A Prospective Study of Tobacco, Alcohol, and the Risk of Esophageal and Gastric Cancer Subtypes

Abstract

Rates of esophageal adenocarcinoma and gastric cardia adenocarcinoma have increased, while rates of esophageal squamous cell carcinoma (ESCC) and gastric noncardia adenocarcinoma have decreased, suggesting distinct etiologies. The authors prospectively investigated the associations of alcohol and tobacco with these cancers in 474,606 US participants using Cox models adjusted for potential confounders. Between 1995/1996 and 2000, 97 incident cases of ESCC, 205 of esophageal adenocarcinoma, 188 of gastric cardia, and 187 of gastric noncardia cancer occurred. Compared with nonsmokers, current smokers were at increased risk for ESCC (hazard ratio (HR) = 9.27, 95% confidence interval (CI): 4.04, 21.29), esophageal adenocarcinoma (HR = 3.70, 95% CI: 2.20, 6.22), gastric cardia (HR = 2.86, 95% CI: 1.73, 4.70), and gastric noncardia (HR = 2.04, 95% CI: 1.32, 3.16). Assuming causality, ever smoking had population attributable risks of 77% (95% CI: 0.55, 0.89) for ESCC, 58% (95% CI: 0.38, 0.72) for esophageal adenocarcinoma, 47% (95% CI: 0.27, 0.63) for gastric cardia, and 19% (95% CI: 0.00, 0.37) for gastric noncardia. For drinkers of more than three alcoholic beverages per day, compared with those whose intake was up to one drink per day, the authors found significant associations between alcohol intake and ESCC risk (HR = 4.93, 95% CI: 2.69, 9.03) but not risk for esophageal, gastric cardia, or gastric noncardia adenocarcinoma.

alcohol drinking, cohort studies, esophageal neoplasms, prospective studies, smoking, stomach neoplasms

Cancers of the stomach and esophagus have high worldwide incidence and mortality. At least two epithelial subtypes occur in the esophagus—esophageal squamous cell carcinoma (ESCC) and esophageal adenocarcinoma. The stomach has at least two subsites—the cardia and the noncardia. Esophageal adenocarcinoma and gastric cardia adenocarcinoma arise in adjacent organs and are difficult to distinguish clinically, and there are no current methods for definitively separating the two cancers (1). This difficulty complicates the understanding of etiologic factors and warrants the examination of these two diseases simultaneously.

Numerous epidemiologic studies have established that tobacco smoking and alcohol are causal factors for ESCC (2, 3). Associations between alcohol and tobacco and gastric noncardia adenocarcinoma are less consistent but generally indicate that smoking is a moderate risk factor (4, 5) and alcohol has no association (6–8). In the past 30 years, the rates of esophageal adenocarcinoma and gastric cardia adenocarcinoma have increased in the United States by over 350 percent and 50 percent, respectively, while the rates of ESCC and gastric noncardia adenocarcinoma have decreased by 35 percent and 25 percent (9–11), such that ESCC and gastric noncardia adenocarcinoma are no longer the predominant forms of esophageal and gastric cancer found in the United States or many Western countries (10, 12, 13).

As the increase in esophageal adenocarcinoma and gastric cardia adenocarcinoma incidence is a recent phenomenon, relatively few previous studies compared the associations between alcohol or tobacco use and esophageal adenocarcinoma and gastric cardia adenocarcinoma risk with those for ESCC and gastric noncardia adenocarcinoma (14–20). A majority were case-control studies, susceptible to both recall and selection bias (14–17, 19, 20). Just one prospective study made this comparison (18), but it lacked complete information on confounders. These studies generally indicated an association between tobacco and esophageal adenocarcinoma and gastric cardia adenocarcinoma risk, but results for alcohol were inconsistent. It is not clear if the risk factors for esophageal adenocarcinoma and gastric cardia adenocarcinoma are similar or distinct. We investigated associations between alcohol and tobacco with esophageal adenocarcinoma and gastric cardia adenocarcinoma risk and compared them with those for ESCC and gastric noncardia adenocarcinoma in a large US prospective cohort, the National Institutes of Health (NIH)–AARP Diet and Health Study.

MATERIALS AND METHODS

The establishment and recruitment procedures of the NIH–AARP Diet and Health Study have been described (21). Between 1995 and 1996, a risk factor questionnaire was mailed to 3.5 million members of AARP, formerly known as the American Association of Retired Persons—a US organization whose membership is open to those greater than 50 years of age, who resided in six US states (California, Florida, Louisiana, New Jersey, North Carolina, and Pennsylvania) and two metropolitan areas (Atlanta, Georgia, and Detroit, Michigan). Of the 617,119 persons who returned the questionnaire (17.6 percent), 566,407 respondents filled out the survey in satisfactory detail and consented to be in the study. We excluded subjects with cancer (except nonmelanoma skin cancer) at baseline (n = 51,219); proxy respondents (n = 15,760); those with energy intake more than two interquartile ranges from the median, which we have used as a marker for questionnaire completeness and accuracy (n = 4,419); those who died or were diagnosed with cancer on the first day of follow-up (n = 10); and those who lacked information on cigarette or alcohol use (n = 20,393). The resulting cohort included 474,606 participants: 282,856 men and 191,750 women. The NIH–AARP Diet and Health Study was approved by the Special Studies Institutional Review Board of the US National Cancer Institute.

Cohort follow-up

As described previously (22), addresses were matched annually with the US Postal Service National Change of Address database. Vital status was ascertained by annual linkage of the cohort to the Social Security Administration Death Master File, cancer registry linkage, and responses to mailings.

Incident cases of cancer were identified by probabilistic linkage between the NIH–AARP cohort membership and the cancer registry databases, each of which has been certified by the North American Association of Central Cancer Registries for meeting the highest standards of data quality. For matching purposes, first and last name, address history, gender, date of birth, and Social Security number (available for 85 percent of our participants) were used. We estimate the sensitivity of case identification to be 89.2 percent and the specificity to be 99.5 percent (22). Cancer sites were identified by anatomic site and histologic code of the International Classification of Diseases for Oncology, Third Edition (ICD-O) (23). We classified 339 tumors with site codes C15.0–C15.9 as esophageal cancer and identified ESCC (97 cases) and esophageal adenocarcinomas (205 cases) by histology. Cases with nonadenocarcinoma or nonsquamous (15 cases) or unspecified (22 cases) histology were excluded. We identified 402 gastric cancer cases and excluded 27 cases with nonadenocarcinoma histology. Cardia tumors (188 cases) had site code C16.0 (defined as the cardioesophageal junction, the esophagogastric junction, and the gastroesophageal junction). We classified tumors with site codes C16.1–C16.9 as noncardia tumors (187 cases). Over 13 percent (25 of 187) of noncardia tumors were an overlapping lesion of the stomach (ICD-O site code C16.8) and 28 percent (53 of 187) were gastric not otherwise specified (ICD-O site code C16.9), raising the possibility that cardia cancers might be present in the noncardia category. However, excluding these 78 subjects from the analysis did not appreciably change the risk estimates (data not shown).

Exposure assessment

The baseline questionnaire inquired about demographics, alcohol, tobacco use, physical activity, and diet by use of a food frequency questionnaire of 124 items (21). Self-administered questionnaires for alcohol intake have high reproducibility (Spearman's rank correlation coefficient of 0.84 after 4 years) and validity with respect to 1-week diet records (Spearman's rank correlation coefficients of 0.92 for men and 0.90 for women) (24). Similarly, questionnaires for smoking have high reproducibility (94 percent agreement for duplicate questionnaires) and validity (92 percent agreement for women and 90 percent agreement for men relative to serum cotinine levels) (25, 26). We calculated alcohol drinks per day from participants' responses for usual frequency (0–>6 times per day) and portion size of wine, liquor, and beer consumption over the past 12 months. One drink corresponded to one serving of the US Department of Agriculture's food guide pyramid: one 12-fluid ounce beer, one 5-fluid ounce glass of wine, or one 1.5-ounce shot of liquor (each approximately 13 g of alcohol) (27, 28). We used categorical variables of alcohol intake: none, up to or including one drink per day, more than one to three drinks per day, and greater than three drinks per day. We lacked data on past alcohol use.

Participants were also asked if they had smoked more than 100 cigarettes during their life (ever smokers), their typical smoking intensity (cigarettes smoked per day), and smoking status (never, former, current). Those who reported quitting within the past year were considered current smokers. We lacked data on smoking duration but did have data on time since quitting. For adjustment, we used the smoke-quit dose variable that combined smoking status and smoking dose into five categories: never smokers, former smokers who smoked one pack or less per day, former smokers who smoked more than one pack per day, current smokers who smoked one pack or less per day, and current smokers who smoked more than one pack per day.

Statistical analysis

Analyses were performed with SAS, version 8.2, software (SAS Institute, Inc., Cary, North Carolina). Potential confounders were tabulated by cigarette smoking status and alcohol intake. An alpha level of less than 0.05 was considered significant, and all tests were two sided. Hazard ratios and 95 percent confidence intervals were calculated by use of Cox proportional hazards regression (29). We found no deviations from the proportional hazards assumption for either cigarette or alcohol use. Follow-up time in person-years (from baseline to diagnosis of esophageal, head and neck, or stomach cancer, date of death, end of study (December 31, 2000), or the date moved out of catchment area) was used as the underlying time metric. Ending follow-up time at the first cancer diagnosis, regardless of site, reduced case numbers slightly but did not appreciably affect the results.

All models were adjusted for sex, body mass index (<18.5, 18.5–<25, 25–<30, 30–<35, and ≥35), the categories of education, vigorous physical activity, and usual daily physical activity, as shown in tables 1 and 2, and continuous measures for total vegetable intake, total fruit intake, energy intake, and age. For the gastric noncardia adenocarcinoma cancer analysis, categorical variables for race/ethnicity, non-Hispanic White, non-Hispanic Black, Hispanic, and Asian/Pacific Islander/Native American, were included in the models. Race/ethnicity was not included for other cancer sites, as more than 93 percent of ESCC, esophageal adenocarcinoma, and gastric cardia adenocarcinoma cases were non-Hispanic White. Inclusion of family cancer history did not affect risk estimates (data not shown). All analyses for cigarette smoking included the four categorical variables for alcohol intake used in table 1. Models for pipe smoking included categorical variables for alcohol consumption and smoke-quit dose (five levels shown in table 3). Alcohol analyses included the categorical smoke-quit dose variables. For the linear trend test with the alcohol variables, a score variable was used, containing median values of alcohol intake within defined categories. Models that analyzed categories of liquor, beer, or wine consumption included the other two types of alcohol as categorical variables, such that total alcohol consumption was accounted for in each model. Missing values for adjusting covariates were included as dummy variables in the models. Excluding those with missing values for covariates reduced the number of cases slightly but did not change point estimates noticeably, as less than 3 percent of the cohort lacked data for each covariate. Assuming a causal relation between smoking and cancer, we calculated percentages of population attributable risk by the method of Bruzzi et al. (30) to estimate the percentage of cancer cases that would not have occurred if members of our cohort had never smoked. We used multivariate-adjusted beta-coefficients for these calculations.

TABLE 1.

Study characteristics by smoking status in 474,606 US members of AARP from 1995/1996 to 2000

	Smoking status*
	Never		Former		Current
	Median	Interquartile range	Median	Interquartile range	Median	Interquartile range
Age (years)	62.3	57.6–66.5	63.1	58.2–66.9	61.1	56.4–65.6
Body mass index (kg/m2)	26.1	23.7–29.3	26.7	24.4–29.8	25.5	23.0–28.5
Total daily calories	1,634.9	1,243.5–2,135.3	1,710.5	1,300.6–2,230.7	1,777.6	1,308.1–2,401.6
Vegetable consumption (servings/day)	3.4	2.3–5.0	3.5	2.3–5.0	3.2	2.1–4.7
Fruit consumption (servings/day)	2.7	1.6–4.1	2.4	1.4–3.8	1.7	0.9–3.0
	Never		Former		Current
	No.	%	No.	%	No.	%
Cohort	173,055	36.5	233,861	49.3	67,690	14.3
Sex
Male	85,740	30.3	161,454	57.1	35,662	12.6
Female	87,315	45.5	72,407	37.8	32,028	16.7
Education
Less than high school	7,584	27.4	14,392	52.1	5,669	20.5
Completed high school	34,320	37.1	42,705	46.1	15,583	16.8
Some post-high school training	50,405	32.0	81,177	51.5	25,961	16.5
Completed college	33,580	37.4	45,689	50.9	10,446	11.6
Completed graduate school	42,945	45.5	43,609	46.2	7,823	8.3
Alcohol intake (drinks per day)
0	51,581	45.1	48,145	42.1	14,700	12.9
>0–1	96,706	38.3	121,972	48.3	33,668	13.3
>1–3	18,883	26.2	42,567	59.0	10,667	14.8
>3	5,885	16.5	21,177	59.3	8,655	24.2
Usual daily physical activity
Sit during the day/little walking	12,476	33.2	19,024	50.7	6,046	16.1
Sit during the day/walk a fair amount	55,877	36.5	75,541	49.3	21,866	14.3
Stand/walk a lot—no lifting	65,721	36.7	89,285	49.9	24,006	13.4
Lift/carry light loads, stairs, hills	30,732	37.7	39,176	48.1	11,606	14.2
Do heavy work/carry loads	4,424	32.7	6,510	48.2	2,586	19.1
Vigorous physical activity
Never	6,900	33.0	9,096	43.5	4,897	23.4
Rarely	21,971	34.1	28,846	44.8	13,555	21.1
1–3 times/month	22,699	35.1	30,220	46.8	11,718	18.1
1–2 times/week	38,018	37.1	49,236	48.1	15,135	14.8
3–4 times/week	47,727	37.5	66,161	52.0	13,353	10.5
≥5 times/week	34,044	37.4	48,556	53.3	8,424	9.3

*

Categories may not add up to 474,606 persons because of small amounts of missing data.

TABLE 2.

Study characteristics by alcohol intake in 474,606 US members of AARP from 1995/1996 to 2000

	Alcohol intake (drinks per day)*
	0		>0–1		>1–3		>3
	Median	Interquartile range	Median	Interquartile range	Median	Interquartile range	Median	Interquartile range
Age (years)	62.9	58.1–66.8	62.2	57.4–66.4	62.9	58.1–66.8	62.7	57.9–66.5
Body mass index (kg/m2)	26.7	24.1–30.4	26.5	23.9–29.4	25.8	23.6–28.3	26.5	24.2–29.1
Total daily calories	1,616.2	1,207.3–2,151.9	1,607.5	1,225.5–2,091.7	1,803.1	1,422.5–2,282.1	2,398.9	1,900.6–3,094.5
Vegetable consumption (servings/day)	3.3	2.1–4.9	3.4	2.3–4.9	3.7	2.5–5.2	3.6	2.4–5.2
Fruit consumption (servings/day)	2.5	1.4–4.1	2.5	1.5–3.8	2.4	1.4–3.6	1.9	1.0–3.2
	0		>0–1		>1–3		>3
	No.	%	No.	%	No.	%	No.	%
Cohort	114,426	24.1	252,346	53.2	72,117	15.2	35,717	7.5
Sex
Male	58,209	20.6	141,564	50.1	52,509	18.6	30,574	10.8
Female	56,217	29.3	110,782	57.8	19,608	10.2	5,143	2.7
Education
Less than high school	10,895	39.4	12,322	44.6	2,631	9.5	1,797	6.5
Completed high school	28,943	31.3	47,709	51.5	10,169	11.0	5,787	6.3
Some post-high school training	36,967	23.5	85,798	54.5	22,699	14.4	12,079	7.7
Completed college	16,692	18.6	48,009	53.5	16,865	18.8	8,149	9.1
Completed graduate school	16,973	18.0	51,987	55.1	18,320	19.4	7,097	7.5
Cigarette smoking status
Never	51,581	29.8	96,706	55.9	18,883	10.9	5,885	3.4
Former	48,145	20.6	121,972	52.2	42,567	18.2	21,177	9.1
Current	14,700	21.7	33,668	49.7	10,667	15.8	8,655	12.8
Pipe or cigar use
No	97,030	25.4	206,022	54.0	53,471	14.0	24,920	6.5
Yes	13,743	17.0	39,715	49.1	17,253	21.3	10,249	12.7
Usual daily physical activity
Sit during the day/little walking	9,779	26.1	20,219	53.9	4,899	13.1	2,649	7.1
Sit during the day/walk a fair amount	34,941	22.8	84,432	55.1	23,006	15.0	10,905	7.1
Stand/walk a lot—no lifting	43,135	24.1	94,466	52.8	27,782	15.5	13,629	7.6
Lift/carry light loads, stairs, hills	19,338	23.7	42,092	51.6	13,456	16.5	6,628	8.1
Do heavy work/carry loads	3,953	29.2	6,318	46.7	1,876	13.9	1,373	10.2
Vigorous physical activity
Never	8,379	40.1	9,054	43.3	1,985	9.5	1,475	7.1
Rarely	18,610	28.9	33,349	51.8	7,332	11.4	5,081	7.9
1–3 times/month	14,738	22.8	35,969	55.7	8,897	13.8	5,033	7.8
1–2 times/week	22,266	21.8	56,514	55.2	15,945	15.6	7,664	7.5
3–4 times/week	27,319	21.5	69,442	54.6	21,333	16.8	9,147	7.2
≥5 times/week	21,741	23.9	46,011	50.6	16,196	17.8	7,076	7.8

*

Categories may not add up to 474,606 persons because of small amounts of missing data.

TABLE 3.

Adjusted hazard ratios and 95% confidence intervals for tobacco use and esophageal and gastric cancer in 474,606 US members of AARP from 1995/1996 to 2000

	Esophageal squamous cell carcinoma*			Esophageal adenocarcinoma*			Gastric cardia adenocarcinoma*			Gastric noncardia adenocarcinoma†
	No. of cases	Hazard ratio	95% confidence interval	No. of cases	Hazard ratio	95% confidence interval	No. of cases	Hazard ratio	95% confidence interval	No. of cases	Hazard ratio	95% confidence interval
Cigarette smoking status
Never	7	1.00	Referent	25	1.00	Referent	30	1.00	Referent	53	1.00	Referent
Former	50	4.35	1.95, 9.72	142	2.82	1.83, 4.34	120	2.14	1.42, 3.22	96	1.22	0.86, 1.73
Current	40	9.27	4.04, 21.29	38	3.70	2.20, 6.22	38	2.86	1.73, 4.70	38	2.04	1.32, 3.16
ptrend		<0.0001			<0.0001			<0.0001			0.0029
Pipe or cigar smoking status‡
Never	77	1.00	Referent	139	1.00	Referent	125	1.00	Referent	143	1.00	Referent
Ever	20	1.04	0.61, 1.77	61	1.06	0.77, 1.44	60	1.30	0.94, 1.79	37	1.01	0.69, 1.48
Quit smoking status
Never smoked	7	1.00	Referent	25	1.00	Referent	30	1.00	Referent	53	1.00	Referent
Stopped ≥10 years ago	29	3.24	1.40, 7.49	107	2.67	1.72, 4.16	89	2.01	1.32, 3.07	70	1.12	0.78, 1.63
Stopped 5–9 years ago	11	6.70	2.57, 17.47	27	4.04	2.33, 7.00	21	2.73	1.55, 4.82	19	1.79	1.05, 3.05
Stopped 1–4 years ago	10	10.30	3.88, 27.35	8	2.21	0.99, 4.93	10	2.39	1.16, 4.92	7	1.18	0.54, 2.62
Current smokers	40	9.52	4.14, 21.85	38	3.71	2.20, 6.25	38	2.87	1.75, 4.73	38	2.05	1.33, 3.18
ptrend		<0.0001			<0.0001			<0.0001			0.0008
Cigarette smoke dose per day
0 packs	7	1.00	Referent	25	1.00	Referent	30	1.00	Referent	53	1.00	Referent
>0–1 pack	47	5.30	2.38, 11.83	80	2.57	1.63, 4.04	83	2.28	1.49, 3.48	80	1.37	0.96, 1.96
>1–2 packs	33	6.02	2.61, 13.90	80	3.69	2.33, 5.84	62	2.50	1.60, 3.91	42	1.41	0.92, 2.15
>2 packs	10	6.17	2.28, 16.66	20	2.64	1.45, 4.81	13	1.52	0.78, 2.96	12	1.29	0.68, 2.45
ptrend		<0.0001			<0.0001			0.0095			0.15
Smoke-quit dose
Never smoked	7	1.00	Referent	25	1.00	Referent	30	1.00	Referent	53	1.00	Referent
Quit ≤1 pack/day	17	2.79	1.15, 6.77	58	2.31	1.44, 3.72	60	2.11	1.35, 3.29	55	1.22	0.83, 1.79
Quit >1 pack/day	33	6.27	2.71, 14.48	84	3.37	2.14, 5.33	60	2.17	1.38, 3.40	41	1.23	0.80, 1.88
Currently smoking ≤1 pack/day	30	11.50	4.95, 26.72	22	3.65	2.04, 6.55	23	2.91	1.67, 5.07	25	1.96	1.20, 3.20
Currently smoking >1 pack/day	10	5.80	2.12, 15.84	16	3.85	2.02, 7.36	15	2.78	1.46, 5.29	13	2.24	1.19, 4.21

*

Adjusted for continuous variables for age at entry into the cohort, fruit intake, vegetable intake, and total energy, as well as for categorical variables for sex, body mass index (<18.5, 18.5–<25, 25–<30, 30–<35, and ≥35), education (less than high school, completed high school, some post-high school training, completed college, completed graduate school), alcohol intake (0 drinks, >0–1 drink, >1–3 drinks, >3 drinks per day), vigorous physical activity (never, rarely, 1–3 times/month, 1–2 times/week, 3–4 times/week, ≥5 times/week), and usual daily physical activity (sit during the day/little walking, sit during the day/walk a fair amount, stand/walk a lot—no lifting, lift/carry light loads, stairs, hills, do heavy work/carry loads).

†

Additionally adjusted for race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, and Asian/Pacific Islander/Native American).

‡

Additionally adjusted for smoke-quit dose (never smoked, quit ≤1 pack/day, quit >1 pack/day, currently smoking ≤1 pack/day, currently smoking >1 pack/day).

RESULTS

Over the course of 2,121,797 person-years of follow-up, 97 subjects were diagnosed with ESCC, 205 with esophageal adenocarcinoma, 188 with gastric cardia adenocarcinoma, and 187 with gastric noncardia adenocarcinoma. The median follow-up time was 4.6 years (interquartile range: 4.5–4.8 years). The median age was 62.5 years, and a majority were male (59.6 percent) and non-Hispanic White (92.6 percent). Most were current or former smokers (tables 1 and 2). Current smokers had received less education, consumed less fruit, and drank more alcohol than did nonsmokers (table 1). Alcohol drinkers consumed fewer servings of fruit, were more likely to be current smokers, and were more likely to have obtained college or graduate education than had nondrinkers (table 2).

Using multivariable-adjusted Cox regression models, we found that current cigarette smoking was associated with increased risk of ESCC (hazard ratio (HR) = 9.27, 95 percent confidence interval (CI): 4.04, 21.29), esophageal adenocarcinoma (HR = 3.70, 95 percent CI: 2.20, 6.22), gastric cardia adenocarcinoma (HR = 2.86, 95 percent CI: 1.73, 4.70), and gastric noncardia adenocarcinoma (HR = 2.04, 95 percent CI: 1.32, 3.16) relative to never smokers (table 3). Former cigarette smokers were also at significantly increased risk for ESCC (HR = 4.35, 95 percent CI: 1.95, 9.72), esophageal adenocarcinoma (HR = 2.82, 95 percent CI: 1.83, 4.34), and gastric cardia adenocarcinoma (HR = 2.14, 95 percent CI: 1.42, 3.22) but not gastric noncardia adenocarcinoma (HR = 1.22, 95 percent CI: 0.86, 1.73) (table 3) relative to never smokers. Assuming a causal relation between smoking and each cancer site, we estimate that ever smoking accounted for 77 percent of ESCC (95 percent CI: 0.55, 0.89), 58 percent of esophageal adenocarcinoma (95 percent CI: 0.38, 0.72), 47 percent of gastric cardia adenocarcinoma (95 percent CI: 0.27, 0.63), and 19 percent of gastric noncardia adenocarcinoma (95 percent CI: 0.00, 0.37) in this cohort.

Among alcohol drinkers, those in the highest category of alcohol intake (>3 drinks per day) were at increased risk of developing ESCC (HR = 4.93, 95 percent CI: 2.69, 9.03) (table 4) compared with moderate drinkers who drank up to one drink per day. However, nondrinkers were at significantly increased risk for ESCC relative to those who drank up to one drink per day (HR = 2.06, 95 percent CI: 1.16, 3.68). We observed no association between drinking more than three alcoholic drinks per day and esophageal adenocarcinoma and nonsignificant associations with cardia (HR = 1.57, 95 percent CI: 0.98, 2.52) and noncardia (HR = 0.62, 95 percent CI: 0.30, 1.27) gastric adenocarcinoma.

TABLE 4.

Adjusted hazard ratios and 95% confidence intervals for alcohol intake and esophageal and gastric cancer in 474,606 US members of AARP from 1995/1996 to 2000

	Esophageal squamous cell carcinoma*			Esophageal adenocarcinoma*			Gastric cardia adenocarcinoma*			Gastric noncardia adenocarcinoma†
	No. of cases	Hazard ratio	95% confidence interval	No. of cases	Hazard ratio	95% confidence interval	No. of cases	Hazard ratio	95% confidence interval	No. of cases	Hazard ratio	95% confidence interval
Total alcohol (drinks/day)
0	23	2.06	1.16, 3.68	42	0.96	0.66, 1.38	46	1.19	0.83, 1.70	59	1.30	0.93, 1.82
>0–1	24	1.00	Referent	101	1.00	Referent	86	1.00	Referent	88	1.00	Referent
>1–3	20	2.33	1.28, 4.24	35	0.95	0.64, 1.40	29	0.99	0.65, 1.52	31	1.17	0.77, 1.77
>3	30	4.93	2.69, 9.03	27	1.10	0.69, 1.74	27	1.57	0.98, 2.52	9	0.62	0.30, 1.27
ptrend		<0.0001			0.68			0.12			0.15
Wine (drinks/day)‡
0	55	1.90	1.17, 3.06	93	1.34	0.95, 1.87	76	0.73	0.51, 1.05	89	1.04	0.73, 1.48
>0–1	35	1.00	Referent	95	1.00	Referent	98	1.00	Referent	87	1.00	Referent
>1–3	6	1.13	0.47, 2.69	15	1.24	0.72, 2.14	12	1.00	0.55, 1.83	9	0.91	0.46, 1.81
>3	1	2.75	0.37, 20.41	2	2.84	0.69, 11.58	2	3.01	0.73, 12.31	2	4.29	1.05, 17.61
ptrend		0.81			0.23			0.089			0.15
Beer (drinks/day)§
0	39	1.07	0.63, 1.81	71	0.93	0.66, 1.33	79	1.35	0.94, 1.94	93	1.18	0.83, 1.70
>0–1	35	1.00	Referent	119	1.00	Referent	92	1.00	Referent	80	1.00	Referent
>1–3	9	2.38	1.13, 5.00	6	0.48	0.21, 1.10	9	1.00	0.50, 1.99	9	1.36	0.68, 2.73
>3	14	3.61	1.76, 7.39	9	0.85	0.41, 1.75	8	1.17	0.55, 2.52	5	1.02	0.40, 2.61
ptrend		0.0002			0.46			0.98			0.96
Liquor (drinks/day)¶
0	45	1.27	0.74, 2.18	84	0.85	0.60, 1.21	91	1.36	0.94, 2.00	100	1.03	0.72, 1.47
>0–1	26	1.00	Referent	90	1.00	Referent	65	1.00	Referent	75	1.00	Referent
>1–3	8	1.53	0.69, 3.38	21	1.24	0.77, 2.00	17	1.53	0.89, 2.62	10	0.82	0.42, 1.59
>3	18	4.50	2.39, 8.49	10	0.82	0.42, 1.61	15	2.15	1.20, 3.87	2	0.27	0.07, 1.10
ptrend		<0.0001			0.93			0.023			0.046

*

Adjusted for continuous variables for age at entry into the cohort, fruit intake, vegetable intake, and total energy, as well as for categorical variables for sex, body mass index (<18.5, 18.5–<25, 25–<30, 30–<35, and ≥35), education (less than high school, completed high school, some post-high school training, completed college, completed graduate school), smoke-quit dose (never smoked, quit ≤1 pack/day, quit >1 pack/day, currently smoking ≤1 pack/day, currently smoking >1 pack/day), vigorous physical activity (never, rarely, 1–3 times/month, 1–2 times/week, 3–4 times/week, ≥5 times/week), and usual daily physical activity (sit during the day/little walking, sit during the day/walk a fair amount, stand/walk a lot—no lifting, lift/carry light loads, stairs, hills, do heavy work/carry loads).

†

Additionally adjusted for race/ethnicity (non-Hispanic White, non-Hispanic Black, Hispanic, and Asian/Pacific Islander/Native American).

‡

Additionally adjusted for categories of beer and liquor intake (drinks).

§

Additionally adjusted for categories of liquor and wine intake (drinks).

¶

Additionally adjusted for categories of wine and beer intake (drinks).

When examined by beverage type, those in the highest beer (HR = 3.61, 95 percent CI: 1.76, 7.39) and liquor (HR = 4.50, 95 percent CI: 2.39, 8.49) category were at significantly increased risk of ESCC relative to those who drank up to one drink per day. The protective association observed between drinking up to one alcoholic drink per day and ESCC risk seemed most related to wine consumption (table 4).

As those subjects with more formal education tended to drink more alcohol than those with less formal education, we explored possible effect modification of the alcohol–ESCC association by education. We observed no evidence of an interaction. Among those with a greater than high school education (66 cases), the ESCC HR estimates were 2.16 for nondrinkers, 2.58 for those who drank from more than one to three drinks per day, and 5.13 for those who drank greater than three alcoholic drinks per day.

Previous studies observed significant interactions between alcohol and smoking and risk of ESCC and other cancers of the upper gastrointestinal tract (31). We were unable to test for this interaction because of the very small number of ESCC cases that drank alcohol but did not smoke cigarettes (four of 97 cases). Although 17 cases of esophageal adenocarcinoma, 25 cases of gastric cardia adenocarcinoma, and 34 cases of gastric noncardia adenocarcinoma did not smoke yet drank alcohol, we did not observe a significant interaction between cigarette smoking and alcohol consumption for these sites (data not shown).

DISCUSSION

We evaluated the association between smoking and alcohol consumption and each of the major esophageal and gastric cancer types—ESCC, esophageal adenocarcinoma, gastric cardia adenocarcinoma, and gastric noncardia adenocarcinoma—in a prospective US cohort. In contrast to most previous prospective studies in Western populations (32–35), the large size of the NIH–AARP Diet and Health Study allowed us to examine these associations with each esophageal and gastric cancer type separately. Cigarette smoking significantly increased cancer risk at all four sites. These findings are consistent with previous studies showing a strong association between smoking and increased risk of ESCC, esophageal adenocarcinoma, and gastric cardia adenocarcinoma (15, 20, 36). In contrast, the association between smoking and gastric noncardia adenocarcinoma has been less consistent. Our results showing twofold increased risk in current smokers but no significant association with smoking intensity are consistent with both a meta-analysis (5) and the European Prospective Investigation into Cancer and Nutrition (EPIC) cohort study (4), and they suggest that smoking is a moderate risk factor for gastric noncardia adenocarcinoma.

Consistent with two meta-analyses (37, 38), consuming more than three alcoholic drinks per day was associated with increased risk for ESCC. We observed no association with esophageal adenocarcinoma and a nonsignificant positive association with gastric cardia adenocarcinoma. Previous studies have shown both no association (15, 17, 18, 20) and increased risk (14, 16, 19, 39). We observed a nonsignificant inverse association between alcohol intake and gastric noncardia adenocarcinoma. The majority of previous studies found no association, although two studies observed an inverse association (6, 40).

When analyzed by type of alcoholic beverage consumed, users in the highest quantile of beer and liquor, but not wine, were at significantly increased risk for ESCC. Previous studies indicate that wine drinking is a strong risk factor for ESCC (41); however, only 0.3 percent of the subjects (n = 1,462) in our study consumed more than three drinks of wine per day, limiting power. It is also possible that the relation between wine intake and cancer is confounded by lifestyle, although we did adjust models for smoking, education, liquor and beer intake, fruit and vegetable intake, and physical activity.

In our study, moderate drinking, up to or including one alcoholic drink per day, was associated with a significantly decreased risk of ESCC relative to nondrinkers. Some previous studies of alcohol and ESCC or “upper gastrointestinal tract” cancers have observed increased risk in nondrinkers relative to moderate drinkers (17, 18, 33, 34, 42). In addition, many studies have combined nondrinkers and moderate drinkers into the same referent category and therefore did not examine this question (18, 43–45). It is possible that the inverse association observed in our study is due to an unmeasured confounder. Increased education was associated with increased alcohol intake in our study, consistent with recent cross-sectional studies of alcohol intake in the United States (46, 47). The positive correlation between alcohol intake and increased education suggested that the protective association observed between modest alcohol intake and ESCC might result from confounding by socioeconomic status, a previously observed ESCC risk factor (15). Although educational level is an imperfect proxy of socioeconomic status, we stratified the cohort on this variable and examined the alcohol–ESCC association in less well educated and better educated subjects separately to look for evidence of residual confounding. The association between modest alcohol consumption and ESCC did not vary across educational levels, arguing against residual confounding. Alternatively, it is possible that sick, formerly heavy drinkers had stopped drinking at the time of the baseline questionnaire, resulting in exposure misclassification. Neither excluding the first 2 years of follow-up nor excluding those 57,171 participants who reported poor or fair health at baseline appreciably altered the risk estimates (data not shown).

Esophageal, gastric cardia, and gastric noncardia adenocarcinomas arise from closely adjacent organs. Gastric cardia and esophageal adenocarcinomas are difficult to distinguish because tumors often overgrow the gastroesophageal junction. In our cohort, about 13 percent of gastric noncardia adenocarcinoma cases were coded as overlapping lesions of the stomach (ICD-O site code C16.8), and 28 percent were gastric adenocarcinomas without specific location (ICD-O site code C16.9), raising the possibility of misclassification. Misclassification of tumors could make the incidence and etiology of these cancers appear similar and limit the utility of examining these sites independently. We performed a sensitivity analysis and found that excluding code C16.8 and code C16.9 cases did not affect risk estimates. Furthermore, studies of the National Cancer Institute's Surveillance, Epidemiology, and End Results data using our same site classification system indicate that the rates of esophageal adenocarcinoma are increasing far more quickly than the rates of gastric cardia adenocarcinoma (10, 11), which further demonstrates the utility of these classifications.

Assuming a causal relation between smoking and each cancer site, smoking accounted for approximately 58 percent of esophageal adenocarcinoma (95 percent CI: 0.38, 0.72), 47 percent of gastric cardia adenocarcinoma (95 percent CI: 0.27, 0.63), and 19 percent of gastric noncardia adenocarcinoma (95 percent CI: 0.00, 0.37) risk in this cohort. We found no significant association between alcohol and esophageal adenocarcinoma risk, a nonsignificant positive association with gastric cardia adenocarcinoma risk, and a nonsignificant inverse association with gastric noncardia adenocarcinoma risk. In our study, the impact of tobacco smoking and alcohol drinking on gastric cardia adenocarcinoma etiology is more similar to esophageal adenocarcinoma than to gastric noncardia adenocarcinoma.

Our study had several strengths, including its prospective nature and very large size, allowing us to prospectively examine the association between alcohol and tobacco use and esophageal and gastric cancer in the US population and to compare the risk across esophageal and gastric cancer subtypes. To limit confounding, we adjusted our models for most other esophageal and gastric cancer risk factors. However, our study also had several limitations. We lacked information on Helicobacter pylori infection, past alcohol use, and smoking duration. H. pylori could confound the association between alcohol or smoking and gastric cancer, although the association between H. pylori and either smoking or alcohol use has been inconsistent (48–51). Lack of information on past alcohol use and smoking duration could result in exposure misclassification and/or possibly incomplete adjustment for confounding. Moreover, perhaps because of low response rates to the initial mailing or the AARP membership, participants in our cohort were more educated, less likely to be current smokers, and more likely to be non-Hispanic White than the US population (21), which may limit generalizability to other subpopulations. However, alcohol use by members of the cohort was similar to that in the US population (52).

In conclusion, we found an association between cigarette use and increased risk of each of the four cancer sites examined; alcohol use was associated with only ESCC risk. These results from our large prospective study indicate distinct etiologies for ESCC and the other three sites examined. The associations between alcohol and cigarette use and esophageal and gastric cardia adenocarcinoma risk appeared similar.

Abbreviations

Abbreviations
 
• CI

  confidence interval

 
• ESCC

  esophageal squamous cell carcinoma

 
• HR

  hazard ratio

 
• ICD-O

  International Classification of Diseases for Oncology

 
• NIH

  National Institutes of Health

This research was supported by the Intramural Research Program of the National Cancer Institute, National Institutes of Health.

Conflict of interest: none declared.