|Home | About | Journals | Submit | Contact Us | Français|
The authors investigated the relation between alcohol consumption and lung cancer risk in the Environment and Genetics in Lung Cancer Etiology (EAGLE) Study, a population-based case-control study. Between 2002 and 2005, 2,100 patients with primary lung cancer were recruited from 13 hospitals within the Lombardy region of Italy and were frequency-matched on sex, area of residence, and age to 2,120 randomly selected controls. Alcohol consumption during adulthood was assessed in 1,855 cases and 2,065 controls. Data on lifetime tobacco smoking, diet, education, and anthropometric measures were collected. Adjusted odds ratios and 95% confidence intervals for categories of mean daily ethanol intake were calculated using unconditional logistic regression. Overall, both nondrinkers (odds ratio = 1.42, 95% confidence interval: 1.03, 2.01) and very heavy drinkers (≥60 g/day; odds ratio = 1.44, 95% confidence interval: 1.01, 2.07) were at significantly greater risk than very light drinkers (0.1–4.9 g/day). The alcohol effect was modified by smoking behavior, with no excess risk being observed in never smokers. In summary, heavy alcohol consumption was a risk factor for lung cancer among smokers in this study. Although residual confounding by tobacco smoking cannot be ruled out, this finding may reflect interplay between alcohol and smoking, emphasizing the need for preventive measures.
A positive association between lung cancer and alcohol consumption has been suggested in several studies (1–3). In a pooled reanalysis of data collected in 7 prospective studies with a total of 399,767 participants and 3,137 lung cancer cases, Freudenheim et al. (4) found a slightly greater risk for alcohol consumption of 30 g/day than for no consumption. However, the question of alcohol's effect on lung cancer risk is controversial, and the relation is often attributed to residual confounding by cigarette smoking (5). The protective effect of low alcohol consumption suggested for cardiovascular disease risk (6) and all-cause mortality (7) was also observed for lung cancer risk (8).
Differences in lung cancer risk by type of alcoholic beverage consumed have been reported (3–4, 9). Beer and liquor intake have been found to be associated with higher risk, but results were not consistent (5). In a meta-analysis based on 14 observational studies, Chao (10) found that modest wine consumption may be inversely associated with the risk of lung cancer.
Since alcohol consumption and smoking behavior are correlated and are both modifiable behaviors, the possible detrimental role of alcohol and its joint effect with smoking in lung cancer risk are of particular importance for prevention. To evaluate these issues, we analyzed the association between alcohol consumption and lung cancer in the Environment and Genetics in Lung Cancer Etiology (EAGLE) Study, a large population-based case-control study conducted in Italy.
The EAGLE Study was a population-based case-control study of lung cancer conducted in the Lombardy region of Italy. Details of the study design and subjects’ characteristics are reported elsewhere (11).
Briefly, between April 2002 and February 2005, incident primary lung cancer patients admitted to 13 hospitals within the Lombardy region were identified. The 13 selected hospitals treat approximately 80% of cases from the chosen catchment area, which consisted of 216 municipalities. Inclusion criteria for both cases and controls were: Italian nationality, age 35–79 years, official residence within the catchment area, no severe disease that could impede participation, and ability to understand study procedures and give informed consent. We enrolled 2,100 cases; 95% of cancers were confirmed pathologically or cytologically, and detailed histologic classification was recorded. The remaining 5% of cancers were confirmed using clinical history and imaging.
Population controls were randomly sampled from the Lombardy Regional Health Service database and frequency-matched to cases according to sex, 5-year age group, and area of residence. Family physicians were asked to contact the selected controls to verify the absence of a history of lung cancer or any advanced diseases that would impede participation. At study completion, 2,120 controls were recruited, with a participation rate of 72.4%.
The institutional review board of the US National Cancer Institute and all of the involved institutions approved the study protocol. Each subject signed an informed consent form prior to participation.
Information was collected using both a computer-assisted personal interview and a self-administered questionnaire. Extensive epidemiologic data were gathered to assess the main potential risk factors for lung cancer. The complete questionnaires used in the EAGLE Study are available on the study Web site (http://eagle.cancer.gov/questionnaires/capi-eng-april9-2002-def.pdf).
Briefly, tobacco smoking was evaluated by means of a computer-assisted personal interview assessing: 1) smoking status, categorized as never (smoking <100 cigarettes in a lifetime), former (stopped smoking ≥6 months before interview), or current (still smoking or stopped <6 months before interview) smoking; 2) type of tobacco used (cigarettes, cigars, cigarillos, or a pipe); 3) smoking duration (adding all time periods characterized by tobacco consumption; quitting periods were considered if subjects had stopped smoking for ≥6 months); and 4) smoking intensity (usual tobacco consumption in each time period). Mean lifetime tobacco consumption was quantified as the number of packs of cigarettes (or equivalent) smoked per day, averaging the period-specific consumption levels weighted by the number of years of exposure in the period. Lifetime tobacco smoking was also quantified in pack-years. Information on passive smoke exposure both at home (childhood and adulthood) and at work was also collected.
Diet over the previous year was assessed using a short, self-administered 58-item food frequency questionnaire designed to address specific dietary aspects of interest within the Italian population (vegetables and fruits, fresh red and processed meat, pasta and pizza).
Body mass index was based on the participant's report of height and weight 1 year before enrollment and was calculated as weight divided by the square of height (kg/m2).
Past alcohol consumption for 4 types of alcoholic beverages (wine, beer, liquor, and aperitifs) during 3 periods (ages 18–25 years, 26–45 years, and ≥46 years) was investigated retrospectively, using a self-administered questionnaire. Respondents were offered 11 possible ordinal reporting categories, from “never” to “6 or more times per day.”
We excluded 300 participants (245 cases and 55 controls) who did not complete the alcohol section of the questionnaire, with a resulting study population of 1,855 cases and 2,065 controls. Nonresponders to the alcohol section were less educated than responders among both cases (16% of nonresponders had a high school diploma or university degree vs. 26% of responders) and controls (25% vs. 40%).
Mean daily ethanol intake (in grams) for each period was calculated assuming an ethanol content of 15.7 g for a 150-ml (5.1-ounce) glass of wine, 17.6 g for a 330-ml (11.2-ounce) can or bottle of beer, 13.1 g for a standard portion (75 ml, 2.5 ounce) of aperitif, and 12.9 g for a standard portion (40 ml, 1.4 ounce) of liquor (12). Finally, for both alcohol in general and for each specific beverage, ethanol intake during adulthood (g/day) was computed as the average of the period-specific intakes weighted by the number of years of exposure in the period. Categories of ethanol intake were based on the amount of ethanol contained in 1 glass of wine (~15 g), the alcoholic beverage mainly consumed in Italy. For consumption of less than 15 g/day, we further split the category into 0.1–4.9 g/day (equivalent to approximately 2 glasses of wine per week) and 5–14.9 g/day (equivalent to approximately 1 glass daily), in order to distinguish very light drinking from the light drinking pattern.
Mean values and standard deviations were computed for continuous study variables. Median values and interquartile ranges are reported for highly skewed distributions. Odds ratios and 95% confidence intervals were calculated from unconditional multivariable logistic regression.
Alcohol consumption was modeled as a categorical variable (nondrinker or ethanol intake of 0.1–4.9, 5–14.9, 15–29.9, 30–59.9, or ≥60 g/day). For simplicity, we refer to the alcohol consumption categories as very light, light, moderate, heavy, and very heavy drinking, respectively. In order to obtain a sufficiently large number of cases and controls in the reference category, we used subjects with very light alcohol consumption (0.1–4.9 g/day) as the reference group.
Adjustment for smoking was performed using mean number of cigarettes smoked per day, duration of smoking, and time since quitting. Other tobacco use and passive smoke exposure were also considered. Additionally, body mass index, consumption of fruit and vegetables, consumption of fresh red and processed meat, and education were considered in the regression models.
In order to test for a log-linear relation between alcohol consumption and lung cancer risk, we considered ethanol intake as an ordinal variable, with the 6 categories previously defined (from 0 for nondrinkers to 5 for ethanol intake of ≥60 g/day). The same analysis for trend was performed after exclusion of nondrinkers. We also used the median consumption levels, with similar results (data not shown). Moreover, with the aim of further exploring the dose-response relation between alcohol consumption and lung cancer risk, we fitted a logistic regression model with restricted cubic splines (13) for mean ethanol intake (in g/day) used as a continuous covariate.
Stratified analyses by smoking status were performed. Heterogeneity of the effect of alcohol consumption on lung cancer risk among strata was tested by means of the likelihood ratio test.
Subanalyses by histologic type (adenocarcinoma, squamous cell carcinoma, and small cell carcinoma) were also performed. Odds ratios and 95% confidence intervals were estimated using unconditional polytomous logistic regression, comparing cases with a specific histologic type with all controls. Homogeneity among histology-specific lung cancer risks was assessed using a Wald test with 2 df (14).
In order to analyze the independent effect of the different types of alcoholic beverages, we included wine, beer, liquor, and aperitifs in the logistic model as separate covariates. We evaluated the extent of multicollinearity by calculating the variance inflation factor in a linear regression model considering case-control status as the response and the types of beverages as continuous predictors, as suggested by Allison (15). All variance inflation factor scores were less than 1.5, indicating that multicollinearity was not an issue. Homogeneity among beverage-specific lung cancer risks was assessed using a Wald test with 3 df (14).
All analyses were performed with SAS software, version 9.1 (SAS Institute Inc., Cary, North Carolina). All P values were 2-sided.
As expected from the study design, cases and controls were similarly distributed by age and sex (Table 1). Cases tended to be less educated, were more likely to be current smokers, and had a higher smoking intensity and duration. Moreover, cases had greater weekly consumption of fresh red and processed meat and lower weekly consumption of fruits and vegetables. Among cases, 8% were nondrinkers during adulthood and 12%, 13%, 26%, 30%, and 11% had very low, low, moderate, heavy, and very heavy mean daily ethanol intakes, respectively. Controls contained the same proportion of nondrinkers (7%) as the cases, and 16%, 19%, 25%, 28%, and 5% percent of controls had very low, low, moderate, heavy, and very heavy mean daily ethanol intakes, respectively.
As Table 1 shows, the prevalence of never smoking decreased as alcohol consumption increased in both cases and controls. However, smoking intensities among ever smokers were similar in all categories of alcohol consumption, although intensity was somewhat higher for the highest category of ethanol intake. Among lung cancer cases, nondrinkers were less likely to have a high school diploma or university degree.
Risks of lung cancer associated with mean daily ethanol intake are shown in Table 2. In the multivariable model that included adjustment only for cigarette smoking, the odds ratio for very heavy ethanol intake compared with very light ethanol intake was 1.53 (95% confidence interval (CI): 1.09, 2.15). The odds ratio for nondrinkers with the same reference category was 1.54 (95% CI: 1.08, 2.18). The addition of the other potential confounders (passive smoke exposure, other tobacco use, educational level, body mass index, fresh red and processed meat consumption, fruit and vegetable consumption) did not substantially change the estimates adjusted for cigarette smoking. Given that aperitifs are a mixed category that includes wine, fortified wine, and liquor, we also performed analyses after excluding the contribution of aperitifs to total ethanol intake and obtained very similar results (data not shown).
We did not observe a statistically significant log-linear component in the dose-response relation between alcohol and lung cancer risk (in the model adjusted for all potential confounders, P for trend = 0.33). After we excluded nondrinkers, there was a statistically significant trend toward higher risk with increasing alcohol consumption (in the model adjusted for all potential confounders, P for trend = 0.008).
Figure 1 shows the fitted spline regression curve and its 95% confidence interval, indicating that subjects with very low ethanol intake were at the lowest risk of lung cancer, while nondrinkers and subjects with moderate, heavy, and very heavy ethanol intakes had similar risk levels.
We observed a statistically significant interaction (likelihood ratio test for interaction: P = 0.03) between smoking and ethanol intake (Table 3). A statistically significantly higher risk for nondrinkers compared with subjects with very light ethanol intake was observed in ever smokers (odds ratio = 1.55, 95% CI: 1.03, 2.34) but not in never smokers. In never smokers, alcohol consumption was not significantly associated with lung cancer risk at any dose (after exclusion of nondrinkers, P for trend = 0.46), while in ever smokers a significant dose-response relation was observed (after exclusion of nondrinkers, P for trend = 0.01). The trends observed in ever smokers after exclusion of nondrinkers were similar among different categories of smoking exposure (likelihood ratio test for interaction: P = 0.31) (Figure 2).
In the analysis by histologic type (Table 4), nondrinkers had a statistically significantly higher risk of squamous cell carcinoma than subjects with very low ethanol intake (odds ratio = 2.35, 95% CI: 1.24, 4.49). The risk of squamous cell carcinoma increased significantly as alcohol consumption increased (after exclusion of nondrinkers, P for trend = 0.002), while no clear patterns were observed for either adenocarcinoma or small cell carcinoma. Despite this, no heterogeneity among histology-specific lung cancer risks was observed, except for the ethanol intake category of 30–59.9 g/day (odds ratios were 0.91, 1.88, and 1.40 for adenocarcinoma, squamous cell carcinoma, and small cell carcinoma, respectively; Wald test of heterogeneity: P = 0.01).
The risks associated with specific beverages are shown in Table 5. Because of the small numbers of heavy drinkers of a single beverage, the upper category was collapsed to ≥30 g/day for wine and ≥15 g/day for beer, liquor, and aperitifs. Subjects in the highest category of liquor consumption had an odds ratio of 1.88 (95% CI: 1.00, 3.52) compared with very light drinkers of liquor. The corresponding odds ratios for consumption of aperitifs, beer, and wine were 1.57 (95% CI: 1.00, 2.45), 0.80 (95% CI: 0.51, 1.25), and 1.29 (95% CI: 0.99, 1.66), respectively. These risks were not significantly heterogeneous (Wald test of heterogeneity: P = 0.11).
We also performed analyses evaluating the effect of alcohol in the age groups 18–25, 26–45, and ≥46 years. The effects of alcohol were not materially different by age at consumption (data not shown).
In this study investigating the association of alcohol consumption and lung cancer risk, we observed that both nondrinkers and very heavy drinkers were at significantly greater risk when compared with very light drinkers, suggesting a nonlinear, U-shaped, dose-response relation. Given the association between alcohol drinking and smoking, residual confounding cannot be completely ruled out. It has been suggested that residual confounding from smoking is likely to be stronger in the highest alcohol consumption categories, because of an excess of extremely heavy smokers among these categories of cases (16). This issue is particularly relevant in our study, since, among drinkers, we found a modest excess risk (odds ratio = 1.44, 95% CI: 1.01, 2.07) only for very heavy consumption, in comparison with very light drinking. Moreover, after stratification by smoking status, we observed a significant positive trend in risk with increasing ethanol intake in ever smokers but not in never smokers. Relative risks of this magnitude may be induced purely by residual confounding if the accuracy of the reported smoking data is uncertain (17). However, we think that the extensive amount of data about smoking behavior that was collected in our study (i.e., number of packs of cigarettes smoked per day, duration of smoking, time since quitting smoking, other tobacco use, passive smoke exposure) permitted rigorous control for smoking; note that the low risk estimate in never smokers was imprecise because of the small number of persons in this category.
Alcohol may enhance the carcinogenic effects of cigarette smoke on tissues by inducing the activity of cytochrome P-450 enzymes, which in turn can activate procarcinogens present in alcoholic beverages (18). If a similar mechanism was at work in our study, alcohol was acting in a fairly homogenous way across different levels of smoking exposure.
The protective effect of low and moderate ethanol intake on lung cancer risk has been previously reported (3, 8, 19), especially when recent alcohol consumption has been considered. Rohrmann et al. (8) hypothesized that subjects with lung cancer symptoms may have stopped drinking alcoholic beverages, thus generating an elevated risk of lung cancer among current nondrinkers as compared with low-level drinkers. In our study, this bias is not likely to have played a role, since we considered average daily ethanol intake during adulthood. As Rohrmann et al. pointed out (8), antiinflammatory, antioxidative, and antimutagenic effects of moderate ethanol intake are thought to contribute to the protective effect, although it is not clear why this would be specific for lung cancer, since a protective effect has not been consistently observed for other neoplasms (20).
Red wine, the prevalent alcoholic beverage consumed by the Italian population, is a major source of flavonoids. The reported beneficial effect of flavonoids on lung cancer risk (21) may explain, at least in part, the reduction of risk for light alcohol consumption observed in our study. We could not test this hypothesis, since we had no information on the type of wine (red vs. white) consumed by this population. On the other hand, it is likely that light drinking is just part of a generally healthy lifestyle. In our study, light drinkers smoked less and were more educated than nondrinkers and heavy drinkers. Moreover, we observed an excess of risk for nondrinkers as compared with very light drinkers among less educated subjects but not among more educated subjects (data not shown). We think that although we controlled for the same factors in both analyses, residual confounding by unmeasured lifestyle factors still played a role among the less-educated subjects.
It could be argued that subjects reporting lifelong abstinence from alcohol may actually be former drinkers who have quit drinking. Using data from a prospective survey of US households, Rehm et al. (22) showed that more than half of the respondents who reported lifetime alcohol abstinence in the most recent survey had reported drinking in previous surveys. Moreover, Ferraroni et al. (23) reported that in a sample of volunteers from northern Italy, approximately 30% of subjects declaring that they never drank according to interviewer-administered food frequency questionnaires were actually found to be current drinkers when the interview information was compared with average intake derived from 7-day dietary records, used as a reference method. The resulting misclassification could explain alcohol's apparently protective effect among very light drinkers in our study.
Analyses of alcohol consumption and histologic type of lung cancer have provided inconsistent results in previous studies. In the New York State Cohort Study (24), an increased risk of squamous cell carcinoma in heavy drinkers was suggested, while in a pooled analysis of 7 prospective studies, including the previous one, excess risks in heavy drinkers were observed for adenocarcinoma and small cell carcinoma (4). In our study, we observed a significantly increased risk of squamous cell carcinoma in heavy drinkers versus light drinkers as compared with a nonsignificantly higher risk observed for adenocarcinoma and small cell carcinoma. In spite of this, we did not find significant heterogeneity in the alcohol effect across histologic types. It has been hypothesized that residual confounding by cigarette smoking may explain the different risks by histologic type (24), because of the stronger association of smoking with squamous cell carcinoma than with adenocarcinoma (25). Squamous cell carcinoma may be more susceptible to environmental factors than adenocarcinoma (24, 26).
Studies evaluating the effect of alcohol on lung cancer by beverage type have more frequently observed relations with beer and liquor than with wine, as reported in Chao's meta-analysis (10). Daily and regular use of wine, mainly during meals, is the typical pattern of alcohol consumption in Mediterranean and Southern European populations, whereas consumption of beer and spirits, mainly on weekends, is typical of Northern European and North American populations. In our study, wine accounted on average for 75% of total consumption, beer for 10%, and liquor or aperitifs for 15%. Because of the higher prevalence of wine drinking in our population, it was difficult to disentangle the effects of different types of beverages. Nevertheless, we observed a significant positive trend for wine consumption, while no clear pattern was evident for consumption of beer, aperitifs, or liquor.
Our study had some limitations. First of all, there is a possibility of recall bias due to the retrospective study design, especially regarding collection of data on early exposures. A second limitation is the absence of more detailed information on alcoholic beverages (e.g., consumption of red wine vs. white wine). Moreover, since the food frequency questionnaire used in the EAGLE Study was designed to obtain information on specific dietary aspects of interest within the Italian population, it was relatively limited in scope, and questions on portion size were not asked. Hence, we were unable to adjust for total energy intake in our regression models. Finally, the exclusion of subjects not responding to the alcohol consumption section of the questionnaire could have biased the results, since we observed that nonresponders were mainly cases (82%) and were less educated than responders.
Our study also had several strengths. To our knowledge, it is the largest population-based case-control study to date to have investigated the relation between alcohol consumption and lung cancer risk. Participation rates were high, and detailed information on smoking history and many other risk factors was collected. Cases were rapidly ascertained, eliminating the need to use surrogate participants. The large sample size permitted investigation by histologic subtype with adequate statistical power. Moreover, since alcohol consumption is very common in the Italian population, we had adequate variability to assess its association with lung cancer risk.
In conclusion, in a large population-based case-control study of lung cancer conducted in Italy, heavy alcohol consumption appeared to be a risk factor among smokers. Although residual confounding by tobacco smoking can never be completely ruled out, this finding may reflect a joint effect of alcohol and tobacco, emphasizing the need for improved strategies to help people reduce their alcohol consumption and quit smoking.
Author affiliations: Department of Statistics, University of Milan-Bicocca, Milan, Italy (Vincenzo Bagnardi); Division of Epidemiology and Biostatistics, European Institute of Oncology, Milan, Italy (Vincenzo Bagnardi); Istituto di Statistica Medica e Biometria “G. A. Maccacaro,” Università degli Studi di Milano, Milan, Italy (Giorgia Randi, Adriano Decarli); S. C. Statistica Medica, Biometria e Bioinformatica, Fondazione IRCSS Istituto Nazionale Tumori di Milano, Milan, Italy (Adriano Decarli); Istituto di Ricerche Farmacologiche “Mario Negri,” Milan, Italy (Giorgia Randi); Biostatistics Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, Maryland (Jay Lubin, Sholom Wacholder); EPOCA, Epidemiology Research Center, Università degli Studi di Milano, Milan, Italy (Dario Consonni, Pier Alberto Bertazzi); Fondazione IRCCS Ospedale Maggiore Policlinico, Mangiagalli e Regina Elena, Milan, Italy (Dario Consonni, Pier Alberto Bertazzi); Cancer Prevention Fellowship Program, Office of Preventive Oncology, National Cancer Institute, Bethesda, Maryland (Tram Kim Lam); Genetic Epidemiology Branch, National Cancer Institute, Bethesda, Maryland (Tram Kim Lam, Alisa M. Goldstein, Andrew W. Bergen, Margaret A. Tucker, Neil E. Caporaso, Maria Teresa Landi); Risk Factor Monitoring and Methods Branch, Division of Cancer Control and Population Sciences, National Cancer Institute, Bethesda, Maryland (Amy F. Subar); and Center for Health Sciences, SRI International, Menlo Park, California (Andrew W. Bergen).
Drs. Vincenzo Bagnardi and Giorgia Randi contributed equally to this work.
Support for this study was provided by the Intramural Research Program of the National Institutes of Health, National Cancer Institute (Division of Cancer Epidemiology and Genetics); by the Region of Lombardy, Milan, Italy (Environmental Epidemiology Program); and by the University of Milan-Bicocca, Milan, Italy (“Fondo Ateneo per la Ricerca” 2007).
The authors are indebted to the EAGLE Study investigators, who are listed on the EAGLE Study Web site (http://dceg.cancer.gov/eagle).
Conflict of interest: none declared.