Académique Documents
Professionnel Documents
Culture Documents
https://doi.org/10.1038/s41371-017-0007-0
REVIEW ARTICLE
Abstract
Some debates exist regarding the association of coffee consumption with hypertension risk. We performed a meta-analysis
including dose–response analysis aimed to derive a more quantitatively precise estimation of this association. PubMed and
Embase were searched for cohort studies published up to 18 July 2017. Fixed-effects generalized least-squares regression
models were used to assess the quantitative association between coffee consumption and hypertension risk across studies.
Restricted cubic spline was used to model the dose–response association. We identified eight articles (10 studies)
investigating the risk of hypertension with the level of coffee consumption, including 243,869 individuals and 58,094
1234567890
incident cases of hypertension. We found no evidence of a nonlinear dose–response association of coffee consumption and
hypertension (Pnonlinearity = 0.243). The risk of hypertension was reduced by 2% (relative risk (RR) = 0.98, 95% confidence
interval (CI) 0.98–0.99) with each one cup/day increment of coffee consumption. With the linear cubic spline model, the
RRs of hypertension risk were 0.97 (95% CI 0.95–0.99), 0.95 (95% CI 0.91–0.99), 0.92 (95% CI 0.87–0.98), and 0.90 (95%
CI 0.83–0.97) for 2, 4, 6, and 8 cups/day, respectively, compared with individuals with no coffee intakes. This meta-analysis
provides quantitative evidence that consumption of coffee was inversely associated with the risk of hypertension in a
dose–response manner.
Materials and methods total coffee consumption analysis. If the numbers of cases in
each category were unavailable, these data were inferred on
Literature search strategy the basis of the number of total cases, level-specific exposed
participant numbers, and the reported effect size. If the
We searched the electronic databases PubMed and Embase exposed participant numbers were not reported in each
for all articles of cohort studies investigating the association category, group sizes were assumed to be approximately
between coffee consumption and hypertension that were equal [21]. Cohort study quality was assessed by the
published up to 18 July 2017. Our search included both Newcastle–Ottawa Scale, which allows for a total score of
free-text and Medical Subject Heading (MeSH) terms, such up to nine points summarizing eight aspects of each study
as “coffee [MeSH]”, “caffeine [MeSH]”, “coffee”, “caffein*”, the scale gives a maximum of 9 stars [22]. All retrieved
“hypertension [MeSH]”, “cardiovascular diseases [MeSH]”, articles were screened by two independent reviewers (C.X.
“high blood pressure*”, OR “cardiovascular disease”, and L.C.). Data extraction and quality assessment were
“hypertens*”, “HBP”, and “CVD”. Details of the wide search done by C.X. and checked by L.C. Any disagreements were
terms are shown in Supplementary Table 1. The reference discussed until agreement was reached.
lists of all retrieved articles [9–16] and previous systematic
reviews [17–19] were manually searched for additional Statistical methods
relevant studies. We restricted the search to studies on
humans and written in English. The meta-analysis was To analyze the trend of coffee consumption and risk of
conducted and reported in accordance with the Meta- hypertension, we used both semiparametric and parametric
analysis of Observational Studies in Epidemiology guide- methods. For the semiparametric method, four coffee con-
lines [20]. sumption groups were generated: lowest, third highest,
second highest, and highest. For each included study, the
Study selection lowest and the highest coffee consumption categories cor-
responded to the lowest and highest groups, respectively.
Studies included in this meta-analysis met the following For studies with four exposure categories, the second and
criteria: (1) the study design was cohort; (2) the articles third categories corresponded to the second and third
were published in English; (3) reported relative risks (RRs), highest groups, respectively. For studies with three expo-
odds ratios (ORs), or hazard ratios (HRs) with 95% con- sure categories, the middle category corresponded to either
fidence intervals (CIs) or data to calculate them; (4) the the second or the third highest group, whichever median
exposure was coffee consumption; (5) the outcome was risk coffee consumption amount was most similar. If the study
of hypertension; and (6) provide the frequency and amount had more than four exposure categories, two consumption
of coffee consumption, number of cases, exposed person- groups, other than the lowest and highest, were chosen on
years, or participant numbers for the dose–response analy- the basis of their similarity of the amount of coffee con-
sis. Abstract of only publications, comment, or conference sumption in that category to the second and third highest
articles were excluded. A search for unpublished literature groups. For studies reporting HRs or ORs for hypertension,
was not performed. we assumed that the HRs and ORs were approximately RRs
[23]. A fixed-effects model [24] was used to calculate the
Data extraction and quality assessment summary RR estimates.
For the parametric method, generalized least-squares
Data were extracted by using a pre-designed extraction (GLS) regression was used to estimate study-specific
form. The following information was collected: the first dose–response association. GLS regression model esti-
author; publication year; country where the study was mates the linear dose–response coefficient, taking into
conducted; study name; study design; follow-up period; account the covariance for each exposure category within
number of incident MetS cases and total population; mean each study, because they are estimated relative to a common
or median age of the study population at baseline; coffee referent PA exposure category [25, 26]. A fixed-effects
consumption ranges or median or mean categories at model was used to pool the study-specific dose–response
baseline; RRs, HRs, ORs, and 95% CIs for the association; RR estimates [24]. First, a linear association was assumed;
definition for hypertension; and covariates on which the study-specific RR estimates were calculated per one cup/
analyses were adjusted. When the required data were not day of coffee consumption increment and then pooled. In
reported in the original articles, we emailed authors to addition, we examined possible nonlinear associations by
obtain further details. If a study only reported caffeinated modeling coffee consumption using a restricted cubic spline
coffee consumption instead of total coffee consumption, with three knots located at the 25th, 50th, and 75th per-
caffeinated coffee consumption was also included in the centiles of the distribution [27].
Fig. 1 Flowchart of study
selection for the meta-analysis
To perform the dose–response meta-analysis, we assigned Heterogeneity was tested by Cochran Q and I2 statistics
the median or mean coffee consumption in each category of [29]. For the Q statistic, P < 0.1 was considered statistically
consumption to the corresponding RR for each study. If the significant; and for the I2 statistic, I2 values of ~25%, 50%,
mean or median consumption per category was not reported, and 75% are considered to reflect low, moderate, and high
the midpoint of the upper and lower boundaries in each heterogeneity, respectively. Subgroup analyses were con-
category was assigned as the mean consumption. If the ducted to investigate potential sources of heterogeneity,
highest or lowest category was open-ended, the width of the including study design, gender, age, region, diagnostic
interval was assumed to be the same as in the closest category criteria number of cases, and the covariates adjusted in the
[28]. Only studies reporting risk estimates for at least three analysis (age, smoking, alcohol consumption, physical
coffee consumption exposure levels for risk of hypertension activity, family history of hypertension, education, intake of
were included in this analysis. The P value for nonlinearity sodium, and body mass index). We performed a sensitivity
was calculated by testing the null hypothesis that the coeffi- analysis by excluding one study at a time to assess the
cient of the second spline is equal to zero. stability of results and potential sources of heterogeneity.
Table 1 Basic characteristics of cohort studies investigating the association of coffee consumption and incident hypertension
First author Region Gender (M/W) Follow-up period (years) Age (mean) BMI (kg/m2) Participants (case) Coffee category (cup/day) OR/RR/HR
Chei et al. [16] Singapore M/W 9.5 54.9 22.72 38,592 (13 658) 1 1
2 0.98 (0.94, 1.02)
3.5 0.93 (0.86, 1.00)
Rhee et al. [15] America W 3 62.5 25.97 29,985 (5566) 0 1
1 1.00 (0.93, 1.08)
2–3 0.95 (0.89, 1.02)
≥4 0.99 (0.90, 1.08)
Grosso et al. [14] Poland M/W 5 56.2/55.2 26.6/26.3 2725 (1735) <1 1
1–2 0.86 (0.68, 1.07)
3–4 0.75 (0.58, 0.95)
>4 1.58 (0.85, 3.64)
Uiterwaal et al. [11] Netherlands M 11 40.7 25.34 2297 (493) 0–3 1
3–6 1.08 (0.79, 1.47)
>6 1.03 (0.72, 1.46)
Uiterwaal et al. [11] Netherlands W 11 40.1 24.63 2892 (463) 0–3 1
3–6 0.83 (0.64, 1.07)
>6 0.67 (0.46, 0.98)
Hu et al. [13] Finland M/W 13.2 43.56 25.84 24,710 (2505) 0–1 1
2–3 1.27 (1.07, 1.51)
4–5 1.20 (1.01, 1.42)
6–7 1.21 (1.01, 1.44)
≥8 1.13 (0.94, 1.36)
Palatini et al. [12, 36] Italy M/W 6.4 33.3 (8.4) 25.4 (3.4) 1107 (561) 0 1
1–3 1.27 (1.04–1.56)
≥4 1.24 (0.94–1.66)
Winkelmayer et al. [10] America W 12 55.4 24.8 53,175 (19,364) <1 1
1 1.06 (1.01–1.10)
2–3 1.00 (0.97–1.04)
4–5 0.93 (0.88–0.99)
≥6 0.88 (0.80–0.98)
Winkelmayer et al. (2002) America W 12 36 24.3 87,369 (13 468) <1 1
1 1.06 (1.01–1.13)
2–3 1.00 (0.95–1.04)
4–5 0.91 (0.84–0.98)
C. Xie et al.
Potential publication bias was evaluated by the Egger’s test
0.91 (0.80–1.04)
1.24 (0.87–1.77)
1.49 (1.01–2.20)
1.07 (0.67–1.69)
OR/RR/HR [30] and results were considered to indicate publication bias
at P < 0.10. All analyses involved use of Stata 12.1 (Stata
1
Corp, College Station, TX, USA).
Coffee category (cup/day)
Results
Characteristics of studies
1–2
3–4
≥6
≥5
0
26.3 (2.4)
risk of hypertension
Klag et al. [9]
First author
Fig. 2 Forest plot of relative risks (RRs) and 95% CIs for the asso- cups/day), the pooled RR for incident hypertension was 0.95 (95% CI
ciation between coffee consumption and hypertension risk in cohort 0.91–0.99, I2 = 44.0%) for the highest category of consumption, 0.96
studies. Compared with the lowest category (median consumption 0 (0.89–1.03, I2 = 61.8%) for the second highest, and 1.02 (0.97–1.06,
I2 = 58.9%) for the third highest category of coffee consumption
Fig. 3 Risk of incident
hypertension for each 1cup/day
increment in coffee
consumption. RR relative risk,
CI confidence interval, M men,
W women
(RR = 0.98, 95% CI 0.98–0.99) with each one cup/day history of hypertension, education, intake of sodium, and
increment of consumption (Fig. 3). BMI) adjusted in the analysis (Table 2). In general, the
We found no evidence of nonlinear association between association was consistent in most analyses. The hetero-
coffee consumption and hypertension risk (Pnonlinearity = geneity appeared to be lower in Americans, age >50 years
0.243); therefore, restricted cubic spline was adopted to old and number of case ≥1000 populations, with I2 reduced
model the linear dose–response association. We found a to 0.0%, 12.7%, and 28.3%, respectively. No significant
linear negative correlation between coffee consumption and changes of heterogeneity occurred in other subgroup
risk of hypertension (Fig. 4). Compared to those with no analyses.
coffee consumption, the RR estimated directly from the When performing sensitivity analyses by removing one
cubic spline model was 0.97 (95% CI 0.95–0.99) for two study at a time, none of the individual studies changed the
cups/day, 0.95 (95% CI 0.91–0.99) for four cups/day, 0.92 pooled risk substantially. No publication bias was detected
(95% CI 0.87–0.98) for six cups/day, and 0.90 (95% CI by Egger’s test (P = 0.618).
0.83–0.97) for eight cups/day.
To explore the sources of heterogeneity, we performed The findings from this systematic review and meta-analysis,
subgroup analyses by study design, gender, age, region, based on 247,659 participants and 54,639 incident cases of
diagnostic criteria, number of cases, the covariates (age, hypertension, demonstrate an inverse association between
smoking, alcohol consumption, physical activity, family risk of hypertension and coffee consumption, with a
reduction of 2% per one cup/day increment of coffee con-
sumption. With the linear cubic spline model, the RRs of
hypertension risk were 0.97 (95% CI 0.95–0.99), 0.95 (95%
CI 0.91–0.99), 0.92 (95% CI 0.87–0.98), and 0.90 (95% CI
0.83–0.97) for 2, 4, 6, and 8 cups/day, respectively, com-
pared with individuals with no coffee intakes.
Controversy exists in the current studies investigating the
effects of coffee intake on hypertension risk. In a meta-
analysis of five cohort studies, Zhang et al. found an inverse
J-shaped association between caffeinated coffee intake and
hypertension risk, with the risk increasing with up to three
cups coffee/day compared with less than one cup/day and
then decreasing at higher intakes [18]. On the basis of the
Fig. 4 Linear dose–response association between coffee consumption same original studies, Steffen et al. reported opposite find-
and hypertension modeled by using restricted cubic spline ings, suggesting no statistically significant effect of coffee
C. Xie et al.
● Given the different definitions of coffee exposure among 12. Palatini P, Dorigatti F, Santonastaso M, Cozzio S, Biasion T,
studies, the association between coffee consumption and Garavelli G, et al. Association between coffee consumption and
risk of hypertension. Ann Med. 2007;39:545–53.
risk of hypertension could not be analyzed precisely.
13. Hu G, Jousilahti P, Nissinen A, Bidel S, Antikainen R, Tuomi-
lehto J. Coffee consumption and the incidence of antihypertensive
What this study adds? drug treatment in Finnish men and women. Am J Clin Nutr.
2007;86:457–64.
● The risk of hypertension was reduced by 2% (RR=0.98, 14. Grosso G, Stepaniak U, Polak M, Micek A, Topor-Madry R,
95% CI: 0.98-0.99) with each 1 cup/day increment of Stefler D, et al. Coffee consumption and risk of hypertension in
coffee consumption. the Polish arm of the HAPIEE cohort study. Eur J Clin Nutr.
2016;70:109–15.
● Compared with individuals with no coffee intakes, the 15. Rhee JJ, Qin F, Hedlin HK, Chang TI, Bird CE, Zaslavsky O,
RRs of hypertension risk were 0.97 (95% CI: 0.95-0.99), et al. Coffee and caffeine consumption and the risk of hyperten-
0.95 (95% CI: 0.91-0.99), 0.92 (95% CI: 0.87-0.98), and sion in postmenopausal women. Am J Clin Nutr.
0.90 (95% CI: 0.83-0.97) for 2, 4, 6 and 8 cups/day, 2016;103:210–07.
16. Chei CL, Loh JK, Soh A, Yuan JM, Koh WP. Coffee, tea, caf-
respectively. feine, and risk of hypertension: the Singapore Chinese health
study. Eur J Nutr. 2017; e-pub ahead of print 1 March 2017;
https://doi.org/10.1007/s00394-017-1412-4.
Acknowledgments This research was supported by the National
17. Steffen M, Kuhle C, Hensrud D, Erwin PJ, Murad MH. The effect
Social Science Foundation of China (Grant number 15BSH043). This
of coffee consumption on blood pressure and the development of
research was supported by the National Social Science Foundation of
hypertension: a systematic review and meta-analysis. J Hypertens.
China (Grant number 15BSH043). The National Social Science
2012;30:2245–54.
Foundation of China had no role in the design/conduct of the study,
18. Zhang Z, Hu G, Caballero B, Appel L, Chen L. Habitual coffee
collection/analysis interpretation of the data, and preparation/review
consumption and risk of hypertension: a systematic review and
approval of the manuscript.
meta-analysis of prospective observational studies. Am J Clin
Nutr. 2011;93:1212–9.
Compliance with Ethical Standards 19. Geleijnse JM. Habitual coffee consumption and blood pressure: an
epidemiological perspective. Vasc Health Risk Manage.
Conflict of interest The authors declare that they have no competing 2008;4:963–70.
financial interest. 20. Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD,
Rennie D, et al. Meta-analysis of observational studies in epide-
miology: a proposal for reporting. Meta-analysis of observational
References studies in epidemiology (moose) group. JAMA.
2000;283:2008–12.
1. World Health Organization. A Global Brief on Hypertension: 21. Bekkering GE, Harris RJ, Thomas S, Mayer AMB, Beynon R,
Silent Killer, Global Public Health Crisis: World Health Day Ness AR, et al. How much of the data published in observational
2013; 2013. studies of the association between diet and prostate or bladder
2. Cao X. A call for global research on non-communicable diseases. cancer is usable for meta-analysis. Am J Epidemiol.
Lancet. 2015;385:e5–6. 2008;167:1017–26.
3. Benjamin EJ, Blaha MJ, Chiuve SE, Cushman M, Das SR, Deo R, 22. Wells GA, Shea BJ, O’Connell D, Peterson J, Welch V, Losos M,
et al. Heart disease and stroke statistics-2017 update: a report from et al. The Newcastle–Ottawa Scale (NOS) for assessing the quality
the American heart association. Circulation. 2017;135:e146–603. of non-randomized studies in meta-analysis. Appl Eng Agric.
4. Murray CJ, Lopez AD. Measuring the global burden of disease. N 2014;18:727–34.
Engl J Med. 2013;369:448–57. 23. Orsini N, Li R, Wolk A, Khudyakov P, Spiegelman D. Meta-
5. Kearney PM, Whelton M, Reynolds K, Muntner P, Whelton PK, analysis for linear and nonlinear dose-response relations: exam-
He J. Global burden of hypertension: analysis of worldwide data. ples, an evaluation of approximations, and software. Am J Epi-
Lancet. 2005;365:217–23. demiol. 2012;175:66–73.
6. Butt MS, Sultan MT. Coffee and its consumption: benefits and 24. Mantel N, Haenszel W. Statistical aspects of the analysis of data
risks. Crit Rev Food Sci Nutr. 2011;51:363–73. from retrospective studies of disease. J Natl Cancer Inst.
7. Horst K, Robinson WD, Jenkins WL, Bao DL. The effect of 1959;22:719–48.
caffeine, coffee and decaffeinated coffee upon blood pressure, 25. Bagnardi V, Zambon A, Quatto P, Corrao G. Flexible meta-
pulse rate and certain motor reactions of normal young men. J regression functions for modeling aggregate dose-response data,
Pharmacol Exp Ther. 1935;53:307–21. with an application to alcohol and mortality. Am J Epidemiol.
8. Simon NM. Coffee and hypertension. N Engl J Med. 2004;159:1077–86.
1978;298:1092. 26. Orsini N, Bellocco R, Greenland S. Generalized least squares for
9. Klag MJ, Wang NY, Meoni LA, Brancati FL, Cooper LA, Liang trend estimation of summarized dose–response data. Stata J.
KY, et al. Coffee intake and risk of hypertension: the 2006;6:40–57.
Johns Hopkins precursors study. Arch Intern Med. 27. Greenland S. Dose-response and trend analysis in epidemiology:
2002;162:657–62. alternatives to categorical analysis. Epidemiology.
10. Winkelmayer WC, Stampfer MJ, Willett WC, Curhan GC. 1995;6:356–65.
Habitual caffeine intake and the risk of hypertension in women. 28. Hartemink N, Boshuizen HC, Nagelkerke NJD, Jacobs MAM,
JAMA. 2005;294:2330–5. Houwelingen HCV. Combining risk estimates from observational
11. Uiterwaal CS, Verschuren WM, Bueno-de-Mesquita HB, Ocke M, studies with different exposure cutpoints: a meta-analysis on body
Geleijnse JM, Boshuizen HC, et al. Coffee intake and incidence of mass index and diabetes Type 2. Am J Epidemiol.
hypertension. Am J Clin Nutr. 2007;85:718–23. 2006;163:1042–52.
29. Higgins JPT, Thompson SG, Deeks JJ, Altman DG. Measuring hypertension. HARVEST Study Group (Italy). Hypertension
inconsistency in meta-analyses. Br Med J. 2003;327:557–60. Ambulatory Recording VEnetia STudy. J Hypertens.
30. Begg CB, Mazumdar M. Operating characteristics of a rank cor- 1995;13:965–70.
relation test for publication bias. Biometrics. 1994;50:1088–101. 38. Cornelis MC, El-Sohemy A, Kabagambe EK, Campos H. Coffee,
31. Godos J, Pluchinotta FR, Marventano S, Buscemi S, Li VG, CYP1A2 genotype,and risk of myocardial infarction. JAMA.
Galvano F, et al. Coffee components and cardiovascular risk: 2006;295:1135–41.
beneficial and detrimental effects. Int J Food Sci Nutr. 39. Platt DE, Ghassibesabbagh M, Salameh P, Salloum AK, Haber M,
2014;65:1–12. Mouzaya F, et al. Caffeine impact on metabolic syndrome com-
32. Higdon JV, Frei B. Coffee and health: a review of recent human ponents is modulated by a CYP1A2 Variant. Ann Nutr Metab.
research. Crit Rev Food Sci Nutr. 2006;46:101–23. 2015;68:1–11.
33. Ding M, Bhupathiraju SN, Chen M, van Dam RM, Hu FB. Caf- 40. Palatini P, Ceolotto G, Ragazzo F, Dorigatti F, Saladini F, Pap-
feinated and decaffeinated coffee consumption and risk of type 2 parella I, et al. CYP1A2 genotype modifies the association
diabetes: a systematic review and a dose-response meta-analysis. between coffee intake and the risk of hypertension. J Hypertens.
Diabetes Care. 2014;37:569–86. 2009;27:1594–601.
34. Kang YS. Obesity associated hypertension: new insights into 41. Ferraroni M, Tavani A, Decarli A, Franceschi S, Parpinel M,
mechanism. Electrolyte Blood Press. 2013;11:46–52. Negri E, et al. Reproducibility and validity of coffee and tea
35. Guessous I, Eap CB, Bochud M. Blood pressure in relation to consumption in Italy. Eur J Clin Nutr. 2004;58:674–80.
coffee and caffeine consumption. Curr Hypertens Rep. 42. Salvini S, Hunter DJ, Sampson L, Stampfer MJ, Colditz GA,
2014;16:468. Rosner B, et al. Food-based validation of a dietary questionnaire:
36. Palatini P, Dorigatti F, Santonastaso M, Cozzio S, Biasion T, The effects of week-to-week variation in food consumption. Int J
Garavelli G, et al. Association between coffee consumption and Epidemiol. 1989;18:858–67.
risk of hypertension. Ann Med. 2007;39:545–53. 43. Noordzij M, Uiterwaal CS, Arends LR, Kok FJ, Grobbee DE,
37. Narkiewicz K, Maraglino G, Biasion T, Rossi G, Sanzuol F, Geleijnse JM. Blood pressure response to chronic intake of coffee
Palatini P. Interactive effect of cigarettes and coffee on daytime and caffeine: a meta-analysis of randomized controlled trials. J
systolic blood pressure in patients with mild essential Hypertens. 2005;23:921–8.