are about 6%
of those of their spouses (and 8 times higher in women not married to
smokers)[9]. Given the strength of relationship between active smoking
and, say, lung cancer, exposure to 6% of the dose that is received by an
active smoker could easily produce the level of risk associated with
passive smoking[10]. However the exact factors in cigarette smoke
responsible for its detrimental health consequences are not fully
understood and thus these forms of calculation are highly approximate.
Given the considerable problems with measurement imprecision, confounding
and the small predicted excess risks associated with passive smoking,
conventional observational epidemiology is somewhat limited in addressing
this issue. Randomised controlled trials of exposure to ETS will clearly
not be carried out, but it may be possible to improve understanding
through a form of natural experiment: "Mendelian Randomisation"[11].
Genetic polymorphisms that are associated with poor detoxification of
tobacco smoke carcinogens have been identified. The distribution of these
polymorphisms in the population will not be related to the same
behavioural in socioeconomic confounders that ETS exposure is currently
associated with. Amongst people unexposed to such carcinogens there is no
reason to believe that the polymorphisms would be related to lung cancer
risk. However amongst the ETS-exposed a decrease in the ability to
detoxify the carcinogens should be related to increased lung cancer risk,
if ETS exposure is responsible for increased lung cancer risk. Indeed one
study suggested that a null (non-functional) variant for one such
detoxification enzyme, GSTM1, was associated with an increased risk of
lung cancer in non-smoking women exposed to ETS, but not in non-exposed
non-smoking women[12]. However a later study failed to confirm this
finding[13], reflecting one limitation of Mendelian Randomisation, which
is that large sample sizes are required to produce robust results. However
this is a promising strategy if we really want to know whether passive
smoking increases the risk of various diseases.
Enstrom and Kabat's study demonstrates yet again the huge elevated risks
for lung cancer, COPD and - to a lesser extent - coronary heart disease,
seen among active smokers. Why, then, is there interest in passive
smoking? It is obvious that smokers have large potential health gains from
quitting. Several commercially sensitive reasons particular to passive
smoking exist. First, legal action against tobacco companies for
compensation with respect to health damaging consequences of active
smoking are opposed on the grounds that smokers should have known of the
risks. The same cannot be said of those involuntarily exposed to other
people's smoke. Second, adverse health consequences of passive smoke
encourage smoking bans in work places and other public spaces, which would
lead to reduced cigarette sales. Both strictures imposed on smoking and
the effectiveness of health promotion may increase if it is shown that
those that have no say in their exposure to tobacco smoke are being made
sick. The financial interests in this regard make it unlikely that the
debate about passive smoking and health will go away.
REFERENCES
1 Schnönherr E. Beitrag zur Statistik und Klinik der Lungentumoren. Z
Krebsforsch 1928;27:436-50.
2 Davey Smith G, Phillips AN. Passive smoking and health: should we
believe Philip Morris's "experts"? BMJ 1996;313:929-33.
3 Bayer R, Colgrove J. Science, politics, and ideology in the campaign
against environmental tobacco smoke. Am J Public Health 2002;92:949-954.
4 Vandenbroucke JP. Passive smoking and lung cancer: a publication bias?
BMJ 1988;296:391-392.
5 LeVois ME, Layard MW. Publication bias in the environmental tobacco
smoke/coronary heart disease epidemiologic literature. Regulatory
Toxicology and Pharmacology 1995;21:184-191.
6 Steenland K, Thun M, Lally C, Heath C. Environmental tobacco Smoke and
Coronary Heart Disease in the American cancer Society CPS-II Cohort.
Circulation 1996;94:622-628.
7 Phillips AN, Davey Smith G. How independent are independent effects?
Relative risk estimation when correlated exposures are measured
imprecisely. J Clin Epidemiol 1991;44:1223-31.
8 Lee PN, Forey VA. Misclassification of smoking habits as a source of
bias in the study of environmental tobacco smoke and lung cancer.
Statistics in Medicine 1996;15:591-605.
9 Andersen KE, Carmella SG, Bliss RL, Murphy L. Metabolites of a tobacco-
specific Lung carcinogen in nonsmoking women exposed to environmental
tobacco smoking. Journal of the National Cancer Institute 2001;93(5):378-
381)
10 Taylor R, Cumming R, Woodward A, Black M. Passive smoking and lung
cancer: a cumulative meta-analysis. Australian and New Zealand Journal of
Public Health 2001;25(3):203-211.
11 Davey Smith G, Ebrahim S. 'Mendelian randomization': can genetic
epidemiology contribute to understanding environmental determinants of
disease? Int J Epidemiology 2003;32:1-22.
12 Bennett WP, Alavanja MCR, Blomeke B, Vähäkangas KH, Castrén K, Welsh
JA, Bowman ED, Khan MA, Flieder DB, Harris CC. Environmental tobacco
smoke, genetic susceptibility, and risk of lung cancer in never-smoking
women. Journal of the National Cancer Institute 1999;91(23):2009-2014.
13 Malats N, Camus-Radon AM, Nyberg F, Ahrens W, Constantinescu V, Mukeria
A, Benhamou S, Batura-gabryel H, bruske-Hohfeld I, Simonato L, Menezes A,
Lea S, Lang M, Boffeta P. Lung cancer risk in nonsmokers and GSTM1 and
GSTT1 genetic polymorphism. Cancer Epidemiology, Biomarkers &
Prevention 2000;9:827833.
Competing interests:
None declared
Competing interests: No competing interests