Soy Food Intake and Pancreatic Cancer Risk: The Japan Public Health Center–based Prospective Study

Outcomes in pancreatic cancer are hampered by a lack of reliable tests for early diagnosis and effective treatments, and the prognosis of this condition remains poor. In Japan, the 5-year survival rate was only 7.7% in 2010 (1), and mortality has now increased, ranking it as the fourth leading cause of cancer-related death (2). While previous epidemiologic studies have identified some risk factors, including obesity, smoking, history of diabetes, history of chronic pancreatitis, and family history of pancreatic cancer (3, 4), no clear prevention strategies for pancreatic cancer have yet been established.

Soybeans contain various bioactive compounds, including isoflavones, polypeptides, lectins, saponins, and enzyme inhibitors. These can have both beneficial and harmful effects on human health (5, 6). Soy food intake has been associated with reduced risk of chronic diseases, including cardiovascular diseases; associated risk factors, such as hyperlipidemia and hypertension (7, 8); and prostate and breast cancer (8–10). In contrast, animal experiments have shown that consumption of raw soy flour causes hypertrophy, hyperplasia, and neoplasm of the pancreas in some animals (11). To our knowledge, however, only one epidemiologic study has demonstrated a positive association between intake of a soy food, miso soup, and pancreatic cancer (12).

Here, we aimed to identify the association between soy food intake and pancreatic cancer risk in a large-scale, population-based prospective study in Japan.

The Japan Public Health Center–based Prospective Study (JPHC Study) is a nationwide, population-based longitudinal study that aims to assess the risk of cancer and cardiovascular disease in the Japanese population. Details of the JPHC Study have been described elsewhere (13, 14). Briefly, cohort I in 1990–1994 and cohort II in 1993–1995 enrolled 140,420 baseline participants ages 40–69 years in 11 prefectural public health center (PHC) areas (Fig. 1). The starting point for this study was defined as the 5-year follow-up survey in 1995–1998 due to its more detailed estimation of dietary intake. According to the eligibility criteria, we excluded participants shown in Fig. 1. Consequently, data from 90,185 participants (41,899 men and 48,286 women) were analyzed in this study. During the study period, 5,725 participants (6.3%) emigrated out of their PHC area and 104 (0.1%) were lost to follow-up. This study conformed to the ethical guidelines of the Declaration of Helsinki. The study protocol was approved by the Institutional Review Board of the National Cancer Center, Japan (approval number: 2001-021).

Dietary factors were estimated on the basis of the average intake frequency and amount consumed relative to a standard portion size during the previous year for 138 food items using the food frequency questionnaire (FFQ), a validated, self-administered questionnaire, as reported previously (14, 15). Of these, we classified soy-based foods into the following groups: total soy foods (eight soy food items), fermented soy foods [two items: fermented soybeans (natto) and fermented soybean paste (miso)], nonfermented soy foods [six items: tofu (soy curd) in miso soup, boiled or cold tofu (yudofu and hiyayakko), predrained tofu (yushidofu), freeze-dried tofu (koyadofu), deep fried tofu (aburaage), and soy milk], and tofu (three items: tofu in miso soup, boiled or cold tofu, and predrained tofu; grouped according to similarities in the manufacturing process). Intake of genistein, an isoflavone in soy foods, was calculated on the basis of the FFQ and the Standard Tables of Food Composition in Japan (7th revised version, 2015; ref. 16). Intake of soy foods and genistein was adjusted for total energy intake using the residual method (17), and participants were divided into quartiles of intake for each food group for analysis. Spearman rank correlation coefficient for total soy food intake was 0.45 for men and 0.44 for women in cohort I (18), and 0.52 for men and 0.54 for women in cohort II (19) in assessment of the validity of the FFQ using 28-day dietary records.

Pancreatic cancer incidence was determined from medical records and population-based cancer registries in the PHC areas as reported previously (14). In accordance with the International Classification of Diseases for Oncology (third edition; ref. 20), pancreatic cancer cases were determined by codes C25.0–C25.3 and C25.5–C25.9, but excluded endocrine tumor (C25.4) was excluded due to the difference in etiology. The proportion of pancreatic cancer cases specified by death certificate only was 12.5%.

We calculated person-years of follow-up for each participant from the date they completed the 5-year follow-up survey questionnaire to the date of pancreatic cancer diagnosis, emigration from the study area, death, or the end of follow-up on December 31, 2012 (for one PHC area) or December 31, 2013 (for nine PHC areas), whichever came first as reported previously (14). HRs and 95% confidence intervals (CI) were calculated with Cox proportional hazards models according to the quartile of intake for each food group with the lowest category as a reference, or an increase in intake of 25 g/day (continuous variable). P values for linear trends were calculated by assigning ordinal quartile scores to each quartile of intake in the models. P values for quadratic trends also included a quadratic term in each linear trend model. In model 1, HRs were adjusted for sex, age (continuous), and study area. Model 2 was further adjusted for potential confounders: smoking status (nonsmoker, past-smoker, current-smoker <20, 20–<40, and ≥40 cigarettes/day, or unknown), history of diabetes mellitus (yes or no), family history of pancreatic cancer (yes or no), body mass index (BMI, <20, 20–<25, 25–<30, ≥30 kg/m², or unknown), ethanol intake (0, <150, 150–<300, 300–<450, ≥450 g/week, or unknown), fish intake (quartile), meat intake (quartile), vegetable intake (quartile), fruit intake (quartile), physical activity (≥1 day/week; yes, no, or unknown), coffee intake (≥1 cup/day; yes, no, or unknown), and log-transformed energy intake (continuous). To reduce the possibility of reverse causation, cases of pancreatic cancer identified during the first 3 years of follow-up were excluded from model 2. Moreover, we conducted analyses stratified by sex, smoking status [never (nonsmoker) or ever smoker (current and past smoker)] and BMI (<25 or ≥25 kg/m²) to investigate whether soy food intake affects pancreatic cancer risk according to known risk factors for pancreatic cancer. Furthermore, to reduce the influence of factors associated with total soy food intake shown in Table 1, analyses stratified by age, vegetable intake, and coffee intake were conducted. P values for interaction were calculated using likelihood ratio tests by comparing Cox proportional hazards models with and without a cross-product term, which was derived by crossing the ordinal quartile scores of the corresponding food intake category with the stratified factor. SAS version 9.3 (SAS Institute Inc.) was used in this study.

During 1,433,854 person-years of follow-up (median 16.9 years), 577 cases of pancreatic cancer were identified, 314 in men and 263 in women. Median intake of total soy foods, fermented soy foods, and nonfermented soy foods in the study population was 69.5 g/day, 27.5 g/day, and 35.5 g/day, respectively. At baseline, participants with higher intake of total soy foods tended to be older, more likely to have diabetes, and have higher BMI. There were no sex differences among the intake quartiles for total soy foods (Table 1). Furthermore, those with higher intake of total soy foods tended to do more physical activity, have higher intake of vegetables, not be current smokers, and have lower intake of alcohol and meat.

In the multivariate-adjusted model, total soy food intake showed a statistically significant positive association with pancreatic cancer incidence (HR of the highest quartile vs. the lowest, 1.48; 95% CI, 1.15–1.92; P_{linear trend} = 0.007; Table 2). Although P_{linear trend} was statistically significant in total soy food intake, only the highest category was statistically significantly increased in the point estimation of HR among quartiles compared with the lowest category. However, P for quadratic trend was not statistically significant (Supplementary Table S1). Among soy foods, higher intake of nonfermented soy foods showed a statistically significant positive association with pancreatic cancer (HR, 1.41; 95% CI, 1.09–1.81; P_{linear trend} = 0.008), but fermented soy foods did not (HR, 0.96; 95% CI, 0.73–1.26; P_{linear trend} = 0.982). Similar findings were observed for an increase in intake of 25 g/day. The highest intake of genistein showed a statistically significant positive association with pancreatic cancer (vs. the lowest; HR, 1.33; 95% CI, 1.03–1.73; P_{linear trend} = 0.033). These findings did not change even after exclusion of participants diagnosed within the first 3 years of follow-up. However, we did not observe a statistically significant association for any individual fermented or nonfermented soy food (Table 2; Supplementary Table S2).

In analyses stratified by sex (Table 3), total soy food intake was statistically significantly associated with pancreatic cancer in women (HR, 1.73; 95% CI, 1.19–2.50; P_{linear trend} = 0.004), but not in men (HR, 1.28; 95% CI, 0.90–1.82; P_{linear trend} = 0.329). In analyses stratified by BMI (Table 4), we observed a statistically significant positive association between the highest intake of total soy foods and pancreatic cancer in participants with BMI ≥ 25 kg/m² (HR, 2.00; 95% CI, 1.18–3.40; P_{linear trend} = 0.004), but a nonsignificant association in participants with BMI < 25 kg/m² (HR, 1.32; 95% CI, 0.98–1.79; P_{linear trend} = 0.220). However, there was no statistically significant interaction between total soy food intake and BMI (P_interaction = 0.174). In analyses stratified by other factors, the direction of the association did not differ among strata for each factor, and P values for interaction were not statistically significant in these analyses (Supplementary Tables S3–S7). Furthermore, the positive association between intake of total soy foods, nonfermented soy foods, and genistein and pancreatic cancer was observed in an analysis which excluded participants with a history of diabetes (Supplementary Table S8).

We analyzed the association between soy food intake and pancreatic cancer incidence by detailed estimation of food intake in a large-scale, population-based prospective study in Japan. We found that total soy food intake was associated with increased risk of pancreatic cancer. To date, there is only one epidemiologic study to have examined the association between soy food intake and pancreatic cancer: Hirayama showed that the frequency of miso soup intake was positively associated with pancreatic cancer risk (12). However, we found no association between intake of fermented soy foods (miso and natto) or individual fermented soy foods (miso or natto) and pancreatic cancer risk. Instead, we observed a positive association between nonfermented soy food intake and pancreatic cancer risk.

In contrast to our study, previous epidemiologic studies have shown inverse associations between intake of legumes and pancreatic cancer risk (21, 22). In the Hawaii-Los Angeles Multiethnic Cohort Study (529 pancreatic cancer cases among 183,522 participants during 8.3 years of follow-up), multivariate models showed that higher intake of legumes, including tofu, had a nonsignificant inverse association [highest vs. lowest; relative risk (RR), 0.84; 95% CI, 0.62–1.13; P_trend = 0.099; ref. 21]. The researchers suggested that isoflavones, proteinase inhibitors, saponins, and dietary fiber may underlie the protective effects of legumes against cancer (21, 23). In the Adventist Health Study (40 pancreatic cancer cases among 34,000 participants during 6 years of follow-up), multivariate models showed that both vegetarian protein products including soy products, and beans/lentils/peas had statistically significant inverse associations (high vs. low; RR: 0.15, 95% CI: 0.03–0.89 for vegetarian protein products; RR: 0.03, 95% CI: 0.003–0.24 for beans/lentils/peas; ref. 24). There are several possible explanations for the difference between the present and previous findings. First, the previous studies grouped nonsoy-based legumes with soy products, whereas our study focused on soy foods. These products differ in their nutrient content and heating methods. Legumes such as beans, lentils, and peas contain more carbohydrates and less fat than soybeans and are heated for longer periods, whereas soy contains more protein and fat and is heated for as long as the taste of the soy food is not adversely affected during the manufacturing and cooking process (16). Second, the type and manufacturing of soy foods in these countries may differ from those in Japan. Third, the studies differed in the number of cases and follow-up period, which may also have contributed to these discrepancies.

One possible mechanism for the increased incidence of pancreatic cancer following high intake of soy foods may be that the presence of trypsin inhibitors, which are contained in soybeans and are heat sensitive, prevents trypsin from inhibiting the release of cholecystokinin (25–27), although the effects of trypsin inhibitors and cholecystokinin on the human pancreas are controversial (28, 29). Consumption of raw soy flour has been shown to cause indigestion and malnutrition in most animals, as well as hypertrophy, hyperplasia, and cancerous lesions in the pancreas of some animals, including rats, due to trypsin inhibitor (5, 11, 26, 27, 30). When trypsin inhibitors in soybeans enable the release of cholecystokinin, the binding of cholecystokinin to CCK-B receptors cause GTP-coupled responses that lead to proliferation of both acinar cells and islet cells in humans (29). Moreover, cholecystokinin induces the release of insulin in humans (31), which stimulates cell proliferation and mitogenesis (32). This putative involvement of cholecystokinin may also be supported by our finding that intake of total soy foods was increased in participants with BMI ≥ 25 kg/m². Mechanisms underlying carcinogenesis due to obesity include cell proliferation via insulin and insulin-like growth factor signaling, and direct DNA damage and chronic inflammation due to oxidative stress (29, 33–35). Moreover, in obese mice, cholecystokinin is expressed in the pancreatic islets and cholecystokinin mRNA expression is upregulated 500-fold compared with that in lean mice (29), leading to accelerated cell proliferation via autocrine/paracrine mechanisms.

Alternatively, additives to nonfermented soy foods should also be considered. For example, magnesium is used as a coagulant in the production of soy curd. However, magnesium intake was not associated with pancreatic cancer risk in previous cohort studies (36, 37), and a deficiency in magnesium has been found to be associated with increased risk of pancreatic cancer (38). Magnesium intake is therefore unlikely to explain the increased risk of pancreatic cancer due to nonfermented soy food intake.

In contrast, certain harmful constituents in soy beans such as trypsin inhibitors may be inactivated by soaking, heating, and fermentation. Most traditional Asian soy foods are fermented to make them digestible (5). This may partly explain the lack of association between fermented soy foods and pancreatic cancer risk.

The increased pancreatic cancer risk following higher intake of total soy foods was more evident in women than in men. This may be supported by the antiestrogenic properties of genistein, with a prospective study suggesting that female hormones have a protective role against pancreatic cancer (39).

Analyses stratified by sex, BMI, and smoking status suggested that the effect modification was unclear. Furthermore, analyses stratified by age, vegetable intake, and coffee intake suggested that the effect of these factors as confounders was small because the direction of the association was generally consistent between the strata of each factor compared with analysis of the overall population. The nature of this study does not allow a determination of causality. However, because the results remained relatively unchanged after exclusion of participants diagnosed with pancreatic cancer in the first 3 years of follow-up, we propose that reverse causality is unlikely.

The strengths of this study are due to features of JPHC Study: prospective design with long follow-up period, large general population of participants with high response rate and high proportion of follow-up participants, and availability of food intake estimation on the basis of a detailed, validated FFQ. Nevertheless, several limitations also warrant mention. First, estimation of lifestyle, including food intake, was conducted at a single timepoint using a 5-year follow-up survey. In analysis of joint classifications in total soy food intake combined quartile categories in 5-year and 10-year follow-up survey (70,260 participants, 378 cases), HR of the highest in the 5-year and highest in the 10-year survey participants did not differ; however, participants whose intake changed did not show a consistent tendency (Supplementary Table S9). The effects of food intake may be better understood by accounting for changes in lifestyle habits, which would require longitudinal estimation. Second, the number of incident cases of pancreatic cancer may not have been sufficient for analyses stratified by exposure subgroup and risk factor, and the stratified analysis findings may in part be due to chance. Finally, our association findings may have been affected by residual confounding effects and unmeasured confounding variables.

In conclusion, higher intake of total soy foods, particularly nonfermented soy foods, might increase the risk of pancreatic cancer. This study is the first to report an association between the intake of various soy foods and pancreatic cancer risk. Further studies are required to confirm our findings.

No potential conflicts of interest were disclosed.

Conception and design: N. Sawada, M. Iwasaki, S. Tsugane

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Sawada, T. Shimazu, T. Yamaji, A. Goto, M. Iwasaki, S. Tsugane

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Yamagiwa, N. Sawada, T. Shimazu, T. Yamaji, A. Goto, J. Ishihara, M. Inoue, S. Tsugane

Writing, review, and/or revision of the manuscript: Y. Yamagiwa, T. Shimazu, T. Yamaji, A. Goto, R. Takachi, J. Ishihara, M. Iwasaki, M. Inoue, S. Tsugane

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Tsugane

Study supervision: S. Tsugane

This study was supported by National Cancer Center Research and Development Fund (since 2011), a Grant-in-Aid for Cancer Research from the Ministry of Health, Labour and Welfare of Japan (from 1989 to 2010), and Ministry of Agriculture, Fishery and Forestry, Japan (MAFFCPS-2016-1-1). JPHC members are listed at the following site (as of April 2017): http://epi.ncc.go.jp/en/jphc/781/7951.html.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.