With 458 918 new cases and 432 242 deaths in 2018, pancreatic cancer (PAC) contributed to 2·5 % of all-cause cancers and 4·5 % of all deaths caused by cancer worldwide. In Japan, a total of 43 119 new cases of PAC and 37 358 related deaths were recorded in the same year representing 4·9 % of all-cause incident cancers and 9·1 % of all cancer deaths in the country, almost twice proportions as the worldwide incidence and mortality(Reference Bray, Ferlay and Soerjomataram1,Reference Rawla, Sunkara and Gaduputi2) . It is projected that, over the period between 2018 and 2040, PAC incidence and mortality will increase worldwide by 77·7 % and 79·9 %, respectively(Reference Rawla, Sunkara and Gaduputi2). Given its growing incidence and poor five-year survival rate that hardly exceeds 5 %, identifying modifiable risk factors for PAC has become a public health priority to apply risk prevention programmes(Reference Del Chiaro, Segersvard and Lohr3,Reference Arafa, Eshak and Abdel Rahman4) .
Despite the complex and multifactorial pathogenesis of PAC(Reference Klein5–Reference Muñoz, Chakravarthy and Gong7), previous research has suggested that dietary factors may play aetiological roles(Reference Casari and Falasca8,Reference Pericleous, Rossi and Mandair9) . For example, red and processed meat consumption was shown to increase the risk of PAC due to the carcinogenic effects of N-nitroso compounds(Reference Sen, Seaman and Burgess10–Reference Paluszkiewicz, Smolinska and Debinska13), while consuming fruits, vegetables and whole grains reduced that risk because of the anticarcinogenic effects of Ca, Mg, potassium, α- and beta-carotene and vitamins A, B6 and C contents of these foods(Reference Paluszkiewicz, Smolinska and Debinska13–Reference Wu, Wu and Zheng17). In this regard, it could be suggested that the intake of dairy products might be associated with the reduced risk of PAC because they are rich sources of vitamins A, B12 and D as well as Ca, Mg and Zn(Reference Bao, Ng and Wolpin18–Reference Skinner, Michaud and Giovannucci20). Those vitamins and minerals pose anticarcinogenic properties via inducing cell cycle arrest, apoptosis and differentiation while suppressing angiogenesis, invasion and metastasis(Reference Donaldson21–Reference Ma, Johnson and Trump24). Further, dairy products include large amounts of lactic acid bacteria and conjugated linoleic acids(Reference Buttriss25–Reference Xu, Boylston and Glatz27). Lactic acid bacteria help in enhancing the host’s immune response, inducing antioxidative and antiproliferative functions and detoxifying toxicants formed during food processing, characteristics that have been proven to prevent cancer(Reference Ahmadi, Ebrahimi and Mehrabian28–Reference Shoukat30). In animal models, lactic acid bacteria were able to alleviate pancreatic inflammation, improve glucose tolerance and prevent pancreatic damage, factors that are closely related to the risk of PAC(Reference Wang, Si and Yang31). Alike, conjugated linoleic acids were shown in pre-clinical and human studies to have potential anticarcinogenic effects(Reference den Hartigh32,Reference Whigham, Cook and Atkinson33) . Using human cells in vitro, diets rich in conjugated linoleic acids were shown to reduce PAC penetrance and repress its proliferation(Reference Ding, Mullapudi and Torres34).
Many case–control studies were conducted to detect the retrospective association between dairy intake and PAC(Reference Gold, Gordis and Diener35–Reference Chan, Wang and Holly51). However, in addition to their conflicting findings, the case–control studies were prone to selection bias attributed to the high and rapid fatality of PAC, thus, researchers, to assess dairy intakes, either recruited the survivors who showed low response rates and posed different socio-demographic and clinical characteristics compared with the deceased or interviewed next-of-kin whom data reliability was considered uncertain. Besides, these studies were subject to bias due to the high possibility of changes in dietary habits among cases after PAC diagnosis. Furthermore, the methodological limitations of these case–control studies did not allow the temporal association between dairy intake and PAC to be investigated(Reference Pericleous, Rossi and Mandair52–Reference Michaud, Giovannucci and Willett54). To avoid such biases, the associations between the intakes of different dairy products and the risk of PAC were investigated using a few prospective cohort studies(Reference Stolzenberg-Solomon, Pietinen and Taylor53–Reference Khan, Goto and Kobayashi59). Although the dietary habits of Asian people are substantially different from those in Western countries(Reference Tsugane, Sasaki and Kobayashi60), only two prospective studies assessed the possible association between dairy consumption and the risk of PAC among Asians(Reference Matsumoto, Ishikawa and Nakamura58,Reference Khan, Goto and Kobayashi59) . Both studies were conducted on Japanese people and were limited by the small number of participants and the lack of representativeness; one study included 11 349 residents of thirteen rural areas(Reference Matsumoto, Ishikawa and Nakamura58) and the other study included 3158 residents of one prefecture(Reference Khan, Goto and Kobayashi59).
Since the consumption of dairy products is encouraged in Japan to ensure adequate nutrient intake(Reference Saito, Okada and Tarui61) and given the high incidence of PAC in the country(Reference Bray, Ferlay and Soerjomataram1,Reference Rawla, Sunkara and Gaduputi2) alongside the limitations of previous national studies(Reference Matsumoto, Ishikawa and Nakamura58,Reference Khan, Goto and Kobayashi59) , we used the data of the Japan Collaborative Cohort Study (JACC Study) to investigate the association between the intakes of three dairy products (milk, cheese and yogurt) and the risk of PAC incidence among a large cohort of middle-aged Japanese. This study primarily hypothesised that the intake of dairy products might be inversely associated with the risk of PAC. Then, we conducted a meta-analysis combining the results of the JACC Study with those from previously published prospective cohort studies.
Methods
The Japan Collaborative Cohort Study
Study population and baseline questionnaire
The JACC Study is a prospective cohort study in which baseline data collection was carried out between 1988 and 1990 in forty-five areas in Japan where 110 585 people aged 40–79 years were included. The JACC Study baseline self-administered questionnaire included data about several socio-demographic characteristics, daily walking and leisure physical activity, intakes of common foods and beverages, smoking and alcohol drinking habits and past medical histories(Reference Mori, Sakauchi and Washio62,Reference Tamakoshi, Ozasa and Fujino63) . The follow-up for cancer incidence was conducted in twenty-four areas using population-based and hospital registries or death certificates before it was terminated by the end of 2009 in four areas, 2008 in two areas, 2006 in two areas, 2003 in one area, 2002 in eight areas, 2000 in one area, 1999 in one area, 1997 in four areas and 1994 in one area(Reference Mori, Sakauchi and Washio62). Herein, we excluded people with a positive history of cancer before baseline and people who missed reporting on dairy intake. Eventually, the analysis was confined to 59 774 Japanese people who reported at least one of the three questions assessing dairy intake: 58 656 in milk, 49 302 in cheese and 49 934 in yogurt (Fig. 1).
Fig. 1. Flow chart of the included participants in the JACC Study.
Exposure, outcome and covariates
Data on dairy intake (exposure) were collected using the self-administered FFQ in the JACC Study baseline questionnaire: ‘How frequently do you consume the following items?’. These items included dairy products in the form of ‘milk’, ‘cheese’ and ‘yogurt’ among other common foods. The available responses were as follows: ‘never’, ‘one to two times/month’, ‘one to two times/week’, ‘three to four times/week’ and ‘almost every day’. A validation study among a subsample of the JACC Study’s participants showed good validity and reproducibility of the three investigated items; the Spearman rank correlation coefficients between two frequencies assessed twice apart one year were 0·69 for milk, 0·57 for cheese and 0·54 for yogurt (P-values < 0·001) and between the frequencies and the weighed dietary record were 0·65 for milk, 0·44 for cheese and 0·58 for yogurt (P-values < 0·001)(Reference Date, Fukui and Yamamoto64). The median portion size of the intakes of the three dairy products per day was estimated in the same validation study and was found to be 146 g for milk, 17 g for cheese and 98 g for yogurt(Reference Date, Fukui and Yamamoto64). Therefore, the five frequencies in our study could be roughly converted into the following amounts: ((milk: 0·0, 6·4, 26·8, 64·0 and 128 g/d), (cheese: 0·0, 0·9, 3·6, 8·5 and 17·0 g/d) and (yogurt: 0·0, 4·9, 21·0, 47·0 and 98·0 g/d)).
On the other hand, the incident cases of PAC (outcome) were diagnosed per the tenth revision of the International Statistical Classification of Diseases and Related Health Problems (C25). Cancer incidence was detected using population-based cancer registries supported by a systematic review of hospital-based cancer registries and inpatients’ records of hospitals treating cancer patients(Reference Mori, Sakauchi and Washio62).
Using the same baseline questionnaire, we collected data on participants’ age, sex, weight, height, educational years, perceived stress, smoking and alcohol behaviours, leisure physical activity and walking, history of diabetes, family history of cancer and daily intakes of several foods that served in calculating daily energy intake (covariates).
Statistical analyses
The age and sex-adjusted P-value for significant differences in the participants’ mean values and proportions of socio-demographic characteristics and common risk factors for PAC by their intake of different dairy products were calculated using the linear and logistic regression tests. The Cox proportional regression was used to compute the hazard ratios (HR) and their 95 % CI of the incidence of PAC for the intakes of milk, cheese and yogurt. To obtain statistical power, the two highest intake categories ‘three to four times/week’ and ‘almost every day’ were merged into one category ‘≥ three times/week’.
Person-years of follow-up were calculated from the date of responding to the JACC Study’s baseline questionnaire to the date of PAC diagnosis, death, moving out or end of the study, whichever came first. The HR were adjusted for the following variables: age in years, sex (men and women), BMI (< 25 and ≥ 25 kg/m2), educational years (< 18 and ≥ 18 years), perceived stress (no, mild, moderate and severe stress), smoking habits (never smokers, former smoker of < 20 cigarettes/d, former smoker of ≥ 20 cigarettes/d, current smoker of < 20 cigarettes/d and current smoker of ≥ 20 cigarettes/d), alcohol intake (never, former and current), leisure sports (never, one to two, three to four and ≥ five hours/week), walking (never, < 30, 30–60 and > 60 min/d), history of diabetes (yes and no), family history of cancer (yes and no) and quartiles of daily intakes of meat, vegetables and total energy (g/d). Besides, the possibility of interaction with sex, age, smoking and history of diabetes was examined. SAS version 9.4 software (SAS Institute Inc) was used for statistical analyses.
Ethical consideration
The research ethics committees of Nagoya University School of Medicine and Osaka University approved the protocol of the JACC Study. The study was conducted per the principles of the Declaration of Helsinki.
The meta-analysis
Literature search
First, we searched MEDLINE (PubMed), Embase and Web of Science for potential studies published in English before 31/3/2021 (the last day of data search) using the following terms: (Dairy OR Milk OR Cheese OR Yogurt) AND (Cancer). A full search strategy of PubMed was provided (online Supplementary file 1). Then, we conducted a manual search of the reference lists of retrieved articles and review articles to obtain additional studies. We reported this meta-analysis according to the checklist of PRISMA(Reference Moher, Liberati and Tetzlaff65) and AMSTAR2(Reference Shea, Reeves and Wells66).
Study selection
Studies were selected for analysis if they met the following criteria: (1) the exposure was milk, cheese or yogurt intake, (2) the outcome was PAC and (3) the study design was a prospective cohort. No limitations were set regarding the year of publication; however, no efforts were made to retrieve unpublished data. The following relevant information was extracted from the included studies: the last name of the first author, year of publication, study name, place of study, age and sex of participants, follow-up years, number of incident cases of PAC and covariates included in regression models. The multivariable-adjusted HR with 95 % CI of PAC according to the used categorisations for dairy product intake were also extracted (online Supplementary file 2). The quality of studies was determined using the modified Newcastle–Ottawa Scale based on studies’ selection (representativeness, selection of the non-exposed, ascertainment of exposure and demonstration of the outcome), comparability and outcome (assessment, follow-up length and adequacy)(Reference Wells, Shea and O’Connell67).
Statistical analysis
We used the fixed-effects model to compute the pooled HR with 95 % CI of the included studies(Reference DerSimonian and Laird68) because the test for heterogeneity was not significant according to the I 2 statistic, a measure of inconsistency across studies(Reference Higgins, Thompson and Deeks69). Publication bias was assessed using the regression test for funnel plot asymmetry(Reference Egger, Davey Smith and Schneider70). All analyses were conducted separately on the following dairy products: milk, cheese and yogurt. To explore the impact of each study, we performed a sensitivity analysis by removing studies one by one and combining the remainders in separate analyses. R-3.2.0 statistical package (Metafor: A Meta-Analysis Package for R) was used for analysis(Reference Viechtbauer71).
Results
The Japan Collaborative Cohort Study
In the JACC Study, participants who reported the intakes of milk, cheese and yogurt were younger, with lower BMI, more educated, more physically active and more total energy consumers than their counterparts who reported no intake of the corresponding dairy products (Table 1).
Table 1. Age-sex-adjusted socio-demographic characteristics of participants according to their dairy intakes of milk, cheese and yogurt (JACC Study)
(Numbers; mean values and standard deviations)
* Mean (standard deviation).
Within a mean follow-up period of 13·0 years (median 13·4 years and maximum 21·6 years), a total of 198 incident cases of PAC were diagnosed. The consumption of the highest v. the lowest amounts of milk, cheese and yogurt was not associated with the risk of PAC in the age-and sex-adjusted regression models: HR (95 % CI): 0·91 (0·63, 1·33), 1·01 (0·58, 1·78) and 0·73 (0·41, 1·28), respectively. Adjustment for socio-demographic, clinical and nutritional variables did not change the results: HR (95 % CI): 0·91 (0·62, 1·33) for milk, 0·91 (0·51, 1·62) for cheese and 0·68 (0·38, 1·21) for yogurt. The P-values for trend across the increasing frequencies of the three dairy products were statistically insignificant. Also, the P-values for sex, age, smoking and history of diabetes interactions in the three dairy products were > 0·10 (Table 2).
Table 2. The associations between dairy intakes of milk, cheese and yogurt and the risk of pancreatic cancer (JACC Study)
(Hazard ratio and 95 % confidence intervals)
Model I: Adjusted for age and sex.
Model II: Adjusted further for BMI, education, perceived stress, smoking behaviour, alcohol, leisure sport, walking, history of diabetes, family history of cancer and total meat, vegetables and energy intake.
P-values for sex interaction (milk = 0·80, cheese = 0·85 and yogurt = 0·25).
P-values for age interaction (milk = 0·20, cheese = 0·94 and yogurt = 0·94).
P-values for smoking interaction (milk = 0·30, cheese = 0·11 and yogurt = 0·40).
P-values for history of diabetes interaction (milk = 0·44, cheese = 0·92 and yogurt = 0·42).
The meta-analysis
Herein, we combined our results, in a meta-analysis, with the results of the other prospective cohort studies assessing the associations between the intakes of dairy products and the risk of PAC. After omitting irrelevant and retrospective studies, a shortlist of seven prospective cohort studies was obtained(Reference Stolzenberg-Solomon, Pietinen and Taylor53–Reference Khan, Goto and Kobayashi59) before three studies in the list were excluded; two studies for publishing more recent results from the same data(Reference Stolzenberg-Solomon, Pietinen and Taylor53,Reference Michaud, Giovannucci and Willett54) and one study for defining the exposure as dairy intake as a whole, not as elements of dairy intake(Reference Nöthlings, Wilkens and Murphy55) (Fig. 2). Eventually, four studies were eligible for meta-analysis(Reference Genkinger, Wang and Li56–Reference Khan, Goto and Kobayashi59) which became five after adding the current JACC Study. Of the four added studies, one study was a pooling of fourteen cohorts from North America, Europe and Oceania(Reference Genkinger, Wang and Li56), one study was conducted in Norway(Reference Ursin, Bjelke and Heuch57) and the remaining two studies were conducted in Japan(Reference Matsumoto, Ishikawa and Nakamura58,Reference Khan, Goto and Kobayashi59) . Among the five studies included for this meta-analysis, the assessment of dairy products was distributed as follows: milk in five studies, cheese in two studies and yogurt in four studies. Except for one study that assessed PAC deaths(Reference Matsumoto, Ishikawa and Nakamura58), all studies assessed the risk of PAC incidence. Only one study conducted a sex-specific analysis(Reference Khan, Goto and Kobayashi59) (Table 3). All studies, according to the modified NOS, were of good quality with scores ranging between seven and nine (online Supplementary file 3).
Fig. 2. Prisma chart of the included prospective cohort studies in the meta-analysis.
Table 3. Summary of the prospective cohort studies included in the meta-analysis and investigating the associations between dairy intake and the risk of pancreatic cancer
Alpha-Tocopherol Beta-Carotene Cancer Prevention Study (ATBC); Breast Cancer Detection Demonstration Project Follow-up Study (BCDDP); Canadian National Breast Screening Study (CNBSS); Cancer Prevention Study II Nutrition Cohort (CPS-II); California Teachers Study (CTS); Cohort of Swedish Men (COSM); Health Professionals Follow-up Study (HPFS); Iowa Women’s Health Study (IWHS); Japan Collaborative Cohort Study (JACC); Jichi Medical School Cohort Study (JMS); Melbourne Collaborative Cohort Study (MCCS); The Netherlands Cohort Study (NLCS); New York State Cohort (NYSC); Nurses’ Health Study (NHS); Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO) and the Swedish Mammography Cohort (SMC).
In agreement with the results of the JACC Study, the pooled HR (95 % CI) for milk, cheese and yogurt intakes in the meta-analysis showed no association with the risk of PAC: 0·95 (0·82, 1·11), 1·16 (0·87, 1·55) and 0·91 (0·79, 1·05), respectively. The JACC Study contributed to 16·2 %, 24·6 % and 5·7 % of the meta-analyses’ weights for milk, cheese and yogurt, respectively. The meta-analyses of the three dairy products showed no heterogeneity across studies (I 2 % = 0·00 each). No signs of publication bias were detected in milk and yogurt meta-analyses, while conducting the regression test for publication bias in the cheese meta-analysis was unsuitable due to including two studies only (Table 4) (online Supplementary file 4). Removing the JACC Study from the milk and yogurt meta-analyses did not substantially change the HR (95 % CI): 0·96 (0·81, 1·14) and 0·93 (0·80, 1·07), respectively. The sensitivity analyses by leaving out studies one by one and combining the remainders did not affect the conclusion (online Supplementary file 5).
Table 4. Meta-analysis showing hazard ratios (95 % confidence intervals), weights and heterogeneity and publication bias across studies investigating the association between the intakes of milk, cheese and yogurt and the risk of pancreatic cancer
(Hazard ratio and 95 % confidence intervals)
Discussion
The JACC Study indicated that, within a mean follow-up period of 13·0 years (median 13·4 years), the intakes of milk, cheese and yogurt were not associated with the risk of PAC among middle-aged Japanese, and no dose–response associations were noticed. Combining the results of the JACC Study with those from other prospective cohort studies, in a meta-analysis, did not materially change the findings.
The World Cancer Research Fund and the American Institute of Cancer Research, based on limited evidence, labeled the possible association between consumption of dairy products and the risk of PAC as ‘limited/non-conclusive’. We could not find any association between consuming dairy products and the risk of PAC(72).
Of note, the JACC Study included numerous strengths such as investigating the relationship between consuming several dairy products and the risk of PAC among a large study population, excluding participants with a history of cancer, using a prospective cohort design and lengthy follow-up period, assessing dairy intake using a validated food frequency sheet and adjusting the results for most potential confounders. Still, the JACC Study carried some limitations that should be addressed. First, the number of incident cases of PAC was limited that it did not allow us to stratify the results by potential risk factors for PAC such as sex, age, smoking behaviour and history of diabetes. Yet, this limitation was partly solved by combining the results of the JACC Study with the results of the other four prospective cohort studies via a meta-analysis. Besides, formal interaction tests showed that age, sex, smoking and diabetes did not affect the associations. Second, we obtained no data about the histopathological classifications of PAC cases and their treatment protocols after diagnosis. Third, data on dairy consumption were collected more than 30 years ago. Although dairy consumption per capita in Japan has been increasing since then, the current consumption of dairy products in Japan is much lower than the recommended levels(73,Reference Saito, Okada and Tarui74) . Fourth, it could be speculated that the variation in the PAC ascertainment time across areas because of their different termination times (1994–2009) might have affected the PAC incidence. Yet, the age-adjusted PAC incidence and attributed mortality did not significantly change during this period(Reference Lin, Tamakoshi and Wakai75). Fifth, this study focused on investigating the intake of dietary products as a whole rather than their nutrients; however, the intakes of vitamin D and Ca were shown in previous studies to be not associated with the reduced risk of PAC(Reference Genkinger, Wang and Li56,Reference Fan, Yu and Nan76,Reference Zhang, Huang and Chen77) .
In addition, our meta-analysis posed several strengths such as augmenting the number of incident PAC cases, limiting the inclusion criteria to prospective cohort studies that avoided the methodological limitations of previous case–control studies, including studies of good quality according to the modified Newcastle–Ottawa Scale, and showing no signs of heterogeneity across studies or publication bias.
However, this meta-analysis had some limitations. First, the JACC Study and Genkiger et al study(Reference Genkinger, Wang and Li56) together contributed to most incident cases of PAC and weights of the meta-analyses that were limited by the small number of included studies especially in terms of cheese and yogurt. Second, since exposure was self-reported in all studies, the possibility of non-differential misclassification bias cannot be entirely excluded. Third, it could be argued that two included studies(Reference Ursin, Bjelke and Heuch57,Reference Matsumoto, Ishikawa and Nakamura58) did not adjust their results for smoking, a major risk factor for PAC(Reference Pandol, Apte and Wilson78); however, both studies contributed to small fractions of the meta-analyses weights. Moreover, adjusting for smoking and other socio-demographic, clinical and nutritional factors in the JACC Study did not materially change the results. Fourth, the included studies used different categories for dairy consumption. For example, the highest consumption categories of milk, cheese and yogurt in the Genkiger et al study(Reference Genkinger, Wang and Li56) were ≥ 500, ≥ 50 and ≥ 57 g/d compared with ≥ 64, ≥ 8·5 and ≥ 47 g/d in the JACC Study, respectively. In the previous Japanese studies, Matsumoto et al.(Reference Matsumoto, Ishikawa and Nakamura58) assessed the risk among everyday consumers v. not everyday consumers, while Khan et al.(Reference Khan, Goto and Kobayashi59) compared consuming more than to equal or less than several times per month, and both studies, however, did not calculate the consumed amounts of dairy products. Lastly, our meta-analysis protocol was not a priori registered, although we performed the meta-analysis in a standard way.
In conclusion, consumption of milk, cheese and yogurt was found to be not associated with the risk of PAC among middle-aged Japanese in the JACC Study, and the results did not change in the meta-analyses of prospective cohort studies. Since consuming dairy products was shown to have no role in reducing the risk of PAC, identifying other modifiable risk factors for PAC is important to reduce its burden.
Acknowledgements
This work was supported by Grants-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology of Japan (MEXT) (Monbusho); Grants-in-Aid for Scientific Research on Priority Areas of Cancer; and Grants-in-Aid for Scientific Research on Priority Areas of Cancer Epidemiology from MEXT (MonbuKagaku-sho) (Nos. 61010076, 62010074, 63010074, 1010068, 2151065, 3151064, 4151063, 5151069, 6279102, 11181101, 17015022, 18014011, 20014026, 20390156 and 26293138), Comprehensive Research on Cardiovascular and Life-Style Related Diseases (H26-Junkankitou [Seisaku]-Ippan-001and H29–Junkankitou (Seishuu)–Ippan–003), JSPS KAKENHI Grant Number JP 16H06277 and Grants-in-Aid for China Scholarship Council (CSC file no. 201608050-113).
A. A. (conceptualisation), H. I. and A. T. (resources), H. I. and A. T. (funding acquisition), A. A., E. S. E., J-Y. D., K. S., I. M., H. I. and A. T. (visualisation), A. A. (review literature), A. A. (draft writing), A. A. (data analysis), E. S. E., J-Y. D. and H. I. (supervision) and A. A., E. E., J-Y. D., K. S., I. M., H. I. and A. T. (critical revision and editing).
The authors declare no conflict of interest.
Supplementary material
For supplementary material/s referred to in this article, please visit https://doi.org/10.1017/S0007114521004232
