Coffee may have a substantial impact on health because of its many bioactive compounds with potential therapeutic antioxidant, anti-inflammatory, antifibrotic or anticancer effects (e.g. the alkaloids, caffeine and trigonelline; chlorogenic acids; and the diterpenes, cafestol and kahweol)( Reference Nieber 1 – 3 ). Four meta-analyses showed that coffee consumption was inversely associated with all-cause and CVD mortality( Reference Je and Giovannucci 4 – Reference Grosso, Micek and Godos 7 ). Furthermore, a recent review of meta-analyses suggested that coffee consumption was more often associated with benefit than harm for a range of health outcomes, including all-cause and CVD mortality( Reference Poole, Kennedy and Roderick 8 ). The evaluation was performed on existing studies, which were mainly from Western countries. However, evidence from studies targeting non-white, non-Western populations is still sparse, although lifestyles and disease risks can vary by ethnicity( Reference Loftfield, Freedman and Dodd 9 , Reference Loftfield, Freedman and Graubard 10 ).
Recent studies targeting multi-ethnic, multinational populations reported that coffee consumption was inversely associated with all-cause mortality and such associations did not vary by ethnicity or country( Reference Park, Freedman and Haiman 11 , Reference Gunter, Murphy and Cross 12 ). In two meta-analyses( Reference Je and Giovannucci 4 , Reference Malerba, Turati and Galeone 6 ) (including three Japanese cohort studies( Reference Iwai, Ohshiro and Kurozawa 13 – Reference Tamakoshi, Lin and Kawado 15 )), coffee consumption was also inversely associated with all-cause mortality. A subsequent cohort study in Japan reported that coffee consumption was inversely associated with mortality from all causes, CVD and respiratory disease( Reference Saito, Inoue and Sawada 16 ). Coffee is popular and widely consumed in Asian countries, although the consumption seems to be lower than that in Western countries( 17 ). Past studies in Singapore and Korea reported that a little coffee consumption was inversely associated with hypertension and metabolic syndrome( Reference Chei, Loh and Soh 18 , Reference Kim, Kim and Park 19 ). Thus, studying coffee use among Japanese community-dwelling people will aid in providing further insights into its associations with health.
In the present study, we examined the associations between coffee consumption and all-cause and cause-specific mortality with adjustment for important dietary and lifestyle factors using data from a population-based cohort in a Japanese city. We evaluated the health effects of fewer cups of coffee consumption than that in Western countries because the average consumption is lower in Japan (1–2 cups/person per d).
Methods
Study participants
The Takayama cohort study was initiated in 1992 and targeted all residents aged 35 years or older in Takayama city, Gifu Prefecture, Japan, to examine lifestyle and dietary factors associated with health. The study details were described previously( Reference Shimizu 20 ). In total, 31 552 residents (85·3 %) completed a self-administered questionnaire at baseline, including questions on demographic characteristics, body weight and height, smoking, alcohol drinking, diet and physical activities. We excluded participants who reported a prior diagnosis of cancer, CHD and stroke at baseline (n 2473), which left 29 079 participants (13 355 men and 15 724 women) in the analyses.
Follow-up and end point
The participants were followed from the baseline survey in 1992 to the date of death or the end of follow-up on 1 October 2008. The data on participants who died or moved out of Takayama city were obtained from basic resident or family registration databases. We identified the underlying causes of death using death certificates provided by the Legal Affairs Bureau, which were coded according to the International Classification of Disease, Tenth Revision (ICD-10). The main end points were all-cause mortality and mortality due to cancer (ICD-10: C00–D48), CVD (ICD-10: I00–I99; IHD: I20–I25; stroke: I60–I69) and other causes. Furthermore, respiratory disease (ICD-10: J10–J18 and J40–J47), injury (ICD-10: S00–T98), digestive disease (ICD-10: K00−K93) and infectious disease (ICD-10: A00−B99) were selected as additional end points.
Coffee consumption
Dietary intake including coffee consumption was assessed only at baseline by a validated 169-item semi-quantitative FFQ. The participants were asked to report how often and what amount they consumed for each food and beverage item during the past year. Nutrient intakes were estimated from the information on frequency and portion size using the fifth revised and enlarged edition of the Japanese Standard Tables of Food Composition. Details of the FFQ and the methods used for calculating nutrient intakes were described previously( Reference Nagata, Takatsuka and Shimizu 21 , Reference Shimizu, Ohwaki and Kurisu 22 ). The frequency of coffee consumption during the past year was determined from the questionnaire at baseline and was originally grouped into nine categories: ‘none’, ‘once per month’, ‘2–3 times per month’, ‘once per week’, ‘2–3 times per week’, ‘4–6 times per week’, ‘once per day’, ‘2–3 times per day’ and ‘≥4 times per day’. Participants who drank coffee every day were presumed to answer with the number of coffees they drank in a day. We classified participants who drank coffee occasionally (‘once per month’, ‘2–3 times per month’, ‘once per week’, ‘2–3 times per week’ and ‘4–6 times per week’) into a single category (<1 cup/d). Finally, the participants were categorized as follows: none, <1 cup/d, 1 cup/d, 2–3 cups/d and ≥4 cups/d. We utilized 12 d dietary records over a year to validate the estimated coffee consumption from the questionnaire. Spearman’s correlation coefficients of estimated frequency (the number of cups) between the questionnaire and the records were 0·73 in men and 0·58 in women. We extrapolated the median frequency of coffee consumption using the results from all participants into the data for participants with missing information.
Covariates
We considered the following variables measured at baseline as potential a priori confounders: age; sex; marital status (married or not married (single, divorced/separated or widowed)); years of education (≤8, 9–11, 12–14 or ≥15 years); BMI (<18·5, 18·5–24·9, ≥25·0 kg/m2 or missing); history of diabetes (no or yes); smoking status (never, former or current); alcohol intake (continuous); weekly vigorous exercise (never, 1, 2, 3, 4, 5, 6 or ≥7 times); use of any vitamin supplement (no or yes); total daily energy intake (continuous); daily intakes of vegetables and fruits, and red meat (continuous); and consumption of black tea, green tea, Chinese tea and soda (none, <1 or ≥1 times/d). Alcohol intake was categorized into five categories: non-drinkers or drinkers (<2·9, 2·9–<15·8, 15·8–<44·7 or ≥44·7 mg/d). In the stratified analyses by sex, alcohol intake was categorized into quartiles for men and into three categories for women (i.e. non-drinkers, or drinkers below or above the median for alcohol consumption). Daily intakes of vegetables and fruits, and red meat were adjusted for total energy intake using the residual method of energy adjustment( Reference Willett 23 ).
Statistical analysis
Person-years for each study participant were counted from the date of the baseline survey to the date of death, date of censorship or the end of follow-up (1 October 2008), whichever occurred first. Using Cox proportional hazards models, we estimated age- and sex-adjusted hazard ratios (HR) and 95 % CI for associations of coffee consumption with all-cause and cause-specific mortality using the ‘none’ category as a reference. Tests for linear trends in the associations were performed by assigning scores for the coffee-consumption categories, which started from 0 for the lowest category (none) to 4 for the highest category (≥4 cups/d). We then estimated multivariate HR after adjusting for all relevant confounding variables as mentioned above. We also conducted these analyses after stratifying by sex. In the analyses for cause-specific mortality (i.e. respiratory disease, injury, digestive disease and infectious disease), we collapsed the two higher categories into one category because of a small number of deaths in the category of ≥4 cups/d.
In sensitivity analyses, we first estimated multivariate HR for all-cause mortality after excluding deaths within the first 3 years. To examine the impact of residual confounding from smoking status, we next conducted further analyses adjusting for smoking-years category (never, smoking <10, 11–20, 21–30, ≥30 years or smoking-years missing) instead of the smoking status category in the multivariate models. We also conducted the same analyses with smoking-cessation years category (currently smoking, <5, 6–10, >10 years of smoking cessation or never smoking). Furthermore, considering dietary patterns which are potentially correlated with coffee consumption among Japanese adults( Reference Maruyama, Iso and Date 24 , Reference Shimazu, Kuriyama and Hozawa 25 ), we additionally adjusted for intakes of fish, rice and dairy products. Statistical significance was defined as a two-sided P value of less than 0·05. The statistical software package Stata SE version 14.2 was used for all analyses.
Results
Table 1 shows the baseline characteristics of all participants according to coffee consumption category. Compared with participants who did not drank coffee, those who drank two or more cups of coffee daily tended to be younger, male, married and current smokers. They also tended to have a higher level of education, consume a larger amount of alcohol, have higher intakes of total energy and red meat, and have a lower intake of vegetables and fruits. In addition, they were more likely to drink black tea, green tea, Chinese tea and soda, and were less likely to have diabetes and use any vitamin supplement. After stratification by sex, men and women showed the same tendency except alcohol consumption in men; that is, men who drank two or more cups of coffee daily tended to consume a smaller amount of alcohol (see online supplementary material, Supplemental Table S1).
Baseline characteristics of total participants according to daily coffee consumption: Japanese adults aged ≥35 years (n 29 079) followed from 1992 to 2008, Takayama study
IQR, interquartile range.
During 16 years of follow-up (mean = 14·1 years), 5339 deaths were identified among 410 352 total person-years. Table 2 shows the associations between coffee consumption and mortality from all causes, cancer, CVD and other causes among all participants and stratified by sex. Regarding all-cause mortality, coffee consumption was associated with reduced risk among all participants after adjusting for all potential confounders. Compared with the category of ‘none’, the HR (95 % CI) was 0·93 (0·86, 1·00) for <1 cup/d, 0·84 (0·76, 0·93) for 1 cup/d, 0·81 (0·71, 0·92) for 2–3 cups/d and 1·09 (0·80, 1·50) for ≥4 cups/d (P for trend <0·001). A similar pattern was observed for men, but not for women.
Hazard ratios (HR) for mortality from all causes, cancer, CVD and other causes among total participants and participants stratified by sex, according to daily coffee consumption: Japanese adults aged ≥35 years (n 29 079) followed from 1992 to 2008, Takayama study
Ref., reference category.
* Marital status, years of education, BMI, diabetes, smoking status, alcohol intake, vigorous exercise, use of any vitamin supplement, dietary intake (total energy, vegetables and fruits, red meat) and beverage consumption (black tea, green tea, Chinese tea, soda) were also adjusted for.
Coffee consumption was not associated with mortality from cancer among all participants or women, although drinking 1 cup coffee/d was associated with reduced risk in men (Table 2). Regarding mortality from CVD, coffee consumption was associated with reduced risk among all participants. Compared with the category of ‘none’, the HR (95 % CI) were 0·87 (0·77, 0·99) for <1 cup/d, 0·76 (0·63, 0·92) for 1 cup/d, 0·67 (0·50, 0·89) for 2–3 cups/d and 1·39 (0·77, 2·49) for ≥4 cups/d (P for trend = 0·002). A similar pattern was observed for men and women, although inverse associations among women across all categories did not reach statistical significance. Specifically, coffee consumption was associated with reduced risk for mortality from stroke and IHD among all participants. The HR (95 % CI) for the categories of <1 cup/d, 1 cup/d, 2–3 cups/d and ≥4 cups/d were 0·86 (0·70, 1·05), 0·64 (0·47, 0·87), 0·65 (0·41, 1·02) and 1·11 (0·41, 3·06) for stroke mortality, and 0·68 (0·50, 0·91), 0·69 (0·46, 1·04), 0·50 (0·27, 0·92) and 1·41 (0·50, 3·98) for IHD mortality, respectively. Regarding mortality from other causes, coffee consumption was associated with reduced risk among all participants. The HR (95 % CI) were 0·91 (0·81, 1·03) for <1 cup/d, 0·85 (0·73, 1·00) for 1 cup/d, 0·72 (0·57, 0·90) for 2–3 cups/d and 1·00 (0·58, 1·72) for ≥4 cups/d (P for trend = 0·005). A similar pattern was observed for men, but not for women.
Table 3 shows the associations between coffee consumption and mortality from other specific causes among all participants. In the multivariate models, coffee consumption was significantly associated with reduced risk of mortality from digestive and infectious diseases. Coffee consumption was also inversely associated with mortality from the other specific causes, although the point estimates were not statistically significant.
Hazard ratios (HR) for mortality from other specific causes among total participants, according to daily coffee consumption: Japanese adults aged ≥35 years (n 29 079) followed from 1992 to 2008, Takayama study
Ref., reference category.
* The categories of 2–3 cups/d and ≥4 cups/d were collapsed into the single category of ≥2 cups/d because of a small number of deaths from the specific causes in the category of ≥4 cups/d.
† Age, sex, marital status, years of education, BMI, diabetes, smoking status, alcohol intake, vigorous exercise, use of any vitamin supplement, dietary intake (total energy, vegetables and fruits, red meat) and beverage consumption (black tea, green tea, Chinese tea, soda) were adjusted for.
In the sensitivity analyses (Table 4), even after excluding deaths within the first 3 years, the main findings for all-cause mortality among all participants did not change substantially. Furthermore, adjusting for the categories of smoking-years or smoking-cessation years instead of smoking status also did not change the findings substantially.
Sensitivity of hazard ratios (HR) for all-cause mortality among total participants, according to daily coffee consumption: Japanese adults aged ≥35 years (n 29 079) followed from 1992 to 2008, Takayama study
Ref., reference category.
* Age, sex, marital status, years of education, BMI, diabetes, smoking status, alcohol intake, vigorous exercise, use of any vitamin supplement, dietary intake (total energy, vegetables and fruits, red meat) and dairy beverage consumption (black tea, green tea, Chinese tea, soda) were adjusted for.
Discussion
We used data from a population-based prospective cohort study in a Japanese city (the Takayama study) to examine the associations of coffee consumption with all-cause and cause-specific mortality after adjusting for important dietary and lifestyle factors. The results showed that drinking coffee, even 1 cup/d, was associated with reduced risk of mortality from all causes, CVD and other causes among all participants, although the associations for ≥4 cups/d were no more significant. In the stratified analyses by sex, similar tendencies were observed for men, but not for women. Furthermore, coffee consumption was also inversely associated with mortality specifically from infectious and digestive diseases.
Although we found that drinking 1 cup coffee/d was associated with reduced risk for mortality from all and other specific causes, previous studies indicated that risks were lowest at 3–4 cups/d for all-cause and CVD mortality with no further reductions at >4 cups/d( Reference Crippa, Discacciati and Larsson 5 , Reference Grosso, Micek and Godos 7 , Reference Grosso, Stepaniak and Micek 26 ). In the present study, there were few participants who drank ≥4 cups coffee/d (n 483, 1·7 %); therefore, we may have failed to observe a tendency in these associations.
The present finding of reduced risk for all-cause mortality is consistent with meta-analyses( Reference Je and Giovannucci 4 – Reference Grosso, Micek and Godos 7 , Reference Grosso, Stepaniak and Micek 26 ) and recent large-scale cohort studies in Western countries (targeting multi-ethnic populations)( Reference Park, Freedman and Haiman 11 , Reference Gunter, Murphy and Cross 12 ) and Japan( Reference Saito, Inoue and Sawada 16 ). In the present study, the risks decreased by 16 and 19 % for the categories of 1 cup/d and 2–3 cups/d, respectively, compared with the ‘none’ category. These findings are similar to those from large-scale cohort studies (15 % for 1 cup/d and 20 % for 2–3 cups/d in risk reductions among Japanese Americans compared with the ‘none’ category( Reference Park, Freedman and Haiman 11 ), and 15 % for the 1–2 cups/d category and 24 % for the 3–4 cups/d category in risk reductions compared with the ‘almost-none’ category( Reference Saito, Inoue and Sawada 16 )).
Coffee consumption was inversely associated with reduced risk for mortality from CVD, but not from cancer in the present study. This finding is consistent with three meta-analyses( Reference Crippa, Discacciati and Larsson 5 – Reference Grosso, Micek and Godos 7 ) and previous studies in Japan( Reference Sugiyama, Kuriyama and Akhter 14 – Reference Saito, Inoue and Sawada 16 , Reference Mineharu, Koizumi and Wada 27 ). The protective associations on mortality from CVD are plausible because of the presence of various bioactive compounds in coffee, as mentioned above. By contrast, the association of coffee with cancer mortality may vary by cancer site (e.g. probable protective associations between coffee consumption and liver and endometrial cancers are suggested( 3 , 28 )). Further studies evaluating coffee’s associations with site-specific cancer mortality in non-Western countries are needed.
After stratification by sex, the inverse associations with mortality from all causes, CVD and other causes did not reach statistical significance in women. The potential reason is that women appeared to report coffee consumption less accurately than men. This may result in non-differential misclassifications, just making the estimates towards the null. It is also possible that caffeine, a coffee compound, could have sex-differentiated impacts on health( Reference Poole, Kennedy and Roderick 8 , Reference Temple and Ziegler 29 ). For example, the CVD mortality risk in women is lower than that in men, which is influenced by the interactions between genetic, hormonal and environmental factors( Reference Winham, de Andrade and Miller 30 , Reference Deo and Albert 31 ). These factors could have sex-differentiated impacts on the associations between coffee consumption and CVD mortality.
In addition, coffee consumption seemed to be inversely associated with other specific-cause (non-CVD, non-cancer) mortality, although we found significant inverse associations only for mortality from infectious and digestive diseases. The bioactive compounds in coffee (e.g. polyphenol, caffeine and melanoidins) could exert such beneficial effects on liver function and chronic inflammation( Reference Nieber 1 , Reference Ludwig, Clifford and Lean 2 ). Past studies reported that coffee consumption was associated with lower levels of liver enzymes, C-reactive protein and adiponectin( Reference Gunter, Murphy and Cross 12 , Reference Xiao, Sinha and Graubard 32 – Reference Zhang and Zhang 34 ). Therefore, it is likely that coffee consumption is also inversely associated with mortality from respiratory disease and endocrine/metabolic disease, as suggested by previous studies( Reference Poole, Kennedy and Roderick 8 , Reference Park, Freedman and Haiman 11 , Reference Gunter, Murphy and Cross 12 , Reference Saito, Inoue and Sawada 16 , Reference Freedman, Park and Abnet 35 , Reference Ding, Satija and Bhupathiraju 36 ). The non-significant inverse associations we observed could be due to a small number of cases. Future studies evaluating associations between coffee consumption and specific causes of death are thus awaited.
The major strengths of the present study are its prospective cohort design with 16 years of follow-up, its high participation rate (85·3 %), use of a valid dietary questionnaire and considering important potential confounding factors simultaneously. Therefore, the present findings provide further evidence for the protective effects of coffee consumption among non-white, non-Western populations.
The present study has several limitations. First, it did not provide strong evidence for high coffee consumption, despite the dose–response relationships in the inverse associations between coffee consumption and all-cause mortality and mortality specifically from CVD, infectious and digestive diseases. Second, coffee consumption was assessed at baseline only and may not reflect long-term coffee consumption, although a study of three prospective US cohorts indicated that a single assessment can capture coffee drinking habits over 10–20 years( Reference Ding, Satija and Bhupathiraju 36 ). Indeed, participants with chronic disease may reduce their coffee consumption, but we excluded those participants from the analyses. Moreover, we conducted sensitivity analyses excluding deaths within the first 3 years, which did not change our findings substantially. However, the referent category of none could include those who had quit drinking coffee before baseline because of known or undiagnosed symptoms. Third, we did not collect information on how the participants prepared and served coffee, although its bioactive compounds vary depending on bean type, degree of roasting and preparation method, including quality of beans, grind settings and brewing methods( Reference Poole, Kennedy and Roderick 8 , Reference Corrêa, Monteiro and Mendes 37 ). Furthermore, we cannot differentiate between canned, instant or brewed coffee and cannot evaluate the associations of decaffeinated coffee because drinking decaffeinated coffee was likely to be rare in Japan( Reference Saito, Inoue and Sawada 16 ). These factors may result in moving the HR towards the null. Finally, we cannot rule out the possibility of residual confounding due to smoking status, which would be an important potential confounding factor. However, sensitivity analyses using two different smoking categories did not change the main findings substantially. In large-scale cohort studies, the results for the associations between coffee consumption and mortality were similar across smoking categories (i.e. never, former or current)( Reference Park, Freedman and Haiman 11 , Reference Gunter, Murphy and Cross 12 , Reference Freedman, Park and Abnet 35 ).
Conclusion
In conclusion, after adjusting for important dietary and lifestyle confounders, we demonstrated that drinking coffee, even 1 cup/d, was associated with reduced risk of all-cause mortality and mortality specifically from CVD, infectious and digestive diseases. Thus, a little coffee consumption will contribute to increasing the health and long life of people in Asian countries, such as Japan.
Acknowledgements
Financial support: This work was supported in part by grants from the Ministry of Education, Culture, Sports, Science, and Technology and from the National Cancer Center Research and Development Fund, Japan. The sponsors were not involved in deciding the study design, the collection, analysis and interpretation of data, the writing of the report or the decision to submit this paper for publication. Conflict of interest: C.N. is an advisor to the All Japan Coffee Association. The other authors have no conflicts of interest to disclose. Authorship: M.Y., Y.G., F.M., S.K. and T.U. designed the study and analytical strategy; C.N. obtained data; K.W., M.Y. and C.N. performed analysis and interpretation of data; M.Y. drafted the initial manuscript; K.W., Y.G., F.M., S.K., T.U. and C.N. reviewed and revised the manuscript; C.N. obtained the grants and supervised the study; and all authors approved the final manuscript as submitted. Ethics of human subject participation: This study was approved by the Ethics Committee of Gifu University Graduate School of Medicine.
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S1368980019000764
