The incidence of cancer of the stomach has been declining in many western countries during the last decades, but nevertheless this disease represents a major health concern globally.1, 2 Despite advances in diagnostic procedures and efforts to improve the treatment, the prognosis in these patients has remained poor.3 In 2002, gastric cancer was the second most common cause of cancer death in the world (about 700,000 deaths).2 Infection with Helicobacter pylori (H. pylori) and dietary factors are the main known causative agents,4 but the etiology of this cancer is not fully understood. The identification of highly prevalent risk factors and estimation of their influence may point to targets for preventive actions. Together with improvements in therapy, preventive measures may be effective in reducing the mortality from this cancer. It is therefore important to establish and estimate the strengths of any relations between tobacco smoking, alcohol drinking and risk of gastric cancer. A summary of previous research indicates that tobacco smoking is a moderate risk factor in various parts of the world,5 while there is little support for an association with alcohol.6, 7 However, there is a sparsity of well-designed prospective epidemiological studies. The principal aim of our study was to provide valid evidence for the role of smoking and alcohol drinking in relation to gastric adenocarcinoma. A second aim was to assess the extent to which preventive measures could reduce the incidence of this deadly cancer. We therefore conducted a large, prospective, population-based cohort study to address these issues.
The relations between tobacco, alcohol and risk of gastric cancer need to be established, and any gain from preventive measures should be estimated. We conducted a population-based, prospective cohort study in Nord-Trondelag county in Norway. During 1984–1986, adult residents were invited to a health survey and they answered questionnaires that assessed exposure to tobacco and alcohol, together with potential confounding factors. The exposure assessment regarding alcohol was limited to a 14-day period. New gastric cancers that occurred during follow-up (1984–2002) were identified by linkage to the Norwegian Cancer Registry. Cox proportion hazards regression models were used to calculate hazard ratios (HRs) with 95% confidence intervals (CI), adjusted for sex, education and body mass index. Follow-up of 1,117,648 person-years at risk among 69,962 cohort members revealed 251 gastric cancers, including 224 noncardia cancers. The risk was almost twice as high in daily smokers (HR = 1.88 [CI 95% = 1.33–2.67]) as in never smokers. Independent dose-response relations were found with earlier age at initiation (p = 0.02), frequency (p = 0.00) and duration of smoking (p = 0.00). Attributable risk (AR) of gastric cancer among current smokers was 8.7/100,000 person-years and the corresponding population AR was 18.4%. No statistically significant associations between various degrees of exposure to alcohol and risk of gastric cancer was revealed, but combined high use of cigarettes (>20/day) and alcohol (>5 occasions/14 days) increased the risk of noncardia gastric cancer nearly 5-fold (HR = 4.90 [95% CI = 1.90–12.62]), compared to nonusers. It is concluded that smoking is a dose-dependent risk factor for gastric cancer. Combined high exposure to smoking and alcohol further increases the risk. Successful preventive measures could considerably reduce the incidence of gastric cancer. © 2006 Wiley-Liss, Inc.
A population-based cohort study was conducted within the frame of a health survey in the county of Nord-Trondelag in Norway between the years 1984 and 2002. All inhabitants of the county of ages at least 20 years were invited to participate in a health survey. This survey, entitled “Helseundersokelsen i Nord Trondelag 1” (HUNT-1), was carried out to provide opportunities for prospective follow-up studies, and much information concerning potential risk factors for gastric cancer was available. At their local health centers, all participants completed extensive written questionnaires covering a wide variety of exposures. In addition to exposure information on tobacco smoking and alcohol drinking, the survey included information on defined disorders and diseases, life-style factors, behavioral habits, physical activity, work and employment information, use of medications and aspects of psychosocial well-being. Furthermore, all participating subjects underwent certain physical examinations, including assessment of body weight and height measured by the study personnel. The specific questions asked in the HUNT-1 are presented elsewhere (http://www.hunt.ntnu.no/index.php?side=english/healthinfo/quest).
Written invitations to participate in the survey were mailed to 85,100 adults living in the county of Nord-Trondelag, and the 75,043 (88.2%) who attended the survey were eligible for subsequent follow-up. To avoid selection bias caused by influence of a yet undetected cancer on the exposure prevalence, we excluded 5,081 attending persons whose follow-up time after participation in the health survey was 3 years or less. Thus, 69,962 persons constituted the final cohort that was followed up in our study.
The identification of gastric cancer occurring during the follow-up of the study cohort was made possible through knowledge of the unique personal identity number that is given to every Norwegian citizen at birth, or to immigrants upon immigration. All persons in our study cohort had such identity number and it was linked with the nationwide and virtually complete Norwegian Cancer Registry. This register contains data of all cancers diagnosed regarding its organ, its histological type and its subsite within the affected organ. In the current study, we included patients ever diagnosed with a gastric cancer of the histological type adenocarcinoma, the quite dominating type of malignancy in the stomach. Virtually all (over 98%) of the gastric cancers are histologically confirmed, and the few cases without adenocarcinoma were not included in our study. Regarding subsite within the stomach, we did study cardia cancer and gastric tumors located distal to the cardia, but the small number of cardia cancers prohibited valid results regarding this particular site. The classification of gastroesophageal junction cases was classified by the physicians and pathologists who reported the case to the Cancer Registry. Tumors without adenocarcinoma (10%) were not included in the study.
Information about incident cancer during follow-up was retrieved through register linkage to the Cancer Registry. For complete follow-up and accurate censoring, we also performed register linkages with the Norwegian Central Person Registry that provided information on vital status and emigration. These linkages were made possible by the personal identity number. The cohort members were followed up from the date 3 years after the health examination (between 1984 and 1986) through December 31, 2002, or to the date of death, date of emigration, or date of diagnosis of primary gastric adenocarcinoma or any other cancer (only first primary gastric cancers were included), whichever event occurred first.
Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated by a Cox proportional hazards regression model, using attained age as the underlying timescale.8 Proportional hazards assumption was tested by Schoens's method9 for all potential risk factors and confounders, and we found that the gender variable did not conform to the assumption of proportionality. Thus, gender-stratified Cox regression models were used. Tobacco smoking status was categorized into 4 groups: never smoker (reference), former smoker, current smoker and missing information. The other main study variable, alcohol drinking, was categorized into 5 groups, depending on the subjects' alcohol intake during the 14 days preceding their entry into the cohort: never drank alcohol (reference); drank alcohol occasionally; drank alcohol on 1–4 occasions; drank on ≥5 occasions and missing information. Adjustments were made for the following covariates: sex, body mass index (BMI) at entry into the cohort (in 4 categories: ≤18.5, corresponding to underweight; 18.6–24.9, corresponding to normal weight; 25.0–29.9, corresponding to overweight and ≥30.0, corresponding to obesity), formal education (in 3 categories: ≤9 years, 10–12 years and ≥13 years), physical activity (current activity during leisure time or at work) and salt intake. Since salt intake and physical activity level did not influence the results, these factors were not included in the final model.
We performed combined and separate analyses for cancer of the two main anatomic subsites of the stomach, i.e., gastric cardia (ICD-10 code C16.0) and distal to the gastric cardia (noncardia gastric cancer, ICD-10 all codes including C16 except C16.0). p values for trends were calculated by introducing the variable with continuous scales in the model. Missing values were excluded when p values for trend were calculated. Never and former smokers were excluded from the analysis regarding duration and intensity when calculating the p-value for trend as suggested by Leffondré et al.10 Interactions between smoking and alcohol drinking were formally tested by likelihood ratio method, comparing models with and without the interaction terms. Attributable risk (AR), incidence rates and age-adjusted to the total person-years experienced by the entire cohort, for different groups, were calculated according to the formula: AR = standardized incidence ratio (SIR) among the exposed − SIR among the unexposed, i.e. the difference in SIR between exposed group and unexposed group.11 For all of the aforementioned statistical analyses, SAS 9.1 software (SAS Institute, Cary, NC) was used.
The 69,962 study cohort members were followed up for an average of 16 years, contributing a total of 1,117,648 person-years at risk. During this follow-up, we identified 251 new cases of gastric cancer, of which 224 (89%) were noncardia gastric cancer and 27 were gastric cardia cancer. Some characteristics of the study participants are presented in Table I. There was a male predominance among gastric cancer patients that was stronger among those with cardia adenocarcinoma (74%) than among those with noncardia adenocarcinoma (55%). Almost 80% of the tumors occurred in persons older than 55 years. Compared to the cohort members at large, patients who developed gastric cancer had a shorter period of formal education (Table I).
|Person-years at risk||Cohort members||Gastric cancer (cardia included)||Noncardia gastric cancer|
|Age at entry (years)|
|Attained age (years)|
Tobacco smoking and risk of gastric cancer
The risk of noncardia gastric cancer was increased almost 2-fold among daily smokers (adjusted HR = 1.88 [95% CI = 1.33–2.67]), compared to nonsmokers. On the basis of the point estimates, the associations with smoking found for noncardia and cardia cancers did not differ substantially (data not shown). Cigarette smoking and pipe smoking were associated with a similarly increased risk of gastric cancer (Table II), compared to nonsmokers. With a combination of two or more types of tobacco smoking, further increases in HR were noted (Table II). Earlier age at initiation of daily smoking was associated with an increased risk of noncardia gastric cancer, independent of adjustment for duration of smoking, suggesting a dose-response relation to earlier onset of smoking (p for trend = 0.02). A combination of early smoking start (age <20 years) and long duration of smoking (>30 years) rendered an adjusted HR of 1.91 (95% CI = 1.22–2.99). Increased frequency of smoked cigarettes per day was similarly linked to risk of gastric cancer in an apparently dose-response manner (p for trend = 0.00). Among persons who smoked more than 20 cigarettes daily, the adjusted HR of gastric cancer was 1.99 (95% CI = 1.16–3.42), compared to nonsmokers. There was also an increased risk of gastric cancer with increased duration of smoking (p = 0.00). Analyses of categorical and continuous models revealed that both duration and intensity of smoking contributed to the increased risk to a seemingly similar extent (data not shown). After adjustment for smoking duration, no trend remained with smoking cessation (Table II). The unadjusted risk estimates regarding tobacco smoking were only slightly changed after adjustment for the potentially confounding variables listed in the methods section, including alcohol drinking (data not shown). Thus, no strong confounding effects of the listed covariates were identified. In the nonsmoking cohort, the SIR was 22.9 (95% CI = 17.9–27.8)/100,000 person-years while the corresponding SIR was 31.6 (95% CI = 24.0–39.1)/100,000 person-years among current smokers, and therefore the AR of current smoking was 8.7/100,000 person-years. The corresponding population AR was 18.4% (95% CI = 12.1–24.8%). The results were generally similar among cases of gastric cancer as compared to cases of noncardia gastric cancer (Table II).
|No. of subjects||Gastric cancer (including cardia)||Noncardia gastric cancer|
|n||HR (95% CI)1||n||HR (95% CI)1|
|Current smoking daily||19,759||77||1.88 (1.33–2.67)||70||1.83 (1.27–2.64)|
|Former smoking||12,148||48||1.30 (0.88–1.91)||1.14 (0.75–1.73)|
|Former and current smoking||31,907||125||1.56 (1.14–2.14)||1.46 (1.05–2.04)|
|Type of tobacco|
|Cigarette only||17,676||59||1.76 (1.22–2.54)||54||1.73 (1.18–2.54)|
|Pipe only||1,191||13||2.02 (1.08–3.76)||11||1.80 (0.92–3.53)|
|Others3||12,148||48||1.31 (0.89–1.92)||39||1.14 (0.75–1.74)|
|Missing||12,363||39||1.28 (0.75–2.18)||33||1.12 (0.63–1.98)|
|Age at start of daily smoking4|
|<15 years||1,687||6||2.60 (0.89–7.58)||6||2.99 (1.00–8.89)|
|15–19 years||17,986||61||2.06 (0.98–4.35)||52||2.04 (0.94–4.44)|
|20–24 years||7,076||35||2.04 (0.95–4.38)||29||1.92 (0.86–4.27)|
|25 years or more||3,970||22||1.63 (0.76–3.52)||21||1.71 (0.77–3.76)|
|p value for trend||0.02||0.04|
|Missing||13,580||40||1.10 (0.66–1.83)||34||0.98 (0.57–1.69)|
|Age at start and duration of smoking|
|Age <20 years and duration >30 years||3,676||35||1.91 (1.22–2.99)||30||1.72 (1.06–2.78)|
|Age >20 years and duration >30 years||1,369||10||1.34 (0.69–2.63)||9||1.25 (0.62–2.52)|
|Age <20 years and duration <30 years||14,800||25||1.50 (0.90–2.49)||22||1.50 (0.87–2.57)|
|Age >20 years and duration <30 years||4,093||16||1.36 (0.79–2.35)||15||1.36 (0.77–2.38)|
|Missing||20,361||78||1.46 (1.02–2.09)||66||1.30 (0.89–1.90)|
|Frequency of smoking|
|<10 cigarettes/day||8,667||33||1.52 (0.99–2.31)||28||1.36 (0.87–2.14)|
|10–19 cigarettes/day||14,258||50||1.73 (1.17–2.55)||42||1.55 (1.03–2.35)|
|20 or more cigarettes/day||4,640||19||1.99 (1.16–3.42)||17||1.93 (1.09–3.42)|
|p value for trend (excluding missing)||0.00||0.00||0.01|
|Missing||16,705||62||1.24 (0.84–1.84)||55||1.19 (0.79–1.80)|
|Duration of smoking|
|<10 years||5,939||6||1.30 (0.55–3.06)||6||1.40 (0.59–3.33)|
|10–19 years||9,632||20||1.54 (0.92–2.60)||15||1.27 (0.71–2.29)|
|20–29 years||6,401||27||1.61 (1.01–2.57)||26||1.70 (1.05–2.73)|
|30 years or more||7,944||65||1.66 (1.14–2.41)||55||1.47 (0.99–2.19)|
|p value for trend||0.00||0.04|
|Missing||14,354||46||1.26 (0.79–2.00)||40||1.16 (0.71–1.89)|
|Current smoking cigarettes||18,073||62||1.38 (0.78–2.47)||57||1.39 (0.75–2.56)|
|Exsmoking cigarette||12,529||50||1.00 (0.56–1.78)||40||0.84 (0.45–1.56)|
|<1 year ago||1,600||2||0.54 (0.12–2.34)||2||0.57 (0.13–2.52)|
|1–5 years ago||2,568||8||1.07 (0.45–2.52)||7||1.01 (0.41–2.52)|
|>5 years ago||8,196||39||1.03 (0.57–1.89)||30||0.82 (0.43–1.58)|
|Other||165||1||1.07 (0.14–7.99)||1||1.12 (0.15–8.42)|
|p value for trend||0.53||0.38|
|Missing||13,668||52||1.34 (0.81–2.21)||45||1.24 (0.73–2.11)|
Alcohol drinking and risk of gastric cancer
The relative risk estimates regarding alcohol drinking and risk of gastric cancer were attenuated after adjustment for tobacco smoking status (Table III). Although the point estimates were generally increased (HR = 1.13–1.56), no statistically significant associations remained after multivariable adjustment between various degrees of exposure to alcohol and risk of gastric cancer or noncardia gastric cancer (Table III).
|No. of subjects||Gastric cancer (including cardia)||Noncardia gastric cancer|
|n||HR (95% CI)1||n||HR (95% CI)1|
|Frequency of alcohol drinking during the last 14 days|
|Never drinking alcohol2||6,325||27||Reference||26||Reference|
|Drinking alcohol occasionaly||25,878||109||1.24 (0.80–1.91)||95||1.16 (0.74–1.82)|
|1–4 times||21,667||58||1.30 (0.78–2.16)||51||1.29 (0.76–2.18)|
|≥5 times||3,406||17||1.49 (0.78–2.83)||17||1.66 (0.87–3.20)|
|p value for trend||0.19||0.09|
|Missing data||12,686||40||1.35 (0.73–2.50)||35||1.32 (0.70–2.51)|
|Feeling of intoxication when drinking|
|No||20,691||70||1.10 (0.67–1.77)||65||1.12 (0.69–1.82)|
|Yes||10,172||26||1.47 (0.81–2.69)||23||1.50 (0.80–2.83)|
|Missing||32,774||128||1.36 (0.88–2.10)||110||1.26 (0.80–1.97)|
|Drinking excessively, or at least a bit to much|
|No||35,668||113||1.15 (0.73–1.80)||101||1.13 (0.71–1.79)|
|Possibly or maybe||6,419||31||1.56 (0.88–2.76)||26||1.49 (0.82–2.72)|
|Yes||5,086||15||1.13 (0.57–2.24)||15||1.30 (0.65–2.60)|
|Missing||16,464||65||1.46 (0.89–2.37)||56||1.33 (0.80–2.21)|
Combination of tobacco and alcohol and risk of gastric cancer
Combinations of various levels of tobacco smoking and alcohol drinking in relation to risk of gastric cancer are presented in Table IV. Smoking of more than 20 cigarettes a day combined with alcohol consumption more than 5 times/14 days was associated with a nearly 5-fold increased risk of noncardia gastric cancer (HR = 4.90 [95% CI = 1.90–12.62]), compared to nonusers. The interaction between tobacco smoking and alcohol drinking were not statistically significant regarding total gastric cancer (p = 0.32), noncardia gastric cancer (p = 0.44) or cardia cancer (p = 0.89).
|Exposure status||No. of subjects||Gastric cancer (including cardia)||Noncardia gastric cancer|
|Smoking1||Alcohol2||n||HR (95% CI)3||n||HR (95% CI)3|
|<20 cigarettes/day||Never||793||3||1.34 (0.40–4.51)||3||1.40 (0.41–4.75)|
|Never smoking||<4 times||19,203||56||1.08 (0.66–1.78)||52||1.06 (0.64–1.78)|
|Never smoking||≥5||741||3||1.57 (0.47–5.28)||3||1.69 (0.50–5.73)|
|<20 cigarettes/day||<4 times||20,194||69||1.79 (1.07–2.99)||56||1.57 (0.91–2.68)|
|≥20||<4 times||3,789||12||1.78 (0.85–3.73)||10||1.66 (0.75–3.66)|
|<20 cigarettes/day||≥ 5||1,526||8||2.15 (0.93–4.94)||8||2.33 (1.00–5.40)|
|≥20||≥ 5||667||6||4.38 (1.72–11.17)||6||4.90 (1.90–12.62)|
|Missing data||18,096||72||1.45 (0.87–2.42)||65||1.42 (0.84–2.40)|
This study provides evidence in support of a moderately strong association between tobacco smoking and risk of gastric cancer. Earlier initiation, a higher frequency and a longer duration of smoking resulted in higher estimates of relative risk in a seemingly dose-dependent manner. No clear positive association between alcohol drinking and noncardia gastric cancer was found. However, combined exposure to high levels of tobacco and alcohol further increased the risk estimates.
Strengths of the present study include the prospective exposure assessment, the population-based design, the long and virtually complete follow-up and the comprehensive information on smoking and several potentially confounding variables. Weaknesses include limited statistical power to study cardia cancer separately, sparse details regarding exposure to alcohol and lack of data concerning some potential confounders. The information about alcohol intake was collected only for the last 14 days, which limits the overall assessment of this exposure. The potentially most relevant confounding factor is infection with H. pylori, a strong and well-known risk factor for gastric cancer.12 Previous analyses in the HUNT study have shown that there is a small overrepresentation of H. pylori infection among male smokers (unpublished data); however, it is unlikely that this would introduce major confounding of the results in the current study. No increased alcohol consumption among the H. pylori-positive persons has been identified in this cohort (unpublished data). Moreover, no detailed information on nutritional factors was available, including intake of antioxidative vitamins, which may have a protective effect against gastric cancer.13 But again, no association between such dietary factors and the use of tobacco or alcohol has been found (unpublished data), and thus dietary factors should not have confounded our results.
Although tobacco and alcohol use have previously been recognized as moderate risk factors,5, 14, 15, 16, 17, 18, 19 few prospective and population-based cohort studies have been published. We found a moderately strong association between tobacco smoking and gastric cancer both with a noncardia and cardia location. The strength of the association was in line with observations in a summary of previous studies.20 Several dose-response associations were suggested in the current study and these results add evidence of a causal association. Moreover, the risk estimates of noncardia gastric cancer decreased with time after cessation of smoking, a finding that should encourage preventive actions. No such effect was found for cardia cancer; however, a result compatible with previous research on this specific tumor site.7, 21 The finding of an increased risk of gastric cancer with earlier onset of smoking has not, to our knowledge, been reported previously, but this observation further supports a true association. The consistency with previous literature in combination with a lack of obvious sources of error that could have explained the results of our study suggest that the association between tobacco smoking and risk of gastric cancer is a true one.
Alcohol consumption has been considered to be an unlikely cause of gastric cancer.6, 7, 21, 22, 23, 24 But previous results are partly contradictory and alcohol use may possibly increase the risk of cardia cancer.6, 7, 25 Our lack of evidence of an association between alcohol drinking and risk of gastric cardia cancer is in line with findings in recent large epidemiologic studies.7, 21 The association between alcohol drinking and noncardia gastric cancer became nonsignificant after adjustment for smoking, which indicate confounding by smoking. Moreover, no consistent associations between the amount of alcohol consumed and risk of gastric cancer were revealed, and thus no clear dose-response association was identified. However, our limited details regarding the exposure to alcohol in combination with a considerable risk of misclassification of self-reported alcohol use may have diluted any true associations.
If we consider the associations of tobacco smoking with risk of gastric cancer to be causal, the calculations of ARs in our population-based study would seem justifiable. Although such calculations must be interpreted with caution, our results suggest that the rate of gastric cancer that is caused by smoking among current smokers was about 28% (AR of 8.7 divided by 31.6). The strengths of the studied associations were moderate, but the high prevalence of this exposure in the population indicates that the incidence could be drastically reduced through successful prevention of this exposure, whereby the number of deaths from gastric cancer could be considerably decreased. An obvious challenge that deserves more attention is the question of how preventive measures can be truly effective in reducing the global exposure to tobacco smoke and the high exposure to alcohol.
In conclusion, this large prospective and population-based cohort study suggests that smoking is a dose-dependent, moderately strong risk factor for gastric cancer, while alcohol drinking might not be associated with an increased risk of this cancer. The combination of a high level of tobacco smoking and alcohol drinking is linked with a substantially increased relative risk of this cancer. Theoretically, successful preventive measures could reduce the number of gastric cancer deaths considerably.
We are grateful to the Norwegian Institute of Public Health and the HUNT Research Centre, Verdal, Norway, for performing the HUNT survey and the medical faculty of the Norwegian University of Science and Technology, Trondheim, Norway, for allowing access to the database.