Association of Coffee Drinking With Mortality by Genetic Variation in Caffeine Metabolism: Findings From the UK Biobank

Key Points

Question  Moderate coffee consumption has been inversely associated with mortality; however, does heavy intake, particularly among those with common genetic polymorphisms that impair caffeine metabolism, increase risk of mortality?

Findings  This large prospective cohort study of a half million people found inverse associations for coffee drinking with mortality, including among participants drinking 1 up to 8 or more cups per day. No differences were observed in analyses that were stratified by genetic polymorphisms affecting caffeine metabolism.

Meaning  This study provides further evidence that coffee drinking can be part of a healthy diet and offers reassurance to coffee drinkers.

Importance  Prospective cohorts in North America, Europe, and Asia show consistent inverse associations between coffee drinking and mortality, including deaths from cardiovascular disease and some cancers. However, concerns about coffee, particularly among people with common genetic polymorphisms affecting caffeine metabolism and among those drinking more than 5 cups per day, remain.

Objective  To evaluate associations of coffee drinking with mortality by genetic caffeine metabolism score.

Design, Setting, and Participants  The UK Biobank is a population-based study that invited approximately 9.2 million individuals from across the United Kingdom to participate. We used baseline demographic, lifestyle, and genetic data form the UK Biobank cohort, with follow-up beginning in 2006 and ending in 2016, to estimate hazard ratios (HRs) for coffee intake and mortality, using multivariable-adjusted Cox proportional hazards models. We investigated potential effect modification by caffeine metabolism, defined by a genetic score of previously identified polymorphisms in AHR, CYP1A2, CYP2A6, and POR that have an effect on caffeine metabolism. Of the 502 641 participants who consented with baseline data, we included those who were not pregnant and had complete data on coffee intake and smoking status (n = 498 134).

Exposures  Total, ground, instant, and decaffeinated coffee intake.

Main Outcomes and Measures  All-cause and cause-specific mortality.

Results  The mean age of the participants was 57 years (range, 38-73 years); 271 019 (54%) were female, and 387 494 (78%) were coffee drinkers. Over 10 years of follow-up, 14 225 deaths occurred. Coffee drinking was inversely associated with all-cause mortality. Using non–coffee drinkers as the reference group, HRs for drinking less than 1, 1, 2 to 3, 4 to 5, 6 to 7, and 8 or more cups per day were 0.94 (95% CI, 0.88-1.01), 0.92 (95% CI, 0.87-0.97), 0.88 (95% CI, 0.84-0.93), 0.88 (95% CI, 0.83-0.93), 0.84 (95% CI, 0.77-0.92), and 0.86 (95% CI, 0.77-0.95), respectively. Similar associations were observed for instant, ground, and decaffeinated coffee, across common causes of death, and regardless of genetic caffeine metabolism score. For example, the HRs for 6 or more cups per day ranged from 0.70 (95% CI, 0.53-0.94) to 0.92 (95% CI, 0.78-1.10), with no evidence of effect modification across strata of caffeine metabolism score (P = .17 for heterogeneity).

Conclusions and Relevance  Coffee drinking was inversely associated with mortality, including among those drinking 8 or more cups per day and those with genetic polymorphisms indicating slower or faster caffeine metabolism. These findings suggest the importance of noncaffeine constituents in the coffee-mortality association and provide further reassurance that coffee drinking can be a part of a healthy diet.

Coffee is one of the most commonly consumed beverages worldwide. Epidemiologic studies have generally reported inverse associations with chronic disease, including cardiovascular disease^1,2; diabetes³; Parkinson disease⁴; and liver,⁵ colorectal,⁶ and endometrial⁷ cancer. Furthermore, prospective studies have consistently observed inverse associations with all-cause and cause-specific mortality.^8-16 Such evidence played a major role in the 2015 US Dietary Guidelines Advisory Committee report, which concluded that moderate coffee consumption of up to 5 eight-ounce cups per day can be a part of a healthy diet.¹⁷

Nevertheless, there remain concerns about the effects of coffee drinking. Coffee is a major source of caffeine, and between-person variation in caffeine metabolism is considerable. Clinical and epidemiologic studies have identified genetic variants that affect caffeine metabolism and coffee consumption,¹⁸ and several studies have suggested that coffee drinkers with common genetic polymorphisms affecting caffeine metabolism may be at increased risk of cardiovascular disease.^19,20 However, the impact of common genetic polymorphisms that affect caffeine metabolism on the coffee-mortality association is unknown. There is also concern about heavy coffee drinking (≥6 cups/d) more generally. However, prior studies have had a limited number of participants in this intake range, precluding stable risk estimates.

We used the 500 000-participant UK Biobank cohort to investigate possible effect modification by caffeine metabolism score, defined by previously identified polymorphisms that affect caffeine metabolism.¹⁸

The UK Biobank study design has been described elsewhere.^21,22 In brief, the study mailed invitations to approximately 9.2 million individuals in the United Kingdom’s National Health Service (NHS), ages 40 to 69 years who resided within 40 km of 22 assessment centers across the United Kingdom.²¹ In total, 503 317 individuals visited assessment centers between 2006 and 2010²²; answered comprehensive touchscreen questionnaires, including about diet; received physical examinations; and provided biological samples. Data from 502 641 participants were available for our study. We excluded participants with incomplete data on coffee intake (n = 2269) or smoking status (n = 1866), women who were pregnant or unsure whether they were pregnant (n = 369), and those with no follow-up time (n = 3), resulting in an analytic cohort of 498 134 participants.

The UK Biobank study was approved by the National Information Governance Board for Health and Social Care and the NHS North West Multicenter Research Ethics Committee. Written consent was obtained from all participants. According to the UK Biobank information leaflet for participants, participation is “entirely voluntary” and participants are “free to withdraw at any time later.”^21,22

Follow-up time was counted from the date of assessment center visit until the date of death or the date of censor (ie, January 31, 2016, for England and Wales and November 30, 2015, for Scotland), whichever came first. For cause-specific mortality analyses, individuals who died from other causes were censored at their date of death.

Vital status, date, and underlying (primary) cause of death were provided by the National Health Service (NHS) Information Centre for participants from England and Wales and by the NHS Central Register, Scotland, for those from Scotland. We defined the following 3 broad categories (ie, causes with >500 deaths) of cause-specific mortality using the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10), codes: cancer (C00-D48), cardiovascular disease (I00-I79), and respiratory diseases (J09-J18 and J40-J47). In addition, for cancer and cardiovascular disease, we further defined the common causes of death (ie, causes with >500 deaths) within these broad categories: colorectal cancer (C18-C20), bronchus and lung cancer (C34), female breast cancer (C50), and pancreatic cancer (C25); ischemic heart diseases (I20-I25); and stroke (I60-I69).

Participants were asked “How many cups of coffee do you drink each day (including decaffeinated coffee)?” Participants selected either the number of cups, “Less than 1,” “Do not know,” or “Prefer not to answer.” If participants reported drinking more than 10 cups, they were asked to confirm the response. For the main analysis of coffee and all-cause mortality, we defined coffee intake as follows: 0, less than 1, 1, 2 to 3, 4 to 5, 6 to 7, and 8 or more cups per day. The top 2 categories were collapsed for secondary analyses, to preserve sample size. Coffee drinkers were also asked “What type of coffee do you usually drink?” and were able to select 1 of 6 mutually exclusive responses including “decaffeinated coffee (any type),” “instant coffee,” or “ground coffee (include espresso, filter, etc).”

The baseline questionnaire also assessed potential confounding factors, including smoking, alcohol, tea, race, education, physical activity, and body mass index (BMI) (calculated as weight in kilograms divided by height in meters squared). We created a 25-level detailed smoking variable by combining data on smoking status, lifetime smoking, previous smoking intensity, current smoking intensity, time since quitting for former smokers, type of tobacco smoked previously, and type of tobacco smoked currently. We created a 6-level variable for alcohol drinking by combining data on drinking status and amount consumed. Less than 9% of the cohort lacked data on any single covariate, and indicator variables were created to represent missing data categories in regression models.

All UK Biobank participants were genotyped using genome-wide arrays. The initial 50 000 participants were genotyped using the Affymetrix UK BiLEVE Axiom array, while the remaining were genotyped using the Affymetrix UK Biobank Axiom array. The 2 arrays were extremely similar and can be imputed together. Quality control and imputation to the Haplotype Reference Consortium (HRC), version 1.1, and UK10K reference panels was performed centrally by the Wellcome Trust Centre for Human Genetics as described elsewhere.²³ Genetic data were available from 403 816 participants in our analytic sample after excluding sample outliers based on heterozygosity and missingness, participants with sex discrepancies between the self-reported and X-chromosome heterozygosity, and those potentially related to other participants, based on estimated kinship coefficients for all pairs of samples. We derived 2 genetic “caffeine metabolism” scores (CMS_G) by adding the number of alleles of single base changes, or single-nucleotide polymorphisms (SNPs), previously associated with blood caffeine metabolite levels, and the map near genes with plausible roles in caffeine metabolism.²⁴ All SNPs were available from the HRC, version 1.1, imputation. CMS_G2 included the 2 SNPs presenting with the largest effect sizes in a genome-wide association study (GWAS) of caffeine metabolites²⁴: rs2472297 (near CYP1A2) and rs6968554 (near AHR). CMS_G4 also included rs17685 (in POR) and rs56113850 (near CYP2A6). Corresponding weighted genetic “caffeine metabolism” scores (wCMS_G) were derived by summing the number of alleles multiplied by their β-coefficients. The latter were estimated by z/{sqrt[p(1 − p)]}, where z is the SNP z-score for the paraxanthine to caffeine ratio²⁴ and p is the SNP minor allele frequency. Estimated β-coefficients for rs6968554, rs2472297, rs17685, and rs56113850 were 17.58, 21.58, 3.79, and 19.43, respectively. Weighted scores were then calibrated such that both CMS_G2 and wCMS_G2 ranged from 0 to 4 and both CMS_G4 and wCMS_G4 ranged from 0 to 8, with higher scores predicting faster caffeine metabolism. All 4 SNPs have previously been linked to coffee consumption in the UK Biobank.^24,25 Because rs762551 (CYP1A2*1F) has been examined previously for interactions with coffee and disease outcomes,¹⁹ we conducted a separate analysis of this SNP.`

We used Cox proportional hazard regression models to estimate hazard ratios (HRs) and 95% CIs for mortality. Person-years (ie, calendar time) of follow-up was used as the underlying time metric. We estimated the Spearman partial correlation coefficients between coffee intake and each CMS_G, adjusting for age, sex, smoking status, genotyping array, and the first 5 genetic principal components. We tested the proportional-hazards assumption by comparing the multivariable model with the interaction term between person-time and coffee intake to the model without it; we observed no deviations from this assumption. Statistical tests were 2-sided, and P < .05 was interpreted as statistically significant. We used SAS software (version 9.4; SAS Institute Inc) and the computational resources of the National Institutes of Health’s (NIH) High-Performance Computing Biowulf cluster to conduct analyses.

In our analysis of coffee intake with all-cause mortality, we first adjusted for age and sex, followed by additional adjustment for other potential confounders; adjustment for detailed smoking history had the largest effect on risk estimates. We tested for trends across increasing categories of coffee intake by modeling the midpoint of each intake category as a continuous variable, with the highest intake categories of 8 or more cups per day and 6 or more cups per day assigned values of 8 and 6, respectively. We evaluated the multivariable-adjusted associations of coffee intake with cause-specific mortality and with all-cause mortality by coffee type, including non–coffee drinkers as the reference group in each subgroup analysis. We also ran multivariable-adjusted models stratified by the following potential effect modifiers: wCMS_G2 (0-1, >1 to 2, or >2), CMS_G4 (0-2, >2 to 3, >3 to 4, or >4), age (<55 or ≥55 years), sex (male or female), smoking status (never, former, or current smoker), general health status (excellent/good or fair/poor health), BMI (18.5 to <25.0, 25.0 to <30.0, or ≥30.0), history of diabetes (yes or no), and history of cancer, heart attack, or stroke (yes or no). The thresholds for strata of CMS_G were based on the distribution of the scores; however, we also considered a 1-unit and 2-unit increase in the weighted scores (eTable 1 in the Supplement). We assessed potential effect modification by modeling the cross-product term of the stratifying variable with coffee intake; the P value for heterogeneity corresponds to the likelihood ratio test comparing the multivariable models with and without the cross-product terms for each level of the stratifying variable with coffee intake (continuous) except in the case of genetic caffeine metabolism scores, which were modeled as continuous variables. Finally, to better understand the potential effect of reverse causality, we conducted a lag analysis in which we examined deaths occurring in 0 to less than 3 years, 3 to less than 5 years, and 5 or more years of follow-up.

In the UK Biobank cohort, 110 640 (22%) participants were non–coffee drinkers. Among coffee drinkers, 214 119 (56%), 87 665 (23%), and 74 053 (19%) reported usually drinking instant, ground, and decaffeinated coffee, respectively. Coffee drinkers, compared with non–coffee drinkers, were more likely to be male, white, former smokers, and drink alcohol. Participants drinking 4 or more cups per day, compared with those drinking less coffee and nondrinkers, were more likely to drink instant coffee and be current smokers, whereas participants drinking 1 to 3 cups per day were older, more likely to have a university degree, and more likely to report “excellent” health (Table 1).

Over 10 years of follow-up (median, 7 years) and 3.4 million person-years, 14 225 deaths occurred. After multivariable adjustment, we observed an inverse dose-dependent association (P < .001 for trend) for total coffee intake and all-cause mortality (Table 2). The HRs for drinking less than 1, 1, 2 to 3, 4 to 5, 6 to 7, and 8 or more cups per day were 0.94 (95% CI, 0.88-1.01), 0.92 (95% CI, 0.87-0.97), 0.88 (95% CI, 0.84-0.93), 0.88 (95% CI, 0.83-0.93), 0.84 (95% CI, 0.77-0.92), and 0.86 (95% CI, 0.77-0.95), respectively. Approximately 8294 (58%), 2833 (20%), and 553 (4%) deaths in the cohort were due to cancer, cardiovascular disease, and respiratory disease, respectively. We observed inverse associations for coffee drinking with all-cancer and all-cardiovascular disease mortality. The number of deaths for specific types of cancer and cardiovascular disease were modest, yet we did observe potential inverse associations, without statistical significance, for stroke, colorectal cancer, and female breast cancer. We observed inverse associations for each coffee type with all-cause and cause-specific mortality; associations were generally stronger for ground coffee than for instant or decaffeinated coffee (Table 3).

Next, we evaluated the effect of CMS_G, defined using 2 or 4 known caffeine-related SNPs. Each score was positively and statistically significantly correlated with coffee intake, with Spearman partial correlation coefficients of 0.07 (P < .001) for both CMS_G2 and CMS_G4, reflecting prior UK Biobank reports.^24,25 Using the weighted CMS_G2, based on variants in AHR and CYP1A2, participants with a lower wCMS_G2 or wCMS_G4 (ie, slower caffeine metabolizers) had similar associations for coffee drinking and mortality as participants with a higher wCMS_G2 (P = .75 for heterogeneity) or wCMS_G4 (P = .17 for heterogeneity) (ie, faster caffeine metabolizers), indicating no modification of the association between coffee drinking and all-cause mortality by these common polymorphisms related to caffeine metabolism (Table 4). We found similar results when limiting the CMS_G analyses to those with white and/or British ancestry only (data not shown), using different thresholds (ie, 1- or 2-unit increase) for wCMS_G strata (eTable 1 in the Supplement), stratifying by CYP1A2*1F only (eTable 2 in the Supplement), and using the unweighted CMS_G (eTable 3 in the Supplement). In addition, we observed a positive association between wCMS_G4 and all-cause mortality in multivariable models with (HR for a 1-unit increase in wCMS_G4 = 1.02; 95% CI, 1.01-1.03) and without (HR for a 1-unit increase in wCMS_G4 = 1.02; 95% CI, 1.00-1.03) adjustment for coffee and tea intake.

Associations for coffee drinking and mortality did not meaningfully differ by sex, age group, BMI, history of diabetes, or previous diagnosis of cancer, heart attack, or stroke. We did observe a statistically significant interaction between coffee drinking and self-reported health, with a stronger inverse association among those reporting worse health than those reporting better health (Table 5). Associations also seemed stronger among former and current smokers than never-smokers, with the latter association lacking statistical significance. However, the formal test for interaction was not statistically significant. Finally, inverse associations seemed stronger for deaths occurring within the first 3 years of follow-up, although similar patterns were also observed in later years (eTable 4 in the Supplement).

In this large study of nearly 500 000 people in the United Kingdom, coffee drinking was inversely associated with all-cause mortality, with statistically significant inverse associations observed in participants drinking 1 to 8 or more cups per day. Inverse associations were also observed for ground and instant coffee, although were somewhat weaker for instant coffee, and for cancer and cardiovascular disease mortality, the 2 most common causes of death in the cohort. Common genetic polymorphisms predisposing individuals to slower or faster caffeine metabolism did not modify associations between coffee drinking and mortality, although, as previously reported in the UK Biobank,^24,25 higher CMS_G was significantly correlated with higher coffee intake. We also observed inverse associations for both caffeinated and decaffeinated coffee, further suggesting the importance of noncaffeine compounds in the association.

Our findings are consistent with prior, large, prospective investigations of coffee drinking and all-cause mortality conducted in the United States,^9,11-13 Europe,¹⁰ and Asia,¹⁵ as well as the most recent meta-analyses^14,16 and the 2015 US Dietary Guidelines Advisory Committee report, which concluded that moderate coffee consumption can be a part of a healthy diet.¹⁷ However, it should be noted that none of these prior studies were able to look at potential effect modification by caffeine metabolism. In addition, our findings for all-cancer and all-cardiovascular disease mortality are reflective of recent large prospective studies of certain cancers⁸ and cardiovascular disease.¹³

There has been concern about the health effects of heavy coffee drinking, particularly in participants with common genetic polymorphisms that affect caffeine metabolism. For example, prior studies have suggested that variants in CYP1A2, encoding the enzyme responsible for more than 95% of caffeine metabolism, may alter associations of coffee drinking with cardiovascular-related outcomes, with slower caffeine metabolizers having higher risk of developing hypertension²⁰ or having a myocardial infarction¹⁹ relative to their nondrinking counterparts, whereas faster caffeine metabolizers who drink coffee are at no or lower risk of these outcomes. However, prior large prospective cohorts have lacked genetic data on most participants, such that it was not possible to examine how genetic variation in caffeine metabolism affects associations between coffee drinking and mortality. With genetic data on more than 400 000 participants in the UK Biobank, we were able to stratify each participant based on their genetic susceptibility for slower or faster caffeine metabolism. We found very similar associations between coffee drinking and mortality across levels of examined caffeine metabolism scores. Genetic caffeine metabolism scores are also associated with coffee and dietary caffeine intake behavior^18,24,25 and have therefore been used in Mendelian randomization studies of coffee and health.^26,27 In the present study, we observed a weak positive association between wCMS_G4 and all-cause mortality, which may indicate that slower caffeine metabolism is beneficial. However, results of these Mendelian randomization studies warrant careful interpretation because CMS_G explains only a very small proportion of the variance in coffee intake²⁵ and, importantly, also reflects biological exposure to caffeine,^24,28 which is how we have applied it in the present study.

The UK Biobank presented a unique opportunity to assess potential differences between drinking instant vs ground coffee because 55% of coffee drinkers reported usually drinking instant coffee. Associations for ground coffee and all-cause mortality were generally stronger than those for instant coffee. This observation may be explained by residual confounding owing to unmeasured or poorly measured confounders. For example, instant coffee was more common among participants without a college degree than among college graduates. Alternatively, instant coffees have been shown to have lower amounts of bioactive compounds, including polyphenols, than ground coffees,²⁹ which may have an effect on observed associations. Mechanisms hypothesized to explain the potential protective effect of coffee drinking on mortality risk include reduced inflammation,³⁰ improved insulin sensitivity,³¹ and effects on liver enzyme levels³¹ and endothelial function.³² Further studies with more detailed information on coffee type and preparation may provide insight into the mechanisms underlying coffee–chronic disease associations. For example, information on coffee preparation (ie, filtered, cappuccino, espresso, and latte) is available on a subset of approximately 70 000 UK Biobank participants who completed 24-hour dietary recalls at baseline and will be useful as the cohort matures.

Strengths of this study include its prospective design, large sample size, wide range of coffee intake, and availability of genotype and coffee type data. Limitations include the study’s low participation rate, approximately 5.5%; consequently, the cohort is not demographically representative of the general UK population, with evidence of a “healthy volunteer” selection bias.²² Nevertheless, valid estimation of exposure-disease relationships does not require a representative population,²² and our results reflect those from prior studies in different populations worldwide. Despite the large size of the cohort, there were relatively few deaths from specific types of cancer and cardiovascular disease. However, future analyses of these end points will be possible as the UK Biobank cohort matures.

In the very large UK Biobank cohort, coffee drinking was associated inversely with all-cause mortality, including in those drinking at least 8 cups per day, in both slow and fast metabolizers of caffeine, and in consumers of ground, instant, and decaffeinated coffee. Our results are based on observational data and should be interpreted with caution. Nevertheless, these results provide further evidence that coffee drinking can be part of a healthy diet and may provide reassurance to those who drink coffee and enjoy it.

Accepted for Publication: April 18, 2018.

Corresponding Author: Erikka Loftfield, PhD, Metabolic Epidemiology Branch, Division of Cancer Epidemiology and Genetics, National Cancer Institute Shady Grove, 9609 Medical Center Dr, Ste 6E320, Rockville, MD 20850-9768 (erikka.loftfield@nih.gov).

Published Online: July 2, 2018. doi:10.1001/jamainternmed.2018.2425

Author Contributions: Dr Loftfield had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Loftfield, Yu, Sinha, Freedman.

Acquisition, analysis, or interpretation of data: Loftfield, Cornelis, Caporaso, Yu, Freedman.

Drafting of the manuscript: Loftfield, Cornelis, Sinha.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Loftfield, Cornelis, Yu.

Administrative, technical, or material support: Loftfield, Caporaso, Freedman.

Study supervision: Sinha, Freedman.

Conflict of Interest Disclosures: None reported.

Funding/Support: This research was conducted using the UK Biobank resource (applications 18623 and 21394). The UK Biobank was established by the Wellcome Trust, the Medical Research Council, the UK Department of Health, and the Scottish Government. The UK Biobank has also received funding from the Welsh Assembly Government, the British Heart Foundation, and Diabetes United Kingdom. This work used the computational resources of the National Institutes of Health’s (NIH) High-Performance Computing Biowulf cluster and was supported by the Intramural Research Program of the NIH; Division of Cancer Epidemiology and Genetics, National Cancer Institute, NIH; and the Department of Health and Human Services.

Role of the Funder/Sponsor: The study sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: We thank Michael Stagner, BS, at Information Management Services, who provided technical support, for which he was compensated.