The relation between smokeless tobacco and cancer in Northern Europe and North America. A commentary on differences between the conclusions reached by two recent reviews.

Abstract

Smokeless tobacco is an alternative for smokers who want to quit but require nicotine. Reliable evidence on its effects is needed. Boffetta et al. and ourselves recently reviewed the evidence on cancer, based on Scandinavian and US studies. Boffetta et al. claimed a significant 60-80% increase for oropharyngeal, oesophageal and pancreatic cancer, and a non-significant 20% increase for lung cancer, data for other cancers being "too sparse". We found increases less than 15% for oesophageal, pancreatic and lung cancer, and a significant 36% increase for oropharyngeal cancer, which disappeared in recent studies. We found no association with stomach, bladder and all cancers combined, using data as extensive as that for oesophageal, pancreatic and lung cancer. We explain these differences.
For those cancers Boffetta et al. considered, we compared the methods, studies and risk estimates used in the two reviews.
One major reason for the difference is our more consistent approach in choosing between study-specific never smoker and combined smoker/non-smoker estimates. Another is our use of derived as well as published estimates. We included more studies, and avoided estimates for data subsets. Boffetta et al. also included some clearly biased or not smoking-adjusted estimates. For pancreatic cancer, their review included significantly increased never smoker estimates in one study and combined smoker/non-smoker estimates in another, omitting a combined estimate in the first study and a never smoker estimate in the second showing no increase. For oesophageal cancer, never smoker results from one study showing a marked increase for squamous cell carcinoma were included, but corresponding results for adenocarcinoma and combined smoker/non-smoker results for both cell types showing no increase were excluded. For oropharyngeal cancer, Boffetta et al. included a markedly elevated estimate that was not smoking-adjusted, and overlooked the lack of association in recent studies.
When conducting meta-analyses, all relevant data should be used, with clear rules governing the choice between alternative estimates. A systematic meta-analysis using pre-defined procedures and all relevant data gives a lower estimate of cancer risk from smokeless tobacco (probably 1-2% of that from smoking) than does the previous review by Boffetta et al.

Full text

Page 1

BioMed Central
Page 1 of 10
(page number not for citation purposes)
BMC Cancer
Open Access
Correspondence
The relation between smokeless tobacco and cancer in Northern
Europe and North America. A commentary on differences between
the conclusions reached by two recent reviews
Peter N Lee* and Jan Hamling
Address: P.N. Lee Statistics and Computing Ltd, Surrey, UK
Email: Peter N Lee* - PeterLee@pnlee.co.uk; Jan Hamling - JanHamling@pnlee.co.uk
* Corresponding author
Abstract
Background: Smokeless tobacco is an alternative for smokers who want to quit but require
nicotine. Reliable evidence on its effects is needed. Boffetta et al. and ourselves recently reviewed
the evidence on cancer, based on Scandinavian and US studies. Boffetta et al. claimed a significant
60–80% increase for oropharyngeal, oesophageal and pancreatic cancer, and a non-significant 20%
increase for lung cancer, data for other cancers being "too sparse". We found increases less than
15% for oesophageal, pancreatic and lung cancer, and a significant 36% increase for oropharyngeal
cancer, which disappeared in recent studies. We found no association with stomach, bladder and
all cancers combined, using data as extensive as that for oesophageal, pancreatic and lung cancer.
We explain these differences.
Methods: For those cancers Boffetta et al. considered, we compared the methods, studies and
risk estimates used in the two reviews.
Results: One major reason for the difference is our more consistent approach in choosing
between study-specific never smoker and combined smoker/non-smoker estimates. Another is our
use of derived as well as published estimates. We included more studies, and avoided estimates for
data subsets. Boffetta et al. also included some clearly biased or not smoking-adjusted estimates.
For pancreatic cancer, their review included significantly increased never smoker estimates in one
study and combined smoker/non-smoker estimates in another, omitting a combined estimate in the
first study and a never smoker estimate in the second showing no increase. For oesophageal cancer,
never smoker results from one study showing a marked increase for squamous cell carcinoma were
included, but corresponding results for adenocarcinoma and combined smoker/non-smoker results
for both cell types showing no increase were excluded. For oropharyngeal cancer, Boffetta et al.
included a markedly elevated estimate that was not smoking-adjusted, and overlooked the lack of
association in recent studies.
Conclusion: When conducting meta-analyses, all relevant data should be used, with clear rules
governing the choice between alternative estimates. A systematic meta-analysis using pre-defined
procedures and all relevant data gives a lower estimate of cancer risk from smokeless tobacco
(probably 1–2% of that from smoking) than does the previous review by Boffetta et al.
Published: 29 July 2009
BMC Cancer 2009, 9:256 doi:10.1186/1471-2407-9-256
Received: 12 February 2009
Accepted: 29 July 2009
This article is available from: http://www.biomedcentral.com/1471-2407/9/256
© 2009 Lee and Hamling; licensee BioMed Central Ltd.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0),
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Page 2

BMC Cancer 2009, 9:256http://www.biomedcentral.com/1471-2407/9/256
Page 2 of 10
(page number not for citation purposes)
Background
In 2008, Boffetta et al. [1] published a short review in Lan-
cet Oncology of the evidence relating smokeless tobacco
(ST) to cancer. Included was a table summarizing smok-
ing-adjusted relative risk (RR) estimates with 95% confi-
dence intervals (CI) relating to cancer of the oral cavity,
oesophagus, pancreas and lung in the USA and Northern
Europe taken from 18 studies, together with a further
table of meta-analysis results. The results of the overall
(USA and Nordic countries combined) meta-analyses are
summarized in Table 1, and show a statistically significant
increase of 60–80% for ever smokeless tobacco use for
oral, oesophageal and pancreatic cancer, and a non-signif-
icant 20% increase for lung cancer. Results for other can-
cers were stated to be "too sparse for a quantitative
investigation."
In their review Boffetta et al. [1] give only limited informa-
tion on their "search strategy and selection criteria." While
they make it clear that they restricted attention to papers
published up to September 2007 (including one in press
at that time) they give little information on how they
selected the cancers for detailed study or how they chose
the estimates to be included in their meta-analyses. Thus
they note that results for cancers other than those of the
oral cavity, oesophagus, pancreas, and lung were "too
sparse for quantitative information" without specifying
the amount of data needed for analysis. Furthermore they
state merely that "we included only studies restricted to
non-smokers and studies that included smokers but were
properly adjusted for the possible confounding effect of
tobacco smoking." without giving any indication as to
how they chose from alternative estimates available in a
number of the papers (e.g. by sub-type of cancer, type of
smoking adjustment, type of ST or timing of ST exposure).
A meticulous description of the methods used should
have been included, but was not.
Shortly before the review of Boffetta et al. [1] was pub-
lished, we had started our own review of this evidence, a
review which has recently been published in BMC Medi-
cine [2]. We continued with our review, because our ini-
tial impression of Boffetta et al.'s was that some relevant
data had been missed and that some of the RRs used in
their meta-analyses seemed inappropriate. Although our
review also considered effect estimates that were not
adjusted for smoking, we took particular care to distin-
guish those that were adjusted for smoking. Our smoking-
adjusted meta-analysis estimates are also shown in Table
1. As will be seen, our estimates are substantially lower for
all four cancers considered by Boffetta et al. For oesopha-
geal, pancreatic and lung cancer the estimated increases
are all less than 15% and not statistically significant, while
for oral cancer our estimate of a 36% increase, though sta-
tistically significant, is lower than the 80% increase esti-
mated by Boffetta et al., and disappears when attention is
restricted to studies published since 1990. Our review also
considers a range of other cancers, and Table 1 also
presents meta-analysis estimates for stomach, bladder and
overall cancer. Each is based on at least as many RRs as are
available for oesophageal, pancreatic and lung cancer, and
none shows a significant excess risk in ST users.
Objectives
The objective of this article is to provide a detailed com-
parison of the two reviews [1,2] in order to clarify why
these major differences in risk estimates have occurred.
Attention is restricted to the four cancers considered by
Boffetta et al. [1]
Differences between the estimates from the two reviews
Table 2 (oropharyngeal cancer [3-21]), Table 3 (oesopha-
geal cancer [4,18,22-25]), Table 4 (pancreatic cancer
[4,5,26-30]) and Table 5 (lung cancer [3-5,27,31]) sum-
marize the estimates used in the two reviews [1,2], with
comments on similarities and differences. Based on this
comparison, the details of the methodology given in our
review [2], and the rather brief description of their proce-
dures presented by Boffetta et al. [1], a number of general
observations can be made.
Sources of difference between the two reviews
Derivation of estimates
Whereas Boffetta et al. [1] limited themselves to using RR
estimates given in the source publication, we [2] calcu-
lated an estimate using available methodology [32-35]
where the required RR was not provided but could be
derived from data given in the publication,. We felt this
necessary so as to avoid omitting relevant studies com-
pletely or, when a study provided non-independent
results from subsets of the data, presenting results only for
one of the subsets.
Table 1: Comparison of our smoking-adjusted random-effects
meta-analysis estimates with those of Boffetta et al.
Boffetta et al. [1]Lee and Hamling [2]
CancerNa
RR (95% CI)b
Na
RR (95% CI)b
Oropharyngeal
- published since 1990
Oesophageal
Pancreatic
Lung
Stomach
Bladder
Overall cancer
131.8 (1.1–2.9)
Not given
1.6 (1.1–2.3)
1.6 (1.1–2.2)
1.2 (0.7–1.9)
Not given
Not given
Not given
19
14
7
7
6
8
10
7
1.36 (1.04–1.77)
1.00 (0.83–1.20)
1.13 (0.95–1.36)
1.07 (0.71–1.60)
0.99 (0.71–1.37)
1.03 (0.88–1.20)
0.95 (0.71–1.29)
0.98 (0.84–1.15)
5
6
5
a Number of individual estimates considered in meta-analysis.
b Smoking-adjusted estimates for any ST use.

Page 3

BMC Cancer 2009, 9:256http://www.biomedcentral.com/1471-2407/9/256
Page 3 of 10
(page number not for citation purposes)
Table 2: Comparison of individual and overall (random-effects) estimates for the two reviews – oropharyngeal cancer
ST usea
RefTypeExposureInclusion of smokersb
Reviewc
SexRelative risk (95% CI)Comments
[3]
(CPS-I)
ST
ST
Current
Current
NS
NS
L&H
B
M
M
2.02 (0.53–7.74)
2.0 (0.5–7.7) Estimates agreed
[3]
(CPS-II)
ST
ST
Current
Current
NS
NS
L&H
B
M
M
0.90 (0.12–6.71)
0.9 (0.1–6.7)Estimates agreed
[4]Snuff
Snuff
Ever
Ever
SNS
SNS
L&H
B
M
M
1.10 (0.50–2.41)
1.1 (0.5–2.4) Estimates agreed
[5]Snuff
Snuff
Ever
Ever
SNS
NS
L&H
B
M
M
0.7 (0.5–0.9)
0.8 (0.4–1.7)NS not SNS
[6]SnuffEverSNS L&HM 3.1 (1.5–6.6)Too recent to be included by B
[7]ChewEverSNSL&HM+F 2.05 (1.48–2.83)e
Not included by B
[8]ChewEverSNS L&HM2.00 (1.16–3.47)e
Not included by B
[9]STEverSNSL&HM3.63 (1.02–12.95)e
Not included by B
[10]Snuff
Snuff
Snuff
Ever
Ever
Ever
SNS
NSf
NSf
L&H
B
B
F
F
F
2.67 (1.83–3.90)e
4.2 (2.6–6.7)
1.5 (0.5–4.8)
Whites
Blacks
[11]ST
ST
Ever
Ever
SNS
SNS
B
B
M
M
2.3 (0.2–12.9)
11.2 (4.1–30.7)
Tongue cancer
Mouth cancer
Not included by L&H as no valid smoking adjustmentg
[12]ST
ST
Ever
Ever
NS
NS
L&H
B
F
F
6.2 (1.9–19.8)
6.2 (1.9–19.8)Estimates agree
[13]SnuffEverNSL&HM+F0.67 (0.08–5.75)e
Not included by B
[14]STEverSNSL&H M+F1.04 (0.41–2.68)e
Not included by B
[15]ST
Chew
Ever
Ever
SNS
SNS
L&H
B
M
M
0.96 (0.70–1.33)e
1.0 (0.7–1.4)Chew not ST
[16]Chew
Chew
Ever
Ever
SNS
NS
L&H
B
M
M
1.11 (0.81–1.53)e
2.3 (0.7–7.3)NS not SNS
[17]h
STEverSNSL&HM+F1.43 (0.64–3.21)e
Not included by B
[18]Snuff
Snuff
Ever
Ever
SNS
SNS
L&H
B
M
M
0.98 (0.63–1.50)e
1.4 (0.8–2.4)Oral cancer excluding pharynx
[19]Snuff
Snuff
Ever
Ever
SNS
SNS
L&H
B
M+F
M+F
0.8 (0.5–1.3)
0.8 (0.5–1.3)
Estimate for NS also available
Estimates agree
[20]STEverSNSL&HM1.0 (0.4–2.3)Not included by B
[21]SnuffEverSNSL&HM+F0.7 (0.3–1.3)Not included by B
TotalL&H
B
1.36 (1.04–1.77)
1.8 (1.1–2.9)
19 estimates
13 estimates
a ST = smokeless tobacco; Chew = chewing tobacco; ever exposure includes undefined use.
b NS = never smokers; SNS = smokers and nonsmokers combined (with adjustment for smoking).
c L&H = Lee and Hamling review [2]; B = Boffetta et al. review [1].
d To within rounding error, as B only expressed estimates to one decimal place.
e Estimated from data provided.
f B stated that the results were for never smokers, but L&H consider the result relates to non-current smoking. L&H's estimate is for current and non-current smokers
combined.
g Valid smoking adjustment was impossible in this study as "for users of multiple tobacco products, only the primary product was recorded".
h The results were cited by Gross et al. [17] based on an unpublished report by Perry et al., "Attributable oral cancer risk due to smokeless tobacco use based on a case-
control study at Sinai Hospital in Detroit".

Page 4

BMC Cancer 2009, 9:256http://www.biomedcentral.com/1471-2407/9/256
Page 4 of 10
(page number not for citation purposes)
Restriction to smoking-adjusted estimates
As noted earlier, Boffetta et al. [1] stated that they
"included only studies restricted to non-smokers and
studies that included smokers but were properly adjusted
for the possible confounding effect of tobacco smoking."
Though, as is so often the situation in smoking and health
literature, the term "non-smokers" was not defined, we
have assumed that "lifelong non-smokers" (i.e. never
smokers) was meant. What was meant by "properly
adjusted" was also undefined, and in practice it appears
that any smoking adjustment was accepted, as we could
find no case of a published smoking-adjusted RR that was
not included by Boffetta et al. This is not surprising since,
as noted in our review [2], only a small proportion of
studies took any account of daily consumption or dura-
tion of smoking. As a consequence, the smoking-adjusted
data in Tables 2, 3, 4, 5 taken from our review are also not
restricted on how the adjustment for smoking was carried
out.
Studies included
We included more studies in our review [2] than did Bof-
fetta et al [1] in theirs. Mainly these are studies [7-
9,13,14,17,23,27] where the estimate required calcula-
tion, but three studies [20,21,31] appear to have been
overlooked by Boffetta et al., and there is also a recent
study [6] published after September 2007, the cut-off date
for their literature search. There is only one study [11]
included by Boffetta et al., but not by us. This study did
not provide results for never smokers, and though it
claimed to have presented estimates adjusted for tobacco
use, this appears impossible as the authors stated that "for
users of multiple tobacco products, only the primary
product was recorded." A comparison of ST users with
non-users of tobacco will therefore be biased by smokers
being included only in the group using ST.
ST type
The majority of studies presented results only for one type
of ST, usually either snuff specifically (typical for Scandi-
navian studies) or for overall ST use. A few studies provide
separate RRs for snuff and chewing tobacco. The RRs we
used in our meta-analyses [2] were based on overall ST use
if possible, calculated if necessary from the separate
results. We note that there were two studies [15,30] where
results were available for snuff and chewing tobacco, and
where Boffetta et al. [1] included only the results for chew-
ing tobacco.
ST exposure
The great majority of the studies present only RRs for
either ever use or unspecified use (which both reviews
Table 3: Comparison of individual and overall (random-effects) estimates for the two reviews – oesophageal cancer
ST usea
Ref Type ExposureInclusion of smokersb
Reviewc
Sex Relative risk (95% CI) Comments
[4]Snuff
Snuff
Ever
Ever
SNS
SNS
L&H
B
M
M
1.40 (0.61–3.24)
1.4 (0.6–3.2)Estimates agreed
[22]Snuff
Snuff
Ever
Ever
SNS
NS
L&H
B
M
M
1.00 (0.79–1.27)e
3.5 (1.6–7.6)
Estimate for NS also available
NS not SNS; squamous cell carcinoma not all
oesophageal cancer
[23]Chew
Chew
Ever
Ever
NS
NS
L&H
L&H
M
F
1.18 (0.28–4.90)e
2.69 (0.92–7.87)e
Not included by B
Not included by B
[24]ST
ST
Ever
Ever
NS
NS
L&H
B
M
M
1.2 (0.1–13.3)
1.2 (0.1–13.3)Estimates agree
[18]Snuff
Snuff
Ever
Ever
SNS
SNS
L&H
B
M
M
1.2 (0.7–2.2)
1.2 (0.7–2.2)Estimates agree
[25]Snuff
Snuff
Ever
Ever
SNS
SNS
L&H
B
M+F
M+F
1.31 (0.89–1.92)e
1.4 (0.9–2.3)Squamous cell carcinoma not all oesophageal cancer
TotalL&H
B
1.13 (0.95–1.36)
1.6 (1.1–2.3)
7 estimates
5 estimates
a ST = smokeless tobacco; ever exposure includes undefined use.
b NS = never smokers; SNS = smokers and nonsmokers combined (with adjustment for smoking).
c L&H = Lee and Hamling review [2]; B = Boffetta et al. review [1].
d To within rounding error, as B only expressed estimates to one decimal place.
e Estimated from data provided.

Page 5

BMC Cancer 2009, 9:256http://www.biomedcentral.com/1471-2407/9/256
Page 5 of 10
(page number not for citation purposes)
[1,2] have considered as essentially equivalent to ever
use). Some studies present results for current and former
use, and our procedure was to include the result for cur-
rent use only if an estimate for ever use was not available
or could not be calculated. Since they did not calculate RR
estimates, Boffetta et al. [1] included current rather than
ever use estimates for lung cancer for CPS-II [3]. However,
otherwise the two sets of estimates agree as regards ST
exposure.
Selection of results according to smoking history
There are a number of studies where RRs are available
both for never smokers and for smokers and non-smokers
combined, with adjustment for smoking. In the meta-
analyses shown in Table 1 taken from our review [2], we
have always included the smoker/non-smoker combined
estimate from these studies, on the basis that they pro-
vided greater power, though our review also presents the
results of meta-analyses of RRs specifically for never
smokers. Boffetta et al. [1] appear not to have defined any
rule here. In three such studies [5,22,30] they include
results for never smokers, and in two studies [4,19] the
results for smokers and non-smokers combined, without
any supporting explanation.
Types of cancer
Where results are available by type of cancer, we have
always included estimates for the total cancer being con-
sidered, but this is not the case for Boffetta et al [1]. Thus
whereas, for Table 2, we [2] include RRs for overall
oropharyngeal cancer, if available, only considering can-
cers of particular regions of the oropharynx if these were
the only data presented, Boffetta et al. omitted relevant
results for pharynx cancer in one study [18] and presented
RRs separately for mouth and tongue cancer in another
[11], a study in which results were also available for a
number of other regions of the oropharynx. For oesopha-
geal cancer, there are two studies [22,25] where Boffetta et
al. included results only for squamous cell carcinoma,
omitting those for adenocarcinoma, despite the other
studies in their analysis only presenting results for overall
oesophageal cancer.
Table 4: Comparison of individual and overall (random-effects) estimates for the two reviews – pancreatic cancer
ST usea
RefTypeExposure Inclusion of smokersb
Reviewc
SexRelative risk (95% CI) Comments
[26] ST
ST
Ever
Ever
SNS
SNS
L&H
B
M
M
1.7 (0.9–3.1)
1.7 (0.9–3.1)Estimates agree
[4]Snuff
Snuff
Ever
Ever
SNS
SNS
L&H
B
M
M
1.67 (1.12–2.50)
1.7 (1.1–2.5)
Estimates for NS also available
Estimates agreed
[5]Snuff
Snuff
Ever
Ever
SNS
NS
L&H
B
M
M
0.9 (0.7–1.2)
2.0 (1.2–3.3)NS not SNS
[27]STEver SNS L&HM0.29 (0.09–0.92)e
Not included by B
[28]Chew
Chew
Ever
Ever
NS
NSf
L&H
B
M
M
2.82 (0.85–9.39)
3.6 (1.0–12.8)
Personal communication from Dr Muscat
Estimate actually for non-current smokers
[29]ST
ST
Ever
Ever
NSg
NS
L&H
B
M+F
M+F
1.1 (0.4–3.1)
1.4 (0.5–3.6)Estimate biased as pipe and cigar smokers included in
numerator only
[30]ST
Chew
Ever
Ever
SNS
NS
L&H
B
M+F
M+F
0.65 (0.43–0.97)e
0.6 (0.3–1.4)
Estimate for NS also available
Chew not ST; NS not SNS
TotalL&H
B
1.07 (0.71–1.60)
1.6 (1.1–2.2)
7 estimates
6 estimates
a ST = smokeless tobacco; ever exposure includes undefined use.
b NS = never smokers; SNS = smokers and nonsmokers combined (with adjustment for smoking).
c L&H = Lee and Hamling review [2]; B = Boffetta et al. review [1].
d To within rounding error, as B only expressed estimates to one decimal place.
e Estimated from data provided.
f See comment.
g Never cigarette smokers, with adjustment for other tobacco use.