Esophageal cancers are the sixth most common cancers

Transcription

1 GASTROENTEROLOGY 2009;136: Alcohol Consumption and the Risks of Adenocarcinoma and Squamous Cell Carcinoma of the Esophagus NIRMALA PANDEYA,*, GAIL WILLIAMS, ADÈLE C. GREEN,* PENELOPE M. WEBB,* and DAVID C. WHITEMAN* for the Australian Cancer Study *Queensland Institute of Medical Research, Brisbane, Australia; and the School of Population Health, The University of Queensland, Brisbane, Australia This article has an accompanying continuing medical education activity on page Learning Objective: Upon completion of this CME activity, successful learners will be able to differentiate the effects of alcohol consumption on the risks of different histologic subtypes of esophageal cancer. Successful learners will also be able to describe the way in which tobacco smoking modifies the effects of alcohol with respect to risk of esophageal cancer. See CME quiz on page 1444; see editorial on page Background and Aims: Alcohol has been declared a carcinogen for cancers of the esophagus, although the evidence relates largely to the squamous subtype. Evidence for an effect on adenocarcinomas is scant and inconsistent. Methods: We compared nationwide samples of patients with esophageal adenocarcinoma (EAC) (n 365) or esophagogastric junction adenocarcinoma (EGJAC) (n 426) or esophageal squamous cell carcinoma (ESCC) (n 303) with controls sampled from a population register (n 1580). We used generalized additive models to assess nonlinear effects of self-reported alcohol intake on cancer risk, and calculated odds ratios (ORs) and 95% confidence intervals (CIs) using multivariate logistic and piecewise regression. Results: We observed no association between average weekly alcohol intake and EAC or EGJAC risk. For ESCC, the relationship with alcohol was nonlinear. At intakes of less than 170 g/wk there was no significant association; at greater than this level, there was a significant linear effect (OR, 1.03; 95% CI, per 10 g alcohol/wk). For ESCC, but not EAC or EGJAC, a statistically significant multiplicative interaction between smoking and alcohol was observed (P.02). In analyses by beverage type, ESCC risks, but not EAC or EGJAC, increased linearly with beer intake (OR, 1.05; 95% CI, ). Those who drank modest levels of wine (<50 90 g/wk) or port or spirits (<10 20 g/wk) had significantly lower risks of all 3 cancers than nondrinkers; higher intakes were associated with increased risks of ESCC only. Conclusions: Alcohol intake above the recommended US dietary guidelines significantly increases the risk of ESCC, but not EAC or EGJAC. Smoking modifies the effect of alcohol intake on ESCC risk. Esophageal cancers are the sixth most common cancers worldwide. The very high mortality associated with these cancers 1 coupled with the rapidly increasing incidence of the adenocarcinoma subtype in Western populations accentuates their public health importance. Recent changes in the incidence and distribution of esophageal cancers suggest that the prevalence of exposure to the causal factors has changed. 2 4 Alcohol is classified as a class 1 carcinogen by the International Agency for Research on Cancer, 5 and high levels of alcohol consumption have long been associated with increased risks of esophageal squamous cell carcinomas (ESCC). 6 8 The reported effects of alcohol on esophageal adenocarcinoma (EAC) and esophagogastric junction adenocarcinoma (EGJAC), however, have been less consistent. Although some investigators have found no association, 8 11 others have reported significantly increased risks of EAC associated with alcohol. 12,13 Those few studies investigating associations with specific types of alcoholic beverages (beer, wine, and so forth) and adenocarcinoma risk also have shown inconsistent patterns of association. For example, although most studies have reported no association between beer intake and EAC risk, consumption of hard spirits has been reported to increase significantly, 13 decrease, 11 and not change the risk of EAC. 10,14 The few studies of wine consumption and adenocarcinomas also have reported discordant findings. 10,14 A common approach to the analysis of alcohol consumption in cancer studies has been to categorize participants based on broad measures of alcohol intake. Such an approach has the advantage of deriving an easily interpretable measure of relative risk, and makes no as- Abbreviations used in this paper: CI, confidence interval; EAC, esophageal adenocarcinoma; EGJAC, esophagogastric junction adenocarcinoma; ESCC, esophageal squamous cell carcinoma; OR, odds ratio by the AGA Institute /09/$36.00 doi: /j.gastro

2 1216 PANDEYA ET AL GASTROENTEROLOGY Vol. 136, No. 4 sumption about any underlying trends or patterns of risk. However, categorization may obscure potentially important differences in risk within and across categories of alcohol intake, particularly if the cut-off points for categorization span biologically important thresholds. 15 Modeling the risk of cancer associated with alcohol intake as a continuous measure by using standard linear regression models avoids the potential loss of information owing to categorization, but it constrains the dose effect to be linear. Again, such constraints may mask important differences in risk associated with different levels of alcohol intake. As has been well described, alcohol consumption has a J-shaped relationship with allcause mortality in human beings and there is no reason to discount the possibility that similar nonlinear patterns of risk may be observed for subtypes of esophageal cancer. In this study, we aimed to compare the effects of different patterns of alcohol consumption on the risks of EAC, EGJAC, and ESCC. In particular, we sought to explore the association between alcohol intake and cancer risk without imposing predetermined constraints on the data. Materials and Methods Study Design and Participants We used data from a nationwide case-control study of esophageal cancer conducted in Australia, the details of which have been described in full elsewhere. 19 In summary, eligible case patients were people aged years with a histologically confirmed primary invasive cancer of the esophagus or esophagogastric junction diagnosed between July 1, 2002 (July 1, 2001, in Queensland), and June 30, 2005, in the mainland states of Australia. Patients were recruited either through major treatment centers or state-based cancer registries. A total of 1577 patients with esophageal cancer received an invitation to participate in the study, of whom 1102 patients (858 through clinics and 244 through cancer registries) returned a completed questionnaire (70% of all invited; 35% of all living and deceased persons in mainland Australia who had been diagnosed with incident esophageal cancer). Eight case patients subsequently were deemed ineligible on pathology review and were excluded from the analysis. Final numbers of case participants by histologic types for the present analysis were 365 EAC, 426 EGJAC, and 303 ESCC patients. Potential controls were selected randomly from the Australian Electoral Roll within strata of age (in 5-year age groups) and state of residence to match the distribution of the case series. We aimed for similar numbers of male cases and controls in each stratum of age and state; female controls were oversampled intentionally at all ages to accommodate their simultaneous enrollment in a parallel case-control study of ovarian cancer. 20 Of 3258 potentially eligible control participants who were contacted and invited to participate, 175 were excluded (16 deceased, 61 were too ill, and 98 were unable to read or write in English), and 41 were lost to follow-up evaluation in the interval between initial contact and participation. Of 3042 remaining controls, 1680 (55%) accepted the initial invitation and 1580 returned the completed questionnaires (51% of all potentially eligible controls contacted). Data Collection Demographic information and health survey. Participants self-completed a health and lifestyle questionnaire asking about their education, general health, smoking history, frequency of symptoms of heartburn (a burning pain behind the breastbone after eating), or acid reflux (a sour taste from acid or bile rising up into the mouth or throat), and use of aspirin in the past 5 years. Height and weight 1 year ago (1 year before diagnosis for cases) were elicited to enable calculation of body mass index. Alcohol intake. Participants were asked whether they currently drank alcohol, were life-long nondrinkers, or used to drink alcohol in the past but had stopped. For those who had ever consumed alcohol, we asked the age at which they first started drinking alcohol at least once a month and the age that they stopped drinking (if they had stopped). Participants then were asked to report the number of alcoholic drinks that they usually consumed each week during the age intervals of years, years, and 50 years and older. This was asked separately for 6 classes of alcohol, namely reduced-alcohol beer, regular beer, white wine, red wine, port/sherry, and spirits/liqueurs. The typical weekly intake for each type of alcoholic beverage was recorded on an ordinal scale as none, less than 1, 1, 2 4, 5 6, 7 13, 14 20, 21 27, and 28 or more drinks per week. One drink was defined as a small bottle or can of beer, a medium glass of wine, a small glass of port/sherry, or a nip of spirits/liqueur. Statistical Methods Derivation of alcohol consumption variables. Within each age interval, the average weekly alcohol intake (in grams) for each type of beverage was estimated by multiplying the midpoint value for each intake frequency by the standard drink volume and percentage of alcohol content by weight (in grams). We used the following percentages of alcohol content for each class of beverage: 2.1 g, 3.8 g, 8.3 g, 9.6 g, 14.1 g, and 29.3 g per 100 ml of reduced-alcohol beer, regular beer, white wine, red wine, port/sherry, and spirits/liqueur, respectively. The total weekly alcohol intake at each age interval was calculated by summing the beverage-specific total intakes. The totals for all age intervals then were summed to obtain the total lifetime alcohol intake. We derived a measure of the average weekly consumption over adult

3 April 2009 ALCOHOL AND ESOPHAGEAL CARCINOMAS 1217 life (hereafter referred to as the average weekly alcohol intake) by dividing the total lifetime alcohol intake by the number of weeks in life starting from age 20 to their reference age (age at diagnosis for cases and age at study participation for controls). We used similar algorithms to calculate the average weekly intake of each type of alcoholic beverage. Analytic approach. Statistical analyses were performed for each alcohol variable in 2 stages. First, exploratory analyses were performed on continuous measures of alcohol consumption using generalized additive logistic models adjusting for the potentially confounding variables of age, sex, educational level, smoking, body mass index 1 year before diagnosis (or at initial interview for controls), frequency of reflux and heartburn within the age band 10 years ago, and frequency of aspirin use in the past 5 years. A generalized additive logistic model is an extension of the generalized linear model in which linear predictors are used as functions of the predictor variable(s), allowing a natural fit to the data. Smoothing splines were used here to look for evidence of nonlinearity in the dose patterns of alcohol. We used smoothing splines fixed at 3 degrees of freedom (unpenalized models) to test for significance of nonlinearity against the linear effect. Generalized additive logistic model analyses were performed using the CRAN (Vienna, Austria) package mgcv 21 in R software. 22 For the alcohol variables in which dose patterns were significantly nonlinear, piecewise linear regression was used to obtain multiple slope parameters separated by a change point where the slope changed direction (ie, where smoothed plots showed one significant transition point). 23 The change point was estimated based on the maximum likelihood approach 24 for each evaluation. Last, we performed standard categoric analyses using approximate quintile cut-off points among the control distribution to define intake categories. Risks of cancer were calculated relative to the reference category (never drinkers). Thus, for the average total alcohol intake, participants were categorized as never drinkers, less than 10 g/wk, g/wk, g/wk, g/wk, and 420 g/wk or greater. These categories correspond to less than one standard alcoholic drink per week (where 1 standard alcoholic drink 10 g ethanol by weight), drinks/wk, drinks/day, drinks/day, and 6 or more drinks/day. For both the piecewise regression and the standard categoric analysis, we used multivariate logistic regression (GENMOD procedure) in SAS (version 9.1; SAS Institute, Inc, Cary, NC) to calculate odds ratios (ORs) and 95% confidence intervals (95% CIs). Table 1. Demographic Characteristics of Controls and Cases of EAC, EGJAC, and ESCC Controls (n 1574) EAC (n 362) EGJAC (n 423) ESCC (n 304) Variables Categories n (%) n (%) n (%) n (%) Age, y Mean (SE) 60.5 (0.3) 63.6 (0.5) 63.3 (0.5) 64.7 (0.5) Sex F 538 (34.2) 35 (9.7) 56 (13.2) 131 (43.1) M 1036 (65.8) 327 (90.3) 367 (86.8) 173 (56.9) Further studies None 642 (40.9) 167 (46.4) 170 (40.4) 172 (56.9) Technical college/diploma 347 (22.1) 74 (20.6) 105 (24.9) 53 (17.6) Trade certification/apprenticeship 339 (21.6) 97 (26.9) 101 (24.0) 50 (16.6) University 242 (15.4) 22 (6.1) 45 (10.7) 27 (8.9) Alcohol Nondrinker 176 (11.2) 26 (7.2) 34 (8.0) 39 (12.8) Current drinker 1291 (82.0) 297 (82.5) 337 (79.7) 236 (77.6) Ex-drinker 107 (6.8) 39 (10.8) 52 (12.3) 29 (9.5) BMI last year (36.4) 71 (20.4) 107 (26.2) 165 (57.5) (43.0) 148 (42.5) 168 (41.1) 80 (27.9) (20.7) 129 (37.1) 134 (32.8) 42 (14.6) Heartburn or acid reflux 10 y before diagnosis or interview Never 694 (44.4) 80 (22.3) 118 (28.0) 143 (47.7) Once a month 465 (29.7) 43 (12.0) 63 (15.0) 39 (13.0) A few times a month 221 (14.1) 85 (23.7) 87 (20.7) 40 (13.3) A few times a week 127 (8.1) 82 (22.8) 97 (23.0) 43 (14.3) Daily 57 (3.6) 69 (19.2) 56 (13.3) 35 (11.7) Smoking status Never smoker 710 (45.6) 91 (25.2) 97 (23.0) 77 (25.5) Ex-smoker 208 (13.4) 73 (20.2) 115 (27.3) 93 (30.8) Current smoker 640 (41.1) 197 (54.6) 210 (49.8) 132 (43.7) Frequency of aspirin use in the past 5 years Never 693 (44.1) 169 (47.1) 191 (45.6) 134 (45.0) Once a month 528 (33.6) 95 (26.5) 131 (31.3) 101 (33.9) Once a week 140 (8.9) 45 (12.5) 32 (7.6) 28 (9.4) Weekly or more 209 (13.3) 50 (13.9) 65 (15.5) 35 (11.7) NOTE. Some variables have missing values. Only observations with complete data on all confounding factors were used in the regression analysis (39 controls, 20 EAC, 20 EGJAC, and 27 ESCC cases had at least one of the observations missing and hence were excluded from multivariate regression analysis). There was no difference in the alcohol consumption of these 106 missing observations compared with those with complete observations. BMI, body mass index.

4 1218 PANDEYA ET AL GASTROENTEROLOGY Vol. 136, No. 4 Results Demographic characteristics of cases and controls are described in Table 1. As anticipated, higher proportions of cases were ever smokers and reported having had reflux symptoms. The prevalence of overweight and obesity was higher among EAC and EGJAC cases than controls or ESCC cases. Distribution of Alcohol Intake Among controls, 6% of men and 21% of women reported that they had never consumed alcohol (Table 1); among ever drinkers, the median weekly alcohol intake over their lifetime was 127 and 41 g for men and women, respectively. For male controls who drank, a substantial proportion of the lifetime alcohol intake was the result of beer consumption (median intake of alcohol through beer: men, 76 g/wk; women, 0 g/wk), whereas for women, wine was the major contributor to total alcohol intake (median alcohol intake through wine: women, 17 g/wk; men, 13 g/wk). Median intakes of spirits were considerably lower than intakes of beer or wine in all groups of cases and controls, and intakes of port or fortified wines were negligible (Table 2). Total Alcohol Intake and the Risk of Cancer: Continuous Models We first estimated risks of each cancer associated with average weekly alcohol intake as a continuous measure using both linear and nonlinear regression functions. (Never drinkers were included and assigned a value of 0 g/wk). Although alcohol consumption differed substantially between men and women, the effects were similar when stratified by sex (data not shown) so the combined results are presented. We found no evidence of an alcohol dose effect for either EAC or EGJAC (Figure 1A and B), however, there was a statistically significant nonlinear dose relationship between alcohol consumption and risk of ESCC (P.01 testing for departure from linearity) (Figure 1C). We then used a piecewise regression model with a change point estimated at 170 g/wk (95% CI, g/wk) to explore this relationship further. The slopes for the lines below and above the change point value were significantly different. Below the change point, there was no association between alcohol intake and risk of ESCC (OR, 0.98; 95% CI, per 10 g/wk) (Table 3). Above this threshold, however, we observed a significant 3% increase in risk of ESCC for each extra 10 g/wk of alcohol intake thereafter (Table 3). The apparent flattening of the curve at higher intakes of alcohol was estimated imprecisely owing to small numbers of participants, and was not statistically significant. Although current drinking status was not associated independently with any of these cancers, when we analyzed the continuous dose effect of alcohol separately among current and former drinkers, the highest risks for Table 2. Distribution of Average Weekly Alcohol Intake (Lifetime) Stratified by Sex Among Esophageal Cancer Patients and Matched Population Controls in Australia in Men Women Controls EAC EGJAC ESCC Controls EAC EGJAC ESCC n 1036 n 327 n 367 n 173 n 538 n 35 n 56 n 131 Alcohol characteristics (amounts: g/wk alcohol) Ever drank alcohol 93.7% 93.6% 94.6% 96.0% 79.4% 85.7% 75.0% 75.6% Age started drinking Median 18 y (17, 20) 18 y (17, 20) 18 y (17, 20) 18 y (17, 19) 19.0 y (18, 25) 20.0 y (18, 25) 19.0 y (17, 20) 20.0 y (18, 24) (25th, 75th percentiles) Duration of drinking Median 39.7 y (31.4, 46.8) 39.9 y (33.2, 47.0) 40.1 y (32.9, 46.5) 41.5 y (33.9, 49.0) 30.1 y (20.0, 38.7) 34.2 y (26.0, 47.7) 34.3 y (29.0, 42.2) 38.7 y (30.7, 45.2) (25th, 75th percentiles) Total alcohol Median (25th, (58.2, 233.7) (75.0, 272.1) (60.3, 255.0) (124.3, 408.1) 41 (15.8, 83.4) 27.9 (15.2, 72.0) 45.2 (17.9, 81.2) 50.6 (23.8, 142.4) 75th percentiles) 76.5 (28.1, 173.5) (37.4, 204.6) 75.7 (34.5, 200.8) (68.6, 330.9) 0.0 (0, 8.2) 0.0 (0, 8.5) 0.0 (0, 10.5) 0.0 (0, 48.4) Beer Median (25th, 75th percentiles) 13.3 (0, 42.7) 4.1 (0, 33.1) 6.0 (0, 34.9) 0.0 (0, 34.5) 17.0 (4.9, 43.6) 7.0 (0, 22.7) 18.3 (5.6, 46.9) 11.7 (0, 41.3) Wine Median (25th, 75th percentiles) 0.0 (0, 4.1) 0.0 (0, 3.4) 0.0 (0, 2.6) 0.0 (0, 0) 0.0 (0, 3.6) 0.0 (0, 4.0) 0.0 (0, 0.7) 0.0 (0, 0.9) Port/sherry Median (25th, 75th percentiles) 3.6 (0, 14.6) 1.6 (0, 15.8) 1.6 (0, 21.2) 0 (0, 13.3) 4.0 (0, 13.7) 0 (0, 13.6) 4.7 (0, 14.1) 0 (0, 7.0) Spirits/liqueur Median (25th, 75th percentiles) Note: Alcohol intake are described only for ever drinkers. Alcohol data were missing for a small number.

5 April 2009 ALCOHOL AND ESOPHAGEAL CARCINOMAS 1219 ESCC were observed among current drinkers with intakes greater than 170 g/wk (not shown). When stratified by smoking status, we observed significantly different dose-effect patterns between alcohol intake and ESCC risk (Supplementary Figure 1). Although the effects of alcohol among never and current smokers did not depart from linearity, the magnitudes of the effects were significantly different from each other (P.02). ESCC risk increased by 3% per 10 g/wk of alcohol intake (OR, 1.03; 95% CI, ) among never smokers but by 8% (OR, 1.08; 95% CI, ) among current smokers. Among ex-smokers, however, we observed a 6% reduction in risk per 10 g/wk (OR, 0.94; 95% CI, ) for those who had consumed less than 170 g/wk (95% CI, g/wk). Above this value, the risks of ESCC increased by 2% (OR, 1.02; 95% CI, ) for each additional 10 g/wk of total alcohol intake. We performed additional analyses stratified by frequency of heartburn or reflux symptoms, and found similar patterns of association with alcohol across all strata (data not shown). Beverage-Specific Intake and the Risk of Cancer: Continuous Models We found no evidence of any association (linear or nonlinear) between average lifetime beer intake and risks of EAC or EGJAC. Beer intake was associated linearly with ESCC (Supplementary Figure 2A), with a significant 5% increase in risk per additional 10 g/wk (Table 3). Although the prevalence and intakes of beer were substantially higher in men than women, we observed similar risks when the analysis was restricted to women. For wine, port/sherry, and spirits, we observed significant nonlinear associations between the average weekly intake and each type of cancer (Supplementary Figure 3). The plots suggested that the risks of EAC, EGJAC, and ESCC were reduced significantly among those with very low intakes of sherry or liqueur ( 10 g/wk) and low to moderate intake ( 90 g/wk) of wine. Higher levels of intake of wine and spirits were associated with significantly increased risks of ESCC, but not EAC or EGJAC (Table 3). Qualitatively similar effects were observed within strata of sex (data not shown), however, the risk estimates were no longer statistically significant among women owing to small numbers. In further analyses, we found that for EGJAC and ESCC, similar patterns of nonlinear associations were observed with red and white wine. For EAC, however, we 4 Figure 1. Plots for dose effect (in logit scale) of average lifetime alcohol consumption on (A) EAC, (B) EGJAC, and (C) ESCC using a generalized additive model with 3 degrees of freedom smoothing spline function adjusted for age, sex, body mass index (BMI) 1 year ago, frequency of heartburn or acid reflux, education, frequency of aspirin use in the past 5 years, and smoking.

6 1220 PANDEYA ET AL GASTROENTEROLOGY Vol. 136, No. 4 Table 3. Relative Risks of Cancers of the Esophagus Associated With Average Weekly Alcohol Intake (Lifetime): Piecewise Regression Analysis EAC EGJAC ESCC Average weekly alcohol intake (lifetime) OR 95% CI OR 95% CI OR 95% CI Total alcohol 170 g/wk 0.98 ( ) 0.98 ( ) 0.98 ( ) 170 g/wk 1.00 ( ) 0.99 ( ) 1.03 ( ) Test for difference in slopes P.23 P.15 P.001 Beer Linear dose effect with a single slope 0.99 ( ) 1.00 ( ) 1.05 ( ) Wine Optimal change point for segments 50 g 90 g 60 g First segment 0.84 ( ) 0.92 ( ) 0.79 ( ) Second segment 1.01 ( ) 1.01 ( ) 1.03 ( ) Test for difference in slopes P.001 P.005 P.001 Port/sherry Optimal change point for segments 10 g 10 g 10 g First segment 0.25 ( ) 0.21 ( ) 0.13 ( ) Second segment 1.02 ( ) 0.93 ( ) 0.95 ( ) Test for difference in slopes P.001 P.001 P.001 Spirits/liqueur Optimal change point for segments 10 g 20 g 10 g First segment 0.31 ( ) 0.69 ( ) 0.21 ( ) Second segment 1.02 ( ) 1.00 ( ) 1.03 ( ) Test for difference in slopes P.001 P.001 P.001 NOTE. The ORs presented are the change in risk per additional 10 g/week of alcohol intake. observed a significant nonlinear risk reduction associated with modest intakes of red wine ( 40 g/wk: OR, 0.78; 95% CI, ; 40 g/wk: OR, 1.02; 95% CI, ), but not white wine ( 40 g/wk: OR, 0.90; 95% CI, ; 40 g/wk: OR, 0.99: 95% CI, ). We also compared risks of esophageal cancer among the small proportion of people ( 11%) who drank either red wine only or white wine only. Although the numbers were small, there was no statistical evidence that the patterns of association differed by type of wine. We also considered whether the effects associated with one type of alcoholic beverage may have been confounded by the effects of other types of alcohol. We observed low correlations between intakes of the various types of beverages (highest correlation was 0.20 between wine and spirits), and the associations remained essentially unchanged when the effects of beverage-specific intakes were adjusted for alcohol intake from other types of beverages (not shown). Total Alcohol Intake and the Risk of Cancer: Categoric Models We proceeded to categoric analyses using the cutoff point derived from the piecewise regression model to derive risk estimates for intakes greater than 170 g/wk compared with never drinkers (EAC: OR, 0.72; 95% CI, ; EGJAC: OR, 0.73; 95% CI, ; ESCC: OR, 2.27: 95% CI, ). We repeated these analyses using cut-off points from the literature for comparability (Table 4). No category of total alcohol intake was associated with increased risks of EAC or EGJAC (Table 4). Similar to the regression analyses, we found no evidence that modest alcohol intakes conferred any excess risk of ESCC compared with never drinkers. At higher intakes, risks of ESCC increased significantly to more than 4-fold for those who drank 420 g/wk or more (OR, 4.7; 95% CI, ). We repeated the categoric analyses with an interaction term for alcohol intake and smoking status, which was significant for ESCC only (P interaction.02). In stratified analyses, we found no evidence that total alcohol intake was associated with EAC or EGJAC for any of the smoking strata. For ESCC, the risks associated with alcohol intakes differed significantly by smoking status (Table 5). Average lifetime alcohol intake of 210 g/wk or greater was associated with significantly increased risks among former and current smokers. Highest risks were observed among current smokers whose average alcohol intake was 420 g/wk or more (OR, 21.9; 95% CI, ). Discussion By using a variety of analytic techniques, we have explored the effects of alcohol intake on the 2 most common histologic subtypes of cancer occurring in the esophagus or esophagogastric junction. Regardless of approach, we found no evidence that total alcohol intake was associated with increased risks of esophageal or junctional adenocarcinomas. In contrast, we found significant nonlinear associations between alcohol intake and

7 April 2009 ALCOHOL AND ESOPHAGEAL CARCINOMAS 1221 Table 4. Risk of Cancers of the Esophagus Associated With Lifetime Average Alcohol Consumption: Overall and Stratified by Sex Group Average lifetime alcohol intake/wk EAC EGJAC ESCC ORs 95% CI ORs 95% CI ORs 95% CI All Never drinkers 1.00 Reference 1.00 Reference 1.00 Reference 10 g g g g g P value (trend test) Men Never drinkers 1.00 Reference 1.00 Reference 1.00 Reference 10 g g g g g P value (trend test) Women Never drinkers 1.00 Reference 1.00 Reference 1.00 Reference 10 g g g P value (trend test) ORs were adjusted for age, sex, body mass index 1 year ago, frequency of heartburn or acid reflux, education, frequency of aspirin use in the past 5 years, and smoking. risk of squamous cell carcinomas of the esophagus. At intakes of less than 170 g/wk, there was no association between alcohol intake and ESCC, but thereafter risks increased in a linear fashion with increasing intakes of alcohol. Our approach to studying the effects of alcohol on esophageal cancer differs somewhat from previous reports 8,10 13 because we did not commence with categoric analyses to estimate the effects within broad classes of exposure. Instead, we applied a series of models and graphic applications so as to not constrain the relationship between alcohol and cancer risk to be linear, and not impose arbitrary boundaries on levels of alcohol intake. The output from these approaches suggests that people Table 5. Risk of Cancers of the Esophagus Associated With Lifetime Average Alcohol Consumption: Stratified by Smoking Status Current smoking status Average lifetime alcohol intake/wk EAC EGJAC ESCC ORs 95% CI OR 95% CI ORs 95% CI Never smokers Never drinkers 1.0 Reference 1.0 Reference 1.0 Reference 10 g g g g g Former smokers Never drinkers 1.0 Reference 1.0 Reference 1.0 Reference 10 g g g g g Current smokers Never drinkers 1.0 Reference 1.0 Reference 1.0 Reference 10 g g g g g ORs adjusted for age, sex, body mass index 1 year ago, frequency of heartburn or acid reflux, education, and frequency of aspirin use in the past 5 years.

8 1222 PANDEYA ET AL GASTROENTEROLOGY Vol. 136, No. 4 who consume alcohol at modest levels have no detectable increase in the risk of any cancers of the esophagus. These findings therefore largely accord with the majority of earlier investigations, which have tended to report no increased risks of EAC among those who drink alcohol. 8 10,14 We also found that risks of ESCC were increased only among those who consumed alcohol at levels greater that 170 g/wk, a finding that parallels some earlier studies 8,11,14 but not all. 25 A strength of the present study was the large sample sourced from the Australian population. This enhanced the precision of risk estimates and allowed us to perform stratified analyses to assess effect modification. The large sample size also reduced the likelihood that the observed null associations were the result of a type 2 error. We elicited comprehensive measures of alcohol intake spanning 3 discrete periods of adult life, ensuring that changes in drinking patterns were recorded and integrated into measures of overall alcohol intake. Moreover, we captured intakes of different types of alcohol to assess the possibility of type-specific effects. Because neither participants nor interviewers were informed of the specific hypotheses of the study, we consider differential recall of alcohol consumption between cases and controls to be low. The qualitatively and quantitatively different patterns of risk for adenocarcinomas and squamous cell carcinomas, and the complexity of the nonlinear associations, suggest that biased recall is unlikely to account for all of the observed effects. A limitation of the study was relatively low rates of participation, raising concerns about possibly biased selection of cases and controls. The cases who participated were very similar in age and sex distribution to those registered in the Australian National Cancer Statistics Clearing House (2002), however, further details about the characteristics of nonparticipating cases were not available because of Australian privacy laws. Bias might explain our findings if alcohol intake was associated with survival and hence study participation. Although no information was available for nonparticipants, we were able to compare survival within categories of alcohol intake for each group of cancer patients in our sample. For patients with EAC and EGJAC, we found no evidence that survival differed according to drinking status or amount, but for ESCC, never drinkers had significantly longer survival than former and current drinkers. If patients with long survival were more likely to take part in the study (as might be expected), then our study may have underestimated alcohol consumption among ESCC cases, and thus underestimated the association between alcohol and ESCC. Bias also might arise if the participating controls differed from the source population with respect to their intakes of alcohol. The strikingly different patterns of total and type-specific alcohol intakes for men and women accord with estimates of recent drinking behaviors from a national survey, 26 although the proportion of never drinkers was slightly higher among our controls (11% vs 7.7%). Because the prevalence of other factors known to be associated with EAC, such as obesity and acid reflux, was similar in our control series to other Australian studies, 27,28 we believe the likelihood of biased selection to be no different from previous studies. Assuming the associations reported here are not the result of error, then the question arises as to whether the reduced occurrence of esophageal cancers in those with low to moderate levels of wine consumption (and possibly of other beverages) are causal, or whether the effect is explained by other factors confounding the association, either unmeasured or unknown, or imperfectly adjusted for in the analyses. A biological effect might be plausible because moderate wine drinkers also have been found to have lower risks of lung and other cancers. 29 This finding also accords with several earlier reports, 11,14 as well as 2 community studies that have observed inverse associations between wine intake and Barrett s esophagus and/or esophageal adenocarcinoma (unpublished data author communication 2008, Kubo et al, and Anderson et al). Recently, there has been speculation that constituents in wine may confer protection against carcinogenesis. Resveratrol (3,5,4 -trihydroxy-trans-stilbene), a polyphenol found in grape-skins and wine, has been advanced as one such cancer chemopreventive agent. 30 In vivo and in vitro experiments suggest that resveratrol has anticancer activity across a spectrum of biologic pathways, 31 and this might explain why moderate wine drinkers apparently enjoy lower risks of cancer. 32 Arguing against this explanation was our finding that similarly reduced risks of cancer were observed for those who drank modest levels of white wine as red wine. Residual confounding remains an equally likely explanation for this particular finding because moderate wine drinkers differ from heavy drinkers and nondrinkers across a range of lifestyle and behavioral factors, including sex, education, income, smoking, and body mass index. Although we carefully adjusted for each of these factors, and performed stratified analyses to assess effect modification, it is difficult (if not impossible) to fully control their effects in a retrospective observational design. How do our findings accord with public health advice regarding alcohol consumption? At levels of intake recommended by US dietary guidelines (up to 140 g/wk), 33 we found no evidence of increased risk of esophageal cancer. Drinking guidelines in other countries are less restrictive, and would condone intakes at levels that could place people at significantly higher risks of ESCC. For example, the UK National Health Service advised that men should not regularly drink more than three to four units of alcohol per day. 34 Similarly, guidelines issued in 2001 by the National Health and Medical Research Council of Australia advised that men should consume no more than 4 standard drinks per day, with 1 2 alco-

9 April 2009 ALCOHOL AND ESOPHAGEAL CARCINOMAS 1223 hol-free days per week. 35 Our data suggest that alcohol intakes at these levels are associated significantly with cancer. In summary, we assessed the effect of alcohol intake on esophageal cancer by using a combination of nonlinear smoothing and piecewise linear regression methods. We found that the risk of ESCC was increased significantly with increasing average weekly alcohol intake (from all sources) beyond a threshold level. For beer, however, the risk of ESCC increased with all levels of intake. Risks of ESCC also were increased with high levels of intake of wine, ports, and spirits, but not with lower levels. We found no evidence that alcohol intake at any level was associated with increased risks of adenocarcinomas. Those who drank moderate amounts of wine had lower risks of these cancers, an association that also has been observed previously and that awaits a full explanation. Supplementary Data Note: To access the supplementary material accompanying this article, visit the online version of Gastroenterology at and at doi: /j.gastro References 1. Coory M, Dinh M. Mortality and incidence trends for adenocarcinoma of the oesophagus in Queensland 1982 to 2001: Health Information centre, Queensland Health; Brisbane, Queensland, Pera M, Pera M. Recent changes in the epidemiology of esophageal cancer. Surg Oncol 2001;10: Pera M, Manterola C, Vidal O, et al. Epidemiology of esophageal adenocarcinoma. J Surg Oncol 2005;92: Bosetti C, Levi F, Ferlay J, et al. Trends in oesophageal cancer incidence and mortality in Europe. Int J Cancer 2008;122: International Agency for Research on Cancer. IARC monographs on the evaluation of carcinogenic risks to humans. Alcohol drinking (IARC Monographs, volume 44), 1998, Lyon, IARC. 6. Bagnardi V, Blangiardo M, La Vecchia C, et al. A meta-analysis of alcohol drinking and cancer risk. Br J Cancer 2001;85: Holmes RS, Vaughan TL. Epidemiology and pathogenesis of esophageal cancer. Semin Radiat Oncol 2007;17: Freedman ND, Abnet CC, Leitzmann MF, et al. A prospective study of tobacco, alcohol, and the risk of esophageal and gastric cancer subtypes. Am J Epidemiol 2007;165: Lindblad M, Rodriguez LAG, Lagergren J. Body mass, tobacco and alcohol and risk of esophageal, gastric cardia, and gastric noncardia adenocarcinoma among men and women in a nested case-control study. Cancer Causes Control 2005;16: Wu AH, Wan P, Bernstein L. A multiethnic population-based study of smoking, alcohol and body size and risk of adenocarcinomas of the stomach and esophagus (United States). Cancer Causes Control 2001;12: Lagergren J, Bergstrom R, Lindgren A, et al. The role of tobacco, snuff and alcohol use in the aetiology of cancer of the oesophagus and gastric cardia. Int J Cancer 2000;85: Kabat GC, Ng SK, Wynder EL. Tobacco, alcohol intake, and diet in relation to adenocarcinoma of the esophagus and gastric cardia. Cancer Causes Control 1993;4: Vaughan TL, Davis S, Kristal A, et al. Obesity, alcohol, and tobacco as risk-factors for cancers of the esophagus and gastric cardia adenocarcinoma versus squamous-cell carcinoma. Cancer Epidemiol Biomarkers Prev 1995;4: Gammon MD, Schoenberg JB, Ahsan H, et al. Tobacco, alcohol, and socioeconomic status and adenocarcinomas of the esophagus and gastric cardia. J Natl Cancer Inst 1997;89: Greenland S. Avoiding power loss associated with categorization and ordinal scores in dose-response and trend analysis. Epidemiology 1995;6: Baglietto L, English DR, Hopper JL, et al. Average volume of alcohol consumed, type of beverage, drinking pattern and the risk of death from all causes. Alcohol Alcohol 2006;41: Andreasson S. Alcohol and J-shaped curves. Alcohol Clin Exp Res 1998;22(Suppl):359S 364S. 18. Ringborg U. Alcohol and risk of cancer. Alcohol Clin Exp Res 1998;22(Suppl):323S 328S. 19. Whiteman DC, Sadeghi S, Pandeya N, et al. Combined effects of obesity, acid reflux and smoking on the risk of adenocarcinomas of the oesophagus. Gut 2008;57: Olsen CM, Bain CJ, Jordan SJ, et al. Recreational physical activity and epithelial ovarian cancer: a case-control study, systematic review, and meta-analysis. Cancer Epidemiol Biomarkers Prev 2007;16: Wood SN. Generalized additive models: an introduction with R. Boca Raton, FL: Chapman and Hall/CRC Press, R: a language and environment for statistical computing (version 2.3.1). Vienna, Austria: R Foundation for Statistical Computing. Available from: Berry WD, Feldman S. Multiple regression in practice. Newbury Park, CA: Sage, Pastor R, Guallar E. Use of two-segmented logistic regression to estimate change-points in epidemiologic studies. Am J Epidemiol 1998;148: Hashibe M, Boffetta P, Janout V, et al. Esophageal cancer in Central and Eastern Europe: tobacco and alcohol. Int J Cancer 2007;120: Australian Bureau of Statistics. Apparent consumption of alcohol, Australia, Vol cat. no Released 27/6/05. Canberra: Australian Bureau of Statistics, Talley NJ, Quan C, Jones MP, et al. Association of upper and lower gastrointestinal tract symptoms with body mass index in an Australian cohort. Neurogastroenterol Motil 2004;16: Australian Bureau of Statistics. National health survey: summary of results Vol cat. no Released 27/02/06. Canberra: Australian Bureau of Statistics, Prescott E, Gronbaek M, Becker U, et al. Alcohol intake and the risk of lung cancer: influence of type of alcoholic beverage. Am J Epidemiol 1999;149: Gescher AJ, Steward WP. Relationship between mechanisms, bioavailability, and preclinical chemopreventive efficacy of resveratrol: a conundrum. Cancer Epidemiol Biomarkers Prev 2003; 12: Gusman J, Malonne H, Atassi G. A reappraisal of the potential chemopreventive and chemotherapeutic properties of resveratrol. Carcinogenesis 2001;22: Bianchini F, Vainio H. Wine and resveratrol: mechanisms of cancer prevention? Eur J Cancer Prev 2003;12: U.S. Department of Health and Human Services and U.S. Department of Agriculture. Dietary guidelines for Americans, th ed. Washington, DC: U.S. Government Printing Office, National Health Services. Accessed: October 30, National Health and Medical Research Council (Australia). Australian alcohol guidelines: health risks and benefits. Canberra: National Health and Medical Research Council, 2001.

10 1224 PANDEYA ET AL GASTROENTEROLOGY Vol. 136, No. 4 Received August 26, Accepted December 18, Reprint requests Address requests for reprints to: Nirmala Pandeya, MMedSc, Division of Population Studies and Human Genetics, Queensland Institute of Medical Research, PO Royal Brisbane Hospital, Queensland 4029, Australia. Nirmala.Pandeya@qimr.edu.au; fax: (61) Conflict of interest The authors disclose no conflicts. Funding This study was supported by the Queensland Cancer Fund and the National Health and Medical Research Council of Australia (program no ). David Whiteman and Penelope Webb are supported by Senior Research Fellowships from the National Health and Medical Research Council of Australia. Nirmala Pandeya is supported by a PhD scholarship from the National Health and Medical Research Council of Australia. The funding bodies played no role in the design or conduct of the study; the collection, management, analysis, or interpretation of the data; or preparation, review, or approval of the manuscript.

11 April 2009 ALCOHOL AND ESOPHAGEAL CARCINOMAS 1224.e1 A A B C Average lifetime alcohol consumption (gms of alcohol/week) Average lifetime alcohol consumption (gms of alcohol/week) B Average lifetime beer consumption (gms of alcohol/week) Average lifetime spirits/liqueur consumption (gms of alcohol/week) Supplementary Figure 2. Dose-response effect of lifetime consumption of (A) beer and (B) liqueur on ESCC using a generalized additive model with 3 degrees of freedom smoothing spline function adjusted for age, sex, body mass index 1 year ago, frequency of heartburn or acid reflux, education, frequency of aspirin use in the past 5 years, and smoking Average lifetime alcohol consumption (gms of alcohol/week) Supplementary Figure 1. Plots for dose effect (in logit scale) of average lifetime alcohol consumption on ESCC among (A) never smokers, (B) former smokers, and (C) current smokers using a generalized additive model with 3 degrees of freedom smoothing spline function adjusted for age, sex, body mass index 1 year ago, frequency of heartburn or acid reflux, education, and frequency of aspirin use in the past 5 years.

12 e2 PANDEYA ET AL GASTROENTEROLOGY Vol. 136, No. 4 A B Average lifetime wine consumption (gms of alcohol/week) C Average lifetime wine consumption (gms of alcohol/week) Average lifetime wine consumption (gms of alcohol/week) Supplementary Figure 3. Plots for dose effect (in logit scale) of average lifetime wine consumption on (A) EAC, (B) EGJAC, and (C) ESCC using a generalized additive model with 3 degrees of freedom smoothing spline function adjusted for age, sex, body mass index 1 year ago, frequency of heartburn or acid reflux, education, frequency of aspirin use in the past 5 years, and smoking.