Introduction

The relationship between alcohol consumption and lung cancer risk remains controversial. While an association has been observed in many studies, it is often presumed to be due to residual confounding from cigarette smoking. However, the association has been found even in studies that attempted to carry out careful control of smoking. In a large meta-analysis, the authors concluded that there might be an association, but only for the highest category of alcohol consumption [1]. Another recent study, which pooled seven cohort studies found weak evidence of a harmful association [2]. Still there is a need for additional confirmatory evidence before it can be concluded that consumption of alcohol is an independent risk factor for lung cancer.

If an association can be established, other more subtle issues are still unclear. For example, it is unknown whether different types of alcohol exert the same effect. Beer and spirits are the types most often reported to be associated with lung cancer while light to moderate wine consumption may be protective [3]. Overall, the evidence about the specific alcoholic beverages has been inconsistent [3].

Additionally, there is little evidence on possible effect modification by such important covariates as gender, smoking history, or diet. Finally, it is unclear how alcohol consumption might affect the risk of different histological types of lung cancer [4].

In this context, we report here on the association between lifetime consumption patterns of alcohol and lung cancer in the context of two large population-based case–control studies in Montreal, Canada. Particular attention was given to the statistical modelling of the smoking variables in order to rule out potential confounding. Detailed analyses were carried out to investigate the role that different types of alcohol might play, to assess the effect of alcohol on the different histological types of lung cancer and finally to identify possible interactions between gender, smoking, selected dietary factors, and alcohol.

Materials and methods

The association was investigated in two data sets. Both arose from large population-based case–control studies conducted in Montreal. The first, referred to as Study I and conducted in the early 1980s, was designed to explore the possible associations between hundreds of occupational substances and multiple cancer sites, including lung cancer, in men [5, 6]. The second, referred to as Study II and conducted in the mid-1990s, focussed on the same occupational exposures and lung cancer in men and women. Both studies identified all newly diagnosed lung cancer cases at any Montreal-area hospital, and living in the Montreal area. In Study I, subjects were 35–70 years of age, while in Study II they were 35–75 years of age. In both studies, population controls were randomly selected from the electoral lists. In Canada, electoral lists have been maintained by means of active enumerations of households; they are thought to report nearly complete listings of Canadian citizens residing in Canada. In the first study, controls were frequency matched by age and area of residence to all cancer cases. This was not a problem for the current analysis because the age distribution of lung cancer was roughly similar as that for all cases. In the second study, controls were stratified to the distribution of lung cancer cases by age and sex. Ethical approval was obtained for both studies.

In both Study I and Study II, the subjects or a surrogate respondent (proxy) was interviewed. Data were collected on a large number of variables including ethnicity, socio-economic status, smoking history, selected dietary items, alcoholic beverage consumption, as well as a detailed occupational exposure history. In addition to years of schooling and ethnicity, we derived the median family income of the census tract of residence as a further measure of socio-economic class (SES). For inclusion in the analyses presented here subjects were required to have completed the interview and to have complete alcohol information.

In Study I, 1082 lung cancer cases and 740 population controls were approached. Of these, 699 (64.6%) cases and 507 (68.5%) controls completed the interview and provided complete alcohol information. In Study II, 1432 eligible cases and 2208 eligible controls, of whom 76.4% and 66.5%, respectively, completed the interview and provided complete alcohol information. There was little difference in response rates between men and women in Study II. Compared to subjects included, those excluded due to missing alcohol information had been responded for by a proxy more frequently, but did not differ markedly in other respects.

Both studies collected information on diet via food frequency questionnaires. In the first study, subjects were asked about the usual frequency of consumption (at least twice a week, at least once a month, seldom or never) for ten different vegetable types and one fruit type (see Appendix for specific foods). In Study II, subjects reported their usual frequency of consumption (7 or more per week, 4–6 per week, 1–3 per week, 1–3 per month, never or less than once per month) of 13 vegetable or vegetable-juice types and for 12 fruit or fruit-juice types (see Appendix). For both studies, the average number of fruit and vegetable servings per month was then estimated.

Ascertainment of alcohol consumption

In both studies, subjects were asked, separately for beer, wine and spirits, if there had ever been a period when they had consumed the alcoholic beverage the equivalent of at least once a week, or nearly every day. For those who reported drinking the beverage nearly every day, information on the age started and stopped drinking, and average daily consumption based on the number of typical servings was collected.

Ascertainment of smoking

In both studies, information was collected on age at initiation, age at quitting and average number of cigarettes smoked per day among subjects who had ever smoked regularly, defined as nearly every day or at least once a week, in Study I and II respectively.

Statistical analyses

Unconditional logistic regression was used to estimate the odds ratios (OR) and 95% confidence intervals (CI) for the association between alcohol consumption and lung cancer. All models were adjusted for age (continuous), log of the SES index (continuous), years of schooling (continuous), ethnicity (three categories: French, Anglo or other), respondent status (proxy or self), and smoking. Different aspects of smoking behaviour were available, and following the modelling approach recommended by Leffondré et al. [7] we used a three-variable parameterisation including smoking status (ever/never), cigarette-years and time since quitting in categories (2–5 years, 5–10 years, 10–15 years and >15 years). In some models, we also explored excluding ex-smokers or replacing cigarette-years by separate terms for duration of smoking and number of cigarettes per day [7]. Models were estimated separately for three sets of cases and controls: men from the first study, men from the second study and women.

Tests for linear trend were performed by including the ordinal variable as a continuous covariate in the regression model [8]. To assess the interaction between smoking and alcoholic beverage consumption, two terms were included in the model that were the product of a dichotomous variable representing the drinking frequency (one for 1–6 drinks per week, and one for ≥7 drinks per week) and an ordinal variable representing strata of smoking and scored as the median number of cigarette-years. To assess the interaction between fruit and vegetable consumption, two terms were included in the model that were the product of a dichotomous variable representing the drinking frequency and a continuous variable representing number of servings of fruits and vegetables each week. The p-values reported are those associated with each of those variables. To assess whether there was any residual confounding by cigarette smoking, generalised additive models were used to model the continuous smoking covariates, using four degrees of freedom [9, 10].

Results

Table 1 shows selected characteristics of cases and controls included in Studies I and II. As compared to controls, cases were more often of French ancestry, had a lower family income, had a lower educational level, were less often married, had had a proxy respondent complete the interview for them, were more often current smokers and had smoked more heavily. This held true in Studies I and II, and for both men and women in Study II. The two studies were carried out over ten years apart. Lung cancer cases diagnosed more recently were less likely to be current smokers or had quit smoking earlier.

Table 1 Selected characteristics of cases and controls in Study I and Study II
Full size table

Table 2 shows the alcohol consumption patterns, by type of alcohol as observed in the two studies. Women were much less likely to have consumed more than 7 drinks per week than were men. Beer was by far the most popular alcoholic beverage among men; among women wine was somewhat more popular than the others were. Among men there was an apparent decrease in beer consumption and an increase in wine consumption between the two studies.

Table 2 Alcohol consumption by type of alcohol
Full size table

We investigated how alcoholic beverages consumption patterns varied according to SES and education among controls. Men in the lower strata of SES or education drank beer and spirits more frequently, and drank wine less frequently than those in higher strata. Higher SES or more educated women drank more frequently, but consumed fewer drinks per day than those in the lower strata.

Table 3 displays the adjusted odds ratios between consumption of alcoholic beverages and lung cancer for men from Study I and II, and for women in Study II. In both’studies, men whose consumption averaged seven or more drinks per week had a non-statistically significant elevated risk of lung cancer (OR=1.3 in Study I; OR=1.2 in Study II).

Table 3 Odds ratios between alcohol consumption and lung cancer in men and womena
Full size table

Among those men drinking at least seven drinks per week, there was an upward trend in risks with increasing cumulative amounts of intake in Study I, driven mainly by beer consumption. This trend was also evident when drink-years was modelled as a continuous variable. No such dose- response was seen for men or women in Study II. Duration of drinking was not predictive of lung cancer risk for men or women.

By contrast, there were indications of an inverse relationship between wine and lung cancer. Indeed, light drinking of wine appeared to be protective for men from Study II (Table 3). For men from Study I, although daily wine drinkers were at lower risk than never drinkers, risk increased as the daily amount consumed increased (OR=1.07 per additional daily glass of wine, 95% CI: 0.95–1.20). Adjusting for consumption of other types of alcoholic beverages made no difference in risk associated with wine or spirits and little difference in those associated with beer.

For women, moderate drinking of any type of beverage was associated with a reduction in risk of lung cancer. The greatest ostensible protection was seen among women who drank 1–6 glasses of wine per week. Results changed minimally when adjusting the beverage-specific estimates for each other.

We also investigated transforming drink-years into total lifetime ethanol. However, because the amount of ethanol contained in a standard serving of beer, wine or spirits is similar, the results were very close to those obtained using drink-years. Moreover, we explored using continuous variables, including cigarette-years, represented by 4-degree- of-freedom smoothing spline functions, in the generalized additive models extension of logistic regression [9, 10]. Point estimates, and statistical significance, for beer or wine drinking changed minimally.

For women and men from Studies I and II, subjects who stopped drinking beer, wine or spirits on a daily basis within the 2-year period preceding diagnosis (or interview) were at a pronounced increased risk of lung cancer. The same phenomenon had been observed for quitting smoking as well [7]. Because this might reflect a reverse-causality bias in that cases who were perhaps not feeling well modified their drinking habits, we defined quitting alcohol as those who reported that they had stopped drinking more than 2 years ago. For Study I men, those subjects who stopped drinking beer and spirits were at decreased risk (OR=0.6, 95% CI: 0.4–0.9; and OR=0.7, 95% CI: 0.3–1.3 respectively), while there was no effect for quitting wine (OR=1.1). For Study II men, the ORs were 0.8, 0.9 and 0.8, for quitting beer, spirits or wine respectively. For women, those who had stopped daily consumption of beer, wine or spirits were at increased risk of lung cancer (the ORs were 2.2, 1.9, and 2.5 respectively) though confidence intervals were wide, as few women had quit drinking.

Table 4 presents the effects of overall drinking and beer consumption on predominant histological subtypes [11]. For men from Study I, the harmful effect of overall consumption and of beer was evident for both squamous cell tumours and adenocarcinoma. For men from Study II, point estimates for overall consumption were above 1.0 for all histological subtypes. In women, drinking 1–6 drinks per week, of beer, or in total, appeared protective across all histological subtypes.

Table 4 Type of alcohol and histological type of lung cancer
Full size table

To investigate possible interactions between cigarette smoking and alcohol consumption, we investigated the association between alcoholic beverages consumption and lung cancer in three strata of smoking status, light, moderate and heavy, defined by approximate tertiles of cigarette-years. For women, the light smoker group consisted of women who had never smoked regularly. For men, there were too few never smokers among cases to constitute such a stratification category. There was no strong evidence of any effect modification by cigarette smoking either in Study I men or in women (Table 5). For Study II men, there was some trend toward increased risks among heavier smokers for total alcohol, and beer, especially among men drinking 1–6 drinks per week.

Table 5 Odds ratiosa between alcohol consumption and lung cancer within strata of smokingb
Full size table

Table 6 presents the results of an analogous analysis, but this time with level of fruit and vegetable consumption as the stratification variable. Fruit and vegetable consumption was found to modify the association between beer drinking and lung cancer risk for men and women from Study II. Whereas beer consumption was associated with an increased risk of lung cancer in the low fruits and vegetable consumption stratum, there was no such association for those consuming relatively many servings of fruits and vegetables each week. This interaction was also evident for total beverage consumption, but not for spirits in Study II men. For men from Study I, point estimates for total alcohol and spirits consumption for subjects in the highest stratum of fruit and vegetable intake were lower than for those in other strata. For men from Study I, there was some evidence that fruit and vegetable consumption modified the wine-lung cancer association: subjects in the lowest stratum of fruit and vegetable consumption were at increased risk from consuming 1–7 glasses of wine per week (OR=2.8), and this risk decreased across the strata (middle stratum: OR=1.5, highest stratum: OR=0.9). Fruit and vegetable consumption did not appear to modify the association between wine and lung cancer for men and women from Study II.

Table 6 Odds ratiosa between alcohol consumption and lung cancer within strata of fruit and vegetable consumption
Full size table

Discussion

We investigated the association between alcoholic beverage consumption and lung cancer in two population-based case–control studies. For men, daily drinkers of any alcoholic beverage were at increased risk of lung cancer, reflecting mainly beer consumption, as moderate wine drinking appeared to be protective. Daily consumption of spirits seemed to increase risk among Study I men, but not Study II men, possibly because Study I men consumed greater daily amounts than did Study II men. Among women, regular drinkers had lower lung cancer risk than non-drinkers, with the strongest protective effect seen among moderate regular drinkers. Although our results suggest that alcohol consumption may be harmful, the lack of a consistent effect across the study groups (Study I men, Study II men and women) and beverage types warrants caution about the interpretation of these results, and sheds some doubt on the carcinogenic effect of alcohol.

Concern over previously published results reporting a harmful effect of alcohol on lung cancer centered on residual confounding by smoking. We attempted to control for cigarette smoking in several ways such as: (i) using generalized additive models and modelling cigarette-years as a smooth, rather than a linear function, (ii) using duration of smoking and amount smoked per day rather than cigarette-years, (iii) including categorical terms representing time since quitting, or (iv) examining the alcohol effect in the subjects who smoked the least. Because the point estimates for the various alcohol intake patterns changed minimally, no matter the approach, we are confident that these results are not due to residual confounding due to inadequate modelling of the cigarette smoking effect. However, imperfect smoking history reports could lead to inadequate control for smoking, even though smoking history is thought to be fairly valid and is well-reported by proxy respondents [12–14].

The reference categories used in these analyses consisted of subjects reporting that they had never consumed the specific beverage weekly, and so included complete abstainers and occasional drinkers. From the questionnaires, it was impossible to discriminate between these subgroups, leading to a potential underestimation of the effects of alcohol. Then again, the advantage of this reference group definition is that complete abstainers may not be a representative group of subjects.

About 15 years separated the two studies, which might explain the different findings between Study I and II. Between the early-1980s and mid-1990s diet in Canada has changed, and fruit and vegetable consumption, shown to be protective for lung cancer [15–17], has increased substantially [18]. This trend was evident when comparing reported consumption patterns of men from Study I and II. We observed that fruit and vegetable intake modified the relationship between beer drinking and lung cancer; this might explain why beer drinking was harmful for men in Study I but not in Study II. Similarly, more women ate vegetables and fruits than men, and this might account for the lack of a harmful effect of beer in women. In fact, it appeared that beer-drinking men and women from Study II who consumed relatively few weekly servings of fruit and vegetables were at increased risk. We did not observe the effect modification for Study I men; perhaps because the diet information elicited in Study I was less detailed than that in Study II. Other studies investigating interactions between diet and alcohol consumption have found a stronger, harmful effect of alcohol drinking on lung cancer risk for those consuming the least vegetables [19], vitamin A [20], and carotenoids [21]. This interaction might explain some of the contradictory reports in the literature with respect to alcohol and lung cancer.

It is unknown what biologic mechanism may be responsible for a harmful effect of alcohol, though several have been proposed. Acetaldehyde, ethanol’s primary oxidative metabolite, has been shown to be carcinogenic in animal studies [3]. Alternatively, alcohol consumption has been shown to change lung lipids and levels of inducible enzymes capable of activating procarcinogens and mutagens in animals [22]. Finally, it may be due to the oxidative effects of alcohol [2]. The ostensible interaction observed between fruit and vegetable consumption and alcohol drinking may be because eating fruits and vegetables, which are high in antioxidants, may protect against oxidative DNA damage and in turn protect against lung cancer [15].

While part of the discrepancy in the findings for beer and spirits between men and women might be explained by diet-based variations, physiologically men and women process alcohol differently. Women reap the same benefit from alcohol consumption for coronary heart disease as men do, but with less alcohol [23]. This might be because, in comparison to men, women’s bodies are typically smaller, alcohol is less soluble in women’s fattier bodies, and/or women metabolise alcohol less efficiently [23, 24]. Several other studies have investigated the effect of alcohol on lung cancer in women, though often there were few lung cancer cases. Four studies reported no association [25–28], while three studies reported discrepant effects [29–32]. Recent evidence suggests a slight protective effect for light/moderate drinking of any type of alcohol [2].

We observed a protective effect of moderate wine consumption against lung cancer in both men and women. Previous results for wine consumption have been contradictory. Four case–control [26–29] and one prospective study [27] found wine consumption to be protective, with most effects statistically significant [31, 33, 34]. One case–control [19] and one prospective study [35] reported harmful effects. Altogether, these results have led some to suggest that the relationship between wine drinking and lung cancer risk is U-shaped. Our findings provide some support for such a hypothesis. We found a protective effect for light wine drinking in women, and a weaker protective effect for heavier drinking. Moreover, men from Study I who drank wine on a daily basis had a lower risk than never drinkers, but risk increased with the amount ingested.

Regular, light or moderate alcohol consumption seems to be protective for overall mortality and for cardiovascular diseases [28, 36]. Accumulating evidence, ours included, suggests that moderate wine drinking may also lower the risk of lung and other cancers, though these results remain controversial [37]. The protective effect of wine might be conferred by resveratrol, a component that reduces the metabolic activation of carcinogens, has antioxidant and anti-inflammatory properties, decreases cell proliferation and induces apoptosis [37, 38]. The hypothesized U-shape association between wine consumption and lung cancer risk may be due to a trade-off between the protection granted by resveratrol and the harm caused by ethanol [3].

Conversely, the protective effects of moderate alcohol consumption on cardiovascular mortality might be due to residual confounding by social, behavioural, demographic or other factors [39]. Indeed, in our study, subjects who never drank wine regularly were older, more likely to be current smokers, less likely to be married, less educated, and poorer growing up and currently. However, moderate wine drinking remained protective even after adjustment for these factors.

A major strength of this study is the large number of cases and controls. Few studies have had the capacity to investigate the role of alcohol in men and women separately, to evaluate the effect of different alcoholic beverages on specific histological subtypes, or to assess whether the effect of alcohol differs according to smoking pattern and fruit and vegetable intake levels.

A potential drawback to the work presented relates to the retrospective, self-reported assessment of alcohol intake and other covariates. Population controls were used, which raises the possibility of recall bias. However, the interview focussed on occupational factors; so the subject was less likely to pinpoint alcohol as an important exposure, especially since there was no widespread belief that alcohol consumption was a risk factor for lung cancer.

Self-reported measures of alcohol consumption have been found to be reasonably valid and reliable [40], even when assessing lifetime consumption [41]. Under-reporting was found to be a problem among alcoholics [42], but, introduced little bias for broad measures of consumption [43]. If alcohol consumption was systematically underreported here, higher levels of risk would be ascribed to lower levels of alcohol consumption. While this could explain some of the risk for low levels of consumption, it could not result in the increased risk for daily beer drinkers, nor would this explain the protective effect of moderate wine drinking.

Using proxy respondents might have resulted in misclassification of exposure to alcohol. However, most studies that assessed proxy respondents’ data quality for alcohol use in case–control studies found that self and proxy responses agree well or very well for broad measures of use, and moderately well for the details of that use [44–47]. For men in our study, the proxy was most often the spouse, reportedly the most reliable surrogate respondent [44]. For women, the proxy was the spouse (25%), offspring (40%), or other family member or friend. Limiting analyses to self-respondents gave similar results to those including all respondents (not shown).

In summary, total consumption of alcoholic beverages was found to be harmful for men, and reflected the lower risk associated across all alcohol types for women drinkers. A protective effect for drinking moderate amounts of wine was noted in both men and women. The results for beer and spirits are less clear. On the one hand, the lack of a consistent effect across the three groups (men from Study I, men from Study II and women) seems to mitigate against a causal association. On the other hand, effect modification by diet might explain some of the discrepant findings for beer. According to our data, regular heavy beer drinking was a stronger risk factor for lung cancer among low consumers of fruit and vegetables. For spirits, the findings differed between Study I and II men, possibly because of differences in intake levels across studies.