Alcohol and the Risk of Barrett Esophagus
The five case–control studies participating in BEACON used for this analysis were as follows: the Study of Digestive Health (Brisbane, Australia); the Factors Influencing the Barrett's/Adenocarcinoma Relationship (FINBAR) study (Ireland); the Epidemiology and Incidence of Barrett's Esophagus study (Kaiser Permanente Northern California); the Study of Reflux Disease (western Washington State); and the Epidemiologic Case-Control Study of Barrett's Esophagus (Chapel Hill, North Carolina). The Institutional Review Boards or Research Ethics Committees of each institution approved the acquisition and pooling of data for the present analysis. Participants provided written informed consent to take part in the studies. Full details of recruitment and participation are provided in the primary papers.
We compared patients with Barrett's esophagus (cases) with two separate control groups: gastroesophageal reflux disease (GERD) controls (i.e., participants who had GERD symptoms or endoscopic evidence of acute esophageal inflammation consistent with GERD) and population-based controls. Five studies provided a GERD control group (see Table 1 and reference 11), and four studies provided population-based controls (the Chapel Hill study did not have a population-based comparison group). The definition of Barrett's esophagus in each study included endoscopic evidence of columnar mucosa in the tubular esophagus, accompanied by the presence of specialized intestinal metaplasia in an esophageal biopsy. Overall, data were pooled for 1,432 population-based controls, 1,659 GERD controls, and 1,320 cases. Among those participants with available data on alcohol consumption (1,376 population-based controls, 1,597 GERD controls, and 1,257 cases), we restricted our analyses to white non-Hispanic study participants (1,282 population-based controls, 1,418 GERD controls, and 1,169 cases) owing to low numbers of cases from other ethnic groups (cases: 12 black, 32 Hispanic, 42 other, 2 missing).
The main exposure was average daily alcohol consumption (drinks per day). Two studies asked participants to report their 'usual consumption of alcohol', two studies captured alcohol consumption for the period 1 year or 5 years before interview, and one study assessed usual frequency of consumption of alcohol during the age intervals of 20–29 years, 30–49 years, and ≥50 years; for this study, we used the average across these intervals for the pooled analysis. The beverage-specific analyses for consumption of beer, wine (red or white), and liquor were conducted using data from four studies (the Chapel Hill study did not collect beverage-specific consumption).
Individual-level harmonized clinical, demographic, and questionnaire data for each study participant were merged into a single de-identified data set and included information on study center, case–control status, age at diagnosis (or reference date for controls), sex, ethnicity, body mass index (BMI, weight divided by height squared, kg/m), education, GERD symptoms status (ever vs. never), and cigarette smoking status (never, former, current) and pack-years (continuous). The data were checked for consistency and completeness, and any apparent inconsistencies were followed up with individual study investigators.
We assessed the association between alcohol consumption (in categories: none, >0–<0.5, 0.5–<1.0, 1–<3, 3–<5, 5–<7, ≥7 drinks/day) and the risk of Barrett's esophagus using a two-stage analytic approach. In the first stage, study-specific odds ratios (OR) and corresponding 95% confidence intervals (95% CI) were estimated using unconditional logistic regression models adjusted for age (<50, 50–59, 60–69, ≥70 years), sex, education (school only, technical college/diploma, university/college), BMI (<25, 25–29.9, ≥30 kg/m), smoking status (never, former, current), and GERD symptoms (ever vs. never; population-based control comparisons only). Study-specific analyses for the Chapel Hill study were not adjusted for education. In the second stage, the study-specific adjusted ORs were pooled to create a summary OR, using random-effects models. We used the I statistic to assess the heterogeneity between studies. Larger I values reflect increasing heterogeneity, beyond what is attributable to chance. I values of 25%, 50%, and 75% were used as evidence of low, moderate, or high levels of heterogeneity, respectively. The referent group for the analyses of total alcohol and beverage-specific consumption was nondrinkers of any alcohol type. For each of the beverage-specific analyses, we adjusted for total alcohol consumption in an attempt to investigate the specific effects independent of ethanol content.
We calculated the P-value for trend by fitting alcohol consumption as an ordinal variable and assigning participants the median value for their category of alcohol exposure. Restricted cubic spline models were used to further investigate the continuous association between alcohol consumption and the risk of Barrett's esophagus. These models allow for easy visualization of nonlinear associations and were plotted using a linear scale on the x axis (drinks/day of alcohol) and a logarithmic scale on the y axis for the OR.
Finally, using the same methodology as for the overall analyses, we conducted stratified analyses by categories of sex, BMI, GERD symptoms, and smoking status to assess potential effect modification. We included an interaction term in the full model to assess the statistical significance of the difference in association across strata. All tests for statistical significance were two-sided at α=0.05, and analyses were conducted using Stata 13.0 (StataCorp LP, College Station, TX).
Methods
Study Population
The five case–control studies participating in BEACON used for this analysis were as follows: the Study of Digestive Health (Brisbane, Australia); the Factors Influencing the Barrett's/Adenocarcinoma Relationship (FINBAR) study (Ireland); the Epidemiology and Incidence of Barrett's Esophagus study (Kaiser Permanente Northern California); the Study of Reflux Disease (western Washington State); and the Epidemiologic Case-Control Study of Barrett's Esophagus (Chapel Hill, North Carolina). The Institutional Review Boards or Research Ethics Committees of each institution approved the acquisition and pooling of data for the present analysis. Participants provided written informed consent to take part in the studies. Full details of recruitment and participation are provided in the primary papers.
We compared patients with Barrett's esophagus (cases) with two separate control groups: gastroesophageal reflux disease (GERD) controls (i.e., participants who had GERD symptoms or endoscopic evidence of acute esophageal inflammation consistent with GERD) and population-based controls. Five studies provided a GERD control group (see Table 1 and reference 11), and four studies provided population-based controls (the Chapel Hill study did not have a population-based comparison group). The definition of Barrett's esophagus in each study included endoscopic evidence of columnar mucosa in the tubular esophagus, accompanied by the presence of specialized intestinal metaplasia in an esophageal biopsy. Overall, data were pooled for 1,432 population-based controls, 1,659 GERD controls, and 1,320 cases. Among those participants with available data on alcohol consumption (1,376 population-based controls, 1,597 GERD controls, and 1,257 cases), we restricted our analyses to white non-Hispanic study participants (1,282 population-based controls, 1,418 GERD controls, and 1,169 cases) owing to low numbers of cases from other ethnic groups (cases: 12 black, 32 Hispanic, 42 other, 2 missing).
Data Collection
The main exposure was average daily alcohol consumption (drinks per day). Two studies asked participants to report their 'usual consumption of alcohol', two studies captured alcohol consumption for the period 1 year or 5 years before interview, and one study assessed usual frequency of consumption of alcohol during the age intervals of 20–29 years, 30–49 years, and ≥50 years; for this study, we used the average across these intervals for the pooled analysis. The beverage-specific analyses for consumption of beer, wine (red or white), and liquor were conducted using data from four studies (the Chapel Hill study did not collect beverage-specific consumption).
Individual-level harmonized clinical, demographic, and questionnaire data for each study participant were merged into a single de-identified data set and included information on study center, case–control status, age at diagnosis (or reference date for controls), sex, ethnicity, body mass index (BMI, weight divided by height squared, kg/m), education, GERD symptoms status (ever vs. never), and cigarette smoking status (never, former, current) and pack-years (continuous). The data were checked for consistency and completeness, and any apparent inconsistencies were followed up with individual study investigators.
Statistical Analysis
We assessed the association between alcohol consumption (in categories: none, >0–<0.5, 0.5–<1.0, 1–<3, 3–<5, 5–<7, ≥7 drinks/day) and the risk of Barrett's esophagus using a two-stage analytic approach. In the first stage, study-specific odds ratios (OR) and corresponding 95% confidence intervals (95% CI) were estimated using unconditional logistic regression models adjusted for age (<50, 50–59, 60–69, ≥70 years), sex, education (school only, technical college/diploma, university/college), BMI (<25, 25–29.9, ≥30 kg/m), smoking status (never, former, current), and GERD symptoms (ever vs. never; population-based control comparisons only). Study-specific analyses for the Chapel Hill study were not adjusted for education. In the second stage, the study-specific adjusted ORs were pooled to create a summary OR, using random-effects models. We used the I statistic to assess the heterogeneity between studies. Larger I values reflect increasing heterogeneity, beyond what is attributable to chance. I values of 25%, 50%, and 75% were used as evidence of low, moderate, or high levels of heterogeneity, respectively. The referent group for the analyses of total alcohol and beverage-specific consumption was nondrinkers of any alcohol type. For each of the beverage-specific analyses, we adjusted for total alcohol consumption in an attempt to investigate the specific effects independent of ethanol content.
We calculated the P-value for trend by fitting alcohol consumption as an ordinal variable and assigning participants the median value for their category of alcohol exposure. Restricted cubic spline models were used to further investigate the continuous association between alcohol consumption and the risk of Barrett's esophagus. These models allow for easy visualization of nonlinear associations and were plotted using a linear scale on the x axis (drinks/day of alcohol) and a logarithmic scale on the y axis for the OR.
Finally, using the same methodology as for the overall analyses, we conducted stratified analyses by categories of sex, BMI, GERD symptoms, and smoking status to assess potential effect modification. We included an interaction term in the full model to assess the statistical significance of the difference in association across strata. All tests for statistical significance were two-sided at α=0.05, and analyses were conducted using Stata 13.0 (StataCorp LP, College Station, TX).