|Home | About | Journals | Submit | Contact Us | Français|
Heavy alcohol drinking has been related to pancreatic cancer, but the issue is still unsolved.
To evaluate the role of alcohol consumption in relation to pancreatic cancer, we conducted a pooled analysis of 10 case–control studies (5585 cases and 11 827 controls) participating in the International Pancreatic Cancer Case–Control Consortium. We computed pooled odds ratios (ORs) by estimating study-specific ORs adjusted for selected covariates and pooling them using random effects models.
Compared with abstainers and occasional drinkers (<1 drink per day), we observed no association for light-to-moderate alcohol consumption (≤4 drinks per day) and pancreatic cancer risk; however, associations were above unity for higher consumption levels (OR = 1.6, 95% confidence interval 1.2–2.2 for subjects drinking ≥9 drinks per day). Results did not change substantially when we evaluated associations by tobacco smoking status, or when we excluded participants who reported a history of pancreatitis, or participants whose data were based upon proxy responses. Further, no notable differences in pooled risk estimates emerged across strata of sex, age, race, study type, and study area.
This collaborative-pooled analysis provides additional evidence for a positive association between heavy alcohol consumption and the risk of pancreatic cancer.
Cigarette smoking is an established risk factor for pancreatic cancer , but it explains only a limited proportion of all pancreatic cancers (i.e. about one in five cases) . Among other factors, some aspects of diet, nutrition, and body composition—including obesity, diabetes, and possibly low intake of fruit and vegetables and high intake of (cooked) meat—have been associated with an increased pancreatic cancer risk [3–8]. Heavy alcoholic beverage consumption is a common cause of chronic pancreatitis, mainly in men [9–11], and pancreatitis is another recognized risk factor for pancreatic cancer [4, 5, 12]. Thus, heavy alcohol consumption may be associated with an increased risk of pancreatic cancer.
Most case–control studies [13–34] and at least seven cohort studies conducted in Europe [35–39] and the United States [40, 41] have found no association between alcohol drinking and the risk of pancreatic cancer, and an International Agency for Research on Cancer (IARC) Monograph working group in 2007 concluded that there was an inadequate evidence for a role of alcohol in pancreatic cancer in humans . An excess risk of pancreatic cancer in heavy drinkers, particularly in Blacks, was suggested in a multicentric case–control study conducted in the United States in the 1980s . Some other case–control [44–49] and cohort [50–54] studies that were able to assess alcohol consumption in more detail reported increased risk of pancreatic cancer among participants with the highest level of alcohol consumption. However, most of these studies suffered from small sample sizes and inadequate consideration of potential confounding by tobacco smoking or presence of chronic pancreatitis.
We conducted a pooled analysis of alcohol consumption and pancreatic cancer risk using data from a large series of pancreatic cancer case–control studies within the International Pancreatic Cancer Case–Control Consortium (PanC4) [55, 56]. We examined the effect of type of alcoholic beverages and evaluated whether the association between heavy alcohol intake and pancreatic cancer risk was confounded/modified by tobacco smoking or history of pancreatitis.
We identified 10 case–control studies of pancreatic cancer that collected data on alcohol consumption using structured questionnaires. They included a total of 5585 participants with adenocarcinoma of the exocrine pancreas and 11 827 controls [30, 32, 43, 47, 49, 57–60]. Characteristics of each study and description of the exposure variables are given in the supplemental Appendix S1 (available at Annals of Oncology online). For 155 cases and 150 controls in the Shanghai study , 63 cases in the Toronto study , and 474 cases and 332 controls in the Surveillance of Environmental Aspects Related to Cancer in Humans (SEARCH) study , a proxy respondent was interviewed.
Questions about alcoholic beverage consumption were generally similar across all studies. However, we conducted a careful and detailed examination of the comparability of alcohol-related questions as part of our harmonization of data from the multiple studies included in this pooled analysis. Details about alcohol consumption data for each study are included in the supplemental Appendix S1 (available at Annals of Oncology online).
In order to create uniform variables for alcohol consumption across all studies, we converted the beverage volume of each alcoholic beverage specified in the questionnaires into milliliters. For each type of alcoholic beverage (beer, wine, or liquor), we calculated the cumulative lifetime consumption of pure ethanol multiplying the volume of each alcoholic beverage, by the duration of consumption, and the volume percentage of pure ethanol (5% for beer, 12% for wine, and 40% for hard liquor). We then computed the lifetime number of standardized drinks by dividing the cumulative lifetime consumption of pure ethanol by 14.8 ml (i.e. the mean volume of pure ethanol per drink across all alcoholic beverage types in the 10 studies). Finally, we computed average frequency of consumption of each type of alcoholic beverage by dividing the lifetime number of standardized drinks by the corresponding duration. For total alcohol, we computed average frequency of consumption as the sum of the lifetime number of standardized drinks for each type of alcoholic beverage divided by the longest duration of consumption. When duration by type of alcoholic beverage was not available [43, 58–60], we considered the frequency of consumption only. For two studies that collected data about alcohol use at different ages [57, 59], we also computed a mean frequency of consumption and total duration as the sum of the age-specific durations. For studies that provided information for various alcoholic beverages within the same broad group (e.g. white and red wine or rice and fruit wine; dinner drinks, grappa, whiskey, cognac, and brandy) [30, 49, 57], we considered the sum of the frequencies and the longest duration of consumption within each group.
To estimate the association between alcohol consumption and pancreatic cancer risk, we used a two-stage modeling approach . At the first stage, we assessed the association between alcohol consumption and pancreatic cancer risk for each study by estimating the odds ratios (ORs) and the corresponding 95% confidence interval (CI) using unconditional logistic regression models . These models also included age (<45, 45–49, 50–54, 55–59, 60–64, 65–69, 70–74, ≥75 years), sex, education (≤8th grade, 9th–11th grade, 12th grade or high school graduates, some college or college graduates, ≥1 year of graduate school), race/ethnicity (non-Hispanic White, Hispanic, non-Hispanic Black, other), body mass index (BMI, <20, 20 to <25, 25 to <30, ≥30 kg/m2), history of diabetes (≥1 year before diagnosis), tobacco smoking (never smokers, smokers of products other than cigarettes, excigarette smokers with ≥10 years since quitting, excigarette smokers with 1 to <10 years since quitting, current cigarette smokers of 1 to <20 cigarettes per day, current cigarette smokers of ≥20 cigarettes per day, as well as a continuous term for number of cigarettes), and center, for multicentric studies. At the second stage of analysis, summary effect estimates were computed using random effects models . The study-specific ORs were weighted by the inverse of the sum of their variance and the estimated between-study variance component. Heterogeneity between studies was evaluated using a χ2 statistic . To test for the significance of linear trends in pancreatic cancer risk across levels of different variables of alcohol consumption, we first estimated the trend in each study (by fitting a model including an ordinal variable, with values corresponding to an increasing score for each level of the variable), then we used a Wald test  to estimate the P value of the summary variable, in random effects model.
To investigate whether the effect of heavy alcohol consumption was homogenous in strata of selected covariates, we conducted analyses stratified by sex, age (<65, ≥65 years), tobacco smoking (ever, former, and current smokers of <20 cigarettes or of ≥20 cigarettes per day), race/ethnicity (non-Hispanic White, non-Hispanic Black), study area (North America, Europe), and source of controls (population, hospital). Heterogeneity across strata was evaluated using a χ2 statistic .
We also conducted a sensitivity analysis to evaluate the influence of any one study on the effect estimates. One study at a time was excluded from the analysis model to assess the impact of that study upon the magnitude and statistical significance of the overall summary estimate.
In addition to the two-stage analysis, we conducted an aggregate analysis where data from all studies were pooled into a single large dataset for the analysis . The association between alcohol consumption and pancreatic cancer risk was determined using unconditional multiple logistic regression models , adjusted for study and the potential confounding factors included in the study-specific models. We also included in the models the interaction terms between study and all confounding factors considered. The results were not substantially different from those obtained in the main two-stage analysis approach and therefore are not reported in detail.
Table 1 shows the distribution of sex, age, race, and other selected variables for the total population of 5585 pancreatic cancer cases and 11 827 controls. For both cases and controls, 57% of participants were men, whereas cases were somewhat older than controls (median: 65 years for cases, 63 years for controls). Cases reported higher education, higher BMI, and a greater proportion of current and ex-smokers, of diabetics, and of subjects with pancreatitis than controls.
ORs for pancreatic cancer associated with consumption of total alcohol and for specific alcoholic beverages are presented in Table 2. Compared with abstainers or occasional drinkers (<1 drink per day), risk estimates were near unity for up to 4 drinks per day and above unity for participants who consumed ≥9 drinks per day (OR = 1.6, 95% CI 1.2–2.2). We also considered total alcohol consumption after excluding 692 cases and 482 controls for whom information was based on proxy interview. All ORs were similar to those obtained in the overall pooled dataset, i.e. the OR for ≥9 drinks per day was 1.9 (95% CI 1.4–2.5). Results of sensitivity analyses showed that no single study unduly influenced the magnitude or the statistical significance of the summary estimates. In analyses by type of alcohol, ORs were increased for participants who consumed ≥4 drinks of wine per day (OR = 1.5; 95% CI 1.0–2.1; P trend 0.02), whereas no excess risk was observed for consumption of beer or hard liquor, although the data were sparse (Table 2).
A forest plot of study-specific and pooled ORs for pancreatic cancer risk associated with total alcohol consumption is presented in Figure 1. The pooled OR was 0.9 (95% CI 0.7–1.2) for 1 to <4 drinks per day, 1.2 (95% CI 1.0–1.5) for 4 to <6 drinks per day, and 1.5 (95% CI 1.2–1.8) for ≥6 drinks per day. ORs were elevated for the highest category of alcohol consumption in six studies. In the latter category, there was no evidence of between-study heterogeneity (P = 0.18). Results of analyses excluding participants with pancreatitis (OR = 1.4, 95% CI 1.1–1.7, for ≥6 drinks per day) and data from proxy respondents (OR = 1.6, 95% CI 1.3–1.9, for ≥6 drinks per day) were consistent with the overall results.
Additional analyses of total alcohol consumption (specifically heavy drinking, i.e. ≥6 drinks per day) were stratified by sex, age, tobacco smoking, race/ethnicity, study area, and source of controls (Table 3). The associations appeared somewhat stronger—although not significant—in Black than in White subjects. No notable or significant differences in risk estimates were observed by sex, age, study area, source of controls, or tobacco smoking. Further, no differences in risk estimates were observed by education, BMI, and history of diabetes (data not shown).
Data on duration of alcohol drinking were available in five studies only [30, 32, 47, 49, 57] and showed no consistent associations (OR = 0.9; 95% CI 0.7–1.1, for the highest duration of consumption, ≥40 years) (data not shown).
This collaborative-pooled analysis of data from the PanC4 case–control studies provides additional evidence, and more accurate quantitative estimates than previously available, on the association between heavy alcohol consumption and the risk of pancreatic cancer. Compared with abstainers or occasional drinkers, ORs of pancreatic cancer were at or near unity for participants who drank up to 4 drinks per day, and statistically significantly above unity for higher intakes, with up to a 60% increased risk among extremely heavy alcohol drinkers (≥9 drinks per day). No consistent relation was observed for duration of alcohol drinking using available data. This is not surprising, since for oral cancer and other alcohol-related neoplasms, the association between duration and risk is much less consistent than that with dose [65–67].
Although the 2007 IARC Monograph Working Group concluded that there was inadequate evidence for alcohol carcinogenicity on pancreatic cancer , an association between heavy alcohol consumption and risk of pancreatic cancer has been reported in some case–control [43–49] and cohort [50–54] studies. An IARC Monograph Working Group in 2009 concluded that there was limited evidence for a causal association between alcohol consumption and pancreatic cancer risk , referring to a meta-analysis of 21 case–control and 11 cohort studies  that showed an overall relative risk (RR) of 0.92 for consumption of <3 drinks/day, but of 1.22 (95% CI 1.12–1.34) for higher consumption. Likewise, the ‘Pooling Project of Prospective Studies of Diet and Cancer’ , which included 14 cohort studies for a total of 2187 incident cases, found no association up to 30 g/day of alcohol consumption (~3 drinks/day) but a moderate positive association for the highest category of consumption (RR = 1.22, 95% CI 1.03–1.45). Moreover, in that study, it was not possible to investigate the higher consumption of alcohol (>3 drinks/day) due to the small number of heavy drinkers in most (American) cohort studies. The Pancreatic Cancer Cohort Consortium-nested case–control analysis based on 1530 pancreatic cases found no significant association between total alcohol intake and pancreatic cancer risk; and the OR was increased for the highest consumption though not significantly (OR = 1.38, 95% CI 0.86–2.23 for ≥60 g/day) . However, that study did report a significant excess risk for ≥45 g/day for liquor in men (OR = 2.23, 95% CI 1.02–4.87), but the number of subjects in the various levels of intake and analyses of type of beverages were too limited for definitive conclusions.
Our results based on a reanalysis of original data from a uniquely large pooled dataset provide additional data on heavier alcohol consumption (i.e. 6–>9 drinks per day) and pancreatic cancer risk. We also were uniquely able to evaluate whether the association between (heavy) alcohol consumption and pancreatic cancer risk was modified by pancreatitis or tobacco smoking. More importantly, we were able to uniformly define the modeling of the exposure, confounding, and outcome variables . Finally, our data harmonization and pooling allowed more detailed adjustment for tobacco smoking than in several individual studies and meta-analyses of alcohol and pancreatic cancer risk.
Heavy alcohol consumption is a recognized cause of chronic pancreatitis [9–11, 72], a known risk factor for pancreatic cancer. In our dataset, 242 cases and 95 controls reported a history of pancreatitis. Our results did not substantially change when we excluded these individuals from the analysis. Although several studies queried participants to ascertain physician-diagnosed acute or chronic pancreatitis and number of episodes, it is difficult to assess the validity of (self-reported) information on pancreatitis, because of the potential misclassification between the various forms of acute and chronic pancreatitis, and obstructive pancreatitis resulting from duct blockage due to the presence of a pancreatic tumor [73, 74]. To diminish the potential for reverse causation on effect estimates, pancreatitis diagnosed in the short term (at least 1 year) before diagnosis/interview was not considered.
Heavy alcohol drinking also may have a direct effect on pancreatic carcinogenesis. Acetaldehyde —the main metabolite of ethanol—is a known carcinogen and the induction of pancreatic injury from fatty acid ethyl esters [12, 76] and reactive oxygen species  are possible mechanisms and might explain the association between heavy alcohol drinking and pancreatic cancer.
Cigarette smoking is an established risk factor for pancreatic cancer  and is often positively correlated with alcohol drinking. In our analyses, the association between heavy alcohol drinking and pancreatic cancer was consistently elevated among ever, former, and current (moderate and heavy) smokers. Interestingly, the magnitude of the effect was lowest among heavy current smokers, moderate for never and current moderate smokers, and greatest for ex-smokers. This result argues against the hypothesis that the excess risk among heavy drinkers is due to the correlation between heavy alcohol drinking and heavy tobacco smoking. Furthermore, in subanalyses that included studies with more detailed information about tobacco smoking (frequency, duration, years since quitting, and non-cigarette tobacco use), the magnitude of the association between alcohol consumption and pancreatic cancer risk was similar to the results that we reported for all studies. Thus, residual confounding by tobacco smoking is an unlikely explanation for the observed association between heavy alcohol consumption and pancreatic cancer risk. Our results also show that other potential confounders such as age, sex, study area, education, BMI, and history of diabetes appear to not modify the effect of heavy alcohol consumption on pancreatic cancer risk. The association with heavy alcohol drinking was somewhat—although not significantly—stronger in Blacks than in Whites. This was previously reported in an National Cancer Institute-based study , which contributed about half of the Blacks cases to the present pooled analysis in PanC4.
Both hospital- and population-based controls can introduce selection bias, e.g. by inclusion or exclusion of alcohol-related diseases in hospital controls, and by lower participation of individuals with alcohol dependency in population-based studies. In PanC4, three studies [32, 49, 60] used hospital controls, one study selected controls from healthy individuals accompanying subjects to the hospital/clinic , and six studies [30, 43, 57–59] used general population controls. Our results were consistent regardless of the source of study controls.
Recall bias and misclassification also may have affected our results, particularly because alcohol drinking may be intentionally or unintentionally underreported by participants and proxies. However, our sensitivity analyses showed that it is unlikely that less complete information collected from proxies influenced the overall results. It is more difficult to assess whether differences in social acceptance of alcohol consumption may have influenced participants’ response to alcohol questions during in person interviews. For three studies [32, 49, 59] included in the present analysis, reproducibility and validity of alcohol drinking were assessed and found satisfactory [78–80]. Although similar information was not available for other studies, results from our sensitivity analysis (i.e. excluding each study from the analyses) showed no substantial change in the pooled risk estimates.
We applied the same estimate of ethanol content for each type of alcoholic beverage across all studies. Although the ethanol content of wine and beer is relatively consistent across countries and regions in the world , there is a potential for variation in ethanol content from (hard) liquors. However, given the smaller number of liquor drinkers as compared with wine and beer drinkers, and little difference in the number of exclusive liquor drinkers by case–control status, this is not likely to have substantially influenced our results.
In summary, the results of this pooled analysis of case–control studies in PanC4 support a moderate increased risk of pancreatic cancer with heavy daily alcohol consumption that is in agreement with a previous pooled analysis of cohort studies  and a recent meta-analysis of published case–control and cohort studies . Our data provide no evidence for a role of light or moderate alcohol drinking in pancreatic carcinogenesis, but rather, an increased risk only for heavy drinking, the effects being independent from those of tobacco smoking.
The Italian and Milan studies were supported by the Italian Association for Cancer Research. The Louisiana State University study was supported by the Louisiana Board of Regents Millennium Trust Health Excellence Fund [Project 5: HEF (2000-05, Genetics Studies in the Acadian Population)]. The NCI study was supported by the Intramural Research Program of the National Institute of Health, National Cancer Institute, Division of Cancer Epidemiology and Genetics (N01-CP-51090, N01-CP-51089, N01-CP-51092, N01-CP-05225, N01-CP-31022, N01-CP-05227). The Montreal investigation in the SEARCH study was supported by the Cancer Research Society, the Toronto contribution was supported by the National Cancer Institute of Canada, and the Netherlands investigation was supported by the Dutch Ministry of Public Health, Welfare and Sports (formerly Welfare, Health, and Culture). The University of California-San Francisco (UCSF) study work was supported in part by National Cancer Institute grants (CA098889 to E.J.D., CA59706, CA108370, CA109767, CA89726 to E.A.H.), and by the Rombauer Pancreatic Cancer Research Fund. Cancer incidence data collection in the UCSF study was supported by the California Department of Public Health, the National Cancer Institute’s Surveillance, Epidemiology and End Results Program contract N01-PC-35136 awarded to the Northern California Cancer Center. E.L. was supported by Environmental Cancer Risk, Nutrition and Individual Susceptibility [European Union, Sixth Framework Program (FP6) Network of Excellence] and the Province of Milan. At Catalan Institute of Oncology, E.J.D. was supported by ISCIII of the Spanish Ministry of Health (RETICC DR06/0020).
The authors declare no conflict of interest.
The authors thank Mrs I. Garimoldi for editorial assistance.