Dietary Habits and Past Medical History as Related to Fatal Pancreas Cancer Risk Among Adventists

PAUL K. MILLS, PHD, W. LAWRENCE BEESON, MSPH, DAVID E. ABBEY, PHD, GARY E. FRASER, MD, PHD, AND ROLAND L. PHILLIPS, MD, DRPHt

Epidemiologic studies of diet and pancreas cancer are few, and include ecologic comparisons and a limited number of prospective and case-control studies. Foods and/or nutrients that have been suggested to be associated with increased risk of this cancer include total fat intake, eggs, animal protein, sugar, meat, coffee and butter. Consumption of raw fruits and vegetables has been consistently associated with decreased risk. Dietary habits and medical history variables were evaluated in a prospective study of fatal pancreas cancer among 34,000 California Seventh-day Adventists between 1976 and 1983. Forty deaths from pancreas cancer occurred during the follow-up period. Compared to all US whites, Adventists experienced decreased risk from pancreas cancer death (standardized mortality ratio [SMR] = 72 for men; 90 for women), which was not statistically significant. Although there was a suggestive relationship between increasing meat, egg, and coffee consumption and increased pancreatic cancer risk, these variables were not significantly related to risk after controlling for cigarette smoking. However, increasing consumption of vegetarian protein products, beans, lentils, and peas as well as dried fruit was associated with highly significant protective relationships to pancreas cancer risk. A prior history of diabetes was associated with increased risk of subsequent fatal pancreas cancer, as was a history of surgery for peptic or duodenal ulcer. A history of tonsillectomy was associated with a slight, nonsignificant protective relationship as was history of various allergic reactions. These findings suggest that the protective relationships associated with frequent consumption of vegetables and fruits high in proteaseinhibitor content are more important than any increase in pancreas cancer risk attendent on frequent consumption of meat or other animal products. Furthermore, the previously reported positive associations between diabetes and abdominal surgery and pancreas cancer risk are supported in these data. Cancer 61:2578-2585, 1988.

the US, more than 24,000 people die of pancreas cancer, making this the fifth leading cause of cancer death.' There is some evidence that the incidence of the disease has increased in the last 30 years2 whereas survivorship remains very Due to the extremely poor survival, incidence is tantamount to mortality. Little is known concerning the cause of the disease. Like other cancers of environmental cause, pancreas cancer is prevalent in the Western industrialized nations and shows a sharp increase with age beginning in the seventh decade of life. Men experience an increase
ACH YEAR in

$ Deceased March 8, 1987. From the Department of Preventive Medicine, School of Medicine. Lorna Linda University, Lorna Linda, California. Supported by grant RO 1-CA14703 from the National Cancer Institute (R.L.P.). Address for reprints: Paul K. Mills, PhD, Department of Preventive Medicine, School of Medicine, Lorna Linda University, Loma Linda, CA 92350. Accepted for publication November 30, 1987.

in risk in comparison to women and, in the US, blacks appear to be at higher risk than whites. The only established risk factor besides age, race and sex is cigarette smoking. Although the pancreas is intimately involved in the digestive process, relatively few studies have addressed the role of diet in the cause of pancreas cancer. This is undoubtedly because of the extremely poor survivorship associated with the disease which makes interviewing patients for retrospective studies problematic. Epidemiologic investigations of dietary factors that may relate to the cause of pancreas cancer include ecologic corn par is on^^^^ and a limited number of prospective6 and case-control studies.7-' I Foods and/or nutrients that have been suggested to be associated with increased risk from pancreas cancer incidence or mortality include total fat intake,4 eggs, animal protein, sugar,5 meat,6,' ~ o f f e e , ~ , beef, ~ - ' pork,"," and b ~ t t e r . ~ Foods and/or nutrients consistently associated with decreased risk of pancreas cancer include raw fruits and vegetables?-" To date, however, the data are sparse, in some instances inconsistent (e.g., the coffee-pancreas

'

2578

No. 12

DIET,MEDICAL HISTORY, AND PANCREAS CANCER RISK

M i l l s et al.

2579

cancer relationship) and this field of inquiry should still be considered embryonic. Similarly, prior medical history has been investigated in regard to pancreas cancer risk. DiabetesI2 and gastrectomy appear to be associated with increased risk, while a history of tonsillectomy9 has been associated with decreased risk. In one study, pancreas cancer patients were noted to have fewer allergies of any kind." Seventh-day Adventists (SDA) provide a unique opportunity to study dietary relationships to pancreas cancer. By church proscription Adventists do not smoke, drink alcohol, or eat pork, and approximately 50% follow a lacto-ovo-vegetarian diet. Therefore, not only is the potentially confounding effect of cigarette smoke nearly absent but there is considerable variation in exposure to meat and other dietary habits within this population which is not found in other populations. Among Adventist vegetarians, many substitute soy products, legumes and peas for meat. These vegetables are known to contain large amounts of protease inhibitors, which are thought to be protective against cancer at various sites.l 3 However, some animal studies indicate that these inhibitors also may serve as promoters of pancreas cancer.I4 We, therefore, evaluated dietary and medical history relationships to fatal pancreas cancer risk in a cohort study involving 34,000 California Seventh-day Adventists who completed a detailed Lifestyle Questionnaire in 1976 and who were then followed for cancer incidence and mortality between return of the questionnaire and the end of 1982.

'

Methods

Study Population Identification and Enumeration The Adventist Health Study (AHS) began preliminary identification of California Adventists in late 1973 by obtaining church directories from each of the 437 California SDA churches. A comprehensive list of names and addresses of all California SDA households (N = 63,530) was computerized in mid-1974 and served as the first mailing list for the AHS. In August 1974, a demographic Census Questionnaire was sent via bulk rate mail to every household identified on a church directory. The Census Questionnaire was designed to be completed by one member of the household (usually the head of the household) on behalf of the entire household. Therefore, this person recorded demographic and other data on himself as well as on all other family members so that, to a certain degree, information on the Census Questionnaire was obtained by proxy. The Census Questionnaire was returned by 36,805 households (58%) listing 95,196 persons of all ages and races. One reason that the response rate was not higher was that many households on our mailing list should not

have been included because every member of these households was no longer an active member of the specific church and had not yet been removed from the official church directories. Persons under 25 years of age were removed from further analyses at this point (N = 36,106) resulting in a study population of 59,090. In August 1976, a Lifestyle Questionnaire was mailed first class to every living member (N = 57,841) of the total study population and was returned by 40,398 individuals (response rate, 69.8%). Among the 46,03 1 living non-Hispanic whites who were sent the Lifestyle Questionnaire, 34,556 (response rate, 75.1%) returned a completed form. At this point, it was decided to divide the study population as follows: 1. Non-Hispanic whites (25 years of age or older at the time of Census Questionnaire completion and who also completed the Lifestyle Questionnaire) on whom all causes of mortality (1974 to 1982) and the incidence of cancer (1976 to 1982) would be monitored. This population (N = 34,198) became the incidence population, because cancer incidence determination would be completed only on this group. Exposure information on individuals within this population is available from both the Census and the Lifestyle Questionnaire. Three hundred fifty-eight non-Hispanic whites were not included in the incidence population because of data processing errors that were not corrected until the case ascertainment phase of the study had been completed. The incidence population is further divided into two subgroups-3 1,208 who are baptized into the church (Adventists) and 2990 who are considered non-SDA (primarily nonbaptized spouses of Adventists). 2. All others in the study population (N = 24,892). Exposure for this population is limited, for the most part, to the Census Questionnaire. This report is concerned with fatal pancreas cancer (International Classification of Disease [ICD] = 157) which occurred in the incidence population, 1976-1 982.

Mortality Ascertainment in the Study Population

Fatal outcomes in this population were detected by various mechanisms which overlapped with each other but which would provide for the most comprehensive coverage of mortality during the entire period of followup ( 1974 to 1982). Three primary mechanisms were used. 1. Computerized record linkage with the California death certificate files. 2. Computerized record linkage with the National Death Index. 3. Manual linkage with SDA church records. The last mechanism is probably least productive, in

2580

CANCER June15 1988

Vol. 61

TABLE 1. Observed and Expected* Number of Pancreas Cancer Deaths in the Adventist Population, 1976-1982, by Age and Sex
Men Person years
37,6 I6 8,592 8,164 7,620 6,284 4,4 I7 2,878 I ,58 1 53 1 77 77,760

Women Expected no. of deaths

1.68 I .69 2.83 2.90 4.35 4.16 2.9 1 1.60 0.54 0.08 23.74

Age group
4 4 55-59 60-64 65-69 70-74 75-79 80-84 85-89 90-94 295

Observed no. of deaths

0
1

Obslexp
0.00 0.59 0.7 I 0.00 0.69 0.96 I .03 0.62 5.58 0.00

Person years
50,452 11,682 1 1,643 11,779 10,723 8,394 6,094 3,44 1 1,314 203
1 15,723

Observed no. of deaths

1

Expected no. of deaths

1.18
1.so

Obs/exp
0.85 0.67 2.42 0.86 0.88 0.63 0.49 0.84 1.10 0.00 0.90

2 0 3 4 3 1 3 0 17

I 6 3 4 3 2 2 1 0

2.48 3.48 4.56 4.77 4.1 I 2.39 0.9 I 0.14 25.52

Total

0.72

23

* Expected numbers are based upon age-specific mortality rates for U.S. whites, 1975-84.
that church records are known to be incomplete. However, at beginning of follow-up, all SDA churches in California were requested to notify AHS whenever a church member died. Church clerks supplied the AHS with basic information on the decedent (particularly date of death and place of death), which enabled the AHS to request death certificates from the State. The primary mechanisms used to monitor mortality in our total study population was record linkage with the state death files and with the National Death Index.

Results

Between the return of the Lifestyle Questionnaire and the end of 1982 there were 40 deaths from pancreas cancer in the Incidence Population cohort (17 men, 23 women), of a total of 2776 deaths. The standardized mortality ratio for pancreas cancer is 0.72 for Adventist men and 0.90 for Adventist women. Neither result is statistically significant (Table 1). Current use of meat, poultry or fish is associated with increasing risk (Table 2), which is of borderline significance. Past use of these flesh foods is associated with a Statistical Analysis nonsignificant increase in the point estimates. Similarly, Age- and sex-specific mortality from pancreas cancer current use of eggs is associated with increased risk. among white, non-Hispanic California Adventists was Current or past use of milk or cheese is not associated with significant elevation or depression of the point esticompared with all US whites by applying age- and sexspecific mortality rates for US whites (1 975 to 1984) to mates. Conversely, among the fruits, vegetables and vegetarian protein products examined, increasing conperson-years at risk in the Adventist cohort, using the sumption of the vegetable protein products, beans, lenprogram of Monson.'' Members of the study population tils, or peas and dried fruits are all associated with contributed person years at risk beginning at the time significant protective relationships to fatal pancreatic they returned the Lifestyle Questionnaire and ending cancer risk. Also, use of tomatoes shows a suggestive, December 31, 1982 or at the time of death. Age- and sex-adjusted incidence ratios of pancreas cancer were though nonsignificant, protective relationship to risk, as calculated by maximum likelihood estimates by treating does the consumption of various types of fresh fruit the outcome (fatal pancreas cancer) as a Poisson-distrib(Table 3). Use of soft margarine is associated with a nonsignifiuted variable and conducting Poisson regression for cant elevation in the relative risk, although increasing grouped data using the GLIM statistical package softuse of this type of margarine on bread does not appear to ware.16 Multivariate analyses were conducted by conbe associated with increased risk (Table 3). The use of structing Cox proportional hazards regression models to butter on bread also bears little relationship to fatal panobtain adjusted estimates of hazard ratios. Program 2L in the BMDP package was used for these ana1y~es.l~ creatic cancer risk. Current coffee consumption bears a suggestive, Interactions in the data were evaluated by including all though nonsignificant relationship to risk in these data, first-order cross-product terms in the initial Cox models although past use is associated with decreased point esand examining the change in the log-likelihood on retimates. Current cigarette smoking is associated with inmoval of all interaction terms as a group.18

No. 12

DIET,MEDICAL HISTORY, AND PANCREAS CANCER RISK

M i l l s et al.

258 1

TABLE 2. Age- and Sex-Adjusted RR for Fatal Pancreas Cancer Among Adventists, Age 25 Years and Older, by Selected Dietary Variables, 1976-1982 Dietary variable
~~

TABLE 4. Age- and Sex-Adjusted RR for Fatal Pancreas Cancer Among Adventists, Age 25 Years and Older, by Cigarette and Coffee Variables, 1976-1 982 Dietary variable Current use of regular coffee Frequency of use or category Never <Daily >Daily Never <Daily rDaily Never Past only Current Age- and sex-adjusted RR (95% CI)
1.00 1.41 (0.56-3.58) 2.00 (0.9 1-4.38) Trend P = 0.087 1.00 0.65 (0.22-1.89) 0.71 (0.34-1.48) Trend P = 0.254 1.00 1.53 (0.69-3.38) 5.42 (1.78-16.48)

Frequency of use or category

Age- and sex-adjusted RR (95% CI)

1.oo

No. of deaths1 person years*

13181,325 6147,573 16/56,853 9/64,474 16181,590 15143,765 2217 1,038 4147,489 8/66,946 21173,942 9170,192 514 I , 132 14136,723 7142,892 15/103,129

No. of deaths/ person years

18/113,554 613 1,503 10/38,6 19 22199,566 4/26,598 11/60,112 251143,885 10/40,369 416515

Current use of meat, poultry or fish Current use of eggs Vegetarian protein products Beans, lentils, peas Raisins, dates, other dried fruits

< 1Xlwk 1-2Iwk r3XIwk <1Xlwk I-2XIwk r3XIwk < 1X/wk l-2Xlwk r3X/wk <1X/wk I-ZX/wk r3XIwk <1Xlmo I-2XImo r3X/wk

0.83 (0.40-2.20) 2.18 (1.04-4.58) Trend P = 0.052 I .oo 1.52 (0.67-3.43) 2.46 (1.08-5.63) Trend P = 0.037 1.oo 0.29 (0.10-0.84) 0.40 (0.18-0.91) Trend P = 0.02
1

Past use of regular coffee Cigarettes

0.47 (0.21-1.02) 0.43 (0.16-1.13) Trend P = 0.044

1.oo

.oo

RR: relative risk; CI: confidence interval.

0.47 (0.19-1.15) 0.35 (0.17-0.73) Trend P = 0.009

RR: relative risk; CI: confidence interval. * Total number of person years at risk and cases may vary between exposure variables due to missing information.

creased relative risk (RR), which is highly significant (Table 4). In regards to prior medical history, surgery for peptic ulcer is associated with a significant elevation in relative risk. The strength of this relationship was attenuated after controlling for cigarette smoking but remained of borderline significance (RR = 2.62; confidence interval [CI] = 1.00-6.90). A history of self-reported diabetes was also associated with increased risk and this increase
TABLE 3. Foods not Associated With Pancreas Cancer Risk Food Current use of whole milk Past use of cheese Cannedlfrozen fruit Other fresh fruit Soft margarine on bread Real butter on bread Past use of meat, poultry or fish Past use of eggs Current use of cheese Cooked green vegetables Green salad Soft or tub margarine Past use of whole milk Current use of tomatoes Fresh citrus fruit Fresh winter fruit NS: Not significant; RR: relative risk. RR

persisted after excluding those pancreatic cancers who died within one year after returning the Lifestyle Questionnaire (RR = 3.43, CI = 1.47-7.94). A history of peptic or duodenal ulcer is seen to be associated with a nonsignificant elevation in the relative risk (Table 5). Whereas as history of tonsillectomy is associated with a nonsignificant decrease in risk, history of an appendectomy is associated with a nonsignificant increase in risk. We also evaluated prior history of allergic disorders in relation to pancreas cancer risk. History of asthma or hay fever is associated with reduced estimates of risk, as are all types of allergic reactions except a history of reaction to chemicals. However, none of these point estimates are significantly different from the null value.

Multivariate Analysis of Dietary Relationships

PI

> I , NS

<I, NS

To evaluate the univariate (adjusted for age and sex only) results in a multivariate setting, Cox proportional hazards regression models were constructed. The initial models included terms for age, sex, smoking status and those dietary variables (individually) seen to be associated with risk in the univanate analysis. In addition, interaction was assessed by removing cross-product terms from the model. To avoid the problem of multiple comparisons all interaction terms were removed as a block and the resulting likelihood ratio test showed that these terms did not add significantly to the predictive ability of the model (P> 0.05). After controlling for age, sex and cigarette smoking, increasing consumption of meat, poultry or fish is still seen to be associated with elevated estimates of risk although the relationship is no longer significant (Table 6). Similarly, increasing use of eggs or coffee is no longer associated with significant risk, although the point estimates remain elevated.

2582

CANCER June 15 1988

Vot. 61

Since Adventists who avoid meats often substitute vegetarian protein products in their diet, in the second series of Cox models we included terms for current meat consumption in addition to age, sex, and smoking. Again, none of the interaction terms proved significant. However, increasing consumption of the vegetarian protein products was seen to be associated with a highly protective relationship to risk as was the use of beans, lentils, or peas as well as dried fruit consumption. All three of these items were highly statistically significant in a multivariate setting. However, if current meat, poultry, or fish is considered simultaneously with the vegetarian products, beans, and dried fruit, the excess risk previously observed disappears (RR = 1.00). Inspection of log-log plots did not reveal any violation of the proportional hazards assumption.

TABLE 6. Relative Risks of Fatal Pancreas Cancer Among

Adventists for Selected Exposures as Derived From Coxs Proportional Hazards Regression Models Exposure (highllow) Model I * Current meat, poultry or fish Current eggs Current coffee Model 2 t Vegetarian protein products Beans, lentils and peas Raisins, dates, dry fruit Current use of meat poultry or fish Multivariate-adjusted predicted RR 95% CI

2.26 3.42 2.21

0.72-7.12 0.72- 16.26 0.6 1-7.99

0.15
0.03 0.19
1.oo

0.03-0.89

0.OO 3-0.24
0.04-0.86 0.28-3.6 I

Discussion
Several limitations in the nature of the study population, the quality of the exposure data (particularly di-

* Variables simultaneously included in the proportional hazards model include age, sex, smoking status, and the primary exposure variable of interest ( i e . ,meat, eggs or coffee consumption). t Variables simultaneously included in the proportional hazards model include age, sex, smoking status, meat, poultry, or fish and the vegetarian products, beans, primary exposure variable of interest (ix, or dried fruit).
etary), outcome data, and interpretation of analytic procedures must be discussed. First, as mentioned, only 58% of all Adventist households initially contacted by the Adventist Health Study returned the Census Questionnaire. This rather low response rate may reflect a certain selection bias in regard to those individuals under study in this particular investigation of pancreas cancer. However, this low response rate was attributable to poor response rates among nonwhites and Hispanics in the total study population (average response rates among these ethnic groups was approximately 35%)and for this reason, the entire study population was divided into non-Hispanic whites and all others. The response rate to the Census Questionnaire mailing among non-Hispanic whites was in excess of 75%, which lessens the likelihood that severe selection bias influences the results reported here since the analysis was restricted to non-Hispanic whites in the study population. Secondly, the quality of the dietary data should be addressed. The original dietary data were gathered by a food frequency questionnaire which was self-administered in 1976. In 1984, a sample (N = 623) of the nonHispanic white component of the total study population was recontacted and asked to recall the frequency of consumption of 35 of the foods initially reported in 1976. Also current (1984) habits were elicited. Analyses from this substudy indicate, in general, a high degree of correlation (most correlation coefficients 2 0.6) between these 35 foods as initially recorded and as recalled 8

TABLE 5. Age- and Sex-Adjusted RR for Fatal Pancreas Cancer Among Adventists, Age 25 Years and Older, by Medical Historv and Allersv Variables. 1976-1982 Age- and Sex-Adjusted RR (95% C.I.)
1

Variable Ulcer of the stomach or duodenum Peptic ulcer surgery Diabetes

H x of tonsillectomy

Category

No. of deaths/ person years

261162,254 812 1,748 271173,218 516104 281173,678 819683 15165,907 171119,039 16/118,113 18/66,159 351174,303 2112,308 331152,107 4134,508 301149,341 7131,273

No Yes
No Yes No Yes No Yes No Yes No Yes No Yes No Yes

.oo

1.86 (0.84-4.13)
1 .oo

3.35 (1.27-8.86)
1.oo 3.76 (1.70-8.31) 1.oo 0.70 (0.39-1.41)
1

Hx of appendectomy Hx of asthma Hx of hay fever Hx of Allergy* H x of reaction to poison ivy (or oak) or other plants Hx of reaction to bee sting Hx of reaction to medication Hx of reaction to chemicals

.oo

1.74 (0.86-3.53)
1 .oo

0.87 (0.2 1-3.63) I .oo 0.66 (0.23-1.87)

1.oo

0.99 (0.44-2.27)

No Yes No Yes No Yes

.oo .oo

0.43 (0.06-3.16) I .oo 0.67 (0.07-5.98)

1

1.53 (0.37-6.30)

3611 76,334 1110,279 291153,613 8132,999 351179,252 217363

* Allergic reaction sufficiently severe to warrant the attention of a

medical doctor.

RR: relative risk; CI: confidence interval; Hx: history.

No. 12

DIET,MEDICAL HISTORY, AND PANCREAS CANCER RISK

Mills et al.

2583

years later (personal communication, K. Linsted) indicating a reasonable degree of validity in the dietary data used in this analysis. A third concern is reliance on death certificates for deaths presumably due to pancreas cancer. Previous case-control studies,10have noted that the percentage of pathologically confirmed cases of this disease is approximately 60% to 70%. Indeed the histologic confirmation rate for pancreas cancer in the Third National Cancer Survey was only 70%. Yet a review of death certificates from the Third National Cancer Survey with an underlying cause of pancreatic cancer death compared with hospital diagnoses for 2500 cases of pancreas cancer showed that the confirmation rate for pancreas cancer was 89%,45indicating that death certificate reporting is consistent with in-hospital diagnoses, although those diagnoses may be based on clinical or surgical observation rather than histologically confirmed studies. A review of the medical records of 40 pancreas cancer cases reported upon here which was done as part of the incidence portion of the AHS indicates that 28 (70%) of the cases were histopathologically confirmed. The other cases for which information is available were detected by autopsy (one case), cytology (one case), gross observation (one case), clinical impression only (one case) and other means (eight cases). The small number of cases available for analysis in this study and the number of analytic comparisons that were completed also deserve comment. When dealing with uncommon outcomes such as pancreas cancer, there is always insufficient material for in-depth analysis even when, as in this case, almost 200,000 person years at risk have been accrued after 6 years of follow-up. Since little is known concerning the cause of pancreas cancer, only age, sex, and cigarette smoking were considered as covariates in most of the analyses. Each of the other exposure variables considered were identified on an a-priori basis after carefully reviewing both the experimental literature, as well as the few epidemiologica investigations in man. No a posteriori comparisons were made. However, in view of the multiple comparisons presented, the following results should be interpreted cautiously. In this study the use of animal products, including meat, poultry, and fish as well as eggs was associated with suggestive, although nonsignificant increases in risk of fatal pancreatic cancer. Substantial protective associations, however, were noted between the use of meat substitutes (ie., vegetarian protein products such as gluten, soy, or nuts) and pancreas cancer risk. Similarly, increasing consumption of beans, lentils, and peas as well as dried fruit (e.g.,raisins and dates) showed significant protective relationships to risk. In previous ecologic comparisons, total fat intake has been associated with

increased pancreas cancer risk,4 as have eggs, animal protein, and sugar.5 In another cohort study of diet and pancreas cancer: daily intake of meat also was related to a higher risk of cancer of the pancreas. Recent casecontrol studies of diet and pancreatic cancer have noted increased risks to be associated with frequent consumption of beef and butter and, in particular, with frequent use of fried or grilled meat. We noted no association between the use of margarine or butter and pancreas cancer risk. In animal models, dietary unsaturated fats have been shown to enhance pancreatic carcinogenesis during the postinitiation stage.14 Similarly, raw soya flour enhances the growth of carcinogen-induced pancreatic tumors, more so than high levels of dietary unsaturated fat.I4 Although the action of soya flour seems to operate via the tripsin inhibitors contained in raw soy, the action of the inhibitors seems to be destroyed by heating. Also, the raw soya flour effect seems to be species-specific, and experiments in higher primates indicate that raw soya flour does not act as a promoter of pancreatic carcinogenesis. Vegetarian Adventists consume copious quantities of soy products (in place of meat), and our attention was drawn to the possible deleterious effect that these substances might have on human health. In these data, however, the consumption of vegetable protein products, including soy products, was associated with decreased risk from pancreas cancer. This may be due to the protease inhibitors found in soy as well as in beans, lentils, and peas, which also were associated with decreased risk. We conclude that the deleterious effect of eating soybeans observed in animal studies is directly related to species specificity, because rats, chicks, mice and guinea pigs are affected by the ingestion of raw soybean mealz0but dogs, swine, and calves are not. One study noted that although soy-based protein diets containing trypsin inhibitor caused pancreatic hypertrophy in the rat, there was no such effect in primates even after 5 years of feeding such a diet.2These findings, and ours, are consistent with previous investigations of the potential for trypsin inhibitors to adversely affect the human pancreasz2 In this study there was a suggestive,though nonsignificant, inverse relationship between the consumption of fresh citrus fruit, fresh winter fruit (apples, bananas, pears), and pancreas cancer risk. Previous case-control studies of pancreas cancer noted similar relationships with fresh fruit consumption,-11 suggesting that the ascorbic acid content of citrus fruit, in particular, may play a protective role in pancreatic cancer. However, we noted no protective relationship between daily consumption of cooked green vegetables or green salads and pancreas cancer risk. Daily consumption of regular coffee showed a sugges-

2584

CANCER June15 1988

Vol. 61

tive, although nonsignificant, relationship to pancreas reactions severe enough to warrant the attention of a cancer risk. Past consumption, however, showed a sugmedical doctor. gestive protective relationship. Numerous other studies We considered the possibility of multicollinearity in have examined this relationship with few studies supthese data, because Adventists who choose to digress porting a role for coffee consumption in the etiology of from church teaching by consuming meat or other anipancreas ~ a n c e r , and ~ , ~most ~ not supporting such a mal products may also smoke and, conversely, those ro]e.9.10,24-32 with close adherence to church teachings would conAmong Adventists, coffee drinking may represent an sume vegetarian protein products rather than meat and adherence to an entire constellation of factors inconsiswould not smoke. Meat and other animal product content with church teachings or recommendations (such as sumption would, therefore, be positively correlated with meat and cigarette use), which may be more meaningful smoking and negatively correlated with the consumpin relation to pancreas cancer risk. Current use of cigation of vegetarian protein products, beans, and dried rettes in this study is associated with highly significant fruit. Multicollinearity was evaluated using various risk estimates for pancreatic cancer risk which is consismechanisms. Bivariate correlation coefficients were caltent with several previous investigation^,^^-^^ however, culated between all exposure variables. None exceeded only 3.7% of the incidence population reported current 0.4 in absolute value. Next, stepwise regression was done smoking, and 56.4% of these smokers are non-SDA. in order to calculate the multiple R2 values (ie., the We evaluated prior medical and surgical history since percentage of the variance in one variable explained by previous work in pancreas cancer suggests that diabetes the remaining variables) associated with each of the exmay increase riskI2as does prior abdominal ~urgery,~','~ posure variables. Again the magnitude of the R2 values whereas tonsillectomy' and allergies" may be associated never exceeded 0.5. This implies that variance inflation with decreased risk of this disease. In this prospective factors were within acceptable limits. Therefore, alstudy, in which subjects were free of pancreas cancer though there was reason to suspect that severe multicollinearity may be present in the data on an a priori basis, when they completed the Lifestyle Questionnaire in in fact this did not seem to be so. 1976, self-reported diabetes is associated with a signifiThis prospective study of diet and subsequent fatal cant elevation in risk for subsequent pancreas cancer. pancreas cancer risk among Adventists indicates that This estimate (RR = 3.76) is as large as that observed in the strength of the protective association between the earlier s t ~ d i e s ' ~ and , ~ ~is ,~ highly ' significant. Similarly, regular consumption of vegetarian protein products (ingastric surgery and, in particular, previous gastrectomy cluding soy products) as well as beans, lentils, and peas are associated with increased risk of pancreas cancer. In appears to be stronger than the increase in risk assothis study, a history of surgery for gastric or duodenal ciated with consuming meat or other animal products ulcer was associated with a significantly elevated risk previously r e ~ o r t e d .These ~ . ~ ~results, based on prospeceven after controlling for cigarette smoking. History of tively gathered data in a population characterized by appendectomy also was associated with an increased unique dietary characteristics, are not hindered by validrisk, although this finding was not statistically signifiity concerns commonly encountered in case-control cant. Some workers have suggested that the gastric hytypes of investigations and should be considered in fupoacidity after gastric surgery results in bacterial overture studies of chemoprevention of human cancer. growth, resulting in production of circulating carcinogens, in particular, N-nitroso corn pound^.^',^^ Conversely, a history of tonsillectomy recently has been REFERENCES shown to be associated with a decreased risk of pancreas 1. Silverberg E, Lubera J. Cancer statistics, 1986. Ca 1986; cancer' in a case-control study using both hospital and 36( 1):9-25. population controls. In this study, a history of tonsillec2. DeVesa S, Silverman D. Cancer incidence and mortality trends tomy was associated with a nonsignificant reduction in in the United States, 1935-74. JNatl Cancer Inst 1978; 60545-571. 3. Axtell L, Asire A, Myers M, eds. Cancer Patient Survival. Report risk. Other investigations have noted this protective reNo. 5. Washington D.C.: U S. Gov't Printing Otfice, 1977. (DHEW l a t i ~ n s h i p ,and ~~. it~has ~ been noted that the tonsils are Publication no. (NIH)77-992). a source of antigenic stimulus. 4. Segi M et al. Cancer mortality of selected sites in 24 countries, No. 5 1964-65, Sendai: Japanese Dept of Public Health, Tohokii UniFinally, we examined allergies in relation to subseversity School of Medicine, 1969. quent cancer risk. With the exception of a history of 5. Armstrong B, Doll R. Environmental factors and cancer incireaction to chemicals, each allergic disorder examined dence and mortality in different countries with special reference to dietary practices. Int JCancer 1975; 15617-631. was associated with decreased risk estimates for pancre6. Hirayama T. Smoking in relation to death rates of 265,118 men atic cancer, although none were statistically different and women in Japan. (1972): A report on five years of follow-up. from the null value. Allergy on the questionnaire used in Presented at the American Cancer Society's 14th Cancer Seminar, Clearwater Beach, Florida, 1972. this investigation referred specifically to those allergic

No. 12

DIET,MEDICAL HISTORY, AND PANCREAS CANCER RISK

Mills el d.

2585

7. MacMahon B, Yen S, Trichopoulos D, Warren K, Nardi G. Coffee and cancer of the pancreas. N Engl J Med 1981; 304:630-633. 8. Durbec JP, Chevillotte G, Bidart JM, Berthezene P, Sarles H. Diet, tobacco and risk of cancer of the pancreas: A case-control study. Br J Cancer 1983; 47:463-470. 9. Gold EB, Gordis L, Diener MD et al. Diet and other risk factors for cancer of the pancreas. Cancer 1985; 55:460-467. 10. Mack TM, Yu MC, Hanisch R, Henderson B. Pancreas cancer and smoking, beverage consumption, and past medical history. J Natl Cancer Inst 1986; 76:49-60. 11. Norell SE, Ahlbom A, Erwald R et a/. Diet and pancreatic cancer: A case-control study. A m J Epidemiol 1986; 124:894-902. 12. Kessler I. Cancer mortality among diabetics. J Natl Cancer Inst 1970; 44~673-686. 13. Troll W, Frenkel K, Wiesner R. Protease inhibitors: Their role as modifiers of carcinogenic processes. In: Friedman M., ed. Nutritional and Toxicological Significance of Enzyme Inhibitors in Foods. New York: Plenum Publishing Corp, 1986; 153-65. 14. Roebuck B, Kaplita P, Edwards B, Praissman M. Effects of dietary fats and soybean protein of Azaserine-induced pancreatic carcinogenesis and plasma cholecystokinin in the rat. Cancer Res 1987; 47: 1333-1338. 15. Monson RR. Analysis of relative survival and proportional mortality. Comput Biomed Res 1974; 7:325-332. 16. Baker R, Nelder J. The GLIM System: Release 3. 77. Users Guide. Oxford: Numerical Algorithms Group, 1985; 45-48. 17. Hopkins A. BMDP Statistical Software Manual. Berkeley, CA: University of California Press, 1985; 576-594. 18. Kleinbaum D, Kupper L, Morgenstern H. Epidemiologic Research: Principles and Quantitative Methods. Belmont, CA: Lifetime Learning Publications, 1982; 447-456. 19. McGuinness E, Morgan R, Levison D, Hopwood D, Wormsley K. Interaction of asaserine and raw soya flour on the rat pancreas. Scand JGastroenterol 1981; 16:49-56. 20. Schneeman B, Gallaher D. Pancreatic response to dietary trypsin inhibitor: Variations among species. Fed Proc 44(5): 1496 (Abstract # 6346), 1985. 2 1. Harwood J, Ausman L, King N et al. Effect of long term feeding of soy-based diets on the pancreas of Cebus monkeys. Fed Proc 44(5): 1496 (Abstract # 6349), 1985. 22. Liener I. Significance for humans of biologically active factors in soybeans and other legumes. J A m Oil Chemist Society 1979; 56: 121-129. 23. Nomura A, Stemmerman G, Heilbrun L. Coffee and pancreatic cancer (Letter). Lancet 1981; 2:415. 24. Severson R, Davis S, Pollisar L. Smoking, coffee and cancer of the pancreas (Letter). Br Med J 1982; 285:214. 25. Wynder E, Hall N, Polansky. Epidemiology of coffee and cancer of the pancreas. Cancer Res 1983; 43:3900-3906. 26. Goldstein N. No association found between coffee and cancer of the pancreas (Letter). N Engl JMed 1982; 306:997. 27. Jick H, Dinan B. Coffee and pancreatic cancer (Letter). Lancet 1981; 2:92.

28. Kessler I. Coffee and cancer of the pancreas (Letter). N Engl J Med 1981; 304:1605. 29. Kinlen L, McPherson K. Pancreas cancer and coffee and tea consumption: A case-control study. Br J Cancer 1984; 49:93-96. 30. Whittemore A, Paffenbarger R, Anderson K, Halpern J. Early precursors of pancreatic cancer in college men. J Chronic Dis 1983; 36:25 1-256. 31. Heuch I, Kvale G, Jacobsen B. Use of alcohol, tobacco and coffee and risk of pancreatic cancer. Br J Cancer 1983; 48537-643. 32. Snowdon DA, Phillips RL. Coffee consumption and risk of fatal cancers. A m J Public Health 1984; 74:820-823. 33. Best EW. A Canadian study of smoking and health. Ottawa: Department of National Health and Welfare, 1966; 65-86. 34. Cederlof R, Friberg L, Hrubec Z, Lorich U. The relationship of smoking and some social co-variables to mortality and cancer morbidity: A ten year follow-up in a probability sample of 55,000 Swedish subjects, age 18-69, part 111. Stockholm: Karolinska Institute, 1975. 35. Doll R, Pet0 R. Mortality in relation to smoking: 20 years' observation on male British doctors. Br Med J 1976; 2:1525-1536. 36. Hammond EC. Smoking in relation to the death rates of one million men and women. In: Haenszel W, ed. Epidemiological Approaches to the Study of Cancer and Other Chronic Diseases. US Public Health Service Natl Cancer Inst Monograph 19, 1966; 126-204. 37. Hirayama T. Changing patterns of cancer in Japan with special reference to the decrease in stomach cancer mortality. In: Hiatt HH, Watson JD, Winston JA, eds. Origins of Human Cancer, vol. 4. Cold Spring Harbor, New York Cold Spring Harbor Laboratory 1977; 55-75. 38. Kahn HA. Report of 8-'h years of observation. In: Haenszel W, ed. Epidemiological Approaches to the Study of Cancer and Other Chronic Diseases. US Public Health Service. National Cancer Institute Monograph 19, 1966; 1-125. 39. Caygill C, Hill M, Hall N, Kirkham J, Northfield T. Gastric surgery as a risk factor in human carcinogenesis (Abstr). Dig Dis Sci 1986; (Suppl) 31(1):220. 40. Wynder EL, Mabuchi K, Maruchi N. Fortner JG. Epidemiology of cancer of the pancreas. J Natl Cancer Inst 1973; 50:645-667. 4 1. Ragozzino M, Melton L, Chu C. Subsequent cancer risk in the incidence cohort of Rochester, Minnesota residents with diabetes mellitus. J Chronic Dis 1982; 35: 13- 19. 42. Lund E. Gastric cancer after gastric surgery: An increasing problem in Norway (Letter). Lancet 1983; 1:973. 43. Lin R, Kessler I. A multifactorial model for pancreatic cancer in man. JAMA 1981; 245:147-152. 44. Cassimos C, Sklarunu-Zurukzoglu S, Catrin D. The frequency of tonsillectomy and appendectomy in cancer patients. Cancer 1973; 32:1374-1379. 45. Percy C, Stanek E, Gloeckler L. Accuracy of cancer death certificates and its effect on cancer mortality statistics. Am J Public Health 1981; 7 1~242-250. 46. Ishii K, Nakamura K, Ozaki I, Yamada N, Takeuchi T. Epidemiological problems of pancreas cancer. Jpn J Clin Med 1968; 26: 1839-1842.