Tobacco Documents Online

Tobacco, Alcohol, and Socioeconomic Status and Adenocarcinomas of the Esophagus and Gastric Cardia Marilie D. Gammon, Janet B. Schoenberg, Habibul Ahsan, Harvey A. Risch, Thomas L. Vaughan, Wong-Ho Chow, Heidi Rotterdam, A. Brian West, Robert Dubrow, Janet L. Stanford, Susan T. Mayne, Diana C. Farrow, Shelley Niwa, William J. Blot, Joseph F. Fraumeng Jr. * Background: Incidence rates for adenocarcinomas of the esophagus and gastric cardia have risen steeply over the last few decades. To determine risk factors for these tumors, we conducted a multicenter, population-based, case-control study. Methods: The study included 554 subjects newly di- agnosed with esophageal or gastric cardia adenocarcinomas, 589 subjects newly diagnosed with esophageal squamons cell carcinoma or other gastric adenocarcinomas, and 695 con- trol subjects. Estimates of risk (odds ratios [ORs] and cor- responding 95% confidence intervals [CIs]) were calculated for the four tumor types separately and for esophageal and gastric cardia adenocareinomas combined. Results: Risk of esophageal and gastric eardia adenocarcinomas combined was increased among current cigarette smokers (OR = 2.4; 95% = 1.7-3.4), with little reduction observed until 30 years after smoking cessation; this risk rose with increasing inten- sity and duration of smoking. Risk of these tumors was not related to beer (OR = 0.8; 95% CI = 0.6-1.1) or liquor (OR = 1.1; 95% CI -- 0.8-1.4) consumption, but it was reduced for drinking wine (OR = 0.6; 95% CI = 0.5-0.8). Similar ORs were obtained for the development of noncardia gastric ad- enocarcinomas in relation to tobacco and alcohol use, but higher ORs were obtained for the development of esophageal squamous cell carcinomas. For all four tumor types, risks were higher among those with low income or education. Con- clusions: Smoking is a major risk factor for esophageal and gastric cardia adenocarcinomas, accounting for approxi- mately 40% of cases. Implications: Because of the long lag time before risk of.these tumors is reduced among ex- smokers, smoking may affect early stage carcinogenesis. The increase in smoking prevalence during the first two thirds of this century may be reflected in the rising incidence of these tumors in the past few decades among older individuals. The [J Natl Cancer Inst 1997;89:1277--84] The incidence rates for adenocarcinomas of the esophagus and gastric cardia have risen steeply in the United States and Europe during the past few decades, whereas the incidence of squamous cell carcinoma of the esophagus and of adenocarci- nomas located elsewhere in the stomach have remained stable or have decreased during this time period (1-5). Incidence rates for esophageal and gastric cardia adenocarcinomas are highest among white males, while rates for esophageal squamous cell carcinoma and noncardia gastric adenocarcinomas are highest among black males (1,3). The reasons underly[hg these c0ntrast-~ ing patterns of incidence are unclear. Tobacco smoking and alcohol use are strong risk factors for esophageal squamous cell carcinoma (6), whereas smoking is only weakly related to adenocarcinomas in the lower stomach (7). Several studies (8-18) have reported only a slight excess in risk of esophageal or gastric cardia adenocarcinomas associated with smoking and drinking. It has been suggested that patients with adenocarcinomas of the esophagus and gastric cardia have a higher income and more years of education than those with esophageal squamous cell carcinoma or noncardia gastric ad- enocarcinomas (2,13,15). Esophageal adenocarcinomas are often located in the lower third of the esophagus near the gastroesophageal junction, and distinguishing adenocarcinomas arising in the lower esophagus from those arising in the gastro-esophageal junction are often very difficult (19,20). Tumors arising in the gastro-esophageal junction itself are classified by the Surveillance, Epidemiology, and End Results (SEER)t Program as being located in the gastric cardia (21). Because of similar incidence patterns and anatomic proxim- ity, it has been hypothesized that the origins of esophageal and gastric cardia adenocarcinomas are similar to one another and are distinct from the causes of esophageal squamous cell carci- noma and other gastric adenocarcinomas (3). To clarify this is.sue, we undertook a large collaborative, population-based study to identify risk factors for adenocarcinomas of the esopha- gus and gastric cardia. For comparison, we also determined risk factors for squamous cell carcinoma of the esophagus and non- cardia adenocarcinomas of the stomach. *Affiliations of authors: M. D. Gammon, H. Ahsan, Division of Epidemiol- Applied Cancer Epidemiology Program, New Jersey Department of Health and Senior Services, Trenton; H. A. Risch, R. Dubrow, S. T. Mayne (Department of Epi-demiology and Public Health), A. B. West (Department of Pathology), Yale University School of Medicine, New Haven, CT; T. L. Vaughan, J. L. Stanford, D.C. Farrow, Program in Epidemiology, Fred Hutchinson Cancer Research Center, Seattle, WA, and Department of Epidemiology, University of Washing- ton School of Public Health, Seattle; W.-H. Chow, J. F. Fraumeni, Jr., Division of Cancer Epidemiology and Genetics, National Cancer Institute, Bethesda, MD; H. Rotterdam, Department of Pathology, College of Physicians & Surgeons of Columbia University, New York, NY; S. Niwa, Westat, Inc., Rockville, MD; W.J. Blot, International Epidemiology Institute. Rockville. Correspondence to: Marilie D. Gammon, Ph.D.. 622 W. 168th St., PHIS-107, New York, NY 10032. See "Notes" following "References." © Oxford University Press This article is for individual use only and may not be fi~her reproduced o~ s~ored elecmmically without writt~m permissi~a from d~e copyright I~tder. gruaulhon.r~d r~predue~ may r~sult in financial and other penalilies, (c) NATL CANCEI1. INSTITUTg

Subjects and Methods This mulficente.r, case-control study was conducted in three geographic areas of the United States with populatinn-based rumor registdes---tbe state of Con- necticut, a 15-county area of New Jersey, and a three-county area of west~ru Washington state. The goal of the collaborative effort was to identify, recruit, and interview four population-based case groups of roughly equal size contain- ing subjects newly diagnosed with I) esophageal adenocarcinoma, 2) gastric cardia adenocarcinoma, 3) esophageal squamous cell carcinoma, or 4) other gastric adenocarcinomas. The investigation was performed after approval from our institutional review boards and in accord with an assurance filed with and approved by the U.S. Department of Health and Human Services. Potentially eligible case subjects were English-speaking men and women who were 30-79 years of age and who were diagnosed with primary invasive cancer of the esophagus or stomach from February I, 1993, through January 31, 1995, in Connecticut; from April 1, 1993, through November 30, 1994, in New Jersey; and from March I, 1993, through February 28, 1995, in Washington. All ease subjects diagnosed with adenecareinomas of the esophagus or gastric cardia (target ease subjects) were considered eligible for the study. Those diagnosed with squamous cell carcinoma of the esophagus or adenecarcinomas located elsewhere in the stomach (comparison ease subjects) were sampled by frequency matching to the expected distribution of the target case subjects on the basis of geographic area and 5-year age group in Connecticut, New Jersey, and Wash- ington; on the basis of sex in New Jersey and Washington; and on the basis of race (white or other) in New Jersey. All case subjects were identified by use of established rapid'reporting sys- tems. Pathology reports were obtained, for all potentially eligible ease patients, and initial subject selection was based on the review of these records. Patients with tumors classified as not otherwise specified (NOS), mixed, or undifferen- tiated or with tumors of uncertain histologic type were initially considered eli- gible and approached for study participation, as were those with tumors in an unspecified subsite of the stomach. Final determination of case subject eligibility was based on a systematic review by the study pathologists grI. Rotterdam for New Jersey and A. B. West for Connecticut and Washington) using standardized criteria. The site of tumor origin was determined by a review of pathology slides and medical records, including pathology, radiology, surgery, and endoscopy reports. For tumors that involved the distal esophagus as well as the gas~e cardia or proximal stomach, the site of origin was determined by estimating the location of the tumor's center using endoscopic, surgical, and pathologic data. Histologie slides of diagnostic biopsy and resection specimens and slides of tumor brushings and other cytologic preparations were reviewed in more than 99% of the cases. For those patients diagnosed with an indeterminate site of tumor origin by the initial study pathologist, records were rereviewed by the other study pathologist; disagreements were resolved by consensus. Population-based control subjects were frequency matched to the expected distribution of target ease subjects by 5-year age group and sex. Control subjects who were 30-64 years of age were identified by use of Waksberg's random- digit-dialing (RDD) method (22); those who were 65-79 years of age were identified by means of random sampling of Health Care Financing Administra- tion (HCFA) rosters. Face-to-face interviews were obtained for 554 (80.6%) of the eligible target case subjects, 589 (74.1%) of the eligible comparison ease subjects, and 695 (73.7%) of the eligible control subjects. If the telephone screener response rate of 90.8% is taken into account for the 51.9% of control subjects who were identified by use of RDD, the overall response rate among control subjects was 70.2%. The primary reason for nonparticipation was subject refusal (12% of target case subjects, 17% of eomparisen ease subjects, and 23.3% of control subjects), followed by physician refusal for ease subjects (4% for each group). Interviews were administered directly to the study subject, rather than to the closest next of kin (usually the spouse), for 70.4% of the target case subjects, 67.8% of the comparison case subjects, and 96.6% of the control subjects. For case subjects, the mean length of time between cancer diagnosis and the inter- view was 3.7 months when the interview was conducted with the subject and 8.5 months when the interview was conducted with a proxy. Prior to the interview, written informed consent was obtained from all sub- jects. During the interview, a structured questionnaire was administered by trained interviewers, and the average time to complete the questionnaire was 130 minutes. Information was collected on demographic characteristics, tobacco and alcohol use, other beverage consumption, medical history, use of medications, diet, and oecupational history. The interview also elicited details on usual to- bacco use anytime prior to I year before the interview, including the product type used (cigarettes, cigars, pipes, chewing tobacco, or snuff) as well as the intensity of its use, the age started and stopped, the total duration of use exclud- ing the years stopped, and the years since last use for each type of product; for cigarette smoking, it was determined whether or not filtered cigarettes were used. A never smoker was defined as having smoked less than 100 cigarettes ever or having smoked less than one cigarette per day for 6 months or longer. An ex-smoker ~,as defined as having stopped smoking 2 or more years before the interview. Alcohol consumption patterns were assessed by inquiring about the usual intake anytime prior to 1 year before the interview for each type of beverage separately, i.e., beer, wine, or liquor. A never drinker was defined as having consumed less than one drink per month. One drink was defined as 12 ounces of beer, 4 ounces of wine, or 1 ounce of hard liquor. The three types of alcohol were also combined to form an overall estimate of use. Unconditional logistic regression was used to calculate odds ratios (ORs), as an estimate of the relative risk, and corresponding 95% confidence intervals (CIs) (23) for each of the four tumor types (esophageal adenoeareinoma, gastric eardia adenecarcinoma, esophageal squamous eel/carcinoma, and other gastric adenocareinomas) and for the combined category of esophageal and gastric ¢ardia adenocarcinomas in relation to tubaeco, alcohol, education, and income. All models included as eovariates the frequency-matched factors of geographic center (Connecticut/Washington/New ]rersey, entered as indicator variables), age (in quartiles and entered as indicator variables), sex (female/male), and race (white/black/other, entered as indicator variables). Logistic regression models were also used to adjust for the confounding effects of body mass index (BMI; expressed as weight in kilograms divided by the square of height in meters) (entered as a continuous variable) and income (entered as an ordered categoric variable). The 5% of persons with missing information on income were assigned values derived from simple regression models for case and control subjects that included age, race, sex, and education. Omission of subjects with missing'income information from the logistic models did not materially alter the estimates of effect for tobacco, alcohol, education, or income. Inclusion of the imputed variables for these persons, however, permitted more precise estimates to be calculated for small subgroups of interest. There- fore, results from models that included all subjects are presented. In models that adjusted for the confounding effects of cigarette smoking, the factor was entered as an indicator variable (current smoker/ex-smoker/nousmoker). Similarly, to adjust for the confounding effects of beer, hard liquor, and wine, each factor was entered as a dichotomous variable (ever/never). Subject characteristics that did not confound our results include edacation, family history of cancer, history of other medical conditions such as ulcers, use of various medications, and calorie intake. Tests for trend in the ORs across exposuse strata were calculated using logistic models that included continuous variables and, where appropriate, omitted never users. Potential effect modification between smoking and alcohol use or with other variables, such as age, sex, center, and race, was evaluated assuming a multiplieative model using logistic regression with cross-product terms repre- senting the interaction between the two variables. ORs and corresponding CIs derived from polytomous logistic regression mod- els (23) for the four tumor types were nearly identical to those obtained from standard unconditional logistic regression. Thus, only the values obtained from the latter method are shown. To compare two continuous variables, Pearson's correlation coefficient was calculated (24). Population attributable risk estimates were calculated (25). All reported P values are from two-sided tests. Results Table I shows the distribution of study participants according to demographic characteristics. Approximately 98% of the case subjects with esophageal and gastric cardia adenocarcinomas were white and 85% were males, whereas the corresponding percentages were 93% and 80% for the control subjects and 81% and 73% for the comparison case subjects. The median age at diagnosis was 66 years, with the case subjects being slightly older on average than the control subjects. About half of the study participants were from New Jersey (46.5%), while 30.9% were from Connecticut and 22.6% were from Washington. Table 2 shows the estimates of risk according to tumor type 1278 ARTICLES Journal of the National Cancer Institute, Vol. 89, No. 17, September 3, 1997

Table 1. Distribution of demographic characteristics among case subjects (by tumor type) and control subjects in Connecticut, New Jersey, and western Washington state, 1993-1995 Control subjects Esophageal Esophageal Gastdc cardia squamous Other gastdc adenocarcinoma adenocarcinoma cell carcinoma adenocarcinoma case subjects case subjects case subjects case subjects No. ' % No. % No. % Characteristic (n -- 695) (n = 293) (n = 261) No. % No. % (n = 221) (n = 368) Age, y <57 179 25.8 57-64 178 25.6 65-71 176 25.3 >71 162 23.3 Sex Men 555 79.9 Women 140 20.1 C_reogrnphie center Connecticut 206 29.6 New Jersey 333 47.9 Washington 156 22.5 Race White 646 93.0 Black 34 4.9 Others 15 2.2 76 25.9 65 24.9 34 15.4 65 17.7 48 16.4 56 21.5 53 24.0 61 16.6 79 27.0 71 27.2 74 33.5 93 25.3 90 30.7 69 26.4 60 27.2 149 40.5 245 83.6 223 85.4 176 79.6 254 69.0 48 16.4 38 14.6 45 20.4 114 31.0 80 27.3 82 31.4 83 37.6 117 31.8 138 47.1 113 43.3 99 44.8 . 172 46.7 75 25.6 66 25.3 39 17.7 79 21.5 289 98.6 252 96.6 168 76.0 307 83.4 2 0.7 4 1.5 48 21.7 36 9.8 2 0.7 5 1.9 5 2.3 25 6.8 in relation to indicators of socioeconomic status. A decrease in risk for all types was noted with increasing levels of education or income; the inverse association with income was most pro- nounced for esophageal squamous cell carcinoma. The adjusted (for age, sex, geographic center, race, smoking, alcohol use, BMI, and education) ORs for esophageal adenocarcinoma and for gastric cardia adenocarcinoma were 0.5 (95% CI = 0.3-1.0) and 0.8 (95% CI = 0.4--1.6), respectively, among respondents with incomes of $75 000 per year or more compared with those with incomes of less than $15 000. Similarly, the corresponding adjusted ORs for graduate education compared with having less than a high school diploma were 0.7 (95% CI = 0.3-1.3) and 0.8 (95% CI = 0.4-1.6), respectively. Pearson's correlatiou co- efficient between income and education among control subjects was .54 (P = .01). Table 3 shows the estimates of risk associated with various patterns of cigarette smoking. With adjustments made for the confounding effects 'of age, sex, geographic center, race, BMI, income, and alcohol use, the risk of esophageal and gastric car- dia adenocarcinomas was doubled among current smokers (OR = 2.2; 95% CI = 1.4-3.3 and OR = 2.6; 95% CI = 1.7-4.0, respectively) as well as among ex-smokers (OR = 2.0; 95% CI = 1.4-2.9 and OR = 1.9; 95% CI = 1.3-2.9, respectively). An increase in risk persisted up to 30 years after cigarette use had ceased. The adjusted ORs also increased with increasing years of smoking and with the ttumber of cigarettes smoked per day for Table 2. Adjusted* odds ratios (ORs) and corresponding 95% confidence intervals (CIs) for esophageal adenocacinoma, gastric cardia adenoearcinoma, esophageal squamous cell carcinoma, and other gastric ad~nocareinomas in relation to education and income Esophageal Gastric eardia Esophageal squamous Other gastric adenoearcinoma adenoearcinoma ceil carcinoma adenocareinoma ease subjects case subjects case subjects ease subjects No. of Characteristic control subjects . No. OR (95% CI') No. OR (95% CI) No. OR (95% CI) No. OR (95% CI) Education <12 years'~ 130 65 1.0 56 1.0 88 1.0 117 1.0 12 years . 178 93 1.3 (0.9-2.1) 85 1.3 (0.8-2.0) 67 1.0 (0.6-1.6) 108 0.8 (0.5-1.1) Vocational school 52 25 1.2 (0.7-2.2) 20 1.1 (0.6-2.1) 15 0.7 (0.3-1.5) 25 0.5 (0.3-0.9) Some college 123 53 1.2 (0.7-2.0) 41 1.0 (0.6-1.7) 23 0.6 (0.3-1.1) 56 0.7 (0.4-1.0) College graduate 118 39 1.0 (0.6-1.8) 36 0.9 (0.5-1.6) 19 0.8 (0.4-1.6) 36 0.6 (0.3-1.0) Graduate school 94 18 0.7 (0.3-1.3) 23 0.8 (0.4-1.6) 9 0.7 (0.3-1.9) 25 0.6 (0.3-1.1) Income, $ per y <15 000"~ 93 60 1.0 41 1.0 71 1.0 87 1.0 15 000-29 999 177 87 0.7 (0.4-1.1) 81 1.0 (0.6-1.6) 75 0.7 (0.4-1.1)~ 15 0.8 (0.6-1.3) 30 000-49 999 175 69 0.5 (0.30.8) 65 0.8 (0.5-1.3) 56 0.6 (0.3-1.1) 100 1.1 (0.7-t.7) 50 000-74 030 126 42 0.5 (0.3-1.0) 37 0.7 (0.4-1.4) 9 0.2 (0.1-0.4) 46 0.9 (0.5-1.5) >75 0Q0 124 35 0.5 (0.3-1.0) 37 0.8 (0.4-1.6) 10 0.2 (0.1-0.6) 20 0.5 (0.2-0.9) *Adjusted for age; sex; geographic center (Connecticut, New lersey, and Washington); race; body mass index (weight in kilograms divided by the square of height in meters); eigarera," smoking; and use of beer, wine, and liquor. ~'Reference category. :~Two-sided test for trend; P for trend = .001. Journal of the National Cancer Institute, Vol. 89, No. 17, September 3, 1997 ARTICLES 1279

Table 3. Adjusted* odds ratios (ORs) and corresponding 95% confidence in~rvals (CIs) for esophageal adenocacinoma, gas~c cm'dia adenocarcinoma, esophageal squamous cell carcinoma, and other gastric adenocarcinomas in relation to cigarette smoking Esophageal Gastric cardia Esophageal squamous Other gastric adenncarcinoma adenocareinoma cell carcinoma adenocarcinoma case subjects case subjects case subjects case subjects Cigarette smoking No. of control subjects No. OR (95% CI) No. OR (95% CI) No. OR (95% CI) No. OR (95% CI) Smoking status Never smoke~ 244 63 1.0 53 1.0 22 1,0 106 1.0 Current smoker I55 86 2.2 (1.4-3.3) 85 2.6 (1.7--4.0) 108 5.1 (2.8-9.2) 96 1.8 (1.2-2.7) Ex-smoker 296 144 2.0 (1.4-2.9) 123 1.9 (1.3-2.9) 91 2.8 (1.5-4.9) 164 1..5 (1.1-2.1) Smoking cessation, y Stopped <11 74 44 2.7 (1.6-4,4):[: 45 2.9 (1.8-4.8):[: 47 5.6 (2.9-10.8)~ 50 1,8 (1.2-2.9)~: Stopped 11-20 77 43 2.3 (1.4-3.8) 26 1.6 (0.9-2.8) 24 2.3 (1.1-4.8) 49 1.7 (1.0-2.7) Stopped 21-30 78 31 1.9 (1.1-3.2) 34 2.2 (1.3-3.'/) 8 1.0 (0.4-2.7) 34 1.5 (0.9-2.4) Stopped >30 67 26 1.2 (0,7-2.2) 18 1.1 (0.6-2.0) 12 1.8 (0.8-4.2) 31 1,0 (0.6--1.8) Smoking intensity (No./day) <16 134 49 1.5 (1.0-2,4):1: 42 1.4 (0.9--2.2):]: 44 2.7 (1.4-5.1):]: 82 1.5 (1.0-2.2) 16-20 148 78 2.2 (1.4-3.4) 71 2.2 (1.4-3.4) 62 3.9 (2.1-7.2) 94 1.7 (1.2-2.6) 21-30 71 49 3.1 (1.9-5.1) 46 3.1 (1.9.5.2) 36 5.3 (2.6-10.7) 30 1.4(0.9.2.5) >30 98 54 2.1 (1.3--3.3) 46 2.0 (1.2-3.3) 57 3.9 (2.0-7.6) 52 1.5 (1.0-2.4) Duration of smoking, y <20 119 43 1.4 (0.9-2.2)~: 40 1.6 (1.0-2.6)~: 17 1.8 (0.9--3.7)~: 44 1.0 (0.7-1.6)~: 20-31 112 45 1.7 (1.0-2.8) 41 1.8 (I.I-2.9) 27 2.0 (I.0-4.0) 55 1.6 (1.0-2.4) 32-42 115 75 2.9 (1.8--4.4) 61 2,7 (1.7-4.2) 53 3.3 (1.8-6.1) 69 1.8 (1.2--2.7) :>42 105 . 66 2.4 (1.5-3.7) 66 2.9 (1.8-4.7) 99 5.9 (3.2-10.7) 90 2.1 (1.4-3.1) Pack-years§ <14 115 38 1.4 (0.8.-2.2):1: 24 0.9 (0.5-1.6)~: 23 2.0 (1.0--4.0)~: 51 1.2 (0.8-.-1.8)~: 14-31 111 40 1,6 (1.0-2.6) 52 2.3 (1.4--3.6) 30 2.8 (1.4-5.5) 59 1.5 (1.0-2.4) 32-54 121 76 2,9 (1.8-4.5) 69 2.8 (1.8-4.4) 62 4.5 (2.4-8.5) 74 1.7 (1.2-2.6) >54 104 76 2.8 (1.8-4.4) 60 2.5 (1.5-4.1) . 84 5.8 (3.1-11.0) 74 2.1 (1.3.--3.2) Filter status Filtered only 240 120 2.0 (1.4-2.9) 109 2.1 (1.4-3.1) 95 2.9 (1.7-5.0) 139 1.6 (1.1-2.2) Filtered and nonfiltered 62 29 1.7 (1.0-3.0) 23 1.5 (0.8-2.7) 26 2.7 (1.4-5.6) 28 1.1 (0.6-1.9) Nonfiltered only 148 76 1.9 (1.2-2.9) 73 2.1 (1.3-3.2) 74 3.6 (2.0-6.4) 88 1.5 (1.0-2.3) *Adjusted for age; sex; geographic center (Connecticut, New Jersey, and Washington); race; body mass index (weight in kilograms divided by the square of height in meters); income; and use of beer, wine, and liquor. "[q'he raferenee category for all estimates in this table is "Never smoker" (see "Subjects and Methods" section for explanation of smoker classification). z~/'wo-sided tests for trend; all P for trend ~ .05. §Pack-years -- the number of packs per day multiplied by the number of years smoking. both tumors, but risks were similar in users of filtered and non- filtered cigarettes. For subjects with esophageal squamous cell carcinoma, the adjusted OR associated with current cigarette smoking was 5.1 (95% CI = 2.8-9.2) but the OR dropped to 2.3 (95% CI = 1.1- 4.8) by 11-20 years after smoking cessation and to 1.8 (95% CI = 0.8-4.2) 30 years after quitting smoking. For noncardia gas- tile adenocarcinomas, the ORs were 1.8 (95% CI = 1.2-2.7) and 1.5 (95% CI = 1.1-2.1) for current and ex-smokers, respec- tively. Case subjects with esophageal and gastric cardia adenocarci- nomas were not more likely than control subjects or comparison pipes, chewing tobacco, or snuff (data not shown). Table 4 shows the ORs for alcohol intake. With adjustments made for smoking and other confounders, adenocarcinomas of the esophagus, the gasr.ric cardia, ur ~dier gastz~co,,~o-:'~ dSd .,e~" ~ appear to be associated with drinking beer or hard liquor. In contrast, the risk of esophageal squamous cell carcinoma was doubled in relation to ever use of beer and tripled in relation to ever use of liquor. The risk estimates rose with increasing con- sumption of either product for this type of tumor. Ever drinking of wine was associated with a decreased risk of adenocarcino- mas of the esophagus (OR = 0.6; 95% CI = 0.4-0.8) and gastric cardia (OR = 0.6; 95% CI = 0.5-0.9) as well as a 1280 ARTICLES reduced risk of esophageal squarnous cell carcinoma (OR = 0.6 95% CI = 0.4-0.9) and of noncardia gastric adenocarcinom~. (OR = 0.7; 95% CI = 0.5-0.9). However, there was no appar ent trend in risk with increasing number of drinks of wine pe week. Also shown in Table 4 are the risks associated with a corn bined estimate of alcohol consumption in comparison with neve use of any type of alcohol. Risk in relation to total alcohol intak was decreased 30% (OR = 0.7; 95% CI = 0.5-1.0) for esopt" ageal adenocarcinoma, 30% (OR = 0.7; 95% CI = 0.5-1.1) f¢ gastric cardia adenocarcinoma, and 20% (OR = 0.8; 95% CI -- 0.6-1.1) for nongastrie cardia adenocarcinomas; however, non • z_f t~ :~_d,,~'~;~,,,~ ;,~ ,-;~r w~ ~tatisticalIv significant. In contras the risk of esophageal squamous cell carcinoma was signif eanfly elevated 3.5-fold (OR = 3.5; 95% CI = 1.9-6.2) for ew versus never users of any alcoholic beverage; this risk increase w~_t~ 6si.ng ~evet~ of intake ~eaching an OR of 7.4 (95% CI ." 4.0-13.7) among those who consumed more than 30 alcohol drinks per week. Since the patterns of risk observed in this study were simil for adenocarcinomas of the esophagus and gastric cardia, tl cases were combined for additional analyses (Table 5). Ri associations with the highest levels of income and educatic compared with the lowest, were decreased (OR = 0.6; 95%, = 0.4-1.1 and 0.7; 95% = 0.4-1.3, respectively), but the ~ Journal of the National Cancer Institute, Vol. 89, No. 17, September 3, I~ 2063629686

Table 4. Adjusted* odds ratios (ORs) and corresponding 95% confidence intervals (CIs) for esophageal adenocarcinoma, gastric cardia adenoearcinoma, esophageal squamous cell carcinoma, and other gastric adenocarcinomas in relation to alcohol consumption by typ~ Esophageal Gastric cardia Esophageal squamous Other gastric adenocarcinoma adenocareinoma cell carcinoma adenocarcinoma case subjects ease subjects case subjects case subjects No. of Type of alcohol control subjects No. OR (95% CI) No. OR (95% CI) No. OR (95% CI) No. OR (95% CT) Be~r Ncvcrl" 310 130 1.0 116 1.0 57 1.0 200 1.0 Ever 385 163 0.9 (0.6-1.2) 145 0.8 (0.6--1.2) 164 2.2 (1.4-3.3) 166 0.8 (0.6-1.1) Drinks/wk <2 86 35 0.9 (0.6-1.5) 22 0.6 (0.4-1.1) 22 1.4 (0.7-2.7)~ 42 0.8 (0.5--1.3) 2-4 112 35 0.7 (0.4-1.1) 40 0.9 (0.5-1.4) 18 1.1 (0.5-2.1) 35 0.6 (0.4-1.0) 5-t2 98 30 0.6 (0.3---0.9) 38 0.8 (0.5-1.2) 32 1.7 (0.9-3.0) 35 0.7 (0.4-1.1) >12 89 58 1.1 (0.7-1.7) 43 0.9 (0.6-1.5) 83 2.6 (1.6-4.4) 47 0.9 (0.5--1.4) Liquor Nevert 320 130 1.0 116 1.0 48 1.0 188 1.0 Ever 375 161 1.2 (0.9--1.7) 145 1.0 (0.7-1.4) 173 3.1 (2.0-4.8) 177 1.0 (0.8-1.4) Drinks/wk <2 106 35 0.9 (0.5-1.4) 35 1.0 (0.6-1.5) 19 1.4 (0.8--2.8)~: 58 1.2 (0.8-1.9) 2-4 85 33 1.1 (0.7-1.9) 32 1.1 (0.7-1.9) 18 1.6 (0.8--3.2) 37 1.0 (0.6-1.6) 5-14 116 52 1.2 (0.8-1.9) 51 1.0 (0.6-1.7) 46 2.2 (1.3-3.7) 45 0.8 (0.5--1.2) >14 64 34 1.2 (0.7-2.0) 24 0.8 (0.4-I.4) 82 5.4 (3.1-9.2) 33 0.9 (0.5-1.6) Wine Neve~ 389 205 1.0 176 1.0 149 1.0 258 1.0 Ever 306 88 0.6 (0.4-0.8) 84 0.6 (0.5-0.9) 72 0.6 (0.4-0.9) 108 0.7 (0.5-0.9) Drink~/wk <2 106 34 0.7 (0.5-1.2) 30 0.7 (0.4-1.1) 17 0.6 (0.3-1.1) 44 0.8 (0.5-1.2) 2-3 48 12 0.6 (0.3-1.2) 12 0.6 (0.3-1.2) 8 0.6 (0.3-1.5) 9 0.4 (0.2-0.9) 4-7 98 27 0.6 (0.4-0.9) 20 0.5 (0.3-0.8) 18 0.7 (0.4-1.3) 38 0.8 (0.5-1.2) >7 50 13 0.5 (0.2-0.9) 19 0.8 (0.5-1.5) 25 0.8 (0.4-1.4) 16 0.6(0.3-1.1) Any alcohol NeverS" 172 79 1.0 63 1.0 19 1.0 " 125 1.0 Ever 523 210 0.7 (0.5-1.0) 196 0.7 (0.5-1.1) 195 3.5 (1.9-6.2) 238 0.8 (0.6-1.1) Drinks/wk <5 161 56 0.7 (0.4-1.0) 46 0.6 (0.4-1.0) 16 0.8 (0.4-1.6)~ 74 0.7 (0.5-1.1) 5-11 134 45 0.6 (0.4--0.9) 59 0.8 (0.5-1.3) 25 1.8 (0.9-3.5) 68 0.9 (0.6-1.3) 12-30 138 57 0.7 (0.4-1.1) 52 0.7 (0.4-1.1) 48 2.9 (1.5-5.4) 55 0.7 (0.4-1.0) >30 90 52 0.9 (0.5-1.4) 39 0.7 (0.4-1.2) 106 7.4 (4.0--13.7) 41 0.6 (0.4-1.0) *Adjusted for age; sex; geographic center (Connecticut, New Jersey, and Washington); rac~, body mass index in meters); income; cigarette smoking; and all other types of alcohol use. "~Referenee category (see "Subjects and Methods" section for explanation of drinker classification). ~.'Two-sided tests for trend; all P for trend ~ .05. (weight in kilograms divided by the squar~ of height ductions were not statistically significant. A greater than twofold increase in risk was associated with ever smoking of cigarettes, which persisted up to 30 years after quitting. Attributable risk calculations revealed that 41% of all esophageal and gastric cardia adenocarcinomas combined are attributable to cigarette smoking (data not sllown). As also shown in Table 5, there was no association between adenocarcinomas of the esophagus and gastric cardia and the consumption of beer or liquor, but a reduction inrisk was seen There was little heterogeneity in the ORs for esophageal and gastric cardia adenocarcinomas combined, or for the other tumor types, in relation to smoking, drinking, or socioeconomic status across subgroups categorized by age, sex, geographic center, or race (data not shown). Table 6 shows the estimates of risk for the four tumor types in relation to cigarette smoking stratified by alcohol drinking. The adjusted ORs for subjects who both smoked and drank were not substantially different from those who just smoked. Al- though the risk of esophageal squamous cell carcinoma more than doubled among smokers and wine drinkers (OR = 6.8; 95% CI -- 2.2-21.1) compared with just smokers (OR = 2.8; Ioumal of the National Cancer Institute, Vol. 89, No. 17, September 95% CI = 1.5-5.3), the apparent effect modification was not statistically significant (P = .39). The statistical analyses were repeated using a sample re- stricted to only subjects who were interviewed directly, i.e., excluding subjects with proxy respondents (data not shown). In this restricted sample, the risk estimates for the four tumor types in relation to smoking and drinking varied little from those ob- tained in analyses that included all participants. Thus, only the analyses based on the entire sample are presented. Discussion In this examination of tobacco, alcohol, and socioeconomic factors, we found similar estimates of effect for the development of esophageal and gastric cardia adenocareinomas. When the two tumor types were combined, risk was reduced 40% in rela- tion to an income of $75 000 or more and reduced 30% in relation to a graduate school education. Risk was increased 2.4 times in relation to currently smoking cigarettes, but it was unaffected by alcohol drinking, except for a 40% decrease as- sociated with wine drinking. Risk estimates associated with in- come, education, and smoking status were similar in direction 3, 1997 ARTICLES 1281

Tab|e 5. Adjusted* odds rados (ORs) and 95% confidence intervals (CIs) for esophageal and gastric cardia adenecarcinomas combined in relation to education, income, cigarette smoking, and alcohol consumption Esophageal and gastric cardia adenoearcinoma case subjects No. of Characteristic control subjects No. OR (95% Ca) Education <12 years~" 130 121 1.0 12 year~ 178 178 1.3 (0.9-1.9) Vocational school 52 45 1.2 (0.7-1.9) Some college 123 94 1.1 (0.7-1.7) College graduate 118 75 1.0 (0.6-1.5) Graduate school 94 41 0.7 (0.4-1.3) Income, $ per y <15 000~" 93 101 1.0 15 000-29 999 177 168 0.8 (0.5--1.2) 30 000-49 999 175 134 0.6 (0.4-0.9) 50000-74000 126 79 0.6 (0.4-1.0) >751300 124 72 0.6 (0.4-I. 1 ) Cigarette smoking Never smoker]" 244 116 1.0 Current smoker 155 171 2.4 (1.7-3.4) Ex-smoker 296 267 2.0 (1.5-2.7) Smoking cessation, y Stopped <11 74 89 2.9 (1.9-4.3):1: Stopped 11-20 77 69 1.9 (1.3-3.0) Stopped 21-30 78 65 1.9 (1.3-3.0) Stopped >30 67 44 1.2 (0.8-2.0) Beer Neve~ 310 246 1.0 Ever 385 308 0.8 (0.6-1.1) Drinks/wk <2 86 57 0.8 (0.5-1.2) 2-4 112 75 0.8 (0.6-1.2) 5-12 98 68 0.7 (0.4-1.0) >12 89 I01 1.0 (0.7-1.5) Liquor Neve~ 320 246 1.0 Ever 375 306 1.1 (0.8-1.4) Drink.s/wk <2 106 70 0.9 (0.6-1.4) 2-4 85 65 1.1 (0.7-1.7) 5-14 116 103 I.I (0.8--1.5) >14 64 58 1.0 (0.6-1.5) Wine Nevert 389 381 1.0 Ever 306 172 0.6 (0.5-0.8) Drinka/wk <2 106 64 0.7 (0.5-1.0) 2-3 48 24 0.6 (0.3-1.0) 4-7 98 47 0.5 (0.3-0.8) >7 50 32 0.7 (0.4-1.1) Any alcohol Never" 172 142 1.0 Ever 523 412 0.8 (0.6-1.0) Drinks/wk <5 161 102 0.7 (0.5--0.9) 5-II 134 104 0.7 (0.5-1.0) 12-30 138 109 0.7 (0.5-1.0) >30 ~ o~ N ~ ¢N <-~ 2~ *OR~ are adjusted for age; sex; geographic center (Connecticut, New Jersey, and Washington); body mass index (weight in kilograms divided by the square of height in meters); and all other variables in the table. "~Referenee category (see "Subjects and Methods" section for explanation of smoker and drinker classifications). ~Two-sided test for trend; P for trend =~ .005. to those observed for the development of noncardia gastric ad- enocarcinomas and esophageal squamous cell carcinoma, al- though the risks were higher for the development of esophageal squamous cell carcinoma. The risk associated with wine con- sumption was reduced for all tumor types in our study, but the risks for beer and liquor intake were elevated twofold and three- fold, respectively, for the development of esophageal squamous cell carcinoma. To our knowledge, this investigation is the largest popula- tion-based study of esophageal and gastric adenocarcinomas conducted to date. The study encompassed all incident cases of cancer of the esophagus and stomach identified in three geo- graphic areas of the United States, with a standardized review of pathology specimens and medical records to assign the site of tumor origin and to confirm the histologic type. A comprehen- sive, personal questionnaire, which was administered by trained interviewers, was employed in the data collection. Despite this systematic approach, some limitations of the study must be con- sidered when interpreting our results. One problem relates to the difficulty of determining the exact site of origin for tumors arising near the gastro-esophageal junc- tion. Thus, it is possible that misclassification of adenocarcino- mas of the lower esophagus and the gastric cardia may contrib- ute to the similar patterns of risk noted for these tumors. Although the incidence rates for esophageal and gastric ad- enocarcinomas have risen dramatically over time, these tumors are still relatively uncommon, which limits our ability to identify subgroup effects. To maximize study efficiency, the comparison case subjects (i.e., those with esophageal squamous cell carci- noma or noncardia gastric adenoearcinomas) were frequency matched to the expected distribution of the target case subjects (i.e., those Wi, th esophageal and gastric cardia adenocarcinomas) on the basis of age and geographic area at all three centers; on the basis of sex in New Jersey and Washington; and on the basis of race in New Jersey. Because comparison cases were matched to target cases on the basis of race (and incidence rates for these latter tumors are higher among whites than among blacks), the comparison case subjects in our study do not reflect the under- lying distributions for all cases of esophageal squamous cell carcinoma or noncardia gastric adenocarcinomas, for which in- cidence ra~es are substantially higher among blacks than among whites. Approximately 30% of our case subject interviews were con- ducted with a proxy respondent whose knowledge of the expo- sure status of the case subject may be incomplete (26). However, it is reassuring that analyses restricted to subjects with self- reports yielded results similar to analyses for all study partici- pants, including those with next-of-kin interviews. Others (27) have also found that use of proxy-reported responses for ciga- rette smoking do not yield substantially biased estimates. Our findings from this large case-control study are consistent wltla earlier population-based studies (13,15) implicating ciga- rette smoking as a risk factor for esophageal and gastric cardia adenocarcinomas. The 140% increase in risk among current smokers observed in our data was not as great as that observed for squamous cell carcinoma of the esophagus, but it was slightly higher than the excess risk observed for adenocarcino- mas of the lower stomach. It is noteworthy that the risk for esophageal and gastric cardia adenocarcinomas among ex- smokers persisted at this elevated level for more than 20 years. On the basis of attributable risk calculations, 41% of all esoph- ageal and gastric cardia adenocarcinomas in the three study areas is attributable to cigarette smoking. 1282 ARTICLES Journal of the National Cancer Institute, Vol. 89, No. 17, September 3, 1997

Table 6. Adjusted* odds ratios (ORs) and 95% confidence intervals (CIs) for esophageal adenocarcinoma, gastric cardia adenocarcinoma, esophageal squamous cell carcinoma, other gastric adanocarcinomas in relation to cigarette smoking and according to beer, liquor, and wine consumption Esophageal Gastric cardia Esophageal squamous Other gastric adenocarcinoma adenocarcinoma cell carcinoma adenocarcinoma case subjects case subjects case subjects case subjects Drinking status Smoking status OR 95% CI OR 95% CI OR 95% CI OR 95% CI m Beer Cigarettes No Not 1.0 1.0 Yes 2.0 1.2-3.3 1.8 I. 1-3.0 Yes Not 1.0 1.0 Yes 2.1 1.2-3.6 2.7 1.5--4.9 Liquor Cigarettes No No'~ 1.0 1.0 Yes 2.4 1.5-3.9 2.0 1.2-3.2 Yes Not 1.0 1.0 Yes 1.9 1.1-3.2 3.13 1.6-5.6 Wine Cigarettes No Not 1.0 1.0 Yes 2.4 1.6-3.7 2.0 1.3--3.1 Yes Not 1.0 1.0 Yes 1.7 0.9-3.1 2.7 1.4-5.3 1.0 1.0 2.5 1.1-5.6 1.6 1.1-2.5 1.0 1.0 5.2 2.4-11.5 1.4 0.9-2.3 1.0 2.6 1.1~.2 4.8 2.3-10.2 2.8 1.5-5.3 1.0 6.8 2.2-21.1 1.0 1.0 2.0 1.~3.4 1.3 0.9-1.9 1.0 2.2 1.24.0 *Adjusted for age; sex; geographic center (Connecticut, New Jersey, and Washington); race; body mass index (weight in kilograms divided by the square of height in meters); income; and the other two types of alcohol. tReference category. No decline in the risk of esophageal and gastric cardia ad- enocarcinomas was evident until 30 years after smoking cessa- tion, which is in contrast to the steady decrease in risk observed after quitting for esophageal squamous cell carcinoma. These patterns suggest that smoking may affect an early stage in the induction of esophageal and gastric cardia adenocarcinomas, whereas later stages are involved for esophageal squamous cell carcinoma. As a result, the time trends in smoking prevalence may have contributed to the divergent incidence trends for these tumors. Thus, the declining incidence of esophageal squamous cell carcinoma since the 1970s is consistent with the reduced prevalence of cigarette smoking among American men that be- gan in the 1960s. On the other hand, the rise in the prevalence of smoking among American men from the early part of this cen- tury until the 1960s, allowing for a lag of 30 years, coincides with the rising incidence of esophageal and gastric cardia adeno- carcinomas, at least among older age groups. The recent national decline in smoking prevalence may not yet have had an impact on the trends for esophageal and gastric adenocarcinomas. ~ Although our study found no excess risk of esophageal and gastric cardia adenocarcinomas associated with the consumption of beer or hard liquor, there was an inverse association seen for wine drinking that extended to all four tumor types. If the re- duction in risk is real, there may be a protective ingredient in wine, such as resveratrol (28), that is not present in beer or liquor. However, previous population-based case-control stud- ies in the United States (13,15) have generally found no asso- ciation between wine drinking and the risk of esophageal and gasmc carala aaenocarcinomas. In comparison with these stud- ies, the prevalence of wine consumption in our population-based control subjects appears to have been elevated, and it was higher among those who were male, white, and younger in age and who also reported a higher income, education, and intake of beer and .liquor (data not shown). Thus, it is possible that the reduced risks associated with wine intake in our study are the result of sam- pling bias or perhaps of residual confounding. Among persons in our study who reported both smoking cigarettes and drinking any type of alcohol, risk was not sub- stantially greater for developing adenocarcinomas of the esopha- gus, the gastric cardia, or other gastric sites than for individuals who only reported smoking. The smoking-associated risk of esoph- ageal squamous ceil carcinoma, however, was more than twice as high among drinkers of any type of alcohol than among nondrink- ers, which is consistent with other studies of this tumor (29). Our finding of a reduced risk of esophageal and gastric cardia adenocarcinomas among those with higher income levels con- firms the results of earlier studies that were population-based and conducted in the United States (13,15). The findings for education, however, are inconsistent. Whereas Brown et al. (13) noted a reduction in risk with decreasing education, the results of Vaughan et al. (15) resemble our results in showing slightly increasing risks with decreasing education. In summary, our large population-based, case-control study of esophageal and gastric cardia adenocarcinomas revealed a doubling of risk among current and ex-smokers, a risk that per- sisted for nearly 30 years after smoking cessation, and a non- significant decrease in risk associated with higher incomes. No association was noted for beer or hard liquor intake, but a 40% reduced risk was associated with wine drinking. These patterns were similar to those observed for noncardia gastric adenocar- cinomas. In contrast, the risks of developing esophageal squa- mous cell carcinoma were estimated to be more than five dines higher among current smokers, 2.8-times higher among ex- smokers, and more than three times higher among liquor drink- ers. Our nndmgs suggest that smoking affects an early stage in the development of esophageal and gastric adenocarcinomas and may have contributed to the recent increases reported in the incidence of these tumors, especially among older persons. References (1) Yang PC, Davis S. Incidence of cancer of the esophagus in the US by histologic type. Cancer 1988;61:612-7. Journal of the National Cancer Institute, VoL 89, No. 17, September 3, 1997 ARTICLES 1283

(2) Powell J, McConkey CC. Increasing incidence of adenocarcinoma of the gastric cardia and adjacent sims. Br J Cancer 1990;62:440-3. (3) Blot WJ, Dcvesa SS, Kneller RW, Fraumeni IF Jr. Rising incidence of adenocarcinoma of the esophagus and gastric eat'dim JAMA1991265: 1257-9. (.4) Poweli J, McConkey CC. The rising trend in oesophageal adenocareinoma ar~ gastric cardia. Eur J Cancer Prey 1992;1:265-9. (.5) Blot WJ, Devesa SS, Fraumeni JF Jr. Continuing climb in rates of esoph- ageal adenoearoinoma: an update [letter]. JAMA 1993;270:1320. (6) Day NE, Munoz N. Esophagus. In: Schottenfeld D, Fraumeni JF, editors. Cancer epidemiology and prevention. New York: Oxford Univ Press, 1996: 681-706. (7) Nomura A. Stomach. In: Schottenfeld D, Fraumeni IF Jr., editors. Cancer epidemiology and prevention. New York: Oxford Univ Press, 1996: 707-24. (8) Wu-Williams All, Yu MC, Mack TM. Life-style, workplace, and stomach cancer by subsim in young men of Los Angeles County. Cancer Res 1990; 50:2569-76. (9) Gray JR, Coldman AI, MacDonald WC. Cigarette and alcohol use in pa- tients with adenocarcinoma of the gastric eardia or lower esophagus. Can- cer 1992;69:2227-31. (10) Newcombe PA, Carbone PP. The health consequences of smoking. Cancer Med Clin North Am I992;76:305-3I. (11) Kabat C-C, Ng SK, Wyader EL. Tobacco, alcohol intake, and diet in rein- don to adenocarcinoma of the eso, phagns and gastric cardia..Cancer Causes Control 1993;4:123-32. (12) Palli D, Bianchi S, Dacarli A, Cipriard F, Avellini C, Coeco P, et al. A case-control study of cancers of the gastric cardia in Italy. Br I Cancer 1992;65:263-6. (13) Brown LM, Silverman DT, Pottem LM, Scheenberg JB, Greenberg Swanson GM, et al. Adenocareinomas of the esophagus and esophagogas- ~e junction in white men in the United States: alcohol, tobacco, mad socioeconomic factors. Cancer Causes Control 1994;5:333--40. (14) Gormalez CA, Agudo A, Mourns I, Riboli E, Sanz JM. Tobacco and alcohol intake in relation to adenocarcinoma of the gastric eardia in Spain. Cancer Causes Control 1994;5:88-9. (15) Vanghan TL, Davis S, Kdstal A, Thomas DB. Obesity, alcohol, and to- bacco as risk factors for cancers of the esophagus and gastric cardia: ad- enoearcinoma versus squamous cell carcinoma. Cancer Epidemiol Biomar- kers Prey 1995;4:85-92. (16) Zhang ZF, Kurtz RC, Sun M, Karpeh M Jr, Yu GP, Gargon N, et al. Adenocarcinomas of the esophagus and gastric eardia: medical conditions, tobacco, alcohol, and socioeconomic factors. Cancer Epidemiol Biomark- ers Prey 1996;5:761-8. (17) Ji BT, Chow WH, Yang G, MeLaughlin JK, Gao RN, Zheng W, et al. The influence of cigarette smoking, alcohol, and green tea consumption on the risk of carcinoma of the cardia and distal stomach in Shanghai, China. Cancer 1996;77:2449-57. (18) Garidou A, Tzonou A, Lipworth L, Signorello LB, Kalal~thald V, Tricho- poulos D. Life-style factors and medical conditions in relation to esopha- geal cancer by histologie O/pc in a low-risk population. Int J Cancer 1996; 68:295-9. (19) Kalish RJ, Claney PIE, Orringer MB, Appelman HD. Clinical, epidemio- logic, and mnrphologie comparison between adenoearcinomas arising in Barrett's esophageal muensa and in the gastric eardia. Gastroenterology 1984;86:461-7. (20) Wang HH, Amonioli DA, Goldman H. Comparative features of esophageal and gastric adenocarcinomas: recent changes in type and frequency. Hum Pathol 1986;17:482-7. (21) Surveillance, Epidemiology, and End Results (SEER) Program. Special Public Use Tape (1973-90). National Cancer Institute, DCPC, Surveillance Program, Cancer Statistics Branch, November 1993. (22) Waksberg J. Sampling methods for random digit dialing. J Am Stat Assoe 1978;73:40-6. (23) Hosmer DW, Lemeshow S. Applied logistic regression. New York: lohn Wiley & Sons, 1989. (24) Armitage P. Statistical methods in medical research. Oxford: Blackwell Scientific Publ, 1971. (25) Rothman KJ. Modem epidemiology. Boston: Little, Brown, 1986. (26) Nelson I2~/, Longstreth WT Jr, Koepsell "I'D, van Belle G. Proxy respon- dents in epidemioIogic research. EpidernioI Rev 1990;I2:71-86. (27) McLaughlin JK, Dietz MS, Mehl ES, Blot WJ. Reliability of surrogate information on cigarette smoking by .type of informant. Am J Epidemiol 1987;126:144-6. (28) Jang M, Cal L, Udeani GO, Slowing KV, Thomas CF, Beecher CW, et al. Cancer chemopreventive activity of resveratrol, a natural product derived from grapes. Science 1997;275:218-20. (29) Longuecker MP, Eager SM. Epideminlogie data on alcoholic beverage consumption and risk of cancer. Clin Chim Acta 1996;246:121--41. Notes 1Editor's note: SEER is a set of geographically defined, population-based central tumor registries in the United States, operated by local nonprofit orga- nizations under contract to the National Cancer Institute (NCD. Each registry annually submits.lts cases to the NCI oft a computer tape. These computer tapes arc then edited by the NCI and made available for analysis. Present address: A. B. West, Department of Pathology, The University of Texas, Galveston. Present address: £ L. Stanford, Department of Oncological Sciences, Utah Cancer Registry, University of Utah, Salt Lake City. Supported in part by Public Health Service grants U01-CA57983, 1501- CA57949, and U01-CA57923 and by contracts N02-CP40501 and N01- CN05230 from the National Cancer Institute, National Institutes of Health, De- partment of Hea/th and Human Services. We thank study managers Sarah Greene and Linda Lannom (Westa0 and field supervisors Tom English (New Jersey), Patrieia Owens (Connectleu0, and Berta Nicol-Blades (Washington) for data collection and processing. We also thank Dr. A. Patel fNCI) for administrative guidance and biostatisticians Drs. Daniel Heit- jan and Brace Levin (Columbia Sehcol of Public Health) for helpful conversa- tions on missing data and ~lytomous regression. In addition, we thank the 178 hospitals in Connecticut, New lersey, and Washington for their participation in the study. Manuscript received February 18, 1997; revised June 17, 1997; accepted July 7, 1997. 1284 ARTICLES Journal of the National Cancer Institute, Vol. 89, No. 17, September 3, 1997