Hostname: page-component-788cddb947-w95db Total loading time: 0 Render date: 2024-10-15T10:03:46.712Z Has data issue: false hasContentIssue false

Association of coffee drinking with all-cause mortality: a systematic review and meta-analysis

Published online by Cambridge University Press:  04 August 2014

Yimin Zhao
Affiliation:
Department of Food Science and Nutrition, Zhejiang University, 866 Yuhangtang Road, Hangzhou 310058, People’s Republic of China APCNS Centre of Nutrition and Food Safety, Hangzhou, People’s Republic of China
Kejian Wu
Affiliation:
Department of Food Science and Nutrition, Zhejiang University, 866 Yuhangtang Road, Hangzhou 310058, People’s Republic of China APCNS Centre of Nutrition and Food Safety, Hangzhou, People’s Republic of China
Jusheng Zheng
Affiliation:
Department of Food Science and Nutrition, Zhejiang University, 866 Yuhangtang Road, Hangzhou 310058, People’s Republic of China APCNS Centre of Nutrition and Food Safety, Hangzhou, People’s Republic of China
Ruiting Zuo
Affiliation:
Department of Food Science and Nutrition, Zhejiang University, 866 Yuhangtang Road, Hangzhou 310058, People’s Republic of China APCNS Centre of Nutrition and Food Safety, Hangzhou, People’s Republic of China
Duo Li*
Affiliation:
Department of Food Science and Nutrition, Zhejiang University, 866 Yuhangtang Road, Hangzhou 310058, People’s Republic of China APCNS Centre of Nutrition and Food Safety, Hangzhou, People’s Republic of China
*
*Corresponding author: Email duoli@zju.edu.cn
Rights & Permissions [Opens in a new window]

Abstract

Objective

We aimed to use the meta-analysis method to assess the relationship between coffee drinking and all-cause mortality.

Design

Categorical and dose–response meta-analyses were conducted using random-effects models.

Setting

We systematically searched and identified eligible literature in the PubMed and Scopus databases.

Subjects

Seventeen studies including 1 054 571 participants and 131 212 death events from all causes were included in the present study.

Results

Seventeen studies were included and evaluated in the meta-analysis. A U-shaped dose–response relationship was found between coffee consumption and all-cause mortality (P for non-linearity <0·001). Compared with non/occasional coffee drinkers, the relative risks for all-cause mortality were 0·89 (95 % CI 0·85, 0·93) for 1–<3 cups/d, 0·87 (95 % CI 0·83, 0·91) for 3–<5 cups/d and 0·90 (95 % CI 0·87, 0·94) for ≥5 cups/d, and the relationship was more marked in females than in males.

Conclusions

The present meta-analysis of prospective cohort studies indicated that light to moderate coffee intake is associated with a reduced risk of death from all causes, particularly in women.

Type
Review Article
Copyright
Copyright © The Authors 2014 

Coffee is one of the most popular beverages in the world and the health-related effects of coffee have been frequently studied. Habitual coffee drinking was reported to be inversely related to the risks of type 2 diabetes( Reference Van Dam and Hu 1 ) and chronic liver disease( Reference Lai, Weinstein and Albanes 2 ). As a major dietary source of antioxidants, coffee may also help to improve the resistance of LDL to oxidation and reduce oxidative DNA damage( Reference Misik, Hoelzl and Wagner 3 ).

Results from prospective cohort studies regarding the association of habitual coffee drinking with all-cause mortality were inconclusive( Reference Freedman, Park and Abnet 4 Reference Tamakoshi, Lin and Kawado 6 ). O’Keefe et al.( Reference O’Keefe, Bhatti and Patil 7 ) recommended that moderate intake of coffee, tending towards two or three to as many as four cups daily, would be a better choice for keeping healthy rather than excessive coffee consumption. Besides, the association of coffee intake with all-cause mortality may differ between men and women. Lopez-Garcia et al.( Reference Lopez-Garcia, van Dam and Li 8 ) observed that the significant inverse association of coffee drinking with total mortality was attenuated in men when compared with that in women. Similar results were also found in another large cohort study( Reference Kleemola, Jousilahti and Pietinen 9 ). A recent meta-analysis of prospective cohort studies suggested that moderate coffee intake was associated with a lower risk of CHD in female drinkers but not in men( Reference Wu, Ho and Zhou 10 ). However, the health-related effects of coffee may not always be in favour of women. Ascherio et al.( Reference Ascherio, Zhang and Hernán 11 ) reported that coffee consumption was associated with reduced mortality from Parkinson’s disease in men but not in women due to the interaction between caffeine and use of postmenopausal oestrogens. Therefore, we conducted a meta-analysis of prospective cohort studies to investigate the association of coffee consumption with all-cause mortality and further to elucidate whether this association varied between male and female coffee drinkers.

Methods

Literature search

Two investigators (Y.Z. and K.W.) independently performed a literature search in the PubMed and Scopus databases for eligible publications up to November 2013 using the search query: (coffee AND (death OR mortality)). The search was restricted to articles published in English. An additional manual search was also conducted using reference lists from previous meta-analyses and review articles. To be included, a study had meet all of the following criteria: (i) prospective cohort design; (ii) evaluate the association of habitual coffee drinking with all-cause mortality in a normal population; (iii) provide adjusted risk estimates with 95 % confidence intervals; and (iv) use non/occasional coffee drinkers as the reference category. Studies carried out in people with hypertension, diabetes, CVD or cancer were not to be taken into consideration. For different articles reporting data from the same population, we selected the article with the largest sample size.

Data extraction

The following information for each study was extracted by two investigators (Y.Z. and R.Z.) independently: (i) first author’s surname; (ii) geographic region where the study was done; (iii) mean follow-up time; (iv) age range and gender of population; (v) total number of subjects and number of death events due to all causes; (vi) coffee intake categories; (vii) corresponding maximally adjusted risk estimates with 95 % confidence intervals; and (viii) confounding variables controlled for. If a study reported outcomes for men and women separately, we separated it as two different cohorts. For dose–response meta-analysis, we also extracted the number of death events and subjects or person-years in each category. If the numbers of death events by levels of coffee intake were not provided directly, we estimated them from the total number of subjects along with the non-adjusted risk estimates or, if not reported, the minimally adjusted risk estimates within each category in that study.

In each category, the median or the mean coffee intake was assigned as the dose of average coffee consumption. If the median or the mean coffee intake was not provided either, the midpoint of the upper and lower boundaries was regarded as the average dose. For an open-ended highest category, the average coffee consumption of the category was set at 1·2 times the lower boundary. If the lower boundary of a lowest category was not available, we assumed the lower boundary was zero.

Statistical analysis

Heterogeneity among studies was assessed by the I 2 statistic, which represents the amount of total variability that is attributed to between-study variability( Reference Higgins and Thompson 12 , Reference Jackson, White and Thompson 13 ). To facilitate description, we considered that I 2 values of <25 %, 25–75 % and >75 % indicated light, moderate and high level of heterogeneity among studies, respectively.

We used relative risk (RR) to express different risk estimates in each study. The natural logarithms of the maximally adjusted relative risk and corresponding 95 % confidence interval in each category were used for pooled analyses. The random-effects model developed by DerSimonian and Laird( Reference DerSimonian and Laird 14 ), which provides more conservative results, was employed to calculate the pooled estimates. For categorical meta-analyses we set four coffee intake groups: (i) non/occasional drinkers (the lowest category in each study); (ii) 1–<3 cups/d; (iii) 3–<5 cups/d; and (iv) ≥5 cups/d. This classification is similar to the one used in a previous meta-analysis on coffee( Reference Larsson and Orsini 15 ). If multiple categories from the same study were located in a single group, we combined them into one category. We further conducted subgroup analyses stratified by gender, geographic region and degree of adjustment. For the analyses stratified by degree of adjustment, we predefined two models: (i) model A referred to studies providing risk estimates and 95 % confidence intervals adjusted for age, smoking status, alcohol intake and physical activity; and (ii) model B referred to studies which provided risk estimates and 95 % confidence intervals adjusted for education level, which could reflect the socio-economic status of coffee drinkers, on the basis of model A. Sensitivity analyses were carried out by omitting one study at a time to examine whether the pooled estimates were driven by any single study. Possible publication bias was tested using Egger’s linear regression test( Reference Egger, Smith and Schneider 16 ) and Begg’s rank correlation test( Reference Begg and Mazumdar 17 ).

We further conducted a two-stage random-effects dose–response meta-analysis to evaluate the association between coffee drinking and all-cause mortality. The potential non-linearity was examined by modelling coffee consumption using restricted cubic splines with three knots at fixed percentiles (10 %, 50 %, 90 %) of the coffee intake distribution, as suggested by Harrell( Reference Harrell 18 ). A P value for possible non-linearity was obtained by testing the null hypothesis that the regression coefficient of the second spline was equal to zero( Reference Orsini, Li and Wolk 19 ). We adopted generalized least squares models as proposed by Greenland and Longnecker( Reference Greenland and Longnecker 20 ) and Orsini et al.( Reference Orsini, Bellocco and Greenland 21 ) to calculate study-specific coefficient estimates and variance/covariance matrix. Then, the two regression coefficients of each study were pooled in the multivariate random-effects meta-regression analysis as described by Jackson et al.( Reference Jackson, White and Thompson 13 ). Those studies which provided less than three coffee consumption categories, including the reference category, were excluded from the dose–response meta-analyses.

All P values were two-tailed with a significance level of 0·05. All statistical analyses were conducted using the statistical software package Stata/SE 12·0 for Windows.

Results

Literature search and study characteristics

Figure 1 presents the flowchart of the literature search. Finally, seventeen eligible studies consisting of 1 054 571 participants and 131 212 death events due to all causes were identified and included in the present study( Reference Freedman, Park and Abnet 4 Reference Tamakoshi, Lin and Kawado 6 , Reference Lopez-Garcia, van Dam and Li 8 , Reference Ahmed, Levitan and Wolk 22 Reference Vandenbroucke, Kok and van T 34 ). The mean follow-up time of each study ranged from 7·1 to 25 years with a median of 15 years. Nine studies( Reference Freedman, Park and Abnet 4 Reference Tamakoshi, Lin and Kawado 6 , Reference Lopez-Garcia, van Dam and Li 8 , Reference Iwai, Ohshiro and Kurozawa 26 , Reference Jazbec, Simic and Corovic 27 , Reference Laaksonen, Talala and Martelin 30 , Reference Sugiyama, Kuriyama and Akhter 33 , Reference Vandenbroucke, Kok and van T 34 ) separately reported outcomes for men and women, hence there were eleven cohorts on men and ten cohorts on women. Of the seventeen studies, eight were conducted in the USA( Reference Freedman, Park and Abnet 4 , Reference Liu, Sui and Lavie 5 , Reference Lopez-Garcia, van Dam and Li 8 , Reference Andersen, Jacobs and Carlsen 23 , Reference Gardener, Rundek and Wright 25 , Reference Kahn, Phillips and Snowdon 28 , Reference Klatsky, Armstrong and Friedman 29 , Reference Paganini-Hill, Kawas and Corrada 31 ), six in Europe( Reference Ahmed, Levitan and Wolk 22 , Reference de Koning Gans, Uiterwaal and van der Schouw 24 , Reference Jazbec, Simic and Corovic 27 , Reference Laaksonen, Talala and Martelin 30 , Reference Rosengren and Wilhelmsen 32 , Reference Vandenbroucke, Kok and van T 34 ) and three in Japan( Reference Tamakoshi, Lin and Kawado 6 , Reference Iwai, Ohshiro and Kurozawa 26 , Reference Vandenbroucke, Kok and van T 34 ). Twelve studies reported risk estimates with 95 % confidence intervals adjusted for age, smoking status, alcohol intake and physical activity( Reference Freedman, Park and Abnet 4 Reference Tamakoshi, Lin and Kawado 6 , Reference Lopez-Garcia, van Dam and Li 8 , Reference Ahmed, Levitan and Wolk 22 Reference Iwai, Ohshiro and Kurozawa 26 , Reference Paganini-Hill, Kawas and Corrada 31 Reference Sugiyama, Kuriyama and Akhter 33 ) and eight of them additionally controlled for education level( Reference Freedman, Park and Abnet 4 , Reference Lopez-Garcia, van Dam and Li 8 , Reference Ahmed, Levitan and Wolk 22 Reference Iwai, Ohshiro and Kurozawa 26 , Reference Sugiyama, Kuriyama and Akhter 33 ). The characteristics of all included studies are shown in Table 1.

Fig. 1 Flowchart of the literature search

Table 1 Characteristics of studies included in the present meta-analysis of coffee drinking and all-cause mortality

RR, relative risk; M, men; W, women; Ref., referent category; MI, myocardial infarction; WHR, waist:hip ratio; DBP, diastolic blood pressure; SBP, systolic blood pressure.

All studies

When compared with non/occasional coffee drinkers, the pooled RR for all-cause mortality were 0·89 (95 % CI 0·85, 0·93; I 2=75 %) for 1–<3 cups/d, 0·87 (95 % CI 0·83, 0·91; I 2 =60 %) for 3–<5 cups/d and 0·90 (95 % CI 0·87, 0·94; I 2=19 %) for ≥5 cups/d. There was considerable between-study heterogeneity in each coffee intake group and relevant forest plots are provided in the online supplementary material. No evidence of publication bias was found (for 1–<3 cups/d, Begg’s P=0·21, Egger’s P=0·59; for 3–<5 cups/d, Begg’s P=0·84, Egger’s P=0·80; for ≥5 cups/d, Begg’s P=0·74, Egger’s P=0·61). In the sensitivity analyses, omitting studies one by one did not change the significance of any pooled RR. For the coffee consumption group of 1–<3 cups/d, after excluding Kahn et al.( Reference Kahn, Phillips and Snowdon 28 ), the heterogeneity among studies that combined men and women together reduced from 76 % to 20 %; however, the between-study heterogeneity of all included studies in this coffee intake group was not obviously lowered.

Two studies( Reference Rosengren and Wilhelmsen 32 , Reference Vandenbroucke, Kok and van T 34 ) were not incorporated in the dose–response meta-analysis because of limited numbers of coffee intake categories. We observed a significant non-linear association (P for non-linearity <0·001) between coffee drinking and all-cause mortality (Fig. 2(a)). A moderate degree of heterogeneity was detected among study-specific trends derived from the coefficients of the first and second spline transformations: I 2 first=69 % (95 % CI 53 %, 80 %) and I 2 second=29 % (95 % CI 0 %, 57 %). The slope of the dose–response relationship was approximately U-shaped and the maximum protection was reached at 3–4 cups coffee/d.

Fig. 2 The non-linear association of habitual coffee drinking with all-cause mortality: (a) all studies (P for non-linearity <0·001); (b) cohorts of men only (P for non-linearity <0·001); (c) cohorts of women only (P for non-linearity <0·001). ———, relative risk (RR); — — —, 95 % confidence intervals; – – –, null effect. The results were gained from the two-stage random-effects dose–response meta-analyses

Men and women

As one of the main objectives of the present research was to explore whether the association of coffee intake with total mortality differed between men and women, those studies that combined two sexes together were not included in the subgroup analyses( Reference de Koning Gans, Uiterwaal and van der Schouw 24 , Reference Gardener, Rundek and Wright 25 , Reference Kahn, Phillips and Snowdon 28 , Reference Klatsky, Armstrong and Friedman 29 , Reference Paganini-Hill, Kawas and Corrada 31 ). The results of subgroup analyses are presented in Fig. 3. With regard to male coffee drinkers, the pooled RR for all-cause mortality were 0·91 (95 % CI 0·84, 0·99; I 2=78 %) for 1–<3 cups/d, 0·90 (95 % CI 0·85, 0·97; I 2=55 %) for 3–<5 cups/d and 0·94 (95 % CI 0·84, 1·04; I 2=54 %) for ≥5 cups/d. No evidence of linear trend (P for non-linearity <0·001) was detected for the association of coffee intake and total mortality in men (Fig. 2(b)) and the heterogeneity among study-specific trends estimates were: I 2 first=62 % (95 % CI 25 %, 81 %) and I 2 second=0 % (95 % CI 0 %, 62 %).

Fig. 3 Subgroup analyses of the association between habitual coffee drinking and all-cause mortality, showing pooled relative risks (RR; ■) and 95 % confidence intervals (represented by horizontal bars) from categorical meta-analyses using a random-effects model. Model A, studies providing risk estimates and 95 % CI adjusted for age, smoking status, alcohol intake and physical activity; model B, studies providing risk estimates and 95 % CI additionally adjusted for education level

In female coffee drinkers, the summarized RR for total mortality were 0·84 (95 % CI 0·80, 0·88; I 2=22 %) for 1–<3 cups/d, 0·81 (95 % CI 0·75, 0·87; I 2=39 %) for 3–<5 cups/d and 0·85 (95 % CI 0·80, 0·90; I 2=0 %) for ≥5 cups/d in comparison with non/occasional coffee drinkers (Fig. 3). The slope of the relationship between coffee consumption and total mortality in women was roughly U-shaped (P for non-linearity <0·001, Fig. 2(c)) and the maximum reduction in all-cause mortality was observed at 3–5 cups coffee/d. Moderate heterogeneity among trend estimates of relevant studies was found: I 2 first=41 % (95 % CI 0 %, 73 %) and I 2 second=31 % (95 % CI 0 %, 68 %).

For men who drunk ≥5 cups coffee/d, there were marginally significant inverse associations of coffee intake with total mortality in studies that reported risk estimates with 95 % confidence intervals adjusted for age, smoking status, alcohol intake and physical activity (RR=0·92; 95 % CI 0·83, 1·00; I 2=0 %; model A, Fig. 3) and in those that additionally controlled for education level (RR=0·89; 95 % CI 0·81, 0·98; I 2=0 %; model B, Fig. 3). For men who drunk 1–<3 cups coffee/d or 3–<5 cups coffee/d and for female coffee drinkers in all the three coffee consumption groups, the pooled RR and corresponding 95 % CI of all-cause mortality remained stable in the subgroup analyses stratified by degree of adjustment (models A and B, Fig. 3). In the subgroup analyses divided by geographical region, there was no evidence of significant inverse associations between habitual coffee drinking and all-cause mortality in all the three coffee consumption groups for male drinkers from the USA or European countries (Fig. 3). However, these results should be interpreted with caution due to the limited numbers of studies.

Discussion

The present research did not support the hypothesis that habitual coffee consumption increased the risk of death from all causes, whereas light to moderate coffee drinking was inversely associated with all-cause mortality especially for female drinkers. The significant reduction in total mortality was attenuated in those who drunk large quantities of coffee for both men and women.

Natella et al.( Reference Natella, Nardini and Belelli 35 ) found that one cup of coffee daily helped to incorporate phenolic acids into LDL and increased the resistance of LDL to ex vivo oxidation in healthy individuals. Meta-analyses of prospective cohort studies have suggested U-shaped associations between coffee drinking and the incidence of stroke( Reference Larsson and Orsini 15 ), congestive heart failure( Reference Mostofsky, Rice and Levitan 36 ) and Parkinson’s disease( Reference Hernan, Takkouche and Caamano Isorna 37 ), with the maximum and also statistically significant protection observed in people who drunk three to four cups of coffee daily in all cases. Despite case–control and cohort studies having reported inconclusive results about the relationship between coffee intake and CHD( Reference Wu, Ho and Zhou 10 ), it is plausible that the inverse association between light to moderate coffee consumption and all-cause mortality may be partially attributed to the protection against certain kinds of cardiovascular and cerebrovascular diseases. The associations of moderate coffee consumption with the incidence and mortality of cancer varied by body sites and low to moderate coffee consumption was inversely related to total cancer incidence( Reference Yu, Bao and Zou 38 ). Whether there is a reverse causality between light to moderate coffee drinking and reduced all-cause mortality requires further investigation.

For both men and women, the reduction in total mortality was attenuated in those who drunk large quantities of coffee. Excessive intake of coffee could neutralize its benefits and even impose unfavourable effects on the human body. In healthy individuals, heavy consumption of either filtered coffee or unfiltered coffee could give rise to substantially elevated plasma concentrations of total homocysteine( Reference Urgert, van Vliet and Zock 39 , Reference Grubben, Boers and Blom 40 ). Olthof et al.( Reference Olthof, Hollman and Zock 41 ) assumed that the homocysteine-elevating effect was possibly caused by the chlorogenic acid in coffee. For hypertensive individuals, excessive coffee intake could produce an acute increase in blood pressure and thus could raise the blood pressure above a safe level, but no evidence was found to support an appreciable association between habitual coffee consumption and a higher risk of CVD in hypertensive individuals( Reference Mesas, Leon-Munoz and Rodriguez-Artalejo 42 ). Cafestol and kahweol contained in coffee beans are regarded as cholesterol-raising factors and it is well documented that unfiltered coffee can dose-dependently increase serum concentrations of total and LDL cholesterol( Reference Jee, He and Appel 43 ). Correa et al.( Reference Corrêa, Rogero and Mioto 44 ) further found that daily consumption of four cups of paper-filtered coffee, from which cafestol and kahweol had been removed, could also have an undesirable effect on plasma cholesterol and inflammation profile in healthy individuals independent of its antioxidant status. Moreover, coffee consumption tended to impair the therapeutic effects of some cardioprotective drugs like statins( Reference Ye, Said and Lin 45 ). Therefore, consuming too much coffee should not be recommended in view of keeping healthy.

The present research suggested that female drinkers were more predisposed to the health-promoting effect of coffee than male drinkers. In subgroup analyses stratified by the degree of adjustment for confounding factors, the gender-specific associations remained unchanged. Results from observational studies suggest that sex hormone status and the activity of cytochrome P450 1A2 enzyme (CYP1A2) could possibly be responsible for the modifying role of gender. As the major biologically active component in coffee, caffeine is predominantly metabolized by CYP1A2 in the liver. Women, because of higher oestradiol concentrations, generally have lower activity of CYP1A2 than men( Reference Gunes and Dahl 46 , Reference Steege, Rupp and Kuhn 47 ). Lowcock et al.( Reference Lowcock, Cotterchio and Anderson 48 ) hypothesized that there could be a greater protective effect of coffee among individuals who metabolize caffeine slowly than among fast metabolizers due to a theoretically extended duration of caffeine exposure. Results from Hallstrom et al.( Reference Hallstrom, Melhus and Glynn 49 ) indicated that excessive consumption of coffee was associated with lower bone mineral density in elderly men, but not in women, and bone mineral density was lower in high consumers of coffee with rapid metabolism of caffeine. A meta-analysis of prospective cohort studies indicated that habitual moderate coffee drinking was also related to a lower risk of CHD in female coffee drinkers rather than male drinkers( Reference Wu, Ho and Zhou 10 ). To date, randomized controlled trials aimed at exploring the coffee–gender interaction are limited. One clinical trial by Gavrieli et al.( Reference Gavrieli, Fragopoulou and Mantzoros 50 ) found that two to four cups of coffee daily induced a lower increase of postprandial insulin concentrations in men than in women. However, results from a randomized controlled trial suggested that the effect of caffeine on blood pressure and heart rate did not vary by sex or hormonal status in healthy individuals( Reference Farag, Whitsett and Mckey 51 ). Yet, the underlying mechanism of the observed variation between men and women in the association of coffee with total mortality remains to be established.

A Scottish study revealed that coffee drinkers were younger, healthier and had higher income than tea drinkers( Reference Woodward and Tunstall-Pedoe 52 ). Nevertheless, the links between coffee drinking habits and health consciousness may vary from person to person and from region to region. An array of prospective cohort studies( Reference Freedman, Park and Abnet 4 Reference Tamakoshi, Lin and Kawado 6 , Reference Lopez-Garcia, van Dam and Li 8 , Reference Andersen, Jacobs and Carlsen 23 ) have reported for both men and women that heavy coffee drinkers were likely to smoke more cigarettes, consume more alcoholic beverages and do less physical exercise. Mukamal et al.( Reference Mukamal, Maclure and Muller 53 ) reported that coffee drinkers tended to be free of co-morbidity but were more likely to be frequent smokers in a population who had survived acute myocardial infraction. In the present study, after pooling studies that controlled for smoking status, alcohol intake and physical activity, the association of coffee consumption and total mortality in either men or women was not significantly altered. However, for individual coffee drinkers, particularly those with less health consciousness, unhealthy dietary and lifestyle factors could offset the beneficial effect of coffee to a large extent.

The present results were not in complete agreement with those from two previous quantitative reviews( Reference Malerba, Turati and Galeone 54 , Reference Je and Giovannucci 55 ) which reported only the favourable association of coffee intake with total mortality and claimed no difference between male and female coffee drinkers. This discordance possibly stems from differences in the included studies and in the methods of data processing. Compared with these two studies, we have set more rigorous selection criteria whereby eligible studies had to include non/occasional coffee drinkers as the reference category and provide adjusted risk estimates. We further incorporated two up-to-date large prospective cohort studies( Reference Tamakoshi, Lin and Kawado 6 , Reference Gardener, Rundek and Wright 25 ). Besides, both of the two previous studies used data from extreme exposure categories only, which may lower the statistical validity and introduce heterogeneity as suggested by Yu et al.( Reference Yu, Schmid and Lichtenstein 56 ). We believe that the present study could provide more comprehensive and profound insights into the association of habitual coffee drinking with total mortality and bring about meaningful implications for future studies concerning the health-related effect of coffee consumption.

Some limitations of the present study should be put forward. First, most included studies did not report the coffee preparation methods and combined caffeinated and decaffeinated coffee together. The composition of a cup of coffee depends largely on the preparing method. We did not conduct subgroup analyses stratified by coffee type since only four of the seventeen studies( Reference Freedman, Park and Abnet 4 , Reference Lopez-Garcia, van Dam and Li 8 , Reference Gardener, Rundek and Wright 25 , Reference Paganini-Hill, Kawas and Corrada 31 ) have reported separately the relationship of different types of coffee (caffeinated and decaffeinated coffee) with all-cause mortality. Muley et al.( Reference Muley, Muley and Shah 57 ) have suggested advantages of filtered coffee over boiled coffee and of decaffeinated coffee over caffeinated coffee in the association with the risk of type 2 diabetes. Decaffeinated coffee was also associated with a small reduction in all-cause mortality, but no substantial difference between caffeinated coffee and decaffeinated coffee in the relationship with total mortality was noted( Reference Freedman, Park and Abnet 4 , Reference Lopez-Garcia, van Dam and Li 8 ). Decaffeinated coffee could be a good option for those who experience uncomfortable effects from caffeine stimulation. One survey in the USA showed that only about 10 % people inclusively consume decaffeinated coffee and its use was related to illness in some people but to a healthy lifestyle in others( Reference Shlonsky, Klatsky and Armstrong 58 ). Besides, filtered coffee has been proved to be less cholesterol-raising than unfiltered coffee( Reference Jee, He and Appel 43 ). Unfortunately, information about the preparing methods were absent in most included studies, thus bringing about great difficulties for us to deal with the heterogeneity across the included studies.

Second, the observed association of coffee drinking with all-cause mortality could be biased by unmeasured or unknown potential confounding factors or alternative explanations, notwithstanding that most of the included studies have controlled for major confounders such as age and smoking status, alcohol intake and physical activity.

Third, most included studies only assessed the frequency of coffee intake at baseline by self-report. People who developed hypertension or other risk factors of CVD during the follow-up period may change their drinking habits. Besides, recall bias and misclassification of coffee consumption also imposed unfavourable impacts on the interpretation of the association between coffee drinking and all-cause mortality.

Conclusion

In summary, it is important to take into account both the advantages and disadvantages of coffee consumption. Although association cannot prove causation, light to moderate coffee drinking may indeed help to decrease the risk of death from all causes, particularly in female drinkers.

Acknowledgements

Financial support: This study was funded by the National Natural Science Foundation of China (NSFC; grant number 81273054). NSFC had no role in the design, analysis or writing of this article. Conflict of interest: None. Authorship: Y.Z. designed the study, conducted the literature search and data extraction, analysed the data and wrote the manuscript; K.W. conducted the literature search and analysed the data; J.Z. analysed the data and critically reviewed the manuscript; R.Z. conducted the data extraction and wrote the manuscript; D.L. designed the study and critically reviewed the manuscript. Ethics of human subject participation: Ethical approval was not needed.

References

1. Van Dam, RM & Hu, FB (2005) Coffee consumption and risk of type 2 diabetes. JAMA 294, 97104.CrossRefGoogle ScholarPubMed
2. Lai, GY, Weinstein, SJ, Albanes, D et al. (2013) The association of coffee intake with liver cancer incidence and chronic liver disease mortality in male smokers. Br J Cancer 109, 13441351.CrossRefGoogle ScholarPubMed
3. Misik, M, Hoelzl, C, Wagner, KH et al. (2010) Impact of paper filtered coffee on oxidative DNA-damage: results of a clinical trial. Mutat Res 692, 4248.CrossRefGoogle ScholarPubMed
4. Freedman, ND, Park, Y, Abnet, CC et al. (2012) Association of coffee drinking with total and cause-specific mortality. N Engl J Med 366, 18911904.CrossRefGoogle ScholarPubMed
5. Liu, J, Sui, X, Lavie, CJ et al. (2013) Association of coffee consumption with all-cause and cardiovascular disease mortality. Mayo Clin Proc 88, 10661074.CrossRefGoogle ScholarPubMed
6. Tamakoshi, A, Lin, Y, Kawado, M et al. (2011) Effect of coffee consumption on all-cause and total cancer mortality: findings from the JACC study. Eur J Epidemiol 26, 285293.CrossRefGoogle ScholarPubMed
7. O’Keefe, JH, Bhatti, SK, Patil, HR et al. (2013) Effects of habitual coffee consumption on cardiometabolic disease, cardiovascular health, and all-cause mortality. J Am Coll Cardiol 62, 10431051.CrossRefGoogle ScholarPubMed
8. Lopez-Garcia, E, van Dam, RM, Li, TY et al. (2008) The relationship of coffee consumption with mortality. Ann Intern Med 148, 904914.CrossRefGoogle ScholarPubMed
9. Kleemola, P, Jousilahti, P, Pietinen, P et al. (2000) Coffee consumption and the risk of coronary heart disease and death. Arch Intern Med 160, 33933400.CrossRefGoogle ScholarPubMed
10. Wu, J, Ho, SC, Zhou, C et al. (2009) Coffee consumption and risk of coronary heart diseases: a meta-analysis of 21 prospective cohort studies. Int J Cardiol 137, 216225.CrossRefGoogle ScholarPubMed
11. Ascherio, A, Zhang, SM, Hernán, MA et al. (2001) Prospective study of caffeine consumption and risk of Parkinson’s disease in men and women. Ann Neurol 50, 5663.CrossRefGoogle ScholarPubMed
12. Higgins, J & Thompson, SG (2002) Quantifying heterogeneity in a meta-analysis. Stat Med 21, 15391558.CrossRefGoogle ScholarPubMed
13. Jackson, D, White, IR & Thompson, SG (2010) Extending DerSimonian and Laird’s methodology to perform multivariate random effects meta-analyses. Stat Med 29, 12821297.CrossRefGoogle ScholarPubMed
14. DerSimonian, R & Laird, N (1986) Meta-analysis in clinical trials. Control Clin Trials 7, 177188.CrossRefGoogle ScholarPubMed
15. Larsson, SC & Orsini, N (2011) Coffee consumption and risk of stroke: a dose–response meta-analysis of prospective studies. Am J Epidemiol 174, 9931001.CrossRefGoogle ScholarPubMed
16. Egger, M, Smith, GD, Schneider, M et al. (1997) Bias in meta-analysis detected by a simple, graphical test. BMJ 315, 629634.CrossRefGoogle ScholarPubMed
17. Begg, CB & Mazumdar, M (1994) Operating characteristics of a rank correlation test for publication bias. Biometrics 315, 10881101.CrossRefGoogle Scholar
18. Harrell, FE (2001) Regression Modeling Strategies: With Applications to Linear Models, Logistic Regression, and Survival Analysis. New York: Springer.CrossRefGoogle Scholar
19. Orsini, N, Li, R, Wolk, A et al. (2012) Meta-analysis for linear and nonlinear dose–response relations: examples, an evaluation of approximations, and software. Am J Epidemiol 175, 6673.CrossRefGoogle ScholarPubMed
20. Greenland, S & Longnecker, MP (1992) Methods for trend estimation from summarized dose–response data, with applications to meta-analysis. Am J Epidemiol 135, 13011309.CrossRefGoogle ScholarPubMed
21. Orsini, N, Bellocco, R & Greenland, S (2006) Generalized least squares for trend estimation of summarized dose–response data. Stata J 6, 4057.CrossRefGoogle Scholar
22. Ahmed, HN, Levitan, EB, Wolk, A et al. (2009) Coffee consumption and risk of heart failure in men: an analysis from the Cohort of Swedish Men. Am Heart J 158, 667672.CrossRefGoogle ScholarPubMed
23. Andersen, LF, Jacobs, DR, Carlsen, MH et al. (2006) Consumption of coffee is associated with reduced risk of death attributed to inflammatory and cardiovascular diseases in the Iowa Women’s Health Study. Am J Clin Nutr 83, 10391046.CrossRefGoogle ScholarPubMed
24. de Koning Gans, JM, Uiterwaal, CS, van der Schouw, YT et al. (2010) Tea and coffee consumption and cardiovascular morbidity and mortality. Arterioscler Thromb Vasc Biol 30, 16651671.CrossRefGoogle ScholarPubMed
25. Gardener, H, Rundek, T, Wright, CB et al. (2013) Coffee and tea consumption are inversely associated with mortality in a multiethnic urban population. J Nutr 143, 12991308.CrossRefGoogle Scholar
26. Iwai, N, Ohshiro, H, Kurozawa, Y et al. (2002) Relationship between coffee and green tea consumption and all-cause mortality in a cohort of a rural Japanese population. J Epidemiol 12, 191198.CrossRefGoogle Scholar
27. Jazbec, A, Simic, D, Corovic, N et al. (2011) Impact of coffee and other selected factors on general mortality and mortality due to cardiovascular disease in Croatia. J Health Popul Nutr 21, 332340.Google Scholar
28. Kahn, HA, Phillips, RL, Snowdon, DA et al. (1984) Association between reported diet and all-cause mortality. Twenty-one-year follow-up on 27 530 adult Seventh-Day Adventists. Am J Epidemiol 119, 775787.CrossRefGoogle Scholar
29. Klatsky, AL, Armstrong, MA & Friedman, GD (1993) Coffee, tea, and mortality. Ann Epidemiol 3, 375381.CrossRefGoogle ScholarPubMed
30. Laaksonen, M, Talala, K, Martelin, T et al. (2008) Health behaviours as explanations for educational level differences in cardiovascular and all-cause mortality: a follow-up of 60 000 men and women over 23 years. Eur J Public Health 18, 3843.CrossRefGoogle Scholar
31. Paganini-Hill, A, Kawas, CH & Corrada, MM (2007) Non-alcoholic beverage and caffeine consumption and mortality: the Leisure World Cohort Study. Prev Med 44, 305310.CrossRefGoogle ScholarPubMed
32. Rosengren, A & Wilhelmsen, L (1991) Coffee, coronary heart disease and mortality in middle-aged Swedish men: findings from the Primary Prevention Study. J Intern Med 230, 6771.CrossRefGoogle ScholarPubMed
33. Sugiyama, K, Kuriyama, S, Akhter, M et al. (2010) Coffee consumption and mortality due to all causes, cardiovascular disease, and cancer in Japanese women. J Nutr 140, 10071013.CrossRefGoogle ScholarPubMed
34. Vandenbroucke, JP, Kok, FJ, van T, BG et al. (1986) Coffee drinking and mortality in a 25-year follow up. Am J Epidemiol 123, 359361.CrossRefGoogle Scholar
35. Natella, F, Nardini, M, Belelli, F et al. (2007) Coffee drinking induces incorporation of phenolic acids into LDL and increases the resistance of LDL to ex vivo oxidation in humans. Am J Clin Nutr 86, 604609.CrossRefGoogle ScholarPubMed
36. Mostofsky, E, Rice, MS, Levitan, EB et al. (2012) Habitual coffee consumption and risk of heart failure: a dose–response meta-analysis. Circ Heart Fail 5, 401405.CrossRefGoogle ScholarPubMed
37. Hernan, MA, Takkouche, B, Caamano Isorna, F et al. (2002) A meta-analysis of coffee drinking, cigarette smoking, and the risk of Parkinson’s disease. Ann Neurol 52, 276284.CrossRefGoogle ScholarPubMed
38. Yu, X, Bao, Z, Zou, J et al. (2011) Coffee consumption and risk of cancers: a meta-analysis of cohort studies. BMC Cancer 11, 96102.CrossRefGoogle ScholarPubMed
39. Urgert, R, van Vliet, T, Zock, PL et al. (2000) Heavy coffee consumption and plasma homocysteine: a randomized controlled trial in healthy volunteers. Am J Clin Nutr 72, 11071110.CrossRefGoogle ScholarPubMed
40. Grubben, MJ, Boers, GH, Blom, HJ et al. (2000) Unfiltered coffee increases plasma homocysteine concentrations in healthy volunteers: a randomized trial. Am J Clin Nutr 71, 480484.CrossRefGoogle ScholarPubMed
41. Olthof, MR, Hollman, PC, Zock, PL et al. (2001) Consumption of high doses of chlorogenic acid, present in coffee, or of black tea increases plasma total homocysteine concentrations in humans. Am J Clin Nutr 73, 532538.CrossRefGoogle ScholarPubMed
42. Mesas, AE, Leon-Munoz, LM, Rodriguez-Artalejo, F et al. (2011) The effect of coffee on blood pressure and cardiovascular disease in hypertensive individuals: a systematic review and meta-analysis. Am J Clin Nutr 94, 11131126.CrossRefGoogle ScholarPubMed
43. Jee, SH, He, J, Appel, LJ et al. (2001) Coffee consumption and serum lipids: a meta-analysis of randomized controlled clinical trials. Am J Epidemiol 153, 353362.CrossRefGoogle ScholarPubMed
44. Corrêa, TA, Rogero, MM, Mioto, BM et al. (2013) Paper-filtered coffee increases cholesterol and inflammation biomarkers independent of roasting degree: a clinical trial. Nutrition 29, 977981.CrossRefGoogle ScholarPubMed
45. Ye, Y, Said, GHA, Lin, Y et al. (2008) Caffeinated coffee blunts the myocardial protective effects of statins against ischemia-reperfusion injury in the rat. Cardiovasc Drugs Ther 22, 275282.CrossRefGoogle ScholarPubMed
46. Gunes, A & Dahl, M (2008) Variation in CYP1A2 activity and its clinical implications: influence of environmental factors and genetic polymorphisms. Pharmacogenomics 9, 625637.CrossRefGoogle ScholarPubMed
47. Steege, JF, Rupp, SL & Kuhn, CM (1992) Menstrual cycle effects on caffeine elimination in the human female. Eur J Clin Pharmacol 43, 543546.Google Scholar
48. Lowcock, EC, Cotterchio, M, Anderson, LN et al. (2013) High coffee intake, but not caffeine, is associated with reduced estrogen receptor negative and postmenopausal breast cancer risk with no effect modification by CYP1A2 genotype. Nutr Cancer 65, 398409.CrossRefGoogle Scholar
49. Hallstrom, H, Melhus, H, Glynn, A et al. (2010) Coffee consumption and CYP1A2 genotype in relation to bone mineral density of the proximal femur in elderly men and women: a cohort study. Nutr Metab (Lond) 7, 1220.CrossRefGoogle ScholarPubMed
50. Gavrieli, A, Fragopoulou, E, Mantzoros, CS et al. (2013) Gender and body mass index modify the effect of increasing amounts of caffeinated coffee on postprandial glucose and insulin concentrations; a randomized, controlled, clinical trial. Metabolism 8, 10991106.CrossRefGoogle Scholar
51. Farag, NH, Whitsett, TL, Mckey, BS et al. (2010) Caffeine and blood pressure response: sex, age, and hormonal status. J Womens Health (Larchmt) 19, 11711176.CrossRefGoogle ScholarPubMed
52. Woodward, M & Tunstall-Pedoe, H (1999) Coffee and tea consumption in the Scottish Heart Health Study follow up: conflicting relations with coronary risk factors, coronary disease, and all cause mortality. J Epidemiol Community Health 53, 481487.CrossRefGoogle ScholarPubMed
53. Mukamal, KJ, Maclure, M, Muller, JE et al. (2004) Caffeinated coffee consumption and mortality after acute myocardial infarction. Am Heart J 147, 9991004.CrossRefGoogle ScholarPubMed
54. Malerba, S, Turati, F, Galeone, C et al. (2013) A meta-analysis of prospective studies of coffee consumption and mortality for all causes, cancers and cardiovascular diseases. Eur J Epidemiol 28, 527539.CrossRefGoogle ScholarPubMed
55. Je, Y & Giovannucci, E (2013) Coffee consumption and total mortality: a meta-analysis of twenty prospective cohort studies. Br J Nutr 111, 11621173.CrossRefGoogle ScholarPubMed
56. Yu, WW, Schmid, CH, Lichtenstein, AH et al. (2013) Empirical evaluation of meta-analytic approaches for nutrient and health outcome dose–response data. Res Synth Methods 3, 256268.CrossRefGoogle Scholar
57. Muley, A, Muley, P & Shah, M (2012) Coffee to reduce risk of type 2 diabetes? A systematic review. Curr Diabetes Rev 8, 162168.CrossRefGoogle ScholarPubMed
58. Shlonsky, AK, Klatsky, AL & Armstrong, MA (2003) Traits of persons who drink decaffeinated coffee. Ann Epidemiol 13, 273279.CrossRefGoogle Scholar
Figure 0

Fig. 1 Flowchart of the literature search

Figure 1

Table 1 Characteristics of studies included in the present meta-analysis of coffee drinking and all-cause mortality

Figure 2

Fig. 2 The non-linear association of habitual coffee drinking with all-cause mortality: (a) all studies (P for non-linearity <0·001); (b) cohorts of men only (P for non-linearity <0·001); (c) cohorts of women only (P for non-linearity <0·001). ———, relative risk (RR); — — —, 95 % confidence intervals; – – –, null effect. The results were gained from the two-stage random-effects dose–response meta-analyses

Figure 3

Fig. 3 Subgroup analyses of the association between habitual coffee drinking and all-cause mortality, showing pooled relative risks (RR; ■) and 95 % confidence intervals (represented by horizontal bars) from categorical meta-analyses using a random-effects model. Model A, studies providing risk estimates and 95 % CI adjusted for age, smoking status, alcohol intake and physical activity; model B, studies providing risk estimates and 95 % CI additionally adjusted for education level

Supplementary material: File

Zhao supplementary material

Figures S1-S4

Download Zhao supplementary material(File)
File 422.4 KB