Coffee and Alcohol Consumption and the Risk of Pancreatic Cancer in Two Prospective United States Cohorts¹

Cancer of the pancreas represents the fifth leading cause of cancer-related mortality in the United States (1). Nonetheless, other than cigarette smoke, few environmental factors have been linked to the risk of pancreatic cancer (2). Numerous prospective and case-control studies have been conducted to determine whether coffee, tea, or alcohol increases pancreatic cancer risk, but results have been inconsistent (findings are summarized in Ref. 3). Many factors may contribute to these inconsistent results. Pancreatic cancer is particularly difficult to study in case-control studies because poor survival leads to reduced participation rates for direct interviews with cases, thus the studies rely more on interviews with surrogates, particularly next-of-kin. Moreover, in population-based case-control studies, heavy alcohol and coffee drinkers may have lower participation rates than nondrinkers, and in hospital-based case-control studies, alcohol and coffee may be related to the health conditions afflicting controls that are included for study.

Prospective studies could potentially avoid these selection biases in case or control participation, as well as potential recall bias. However, previous prospective studies (4, 5, 6, 7, 8, 9, 10, 11) on alcohol or coffee intake have generally been limited by the small number of incident cases, which have ranged from 21 to 66, with the exception of two studies (12, 13) based on over 400 pancreatic cancer cases each. To date, 10 prospective studies have examined the relation between coffee intake and pancreatic cancer risk, and all but two of these (4, 12) found no association (5, 6, 7, 8, 9, 10, 11, 13). Additionally, seven prospective studies have examined the influence of alcohol intake in nonalcoholic populations, and the results are largely inconsistent (4, 5, 6, 7, 8, 12, 13). Three of seven of these studies (4, 5, 8) have reported a direct relationship between alcohol intake and risk of pancreatic cancer, and a fourth found an increase in risk among whiskey drinkers (12). Of the remaining three studies, one reported a strong but statistically nonsignificant positive association among past drinkers (7), and only two observed no associations between alcohol and pancreatic cancer (6, 13).

Despite the lack of association for coffee intake and pancreatic cancer in most previous cohort studies, a recent cohort study of 33,976 United States women reported statistically significant elevations in risk of pancreatic cancer for coffee consumption, independent of age and cigarette smoking (4). In that study, consumption of 18+ cups of coffee/week was associated with a 2-fold elevation in pancreatic cancer risk when compared with <7 cups/week. Alcohol was also observed to increase risk in the same study, although the association was not statistically significant (more than two drinks of alcohol/week increased pancreatic cancer risk by 65% when compared with nondrinkers).

We examined the relationships of coffee and alcohol intakes and pancreatic cancer risk in the HPFS³, a prospective cohort of men, and in the NHS, a prospective cohort of women. The long follow-up period, the substantial number of cases (n = 288), and the availability of repeated dietary assessments allowed for alternative analyses to examine temporal relationships between coffee and alcohol and the risk of pancreatic cancer. In addition, we examined the intakes of decaffeinated coffee, tea, caffeine, and alcoholic beverage type in relation to pancreatic cancer risk.

Two ongoing cohort studies provided data for our analyses, the HPFS and the NHS. The HPFS was initiated in 1986 when 51,529 United States men 40–75 years of age responded to a mailed questionnaire. Fifty-eight percent of the men in the HPFS cohort are dentists, and the other professions include optometrists, osteopaths, podiatrists, pharmacists, and veterinarians. The NHS began in 1976 when 121,700 female registered nurses 30–55 years of age responded to a mailed questionnaire. Detailed information on individual characteristics and habits such as age, marital status, weight, height, medical history, medication use, smoking status, and physical activity was obtained through the mailed questionnaires at baseline and subsequently every 2 years.

For each of the follow-up questionnaires, up to six mailings were sent to nonrespondents. Most of the deaths in this cohort were reported by family members or by the postal service in response to the follow-up questionnaires. In addition, the National Death Index was searched for nonrespondents; this method has been shown to have a sensitivity (the probability that a deceased person is found in the Index, given that the person is deceased) of 98% (14).

In 1986, the baseline questionnaire for the HPFS cohort included a 131-item semiquantitative FFQ. In the NHS, a 61-item FFQ was mailed to all of the participants of the study in 1980. Both the HPFS and the NHS initial FFQs included questions on intakes of coffee and alcohol (wine, beer, and liquor) at baseline. On the initial questionnaires, we also inquired if a participant’s alcohol use (and coffee in the NHS) had greatly increased or decreased over the past 10 years. In 1984, a more comprehensive FFQ (116-items) was mailed to the NHS cohort. This questionnaire resembled the FFQ administered to the HPFS in 1986 and included a question on decaffeinated coffee consumption (which was not asked on the 1980 questionnaire). In addition, both cohorts completed detailed FFQs in 1986, 1990, and 1994.

On the questionnaire, participants were asked to report their average frequency of intake over the previous year for a specified serving size of each food. Individual nutrient intakes were calculated by multiplying the frequency of each food consumed by the nutrient content of the specified portion size (obtained from the United States Department of Agriculture and supplemented by other publications) and then by summing the contributions from all of the foods. Total alcohol intake was the sum of the values for all of the three beverages; a 12-oz can or bottle of beer was assumed to contain 13.2 g of alcohol, a 4-oz glass of wine, 10.8 g, and a standard drink of liquor, 15.1 g (United States Department of Agriculture). Total caffeine intake was calculated by summing the amount of caffeine in coffee (77%), tea (13%), soda (6.6%), decaffeinated coffee (3%), chocolate (0.3%), and candies (0.1%) consumed by the study participants.

Validity studies of the FFQ, compared with 1-week diet records, among 127 men from the HPFS and 173 women from the NHS indicated that most foods and beverages are reported reasonably well. These studies are reported in detail elsewhere (15, 16). Correlations between the average intake assessed by two 1-week diet records completed 6 months apart and 1-FFQ were 0.93 and 0.77 for coffee and tea, respectively, in the HPFS (15), and 0.78 and 0.93 for coffee and tea, respectively, in the NHS (16). For alcohol intake, the correlation between multiple dietary records and a FFQ was 0.90 in the NHS and 0.86 in the HPFS (17).

Smoking status and history of smoking were obtained at baseline and in all of the subsequent questionnaires in both cohorts. Current smokers also reported intensity of smoking (average number of cigarettes smoked/day) on each questionnaire. Past smokers reported when they last smoked on the baseline questionnaire. The time since quitting was also calculated for those who quit during follow-up. On the 1976 questionnaire, the NHS participants were asked at what age they started smoking. In the HPFS, each participant reported on the baseline questionnaire the average number of cigarettes smoked/day for each decade of life. From these data, categories were derived for age at start of smoking, based on the earliest decade where smoking was recorded, because there was no question (on the HPFS questionnaires) asking at what specific age the participant had initiated smoking.

Participants were asked about history of diabetes at baseline and in all of the subsequent questionnaires. In 1986 (HPFS) and in 1982 (NHS), and biennially thereafter, participants were asked about their history of cholecystectomy. BMI (kg/m²) was calculated from self-reported height and weight (asked at baseline).

In both cohorts, participants were asked to report specified medical conditions including cancers that were diagnosed in the 2-year period between each follow-up questionnaire. Whenever a participant (or next of kin for decedents) reported a diagnosis of pancreatic cancer, we asked for permission to obtain related medical records or pathology reports. If permission to obtain records was denied, we attempted to confirm the self-reported cancer with an additional letter or phone call to the participant. If the primary cause of death as reported by a death certificate was a previously unreported pancreatic cancer case, we contacted a family member to obtain permission to retrieve medical records or at least to confirm the diagnosis of pancreatic cancer. In the participants included in these analyses, 130 new cases of pancreatic cancer were diagnosed in the HPFS between 1986 and 1998, and 158 pancreatic cancer cases were diagnosed in the NHS between 1980 and 1996.

Of the surviving 121,700 NHS cohort members, 98,770 responded to the 1980 dietary questionnaire. We excluded participants with cancer (other than nonmelanoma skin cancer) diagnosed before baseline (1986 for the HPFS; 1980 for the NHS) and all of the individuals with implausibly high or low scores for total energy intake on the baseline questionnaires. As a result, 47,794 men and 88,799 women were eligible for analysis. For men and women, we computed person-time of follow-up for each participant from the return date of the baseline questionnaire to the date of pancreatic cancer diagnosis, death from any cause, or the end of follow-up (January 31, 1998 for men and June 30, 1996 for women), whichever came first. Incidence rates of pancreatic cancer were calculated by dividing the number of incident cases by the number of person-years in each category of dietary exposure. We computed the RR for each of the upper categories of intake by dividing the rates in these categories by the rate in the lowest category of intake.

For each cohort, RRs adjusting for potential confounders were estimated using pooled logistic regression analyses with 2-year time increments (18). This approach is asymptotically equivalent to the Cox regression model with time-dependent covariates, given short time intervals and low probability of the outcome within the interval (18). In these models, age was categorized into 5-year age groups, total energy intake and BMI were divided into quintiles, and cigarette smoking was categorized as follows (based on a previous analysis of these cohorts; Ref. 19): never-smoker, quit ≥15 years ago, quit <15 years ago and smoked ≤25 pack-years in past 15 years, quit <15 years ago and smoked >25 pack-years in past 15 years, current smoker with ≤25 pack-years in past 15 years, and current smoker with >25 pack-years in past 15 years. In addition, we controlled for history of diabetes and cholecystectomy because they have been previously associated with pancreatic cancer risk in other studies (20, 21, 22). We also repeated our analyses, updating the exposure of interest cumulatively, using data from follow-up questionnaires. For the cumulative updated analyses in the NHS, we used data from the 1980 questionnaire to allocate person-time to each of the quintiles of exposure between 1980 and 1984, the average of 1980 and 1984 for the interval from 1984 to 1986, and the average of 1980, 1984, and 1986 for the interval from 1986 to 1990, and so forth. Cumulative updating was similarly performed in the HPFS. Tests for trend were performed using continuous variables, and two-sided Ps are reported.

We pooled the data from the two cohorts using a random-effects model for the log of the RRs (23). Tests of heterogeneity using the Q statistic (23) were obtained for the main exposures (as continuous variables) to evaluate the overall trend in each cohort (and thus by gender) before pooling.

In the HPFS and the NHS cohorts, 288 incident cases of pancreatic cancer were documented during 1,907,222 person-years of follow-up. Compared with nondrinkers of coffee, men and women who had a high intake of coffee (four or more cups/day) were more likely to be current smokers, had higher lifetime cigarette use (pack-years), and drank more alcohol. However, heavy coffee drinkers were less likely to have a history of diabetes or cholecystectomy (Table 1). Similarly, men and women with a high alcohol intake (30 or more grams of alcohol/day) smoked more and drank more coffee but had a lower prevalence of diabetes or history of cholecystectomy (Table 1).

Consumption of caffeinated coffee was not associated with an increased risk of pancreatic cancer in the pooled analysis of the two cohorts. Among men in the HPFS, caffeinated coffee drinkers had a significantly lower risk of pancreatic cancer than those who did not drink any coffee, after adjusting for smoking and other pancreatic risk factors (P for continuous variable, 0.04; Table 2). Among women in the NHS, the multivariate RR for intake of four or more cups of coffee a day, compared with none, was 0.88 (95% CI, 0.56–1.38). The pooled RR between both cohorts was 0.62 (95% CI, 0.27–1.43) for four or more cups of regular coffee a day, compared with none (Table 2). Intakes of decaffeinated coffee and tea were not related to the risk of pancreatic cancer in either cohort (Table 2). Compared with nondrinkers of coffee or tea, the pooled RR was 0.93 (95% CI, 0.45–1.91) for four or more cups of decaffeinated coffee and 1.04 (95% CI, 0.71–1.54) for more than one cup of tea/day. Controlling for alcohol intake in these multivariate models did not alter the RRs presented in Table 2.

Total caffeine intake (which includes coffee, tea, other caffeinated drinks, and caffeine from food sources) was also examined in relation to the risk of pancreatic cancer. Compared with the bottom quintile of intake, the top quintile was associated with a weak inverse association in the men, which was not statistically significant (Table 2). In the NHS, no association was apparent between total caffeine intake and risk of pancreatic cancer (Table 2). The pooled RR of pancreatic cancer for the highest quintile of caffeine intake, compared with the lowest, was 0.82 (95% CI, 0.57–1.17).

We examined possible differences in the associations of specific coffee brewing methods with the risk of pancreatic cancer. The coffee brewing method used was asked on the 1990 questionnaires in both the cohorts, and, consequently, the follow-up periods for these analyses start in 1990. We observed no association with the type of brewing method (filtered, instant, or expresso), controlling for amount of coffee consumed (data not shown).

Caffeinated soda consumption was also examined in these cohorts. No significant association was observed for sugared or artificially sweetened caffeinated soda and the risk of pancreatic cancer (≥1 can of regular soda/day compared with none: multivariate RR, 1.32; 95% CI, 0.47–3.67 in the NHS; and RR, 1.19; 95% CI, 0.43–3.30 in the HPFS; ≥1 can of low calorie soda/day compared with none: multivariate RR, 0.83; 95% CI, 0.40–1.73 in the NHS; and RR, 0.67; 95% CI, 0.27–1.66 in the HPFS).

We examined the association between alcohol intake and pancreatic cancer in both cohorts. In the HPFS, the multivariate RRs of pancreatic cancer were nonsignificantly elevated for each category of alcohol intake compared with abstinence (Table 3). In the NHS, no association was observed for alcohol intake. In the pooled analysis of both cohorts, alcohol intake was not associated with risk of pancreatic cancer (continuous P, 0.94; Table 3). In addition to these analyses of intakes of total alcohol intake, we examined individual alcoholic beverages, including wine, beer, and liquor. No associations were observed between these specific beverages and risk of pancreatic cancer in either cohort (data not shown). Multivariate RRs for every additional drink/day in the pooled analyses were 1.08 (95% CI, 0.88–1.33) for beer, 0.91 (95% CI, 0.70–1.19) for wine, and 0.96 (95% CI, 0.81–1.15) for hard liquor. Additional control for coffee intake in the multivariate analyses did not change the RRs presented in Table 3.

We excluded from all of our analyses those men and women who abstained from alcohol and who had reported at baseline having substantially reduced their alcohol intake in the previous decade. In another analysis, we created two separate categories for current abstainers: those who reported past heavy drinking, and those who never drank alcohol. In both cohorts, past heavy alcohol drinking was not significantly associated with risk of pancreatic cancer (compared with never drinkers, multivariate RR, 1.62; 95% CI, 0.75–3.50; and multivariate RR, 1.11; 95% CI, 0.52–2.36 in the HPFS and NHS, respectively).

We performed alternative analyses in addition to the baseline analyses for the intakes of coffee and alcohol (Table 4). If recent intake of coffee or alcohol was of greatest etiological importance, then we would expect a stronger association in the analyses in which intake is updated every 4 years with the most recent food frequency questionnaires (simple update); however, this pattern was not observed for either coffee or alcohol. The cumulative update analyses average all of the previous FFQs for the subsequent follow-up period and minimize the measurement error for long-term average consumption. For these analyses, we observed similar estimates to those obtained using baseline diet (data not shown).

Although we excluded participants with cancer at baseline, individuals could have altered their dietary patterns because of preclinical manifestations of disease. To address this issue, we repeated our analyses (using baseline diet) after excluding the first 2 years of follow-up. Such lag analyses did not lead to substantially different findings for coffee or alcohol. Similarly, removing women in the NHS who reported having changed their coffee intake substantially in the previous decade (on the 1980 dietary questionnaire) did not change the findings for coffee intake (this question was not asked in the HPFS). In an additional analysis in the NHS, we added an 8-year lag period between diet (1980) and follow-up period (1988–1996), but we observed no association for either coffee or alcohol intake and risk of pancreatic cancer (data not shown).

We assessed the effect of consistent coffee or alcohol use over a 4-year induction period using two consecutive questionnaires (1986 and 1990 in the HPFS; 1980 and 1984 in the NHS). In each analysis, we excluded individuals who changed their frequency of coffee or alcohol intake and started follow-up after the second questionnaire was returned. No associations were observed using these analyses for coffee intake and risk of pancreatic cancer. For alcohol, the association was inverse when using consistent drinkers in the HPFS.

We examined intakes of coffee and alcohol stratified by age (<55 and ≥55 years of age), BMI (less than median and equal or above median), and smoking status (never, past, and current; Table 5). BMI, age, and smoking status did not modify the associations between coffee and pancreatic cancer. A nonsignificant elevation in risk of pancreatic cancer was observed among alcohol consumers less than 55 years old; however, no similar elevation in risk was observed among those 55 or more years old. Although RRs of pancreatic cancer were different by BMI strata for alcohol intake, these differences were not statistically significant, and no trend was apparent for the increased risk among those in the low BMI stratum. The alcohol intake and pancreatic cancer associations were similar for never-smokers and past smokers. Among current smokers, those men who drank more than 5 g of alcohol a day had an elevated risk of pancreatic cancer, but these associations were statistically nonsignificant (compared with <5 g of alcohol/day, multivariate RRs, 3.98; 95% CI, 0.8–18.9; and multivariate RR, 2.14; 95% CI, 0.4–10.5 for 5–14.9 g/day and ≥15 g/day, respectively). In the NHS, a nonsignificant increase in risk was also observed among current smokers drinking alcohol (RR, 1.65; 95% CI, 0.77–3.54 for 15 or more g/day of alcohol, compared with nondrinkers).

We found no statistically significant associations for intakes of coffee or alcohol and the risk of pancreatic cancer in men and women living throughout the United States. Our findings were similar whether we used baseline, current, or cumulative-averaged intakes of alcohol or coffee, as well as after we removed the first 2 years of follow-up or restricted the analyses to those with consistent intakes during two consecutive dietary questionnaires. Furthermore, no associations were observed for intakes of caffeine, decaffeinated coffee, tea, or different types of alcoholic beverages. Although there was a suggestion that alcohol may increase the risk of pancreatic cancer among current smokers, we had limited statistical power to examine this relationship.

The lack of an association between coffee intake and risk of pancreatic cancer in this study is consistent with most previous data. Reviewing the available studies that have assessed the influence of coffee or alcohol consumption (24, 25), the IARC concluded that there was little evidence to support a causal relation between coffee or alcohol and risk of pancreatic cancer.

Because we had repeated dietary assessments over time, we were able to examine the effects of recent and long-term coffee intake and confirm that neither one yielded a significant effect on pancreatic cancer. In contrast to the recent prospective study of women in Iowa, we did not confirm a positive association between coffee intake and pancreatic cancer in our cohort of women (4). In addition, no positive association was observed in the age strata of 55 years and older, which most closely resembles the ages of women found in the Iowa Study (4). The RR reported for 18 or more cups of coffee a week in the Iowa study (RR, 2.15; 95% CI, 1.08–4.30) is significantly different to our finding for a similar coffee intake (RR, 0.62; 95% CI, 0.27–1.43). The IWHS findings (4) are based on a total of 66 cases of pancreatic cancer. In contrast, our results are based on 288 cases (158 women). The IWHS also reported an elevated risk of pancreatic cancer among alcohol drinkers; however, this finding was not statistically significant, and it was statistically compatible with our findings. Given that the elevated risks observed in the IWHS were also present in analyses restricted to never-smokers, it is unlikely that inadequate control of smoking is responsible for their results. Although a FFQ very similar to ours was used in the IWHS, alcohol intake appeared to be poorly reported in the IWHS (r = 0.32 in validation study), and no validation information was provided for coffee intake (4). Given the very high correlations observed in our validation studies for alcohol and coffee, it is unlikely that measurement error is responsible for the lack of associations in this study, although this cannot be ruled out completely. A longer follow-up with additional cases should help elucidate whether the findings of the IWHS are the result of chance.

In this study, we had data on regular versus decaffeinated coffee and also on the method typically used for brewing coffee. Because the brewing method may affect the concentration of certain compounds found in coffee, such as methylglyoxal (26) and caffeine (27), it may be an important aspect of coffee intake. In vitro studies show that caffeine can inhibit DNA repair (28, 29) and induce mitotic events before DNA replication is completed (30). Human physiological studies indicate that moderate caffeine consumption is unlikely to have an effect on health (27). The brewing method was not related to risk of pancreatic cancer, and we did not observe any effect of total caffeine intake (calculated from all of the caffeinated beverages and foods).

In previous prospective studies, findings have been less consistent for alcohol than for coffee intake. Prospective studies that have reported a relation between alcohol (or an alcoholic beverage) and pancreatic cancer tended to have longer follow-up periods (8, 12, 17, and 20 years; Refs. 4, 5, 12, 13) than those studies reporting no associations (6, 9, and 12 years; Refs. 6, 7, 8). Because dietary exposures in case-control studies mostly reflect recent diet, the lack of association for alcohol intake might suggest that alcohol plays a role at an early stage in carcinogenesis (e.g., initiation). However, our data do not support an early or late effect of alcohol intake on risk of pancreatic cancer. Past heavy users of alcohol were not at an elevated risk of pancreatic cancer in either the NHS or the HPFS cohorts. Moreover, allowing for an 8-year induction period did not demonstrate any significant association with alcohol intake.

In a previous case-control study (31), an interaction was observed between smoking status and alcohol intake such that alcohol drinking did not increase risk of pancreatic cancer among nonsmokers or low-dose smokers, but it did substantially increase risk among cigarette smokers. Alcohol intake was not associated with risk of pancreatic cancer overall in this study, but RRs were nonsignificantly elevated in both cohorts among current smokers drinking alcohol, compared with smokers not drinking. Because of the relatively small number of cases in this subset of participants and, thus, lack of statistical power, these findings should be interpreted with caution, and we cannot exclude a moderate increase in these groups.

The null relationship between tea intake and risk of pancreatic cancer is consistent with data from most prospective studies (4, 5, 6, 7, 32, 33), with the exception of one study that reported a significant inverse association (34). Similarly, case-control studies have consistently found no association between tea and pancreatic cancer risk (3).

The lack of associations reported here are unlikely to be the result of misclassification of exposure. Validation studies of the dietary assessment questionnaires (FFQ) used in these two cohorts have demonstrated that coffee and alcohol are highly correlated with reports from detailed dietary records (15, 16). In addition, we observed similar results in those analyses that used the average of the multiple dietary assessments, minimizing measurement error. Furthermore, findings were similar after restricting analyses to those who reported consistent coffee or alcohol intake over consecutive questionnaires. Given the high response rates obtained in the NHS and HPFS cohorts, neither nonresponse rates nor differential follow-up are likely to account for our observations.

In conclusion, the risk of pancreatic cancer was not elevated with increasing consumption of alcohol, coffee, decaffeinated coffee, or tea in two large prospective cohort studies of men and women living in the United States. Specifically, the associations did not change whether examining distant or recent past intakes of alcohol or coffee or when allowing for a lag between diet assessment and cancer diagnosis. Similarly, total caffeine intake from beverages and diet was not associated with the risk of pancreatic cancer. With fewer than 100 pancreatic cancer cases who are current smokers, our findings are still inadequate to offer a strong conclusion for the potential interaction between smoking and alcohol. Ultimately, our data are consistent with conclusions of the IARC Working Group (24, 25) that neither coffee nor alcohol intakes are associated with an increase in pancreatic cancer risk.

We thank the participants of the HPFS and the NHS for their continuing outstanding dedication and commitment to the study and Alvin Wing, Kathleen Markham, and Mildred Wolff for their assistance.