AMERICAN JOURNAL o r EPIDEMIOLOGY
VOL 131, No. 3
Copyright «5 1990 by The Johns Hopkins University School of Hygiene and Public Health
Printed m U S A,
All nghu reserved
NUTRITIONAL EPIDEMIOLOGY OF CANCER OF THE ESOPHAGUS SAXON GRAHAM,1 JAMES MARSHALL,' BRENDA HAUGHEY,1 JOHN BRASURE,1 JO FREUDENHELM,1 MARIA ZIELEZNY,1 GREGG WILKINSON,' AND JAMES NOLAN4 Graham, S. (Dept of Social and Preventive Medicine, State U. of New York at Buffalo, Buffalo, NY 14214), J. Marshall, B. Haughey, J. Brasure, J. Freudenheim, M. Zielezny, G. Wilkinson, and J. Nolan. Nutritional epidemiology of cancer of the esophagus. Am J Epidemiol 1990;131:454-67. This study of 178 cases of cancer of the esophagus from three counties in western New York, as compared with sex- and age-matched neighborhood controls in 1975-1986, replicated 3ome earlier findings, particularly with regard to the increased risks associated with use of cigarettes and alcohol. The concentration of alcohol in an alcoholic beverage apparently did not affect risk: Beer carried a substantial risk, whereas less-dilute forms of alcohol carried no risk. These findings also suggest that the risk of cancer of the esophagus increases with ingestion of foods containing retinol but not carotene. Although increased risks were found to be associated with increases in total calories and fat ingested, as well as calcium, they appeared to be confounded with the risk associated with retinol, as distinct from carotene. Inasmuch as a difference in risk associated with retinol and carotene has been shown in a few previous inquiries dealing with esophageal cancer and cancer at other sites, a need for further investigations distinguishing risks associated with the two compounds is apparent carotene; diet; esophageal neoplasms; esophagus; nutrition; vitamin A
The low incidence of cancer of the esophagus presents problems to the investigator attempting to amass enough cases for epidemiologic analysis. Nevertheless, a number of researchers, including ourselves, have conducted inquiries providing evi-
dence that both tobacco and alcohol are involved as risk factors (1-3). A few additional studies have examined the relation between the ingestion of various foods, as well as nutritional factors, and the risk of cancer of the esophagus. These studies have raised questions as to a possible risk-
Received for publication -March 10, 1989, and in enhancing role for a high ingestion of fats final form July 6,1989. and a low ingestion of vegetables and fac1 Department of Social and Preventive Medicine, tors associated with these such as Carotene State University of New York at Buffalo, Buffalo, NY. , , . . , ,„ „. „ , . , ,. ' School of Nursing, State University of New York M d ascorbic acid (3-7). We wished to reexat Buffalo, Buffalo, NY. amine the roles of tobacco and alcohol and 1 Epidemiological Resources, Inc., Chestnut HUL t n e interaction of these factors in the risk • 'Department of Medicine. Erie County Medical Center, Buffalo, NY. Reprint requests to Dr. Saxon Graham, Department of Social and Preventive Medicine, State University of New York at Buffalo, 2211 Main Street, Buffalo, NY 14214. This investigation was supported by US Public Health Service Grant CA 11535 awarded by the National Cancer Institute, Department of Health and Human Services.
of thia
disease, and to add to the sparse literature dealing with nutritional and die tary factors in the epidemiology of cancer -,, , n, c JL J of t h e esophagus. Our findings, based On a case-control study in three western New York counties, replicated a number of ear.. .,. , , , ., h e r results ^ Presented new evidence on carotene as distinct from retinol.
454
DIET IN ESOPHAGEAL CANCER
455
omitted because they comprise such a small percentage of the population of these counThis study was designed to 1) eliminate ties that a definitive analysis by race would the biases inherent in studying cases and be impossible. About 10 percent (n = 71) controls from only one or two hospitals or of those patients found refused, but a more communities, 2) provide as many subjects important selection factor was the 25 peras possible to facilitate multivariate anal- cent whose physicians felt they should not yses of risk associated with various foods be interviewed. A median period of 2 and nutrients, and 3) ascertain diet histo- months elapsed between time of diagnosis ries in considerable detail so as to provide and time of interview. A total of 178 (24 estimates of several nutrients as well as percent of all those identified) usable interviews were completed. Because of the intotal calories ingested. We drew upon pathologically confirmed hibiting factors noted above, the series we incident esophageal cancer cases from all were able to study should not be construed except five small hospitals in the counties as population-based, but rather as a series of Erie, Niagara, and Monroe, New York of persons who were identified in all but (containing the cities of Buffalo, Lacka- five small hospitals, were from all commuwanna, Niagara Falls, and Rochester; a nities in the three study counties, and were large number of villages and towns, includ- successfully interviewed. The potential for ing Kenmore, Williamsville, Lancaster, bias in the group interviewed is not known Hamburg, and East Aurora; and a sizeable with any certainty but will be discussed rural population), and neighborhood con- below. trols. A lengthy (2.5-hour) interview schedControls were matched on age (within 5 ule was developed to assess an approxi- years), sex, and neighborhood with cases. mation of total nutrient intake and total We wanted to match on neighborhood becaloric intake. cause this would maximize the similarity Our nurse-interviewers identified 743 between cases and controls in terms of socases during 1975-1986 via records in hos- cioeconomic status, a factor shown in the pital pathology departments and surgical past to be associated with esophageal canservices. Methods remained the same cer in Buffalo (8). By matching on neighthroughout this period. After identification, borhood, we could lessen the effects of sopatients' physicians were contacted for cioeconomic status so that our focus could written approval of our request for patients be more purely upon diet. Each control for to participate in the extensive interview. a given case was interviewed as soon as We excluded individuals who were physi- possible after that case, a median elapsed cally or mentally incapable of completing a time of 2 months later. long interview or who had a language barControls were selected by a strict protorier or hearing problem (n = 59). Twenty- col requiring the interviewer to visit sysfour percent (n = 182) were excluded be- tematically addresses close to the case's cause they had died of their disease, had residence. Interviews were attempted at had a previous cancer, or had been diag- all times of day on both weekdays and nosed more than 12 months before our at- weekends. Our control series ultimately tempted contact. The long process of ob- comprised a series of persons who, upon taining written permission from physicians contact, were willing to participate in a 2.5to interview patients contributed to delays hour interview. To maximize the likelihood in attempting contact. Interviews with sur- of a control's completing the lengthy interrogates for patients unavailable were not view, interviewers carefully explained this undertaken because of the questionable va- requirement to prospective subjects. This lidity of such data, especially in the mea- may have been a factor in the high rate of surement of diet. Fifty-one blacks were refusals—52 percent—as well as the fact MATERIALS AND METHODS
456
GRAHAM ET AL.
that once initiated, most interviews were completed. The interview schedule was developed after a substantial trial period which included a number of pilot testa. We used the food frequency approach, which we had found to be the preferred method in earlier research, and inquired as to usual diet in the period prior to 1 year before symptom onset. This was to lessen the possibility that symptoms might have altered diet (9). Respondents were provided with photographs with which to estimate the usual size of their servings. Information was obtained on demographic traits, other factors already known as possible risk factors for cancer of the esophagus, and diet. Our analysis took place in several stages. First, because of the nutritional complexity of foods, we felt it important to analyze the risk associated with individual foods. In addition, however, we wished to measure the ingestion of specific nutrients, and to this end, we used the tables in US Department of Agriculture handbooks nos. 8 (10) and 456 (11) to estimate the amount of vitamin A, ascorbic acid, fats, protein, other nutrients, and total calories in the diets of patients and controls. To assess dietary fiber and for estimates of nutrients not available elsewhere, we utilized the tables of Paul and Southgate and Pennington's Dietary Nutrient Guide (12, 13). Since our study design used matched controls, we initially did both matched and unmatched analyses. The results were found not to differ. Because of the small size of our series—136 males and 42 females—and because a considerable loss of data deriving from missing information is entailed in matched analyses, especially in multivariate analyses, we used unmatched, pooled analyses. Breslow and Day (14) have shown that if factors upon which subjects are matched are controlled analytically, matched analysis can be replaced by unmatched analysis. This we did by adjusting on age and socioeconomic status as measured by educational level (education and
socioeconomic status are significantly associated (15)). The large difference in amounts and types of foods ingested between males and females demanded that odds ratios be calculated for exposure to various nutrients among quartiles and textiles whose ranges were specific to each sex. In supplementing our contingency table analyses to control greater numbers of factors, we used logistic regression procedures in which a set of dummy variables representing the quantile levels was tested in models containing the continuous variables of age and years of education as covariates (14, 16). Because analyses by sex revealed essentially the same relations (see tables 1 and 2), because there is no verified sex-linked endogenous factor in this disease, and because we needed the largest numbers possible for multivariate analysis, the sexes were combined and all subsequent analyses were adjusted for sex. Quartiles for these analyses were based on exposures for both sexes combined. To assess the likelihood that odds ratios discovered suggest a dose response, we used a test of trend in which the p value for trend is the p value for the exposure as a continuous variable when added to the logistic model. Studies of the reliability of data were an integral part of the Western New York Diet and Cancer Studies of which the esophageal research was a part. Cancer of the mouth, stomach, colon, and rectum also were studied using the same case-neighborhood control design, and reliability assessments were carried out on the cases and controls as a whole rather than on specific sites. This was necessitated by the design of the reliability inquiries, which were carried out on samples of the total interviewed. Details on these are found in an earlier publication (17): They included a telephone reinterview of a 10 percent sample which showed close agreement on mean frequency of eating specific foods among 265 case and 195 control pairs. Other studies consisted of a comparison of answers by 158 study subjects to
DIET IN ESOPHAGEAL CANCER
responses of their spouses to questions dealing with the study spouse's diet (18), and a comparison of responses from 323 subjects given 5-8 years apart (19). We found enough agreement in all of these inquiries to lend some credibility to our findings. Because of the widespread concern over the adequacy of reports of diet and because of the vital importance of obtaining reliable data to test hypotheses regarding diet and cancer, a number of other reliability studies have been completed by other investigators in the last 10 years. They have used various types of strategies, including interviews of spouses, reinterviews, comparisons of data from diet diaries with food-frequency approaches, and the like. These investigations suggest that reliability is sufficient to allow estimation of amounts ingested in groups of cases and controls (18-21). Providing absolutely accurate point estimates of ingestion via the food frequency interview or questionnaire approach is not feasible. But our ability to ascertain the quartile or quintile of nutrient intake in which groups of subjects fall is adequate to test the hypothesis that there is a dose-response relation between the ingestion of a given food or nutrient and cancer risk. RESULTS
As might be expected in a study of cases and controls matched on neighborhood of residence, cases and controls were almost identical in educational attainment, which is strongly associated with socioeconomic status (22). There were about 11 percent more Catholics among the cases than among the controls and a smaller proportion of Protestants and Jews among the. cases than among controls. There were somewhat larger proportions of Americans with ancestors from Ireland and Poland and smaller proportions with ancestors from Italy among cases as compared with controls. The two groups differed little in proportions deriving from England and Germany, with the German Americans
457
having the largest representation (about 32 percent in both cases and controls). Table 1 presents odds ratios for the risk of esophageal cancer associated with various pack-years of cigarette smoking throughout life for males and females and (adjusted for sex) for the total population. All analyses were also adjusted for age and education. Note that the risk of esophageal cancer increases with increasing use of cigarettes for each sex and for the total population in logistic regression analyses. No increase in risk was associated with use of pipe tobacco or cigars. Similarly, as table 2 shows, there was an increase in risk with increasing ingestion of beer for each sex and for the total population studied in logistic regression analyses, also adjusting for age and education. Because the results for the two sexes are similar, as shown in tables 1 and 2, and reflect findings adjusting for sex, age, and education for the total population, and because they are similar for virtually all of the analyses conducted in this investigation, for the sake of economy we henceforth present data for the total population only, adjusted on sex and other factors. The risk associated with beer was the highest for any kind of alcoholic beverage, but risk varied with the type of beverage consumed. Although we had hypothesized that the less dilute means of taking alcohol would be associated with higher risk, this was not the case. No risk was associated with ingestion of wine. Moreover, there was no increase in risk associated with drinking the more concentrated forms of alcoholic potables, cocktails, distilled spirits "on the rocks," spirits without any dilution, or drinking "long drinks," e.g., with 4-5 ounces of water, seltzer, etc. Thus, one of the most dilute of alcoholic potables, beer, carried the highest risk (table 3). The risk associated with total alcohol ingestion, adjusting for sex, age, education, and cigarette smoking, was smaller but reflected that of beer. The same results were obtained when risk was examined for the various forms in
458
GRAHAM ET AL. TABLE 1
Relative nsk of esophageal cancer associated with lifetime cigarette smoking, western New York, Pack-yean of cigarette smoking (lifetime)
No.
%
0 1-27 28-47 48-124
32 40 49 55
18.2 22.7 27.8 31.3
51 46 41 36
29.3 26.4 23.6 20.7
176
100.0
174
100.0
Cases
197S-1986\
Control* No
OR*
9 8 * CI*
1.00 1.38 1.92* 2.49"
0.74-2.54 1.04-3.55 1-34-1.64
%
Total population
Pod
= 0.003
Males 0-2 3-31 32-50 51-120
23 34 36 42
17.0 25.2 26.7 31.1
43 31 33 25
32.6 23.5 25.0 18.9
136
100.0
131
100.0
11 11 19
26.8 26.8 46.4
16 16 10
38.1 38.1 23.8
41
100.0
42
100.0
1.00 1.93 2.02* 3.02"
0.95-3.94 1.01-4.04 1.48-6.18 -0.005
Females 0 1-26 27-124
1.00 1.00
0.33-2.98 0.94-8.43 = 0.25
•p
VOL 131, No. 3
Copyright «5 1990 by The Johns Hopkins University School of Hygiene and Public Health
Printed m U S A,
All nghu reserved
NUTRITIONAL EPIDEMIOLOGY OF CANCER OF THE ESOPHAGUS SAXON GRAHAM,1 JAMES MARSHALL,' BRENDA HAUGHEY,1 JOHN BRASURE,1 JO FREUDENHELM,1 MARIA ZIELEZNY,1 GREGG WILKINSON,' AND JAMES NOLAN4 Graham, S. (Dept of Social and Preventive Medicine, State U. of New York at Buffalo, Buffalo, NY 14214), J. Marshall, B. Haughey, J. Brasure, J. Freudenheim, M. Zielezny, G. Wilkinson, and J. Nolan. Nutritional epidemiology of cancer of the esophagus. Am J Epidemiol 1990;131:454-67. This study of 178 cases of cancer of the esophagus from three counties in western New York, as compared with sex- and age-matched neighborhood controls in 1975-1986, replicated 3ome earlier findings, particularly with regard to the increased risks associated with use of cigarettes and alcohol. The concentration of alcohol in an alcoholic beverage apparently did not affect risk: Beer carried a substantial risk, whereas less-dilute forms of alcohol carried no risk. These findings also suggest that the risk of cancer of the esophagus increases with ingestion of foods containing retinol but not carotene. Although increased risks were found to be associated with increases in total calories and fat ingested, as well as calcium, they appeared to be confounded with the risk associated with retinol, as distinct from carotene. Inasmuch as a difference in risk associated with retinol and carotene has been shown in a few previous inquiries dealing with esophageal cancer and cancer at other sites, a need for further investigations distinguishing risks associated with the two compounds is apparent carotene; diet; esophageal neoplasms; esophagus; nutrition; vitamin A
The low incidence of cancer of the esophagus presents problems to the investigator attempting to amass enough cases for epidemiologic analysis. Nevertheless, a number of researchers, including ourselves, have conducted inquiries providing evi-
dence that both tobacco and alcohol are involved as risk factors (1-3). A few additional studies have examined the relation between the ingestion of various foods, as well as nutritional factors, and the risk of cancer of the esophagus. These studies have raised questions as to a possible risk-
Received for publication -March 10, 1989, and in enhancing role for a high ingestion of fats final form July 6,1989. and a low ingestion of vegetables and fac1 Department of Social and Preventive Medicine, tors associated with these such as Carotene State University of New York at Buffalo, Buffalo, NY. , , . . , ,„ „. „ , . , ,. ' School of Nursing, State University of New York M d ascorbic acid (3-7). We wished to reexat Buffalo, Buffalo, NY. amine the roles of tobacco and alcohol and 1 Epidemiological Resources, Inc., Chestnut HUL t n e interaction of these factors in the risk • 'Department of Medicine. Erie County Medical Center, Buffalo, NY. Reprint requests to Dr. Saxon Graham, Department of Social and Preventive Medicine, State University of New York at Buffalo, 2211 Main Street, Buffalo, NY 14214. This investigation was supported by US Public Health Service Grant CA 11535 awarded by the National Cancer Institute, Department of Health and Human Services.
of thia
disease, and to add to the sparse literature dealing with nutritional and die tary factors in the epidemiology of cancer -,, , n, c JL J of t h e esophagus. Our findings, based On a case-control study in three western New York counties, replicated a number of ear.. .,. , , , ., h e r results ^ Presented new evidence on carotene as distinct from retinol.
454
DIET IN ESOPHAGEAL CANCER
455
omitted because they comprise such a small percentage of the population of these counThis study was designed to 1) eliminate ties that a definitive analysis by race would the biases inherent in studying cases and be impossible. About 10 percent (n = 71) controls from only one or two hospitals or of those patients found refused, but a more communities, 2) provide as many subjects important selection factor was the 25 peras possible to facilitate multivariate anal- cent whose physicians felt they should not yses of risk associated with various foods be interviewed. A median period of 2 and nutrients, and 3) ascertain diet histo- months elapsed between time of diagnosis ries in considerable detail so as to provide and time of interview. A total of 178 (24 estimates of several nutrients as well as percent of all those identified) usable interviews were completed. Because of the intotal calories ingested. We drew upon pathologically confirmed hibiting factors noted above, the series we incident esophageal cancer cases from all were able to study should not be construed except five small hospitals in the counties as population-based, but rather as a series of Erie, Niagara, and Monroe, New York of persons who were identified in all but (containing the cities of Buffalo, Lacka- five small hospitals, were from all commuwanna, Niagara Falls, and Rochester; a nities in the three study counties, and were large number of villages and towns, includ- successfully interviewed. The potential for ing Kenmore, Williamsville, Lancaster, bias in the group interviewed is not known Hamburg, and East Aurora; and a sizeable with any certainty but will be discussed rural population), and neighborhood con- below. trols. A lengthy (2.5-hour) interview schedControls were matched on age (within 5 ule was developed to assess an approxi- years), sex, and neighborhood with cases. mation of total nutrient intake and total We wanted to match on neighborhood becaloric intake. cause this would maximize the similarity Our nurse-interviewers identified 743 between cases and controls in terms of socases during 1975-1986 via records in hos- cioeconomic status, a factor shown in the pital pathology departments and surgical past to be associated with esophageal canservices. Methods remained the same cer in Buffalo (8). By matching on neighthroughout this period. After identification, borhood, we could lessen the effects of sopatients' physicians were contacted for cioeconomic status so that our focus could written approval of our request for patients be more purely upon diet. Each control for to participate in the extensive interview. a given case was interviewed as soon as We excluded individuals who were physi- possible after that case, a median elapsed cally or mentally incapable of completing a time of 2 months later. long interview or who had a language barControls were selected by a strict protorier or hearing problem (n = 59). Twenty- col requiring the interviewer to visit sysfour percent (n = 182) were excluded be- tematically addresses close to the case's cause they had died of their disease, had residence. Interviews were attempted at had a previous cancer, or had been diag- all times of day on both weekdays and nosed more than 12 months before our at- weekends. Our control series ultimately tempted contact. The long process of ob- comprised a series of persons who, upon taining written permission from physicians contact, were willing to participate in a 2.5to interview patients contributed to delays hour interview. To maximize the likelihood in attempting contact. Interviews with sur- of a control's completing the lengthy interrogates for patients unavailable were not view, interviewers carefully explained this undertaken because of the questionable va- requirement to prospective subjects. This lidity of such data, especially in the mea- may have been a factor in the high rate of surement of diet. Fifty-one blacks were refusals—52 percent—as well as the fact MATERIALS AND METHODS
456
GRAHAM ET AL.
that once initiated, most interviews were completed. The interview schedule was developed after a substantial trial period which included a number of pilot testa. We used the food frequency approach, which we had found to be the preferred method in earlier research, and inquired as to usual diet in the period prior to 1 year before symptom onset. This was to lessen the possibility that symptoms might have altered diet (9). Respondents were provided with photographs with which to estimate the usual size of their servings. Information was obtained on demographic traits, other factors already known as possible risk factors for cancer of the esophagus, and diet. Our analysis took place in several stages. First, because of the nutritional complexity of foods, we felt it important to analyze the risk associated with individual foods. In addition, however, we wished to measure the ingestion of specific nutrients, and to this end, we used the tables in US Department of Agriculture handbooks nos. 8 (10) and 456 (11) to estimate the amount of vitamin A, ascorbic acid, fats, protein, other nutrients, and total calories in the diets of patients and controls. To assess dietary fiber and for estimates of nutrients not available elsewhere, we utilized the tables of Paul and Southgate and Pennington's Dietary Nutrient Guide (12, 13). Since our study design used matched controls, we initially did both matched and unmatched analyses. The results were found not to differ. Because of the small size of our series—136 males and 42 females—and because a considerable loss of data deriving from missing information is entailed in matched analyses, especially in multivariate analyses, we used unmatched, pooled analyses. Breslow and Day (14) have shown that if factors upon which subjects are matched are controlled analytically, matched analysis can be replaced by unmatched analysis. This we did by adjusting on age and socioeconomic status as measured by educational level (education and
socioeconomic status are significantly associated (15)). The large difference in amounts and types of foods ingested between males and females demanded that odds ratios be calculated for exposure to various nutrients among quartiles and textiles whose ranges were specific to each sex. In supplementing our contingency table analyses to control greater numbers of factors, we used logistic regression procedures in which a set of dummy variables representing the quantile levels was tested in models containing the continuous variables of age and years of education as covariates (14, 16). Because analyses by sex revealed essentially the same relations (see tables 1 and 2), because there is no verified sex-linked endogenous factor in this disease, and because we needed the largest numbers possible for multivariate analysis, the sexes were combined and all subsequent analyses were adjusted for sex. Quartiles for these analyses were based on exposures for both sexes combined. To assess the likelihood that odds ratios discovered suggest a dose response, we used a test of trend in which the p value for trend is the p value for the exposure as a continuous variable when added to the logistic model. Studies of the reliability of data were an integral part of the Western New York Diet and Cancer Studies of which the esophageal research was a part. Cancer of the mouth, stomach, colon, and rectum also were studied using the same case-neighborhood control design, and reliability assessments were carried out on the cases and controls as a whole rather than on specific sites. This was necessitated by the design of the reliability inquiries, which were carried out on samples of the total interviewed. Details on these are found in an earlier publication (17): They included a telephone reinterview of a 10 percent sample which showed close agreement on mean frequency of eating specific foods among 265 case and 195 control pairs. Other studies consisted of a comparison of answers by 158 study subjects to
DIET IN ESOPHAGEAL CANCER
responses of their spouses to questions dealing with the study spouse's diet (18), and a comparison of responses from 323 subjects given 5-8 years apart (19). We found enough agreement in all of these inquiries to lend some credibility to our findings. Because of the widespread concern over the adequacy of reports of diet and because of the vital importance of obtaining reliable data to test hypotheses regarding diet and cancer, a number of other reliability studies have been completed by other investigators in the last 10 years. They have used various types of strategies, including interviews of spouses, reinterviews, comparisons of data from diet diaries with food-frequency approaches, and the like. These investigations suggest that reliability is sufficient to allow estimation of amounts ingested in groups of cases and controls (18-21). Providing absolutely accurate point estimates of ingestion via the food frequency interview or questionnaire approach is not feasible. But our ability to ascertain the quartile or quintile of nutrient intake in which groups of subjects fall is adequate to test the hypothesis that there is a dose-response relation between the ingestion of a given food or nutrient and cancer risk. RESULTS
As might be expected in a study of cases and controls matched on neighborhood of residence, cases and controls were almost identical in educational attainment, which is strongly associated with socioeconomic status (22). There were about 11 percent more Catholics among the cases than among the controls and a smaller proportion of Protestants and Jews among the. cases than among controls. There were somewhat larger proportions of Americans with ancestors from Ireland and Poland and smaller proportions with ancestors from Italy among cases as compared with controls. The two groups differed little in proportions deriving from England and Germany, with the German Americans
457
having the largest representation (about 32 percent in both cases and controls). Table 1 presents odds ratios for the risk of esophageal cancer associated with various pack-years of cigarette smoking throughout life for males and females and (adjusted for sex) for the total population. All analyses were also adjusted for age and education. Note that the risk of esophageal cancer increases with increasing use of cigarettes for each sex and for the total population in logistic regression analyses. No increase in risk was associated with use of pipe tobacco or cigars. Similarly, as table 2 shows, there was an increase in risk with increasing ingestion of beer for each sex and for the total population studied in logistic regression analyses, also adjusting for age and education. Because the results for the two sexes are similar, as shown in tables 1 and 2, and reflect findings adjusting for sex, age, and education for the total population, and because they are similar for virtually all of the analyses conducted in this investigation, for the sake of economy we henceforth present data for the total population only, adjusted on sex and other factors. The risk associated with beer was the highest for any kind of alcoholic beverage, but risk varied with the type of beverage consumed. Although we had hypothesized that the less dilute means of taking alcohol would be associated with higher risk, this was not the case. No risk was associated with ingestion of wine. Moreover, there was no increase in risk associated with drinking the more concentrated forms of alcoholic potables, cocktails, distilled spirits "on the rocks," spirits without any dilution, or drinking "long drinks," e.g., with 4-5 ounces of water, seltzer, etc. Thus, one of the most dilute of alcoholic potables, beer, carried the highest risk (table 3). The risk associated with total alcohol ingestion, adjusting for sex, age, education, and cigarette smoking, was smaller but reflected that of beer. The same results were obtained when risk was examined for the various forms in
458
GRAHAM ET AL. TABLE 1
Relative nsk of esophageal cancer associated with lifetime cigarette smoking, western New York, Pack-yean of cigarette smoking (lifetime)
No.
%
0 1-27 28-47 48-124
32 40 49 55
18.2 22.7 27.8 31.3
51 46 41 36
29.3 26.4 23.6 20.7
176
100.0
174
100.0
Cases
197S-1986\
Control* No
OR*
9 8 * CI*
1.00 1.38 1.92* 2.49"
0.74-2.54 1.04-3.55 1-34-1.64
%
Total population
Pod
= 0.003
Males 0-2 3-31 32-50 51-120
23 34 36 42
17.0 25.2 26.7 31.1
43 31 33 25
32.6 23.5 25.0 18.9
136
100.0
131
100.0
11 11 19
26.8 26.8 46.4
16 16 10
38.1 38.1 23.8
41
100.0
42
100.0
1.00 1.93 2.02* 3.02"
0.95-3.94 1.01-4.04 1.48-6.18 -0.005
Females 0 1-26 27-124
1.00 1.00
0.33-2.98 0.94-8.43 = 0.25
•p
