Oral use of Swedish moist snuff (snus) and risk for cancer of the mouth, lung, and pancreas in male construction workers: a retrospective cohort study

by: Juhua Luo, Weimin Ye, Kazem Zendehdel, Johanna Adami, Hans-Olov Adami, Paolo Boff etta, Olof Nyrén


Summary
Background Although classifi ed as carcinogenic, snuff is used increasingly in several populations. Scandinavian moist snuff (snus) has been proposed as a less harmful alternative to smoking, but precise data on the independent associations of snus use with site-specific cancers are sparse. We aimed to  assess the risks for cancer of the oral cavity, lung, and pancreas.

Methods Detailed information about tobacco smoking and snus use was obtained from 279 897 male Swedish construction workers in 1978–92. Complete follow-up until end of 2004 was accomplished through links with population and health registers. To distinguish possible eff ects of snus from those of smoking, we focused on 125 576 workers who were reported to be never-smokers at entry. Adjusted relative risks were derived from Cox proportional hazards regression models.

Findings 60 cases of oral, 154 of lung, and 83 of pancreatic cancer were recorded in never-smokers. Snus use was independently associated with increased risk of pancreatic cancer (relative risk for ever-users of snus 2·0; 95% CI 1·2–3·3, compared with never-users of any tobacco), but was unrelated to incidence of oral (0·8, 95% CI 0·4–1·7) and lung cancer (0·8, 0·5–1·3).

Interpretation Use of Swedish snus should be added to the list of tentative risk factors for pancreatic cancer. We were unable to confi rm any excess of oral or lung cancer in snus users.

---

Introduction
Use of snuff has become increasingly popular in several countries, but Sweden has the highest consumption, predominantly in the form of moist snuff (snus). The habit is especially gaining popularity in adolescents and women.1 At present, however, the majority of users are men; at least 23% of Swedish men used snus in 2002.2

About 30 carcinogens have been identifi ed in smokeless tobacco, and the tobacco-specifi c nitrosamines, formed from nicotine and related tobacco alkaloids, are thought to be particularly important.3 The tobacco-specifi c nitrosamines with the greatest proportions in snuff (4-(nitrosomethylamino)-1-(3-pyridyl)- 1-butanone [NNK] and N´-nitrosonornicotine [NNN]), have been implicated in the cause of tobacco related cancers.4–6 Comparative studies have generally shown lower concentrations of tobacco-specifi c nitrosamines in Swedish snus than in American snuff ,7 leading to a perception that the use of Swedish snus is a suitable alternative to smoking. Indeed, with a few exceptions,8–10 studies of Scandinavian snus have shown no risk associated with use of this form of tobacco.7 The Scandinavian experience diff ers from that in South Asia11 and elsewhere,12,13 where smokeless tobacco is an established risk factor for oral cancer. This inconsistency might be attributable to methodological aspects, such as inadequate control for confounding by cigarette smoking and alcohol use, which are strong risk factors for oral cancer.

Because of NNK’s specifi city for the lung in rodent cancer models,14,15 lung cancer should be another concern in relation to smokeless tobacco.  However, few studies have addressed this risk in human beings. The only study of Scandinavian snus and lung cancer showed a non-significantly decreased risk in snus users,10 raising questions about residual confounding due to smoking. Epidemiological evidence10,16–18 suggests that the use of smokeless tobacco, including Scandinavian snus,10 might increase the risk of pancreatic cancer, but published data are based on few snus-exposed cases.

With a growing awareness of the health hazards associated with smoking, snus could become increasingly popular,19,20 and the habit might spread to people who would otherwise refrain from tobacco use. Therefore, valid and precise epidemiological data on health risks associated with use of snus are urgently needed. We consequently did a prospective study in Swedish construction workers, with a high prevalence of exposure to snus, to address the association of snus use with oral, lung, and pancreatic cancer.

---

Methods
Setting and participants
The background of the Swedish construction worker cohort has been described previously.21 Briefly, from 1969 through 1992, preventive health check-ups were offered to all workers in the Swedish building industry, and from 1971, the collected data were compiled in a computerised central register. Each record also contained the participant’s National Registration Number, a unique personal identifier assigned to every Swedish resident at birth or immigration. This identifi er includes the date of birth. Because of ambiguities in the coding of smoking status in the questionnaires used during 1971–75 (Zendehdel K, et al, unpublished), we restricted our analysis to workers with at least one visit in the 1978–92 period, when information on smoking and snus use was obtained through personal interviews by nurses. Because the group contained few women, we limited our analyses to men. Links with nationwide registers of the total population, emigration, and death enabled us to exclude records with incorrect National Registration Numbers (which could not be found in any of these registers), and men with a death or emigration date before entry. Links with the Swedish Cancer Register led to exclusion of men with cancer before entry. We also excluded men with incomplete tobacco exposure data.

---

Procedures
We only used exposure information obtained at the first visit, which defi ned entry into the cohort: snus user status (never, previous, or current), grams of snus per day (<10 g or ≥10 g), smoking status (never, previous, or current), grams of smoking tobacco per day (continuous), and body-mass index (BMI; <25, 25–29, or ≥30). The quality of exposure data has been reviewed previously and was deemed satisfactory.21

Follow-up was done through linking of records to the nationwide, and essentially complete, population and health registers previously mentioned. For correct censoring, dates of death were obtained from the Causes of Death register, and dates of emigration came from the Register of Domestic and International Relocations. The Cancer Register, established in 1958, codes malignant neoplasms according to the International Classification of Diseases, 7th edition, and includes information on more than 98% of all diagnosed cases in Sweden.22,23 We used codes 140, 141, 143, and 144 for incident cases of oral cancer (not including cancers of the salivary glands, pharynx, or larynx), code 162 for lung cancer, and code 157 for pancreatic cancer. Each cohort member contributed person-time from the date of entry until the date of any first cancer diagnosis, migration, death, or December 31, 2004, whichever occurred first.

---

Statistical analysis
All three cancers are highly age dependent. Therefore, we investigated age distributions in each exposure category. The associations between exposure variables and risk of cancer were expressed as relative risks (RRs) derived from Cox proportional hazards regression models, with attained age (continuous) as time scale. Initially, we fi tted models in which the relative risks associated with smoking were adjusted for snus use, and in which relative risks linked to snus use were adjusted for smoking. To better control the strong confounding effect of smoking in our analyses of snus, we fitted models restricted to never-smokers. We adjusted for BMI in all our models. However, since BMI could conceivably be in the causal pathway, we also did analyses unadjusted for this factor. Tests for linear trend were done by creating a continuous variable from the median of the categories.

The assumption of proportional hazards was tested on the basis of the cumulative sums of Martingale residuals with the Kolmogorov-type supremum test,24 in which 1000 realisations were used. Results indicated that the proportional assumption was satisfi ed for all models.

---

Role of the funding source
The funding source had no role in the study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had fi nal responsibility for the decision to submit the paper for publication.

---

Results
The figure shows the numbers of eligible workers included in and excluded from the group for our analysis. Characteristics of the 279 897 men in our cohort, including smoking and snus use, are shown in table 1. Average age at entry was 35 years (SD 13). These men were followed-up for an average of 20 years (SD 6). At time of entry, 31% of the cohort members used or had previously used snus. The proportion of ever-smokers was greater for men older than 30 years than in younger men, whereas snus use was more common in those younger than 30 years, refl ecting the spreading habit in the Swedish male population.

258 incident cancers of the oral cavity, 2216 of the lung, and 468 of the pancreas were recorded during follow-up. Of these, 60 oral, 154 lung, and 83 pancreatic cancers occurred in the 125 576 never-smokers.

We confirmed that tobacco smoking was a strong risk factor for all the studied cancers (table 2). The Cox regression models, which also included men who used snus simultaneously, were adjusted for attained age, BMI, and snus use. Removal of BMI from the models had little eff ect on the results (data not shown).

In analyses that included all cohort members, irrespective of smoking and snus user status, the adjusted relative risks for cancer in ever-users of snus, compared with never-users, were 0·7 (95% CI 0·5–0·9) for oral, 0·7 (0·6–0·7) for lung, and 0·9 (0·7–1·2) for pancreatic cancer. In analyses restricted to men who were never-smokers, ever-use of snus was associated with a significant increase of the risk for pancreatic cancer, compared with the risk in never-users of any tobacco (table 3). We also noted a significant dose-risk trend for pancreatic cancer with increasing amount of snus use (p=0·01). However, the point estimates for the two dose categories above zero (1–9 g and ≥10 g snus per day) did not diff er greatly from each other. We did not observe an increased risk of oral cancer or lung cancer in men who used snus but did not smoke. Repeated analyses without adjustments for BMI produced similar results (data not shown).

---

Discussion
The main finding of this large cohort study was an increased risk of pancreatic cancer in never-smoking snus users compared with never-users of any tobacco, with some evidence for a dose-risk association. We did not detect any excess risk for cancer of the oral cavity or lung.

Our fi nding is at odds with the perception that use of Swedish moist snus has no demonstrable carcinogenic risk.7 If valid, it will have important public-health implications, since snus has been proposed as a way to reduce harm in nicotine addicts.19,20 The increase in risk is, however, in line with that reported in a cohort study from Norway10—the only published Scandinavian study on the association between use of smokeless tobacco and risk of pancreatic cancer. In that study, a significant 70% excess incidence was noted in ever-users relative to never-users of smokeless tobacco, after adjustment for smoking and alcohol use.10,25 Some of the tobacco consumption was in the form of local chewing tobacco (skrå). In our cohort, the participants reported specifically about snus use, and use of other smokeless tobacco products was probably negligible. Results of several American studies of smokeless tobacco support our findings16–18 although some do not.26,27

The excess risk was noticeable only in an analysis restricted to the never-smoking stratum. This analysis was defined a priori to eliminate residual confounding by smoking dose. Previous evidence, reinforced by observed data in the present study (not shown), suggests that individuals who combine smoking with snus use smoke less and might increase their overall chances of subsequent abstinence, compared with those who only smoke.28 Indeed, although fi ndings of a Swedish casecontrol study8 showed no signifi cant relation between use of snus and overall risk of head and neck cancer in multivariate-adjusted analyses, snus use among neversmokers was associated with an almost fi ve-times increased risk. In the Norwegian cohort study mentioned previously10 a 20% reduction in risk of lung cancer was noted in multivariate-adjusted analyses, again suggesting residual negative confounding. The shift from a similar inverse association with lung cancer in our multivariateadjusted analysis to a null result in the analysis restricted to never-smokers is in good agreement with the Norwegian data and provides further support for the concern about confounding. Hence, we believe that the estimate for snus in never-smokers is less biased than an estimate obtained in an overall analysis that also includes smokers and in which control for confounding by smoking is attempted through multivariate modelling. The absence of association with lung cancer in this stratum, in eff ect, confirms the absence of important confounding by smoking.

Efficient adjustment for smoking dose in snus-using smokers is expected to nullify any positive consequences of snus use conferred through its purported anti-smoking effects. The signifi cant risk reductions for all three studied cancers among snus users noted in our conventional models that included the entire group, despite our attempts to adjust for smoking dose, suggest that the net eff ect of snus use in the studied population might be a reduced risk of cancer.

The apparent specifi city for the pancreas as the target organ is biologically plausible. First, the carcinogenicity of tobacco-specific nitrosamines is remarkably organspecific in animal experiments.6 Although the lung and upper respiratory tract dominate as target organs, rats develop pancreatic adenocarcinoma when exposed to NNK or its metabolite 4-(methylnitrosamino)-1-(3- pyridyl)-1-butanol (NNAL) in drinking water.14 Second, measurable amounts of NNK and NNAL have been documented in human pancreatic juice, in the case of NNK at significantly higher concentrations in smokers than in non-smokers.29 Third, it is well established that NNK metabolites bind to DNA and induce activating point mutations in the RAS gene—mutations that are observed in 50–90% of all pancreatic adenocarcinomas.30 Fourth, NNK acts as an agonist on β-adrenergic receptors, which activate signal transduction pathways that induce the formation of arachidonic acid and its mitogenic metabolites.30 Fifth, Swedish data suggest a causal link between snus use and risk of type 2 diabetes,31 and increasing evidence implicates insulin resistance and abnormal glucose metabolism as risk factors for development of pancreatic cancer.32

The absence of an increased risk for oral cancer in snus users confi rms the negative results of published work on this particular type of smokeless tobacco.8,10,33,34 However, residual negative confounding from smoking dose cannot be confi dently excluded in these studies, as discussed above. An International Agency for Research on Cancer working party recently concluded, mainly on the basis of American and Asian data, that sufficient evidence exists that smokeless tobacco causes oral cancer in human beings.13 With only ten cases among ever-users of snus in the never smoker stratum, oral cancer was the least common cancer of the three studied in our analysis, making the estimates liable to chance variations.

In accord with our fi ndings, previous epidemiological evidence on smokeless tobacco and lung cancer in developed countries has been essentially negative,10,26,35 with few exceptions,36 despite the strong link between exposure to tobacco-specifi c nitrosamines and formation of lung tumours in rodents.6 The reasons for the discrepancy between animal and human data remain to be clarified; in our study, confounding from smoking dose is an unlikely explanation.

Our study has several strengths but also some limitations. An important strength is the cohort design, which essentially precludes the possibility that the cancer outcome could have affected the initial reports about, or the actual use of, the tobacco products of interest. One disadvantage of this design is that individuals’ tobacco-use habits might have changed during follow-up. The repeat visits during follow-up varied in number and timing, and therefore were sensitive to self-selection bias. However, we used the smoking information recorded at these visits to investigate whether workers who were initially classifi ed as never-smoking snus users might diff er from those who were classifi ed as never-users of any tobacco. We found that 2132 of 17 634 (12%) of never-smoking snus users were later recorded at some point in time as former or current smokers. The corresponding proportion in never-users of any tobacco was 2824 of 39 469 (7%). We used these data and the effect sizes derived from tables 2 and 3 in a sensitivity analysis according to Schneeweiss.37 The suggested misclassification of smoking status aff ected our reported estimates no more than trivially (data not shown). In accord with a recent Swedish study that reported a high probability of continuing snus use once the habit has been initiated,38 our data from the repeat visits suggested that dose of snus remained stable over time (data not shown).

Another strength is the completeness of follow-up. Furthermore, the large cohort size and the high prevalence of exposure to snus made it possible to obtain meaningful estimates in never-smokers. However, the statistical precision is still a weak point; the estimates for the three types of cancer in never-smoking ever-users of snus were based on few cases, with considerable risk for type 2 error in analyses for oral and lung cancer. The scarcity of information about covariates in our database needs careful consideration. The restriction to male construction workers allays concerns about confounding by sex, socioeconomic status, and occupational exposures. Furthermore, it is hard to imagine any negative confounding that would have hidden a true association of snus with risk for oral and lung cancer. In the case of pancreatic cancer, we were unable to identify any established or suspected risk factor39 other than smoking that might be linked to snus use, although confounding by dietary factors is a possibility. Another, more speculative, confounding factor could be passive smoking, but such an eff ect seems unlikely in view of the strength of the association and the absence of an increased risk for lung cancer.

At present, our results can be confi dently generalised only to Swedish male construction workers. Although our relative risk estimates—if unbiased and unconfounded—might refl ect a biological relation that can be generalised to other populations, measures that depend on the underlying baseline risk and exposure prevalence rates (eg, risk diff erence, numbers needed to harm, population attributable risk percentage, etc) could differ substantially between population groups. These measures are typically the ones that are most important for public-health consequences.

We conclude that our fi ndings are probably internally valid. Although we have some reservations about statistical power, the oral use of snus does not seem to be linked to the risk for cancer of the oral cavity or lung, in agreement with some but contrary to other previous work on oral cancer. However, the habit seems, with slightly greater certainty, to be associated with an increased risk of pancreatic cancer. The overall consistency of combined available evidence suggests that the association with pancreatic cancer is real, but perhaps weaker than that noted for smoking. Therefore, oral use of snus should be added to the list of tentative risk factors for pancreatic cancer. The Swedish snus investigated in this cohort, despite its low concentrations of tobacco-specific nitrosamines in comparison with many other smokeless tobacco products, might not be an entirely safe product. Because of the special characteristics of the cohort, additional studies in populations with other patterns of use, not the least in women, are desirable—albeit difficult to accomplish, in view of the sample sizes needed—to put the implications for public health in perspective.

---

Lancet 2007; 369: 2015–20

Published Online May 10, 2007. DOI:10.1016/S0140- 6736(07)60678-3

Department of Medical  Epidemiology and Biostatistics
(J Luo MSc, W Ye MD,  K Zendehdel MD, J Adami MD,  Prof H-O Adami MD,  Prof P Boff etta MD,  Prof O Nyrén MD)
 
Department of Medicine, Clinical Epidemiology Unit
(J Adami)

Karolinska Institutet,  Box 281, SE 171 77, Stockholm,  Sweden; Cancer Institute  Research Center, Medical  Sciences/University of Tehran,  Tehran, Iran
(K Zendehdel)

Department of Epidemiology,  Harvard School of Public  Health, Boston, MA, USA
(Prof H-O Adami)

International Agency for  Research on Cancer, Lyon,  France
(P Boff etta)

Correspondence to:
Prof Olof Nyrén
olof.nyren@ki.se

---

Contributors
JL participated in the conception and design of the study, analysis of the data, and drafting the manuscript. WY participated in the conception and design of the study and in the interpretation of results. KZ assisted with data analysis. JA coordinated the data collection. HOA and PB provided scientific suggestions. ON was the lead author in the overall conceptualisation and design of the study, and provided overall supervision for the article. Raw data were reviewed by JL, WY, and ON. All authors took part in reviewing and editing the entire manuscript, and approved the fi nal version of the manuscript.

Conflict of interest statement
We declare that we have no confl ict of interest.

Acknowledgments
This study was partly funded by a grant from the Swedish Cancer Society (4409-B00-01XAC).

---

References
1. Rodu B, Nasic S, Cole P. Tobacco use among Swedish
schoolchildren. Tob Control 2005; 14: 405–08.

2. Stegmayr B, Eliasson M, Rodu B. The decline of smoking in
northern Sweden. Scand J Public Health 2005; 33: 321–24.

3. Hoff mann D, Djordjevic MV, Fan J, Zang E, Glynn T, Connolly GN.
Five leading U.S. commercial brands of moist snuff in 1994:
assessment of carcinogenic N–nitrosamines. J Natl Cancer Inst
1995; 87: 1862–69.

4. Hecht SS. DNA adduct formation from tobacco-specifi c
N-nitrosamines. Mutat Res 1999; 424: 127–42.

5. Hoff mann D, Djordjevic MV. Chemical composition and
carcinogenicity of smokeless tobacco. Adv Dent Res 1997; 11: 322–29.

6. Hecht SS. Biochemistry, biology, and carcinogenicity of
tobacco-specifi c N-nitrosamines. Chem Res Toxicol 1998; 11: 559–603.

7. Rodu B, Jansson C. Smokeless tobacco and oral cancer: a review of
the risks and determinants. Crit Rev Oral Biol Med 2004; 15: 252–63.

8. Lewin F, Norell SE, Johansson H, et al. Smoking tobacco, oral
snuff , and alcohol in the etiology of squamous cell carcinoma of the
head and neck: a population-based case-referent study in Sweden.
Cancer 1998; 82: 1367–75.

9. Bjelke E SL. Chew tobacco and use of snuff : relationship to cancer of
the pancreas and other sites in two prospective studies. Proceedings
of the 13th International Cancer Congress, 1982; 207 (abstr).

10. Boff etta P, Aagnes B, Weiderpass E, Andersen A. Smokeless
tobacco use and risk of cancer of the pancreas and other organs.
Int J Cancer 2005; 114: 992–95.

11. Gupta PC, Ray CS. Smokeless tobacco and health in India and
South Asia. Respirology 2003; 8: 419–31.

12. IARC Working Group. Tobacco habits other than smoking;
betel-quid and areca-nut chewing; and some related nitrosamines.
IARC Monogr Eval Carcinog Risk Chem Hum 1985; 37: 1–268.

13. Cogliano V, Straif K, Baan R, Grosse Y, Secretan B, El Ghissassi F.
Smokeless tobacco and tobacco-related nitrosamines. Lancet Oncol
2004; 5: 708.
14. Rivenson A, Hoff mann D, Prokopczyk B, Amin S, Hecht SS.
Induction of lung and exocrine pancreas tumors in F344 rats by
tobacco-specifi c and Areca-derived N-nitrosamines. Cancer Res 1988;
48: 6912–17.

15. Prokopczyk B, Rivenson A, Hoff mann D. A study of betel quid
carcinogenesis. IX. Comparative carcinogenicity of 3-(methylnitrosa
mino)propionitrile and 4-(methylnitrosamino)-1-(3-pyridyl)-
1-butanone upon local application to mouse skin and rat oral
mucosa. Cancer Lett 1991; 60: 153–57.

16. Alguacil J, Silverman DT. Smokeless and other noncigarette tobacco
use and pancreatic cancer: a case-control study based on direct
interviews. Cancer Epidemiol Biomarkers Prev 2004; 13: 55–58.

17. Zheng W, McLaughlin JK, Gridley G, et al. A cohort study of
smoking, alcohol consumption, and dietary factors for pancreatic
cancer (United States). Cancer Causes Control 1993; 4: 477–82.

18. Muscat JE, Stellman SD, Hoff mann D, Wynder EL. Smoking and
pancreatic cancer in men and women.
Cancer Epidemiol Biomarkers Prev 1997; 6: 15–19.

19. Rodu B, Stegmayr B, Nasic S, Asplund K. Impact of smokeless
tobacco use on smoking in northern Sweden. J Intern Med 2002;
252: 398–404.

20. Ramström L. Reduction of smoking and smoking–related health
hazards by promoting alternative nicotine delivery systems. UN
Focal Point on Tobacco and Health. Geneva: ICAA and EMASH,
1997.

21. Nyren O, Bergstrom R, Nystrom L, et al. Smoking and colorectal
cancer: a 20-year follow-up study of Swedish construction workers.
J Natl Cancer Inst 1996; 88: 1302–07.

22. Mattsson B, Wallgren A. Completeness of the Swedish Cancer
Register. Non-notifi ed cancer cases recorded on death certifi cates in
1978. Acta Radiol Oncol 1984; 23: 305–13.

23. Ekstrom AM, Signorello LB, Hansson LE, Bergstrom R,
Lindgren A, Nyren O. Evaluating gastric cancer misclassifi cation: a
potential explanation for the rise in cardia cancer incidence.
J Natl Cancer Inst 1999; 91: 786–90.

24. Lin D, Wei L, Ying Z. Checking the Cox model with cumulative
sums of martingale-based residuals. Biometrika 1993; 80: 557–72.

25. Boff etta P, Aagnes B, Weiderpass E, Andersen A. Response to
comments by Drs. Rutqvist, Lewin, Nilsson, Ramstrom, Rodu and
Cole further to the publication of the manuscript “smokeless
tobacco use and risk of cancer of the pancreas and other organs”.
Int J Cancer 2006; 118: 1586–87.

26. Williams RR, Horm JW. Association of cancer sites with tobacco
and alcohol consumption and socioeconomic status of patients:
interview study from the Third National Cancer Survey.
J Natl Cancer Inst 1977; 58: 525–47.

27. Farrow DC, Davis S. Risk of pancreatic cancer in relation to medical
history and the use of tobacco, alcohol and coff ee. Int J Cancer 1990;
45: 816–20.

28. Gilljam H, Galanti MR. Role of snus (oral moist snuff ) in smoking
cessation and smoking reduction in Sweden. Addiction 2003; 98:
1183–89.

29. Prokopczyk B, Hoff mann D, Bologna M, et al. Identifi cation of
tobacco-derived compounds in human pancreatic juice.
Chem Res Toxicol 2002; 15: 677–85.

30. Schuller HM. Mechanisms of smoking-related lung and pancreatic
adenocarcinoma development. Nat Rev Cancer 2002; 2: 455–63.

31. Persson PG, Carlsson S, Svanstrom L, Ostenson CG, Efendic S,
Grill V. Cigarette smoking, oral moist snuff use and glucose
intolerance. J Intern Med 2000; 248: 103–10.

32. Gapstur SM, Gann PH, Lowe W, Liu K, Colangelo L, Dyer A.
Abnormal glucose metabolism and pancreatic cancer mortality.
Jama 2000; 283: 2552–58.

33. Rosenquist K, Wennerberg J, Schildt EB, Bladstrom A,
Hansson BG, Andersson G. Use of Swedish moist snuff , smoking
and alcohol consumption in the aetiology of oral and oropharyngeal
squamous cell carcinoma. A population-based case-control study in
southern Sweden. Acta Otolaryngol 2005; 125: 991–98.

34. Schildt EB, Eriksson M, Hardell L, Magnuson A. Oral snuff ,
smoking habits and alcohol consumption in relation to oral cancer
in a Swedish case-control study. Int J Cancer 1998; 77: 341–46.

35. Henley SJ, Thun MJ, Connell C, Calle EE. Two large prospective
studies of mortality among men who use snuff or chewing tobacco
(United States). Cancer Causes Control 2005; 16: 347–58.

36. Accortt NA, Waterbor JW, Beall C, Howard G. Cancer incidence
among a cohort of smokeless tobacco users (United States).
Cancer Causes Control 2005; 16: 1107–15.

37. Schneeweiss S. Sensitivity analysis and external adjustment for
unmeasured confounders in epidemiologic database studies of
therapeutics. Pharmacoepidemiol Drug Saf 2006; 15: 291–303.

38. Furberg H, Lichtenstein P, Pedersen NL, Bulik C, Sullivan PF.
Cigarettes and oral snuff use in Sweden: prevalence and transitions.
Addiction 2006; 101: 1509–15.

39. Ekbom A, Hunter D. Pancreatic cancer. In: Adami HO, Hunter D,
Trichopoulos D, eds. Textbook of cancer epidemiology. New York:
Oxford University Press, 2002: 233–47.