Alcohol intake and pancreatic cancer risk: An analysis from 30 prospective studies across Asia, Australia, Europe, and North America

Abstract

Background: Alcohol is a known carcinogen, yet the evidence for an association with pancreatic cancer risk is considered as limited or inconclusive by international expert panels. We examined the association between alcohol intake and pancreatic cancer risk in a large consortium of prospective studies.

Methods and findings: Population-based individual-level data was pooled from 30 cohorts across four continents, including Asia, Australia, Europe, and North America. A total of 2,494,432 participants without cancer at baseline (62% women, 84% European ancestries, 70% alcohol drinkers [alcohol intake ≥ 0.1 g/day], 47% never smokers) were recruited between 1980 and 2013 at the median age of 57 years and 10,067 incident pancreatic cancer cases were recorded. In age- and sex-stratified Cox proportional hazards models adjusted for smoking history, diabetes status, body mass index, height, education, race and ethnicity, and physical activity, pancreatic cancer hazard ratios (HR) and 95% confidence intervals (CI) were estimated for categories of alcohol intake and in continuous for a 10 g/day increase. Potential heterogeneity by sex, smoking status, geographic regions, and type of alcoholic beverage was investigated. Alcohol intake was positively associated with pancreatic cancer risk, with HR30-to-<60 g/day and HR≥60 g/day equal to 1.12 (95% CI [1.03,1.21]) and 1.32 (95% CI [1.18,1.47]), respectively, compared to intake of 0.1 to <5 g/day. A 10 g/day increment of alcohol intake was associated with a 3% increased pancreatic cancer risk overall (HR: 1.03; 95% CI [1.02,1.04]; pvalue < 0.001) and among never smokers (HR: 1.03; 95% CI [1.01,1.06]; pvalue = 0.006), with no evidence of heterogeneity by sex (pheterogeneity = 0.274) or smoking status (pheterogeneity = 0.624). Associations were consistent in Europe-Australia (HR10 g/day = 1.03, 95% CI [1.00,1.05]; pvalue = 0.042) and North America (HR10 g/day = 1.03, 95% CI [1.02,1.05]; pvalue < 0.001), while no association was observed in cohorts from Asia (HR10 g/day = 1.00, 95% CI [0.96,1.03]; pvalue = 0.800; pheterogeneity = 0.003). Positive associations with pancreatic cancer risk were found for alcohol intake from beer (HR10 g/day = 1.02, 95% CI [1.00,1.04]; pvalue = 0.015) and spirits/liquor (HR10 g/day = 1.04, 95% CI [1.03,1.06]; pvalue < 0.001), but not wine (HR10 g/day = 1.00, 95% CI [0.98,1.03]; pvalue = 0.827). The differential associations across geographic regions and types of alcoholic beverages might reflect differences in drinking habits and deserve more investigations.

Conclusions: Findings from this large-scale pooled analysis support a modest positive association between alcohol intake and pancreatic cancer risk, irrespective of sex and smoking status. Associations were particularly evident for baseline alcohol intake of at least 15 g/day in women and 30 g/day in men.

Fig 1. Association between alcohol intake and the risk of pancreatic cancer, overall and by sex.

Abbreviations: HR: hazard ratio, CI: confidence interval; ¹ Cox proportional hazard models were adjusted for smoking status, smoking duration, smoking intensity, time since smoking cessation, diabetes status, body mass index, height, education, race and ethnicity, and physical activity. Analyses in continuous were further adjusted for an indicator variable for alcohol drinking status. Models were stratified by age at baseline, year of baseline questionnaire completion, study, country (in EPIC [59]), and sex (overall models only); ² P-value for the Wald test statistics compared with a X² distribution with degrees of freedom equal to the number of alcohol intake categories minus one, not including the category of non-drinkers (<0.1 g/day); ³ P-value for alcohol consumption modelled as a continuous variable for a 10 g/day increase, with inclusion in the model of an indicator variable expressing the alcohol drinking status; ⁴ Heterogeneity across studies was tested by adding interaction terms between alcohol intake modeled in continuous and each study level, then comparing the Wald test statistics for significance to a X² distribution with the number of degrees of freedom equal to the number of studies minus one, in a model including an indicator variable expressing the alcohol drinking status; ⁵ Heterogeneity by sex was tested by adding interaction terms between alcohol intake modelled in continuous and sex, then comparing the Wald test statistics for significance to a X² distribution with one degree of freedom, in a model including an indicator variable expressing the alcohol drinking status. The alcohol and pancreatic cancer dose–response relationship using restricted cubic splines among participants with alcohol intake lower than 100 g/day showed no departure from linearity, neither overall (p_nonlinearity = 0.345) nor in women (p_nonlinearity = 0.355) or men (p_nonlinearity = 0.633) (Fig B in S1 File).

Fig 2. Association between alcohol intake and the risk of pancreatic cancer by smoking status (never, former, current smoker).

Abbreviations: HR: hazard ratio, CI: confidence interval; ¹ ATBC recruited only current smokers; ² Cox proportional hazard models were adjusted for smoking duration, smoking intensity, time since smoking cessation, diabetes status, body mass index, height, education, race and ethnicity and physical activity. Analyses in continuous were further adjusted for an indicator variable for alcohol drinking status. Models were stratified by age at baseline, year of baseline questionnaire completion, study, country (in EPIC [59]) and sex. Models included interaction terms between alcohol intake and smoking status, keeping the 0.1 to  <5 g/day category as reference, while participants without information on their smoking status were excluded; ³ P-value for the Wald test statistics compared with a X² distribution with four degrees of freedom, not including the category of non-drinkers (<0.1 g/day); ⁴ P-value for alcohol consumption modelled in continuous, in a model including an indicator variable expressing alcohol drinking status; ⁵ Heterogeneity by smoking was tested comparing the Wald test statistics for interaction between alcohol intake and smoking level to a X² distribution, with either four degrees of freedom not including the category of non-drinkers (<0.1 g/day) for analyses in categories, or one degree of freedom in a model including an indicator variable expressing alcohol drinking status for analyses in continuous.

Fig 3. Association between alcohol intake and the risk of pancreatic cancer by geographic region (Europe/Australia, North America, Asia).

Abbreviations: HR: hazard ratio, CI: confidence interval; ¹ Geographic region was coded as Europe (ATBC, COSM, GS, EPIC, MCCS, NLCS, SMC, SNMC and WLHS), North America (BCDDP, CARET, CLUEII, CNBSS, CPSII, CTS, HPFS, IWHS, MEC, NHS, NHSII, NIH-AARP, NYSC, PLCO, VITAL, and WHI) and Asia (JPHCI, JPHCII, SCHS, SCS and SMHS); ² Cox proportional hazard models were adjusted for smoking status, smoking duration, smoking intensity, time since smoking cessation, diabetes status, body mass index, height, education, race and ethnicity, and physical activity. Analyses in continuous were further adjusted for an indicator variable for alcohol drinking status. Models were stratified by age at baseline, year of baseline questionnaire completion, study, and country (in EPIC [59]) and sex; ³ For analyses in categories, p-value compared the Wald test statistics with a X² distribution with degrees of freedom equal to the number of alcohol intake categories minus one, not including the category of non-drinkers (<0.1 g/day). In continuous analyses, it was the p-value for alcohol consumption in continuous in a model including an indicator variable expressing alcohol drinking status; ⁴ Heterogeneity across studies within each geographic region was tested adding interaction terms between alcohol intake modelled in continuous and each study level, then comparing the Wald test statistics for significance to a X² distribution with the number of degrees of freedom equal to the number of studies minus one, in a model including an indicator variable expressing alcohol drinking status; ⁵ Heterogeneity by sex within each geographic region was tested adding interaction terms between alcohol intake in continuous and sex, then comparing the Wald test statistics for significance to a X² distribution with one degree of freedom in a model including an indicator variable expressing alcohol drinking status; ⁶ Heterogeneity by geographic region was tested adding interaction terms between alcohol intake in continuous and geographic region, then comparing the Wald test statistics for significance to a X² distribution with two degrees of freedom in a model including an indicator variable expressing alcohol drinking status.

Fig 4. Association between alcohol intake from different alcoholic beverages and the risk of pancreatic cancer.

Abbreviations: HR: hazard ratio, CI: confidence interval; ¹ Information on type of alcoholic beverages was not available in NYSC; ² Cox proportional hazard models were adjusted for alcohol intake from the other type of beverage than the one under evaluation, smoking status, smoking duration, smoking intensity, time since smoking cessation, diabetes status, body mass index, height, education, race and ethnicity, and physical activity. Analyses in continuous were further adjusted for an indicator variable for alcohol drinking status based on total alcohol intake. Models were stratified by age at baseline, year of baseline questionnaire completion, cohort and country (in EPIC [59]) and sex; ³ For analyses in categories, p_Wald compared the Wald test statistics with a X² distribution with degrees of freedom equal to the number of the given type of beverage categories minus one, not including the category of non-drinkers (<0.1 g/day); ⁴ In continuous analyses, p_trend was the p-value for the given type of beverage modelled as a continuous variable for a 10 g/day increase, in a model including an indicator variable expressing alcohol drinking status; ⁵ Heterogeneity across studies for a given type of beverage was tested adding interaction terms between the given type of beverage modeled in continuous and each study level, then comparing the Wald test statistics for significance to a X² distribution with the number of degrees of freedom equal to the number of studies minus one, in a model including an indicator variable expressing alcohol drinking status; ⁶ Heterogeneity by geographic region for a given type of beverage was tested adding interaction terms between the type of beverage and geographic region, then comparing the Wald test statistics for significance to a X² distribution with two degrees of freedom, in a model including an indicator variable expressing alcohol drinking status; ⁷ Heterogeneity by sex for each type of alcoholic beverage was tested adding interaction terms between alcohol intake and sex, then comparing the Wald test statistics for significance to a X² distribution with one degree of freedom, in a model including an indicator variable expressing alcohol drinking status.