CoffeariumBeta
  1. Plant
  2. Farm
  3. Process
  4. Trade
  5. Roast
  6. Brew
  7. Cup
Science · Health

Coffee and long-term health

What large reviews, genetic studies, trials and health authorities report about habitual coffee drinking and mortality, heart disease, diabetes, liver disease, cancer and more – and why most of it shows association, not cause.

Across the chain: Science

Source-checkedby Coffearium verification pass· Last source-checked 21 Oct 2026· Prepared with AI assistance· 12 sources (12 tier A)· 3 contested· Updated 21 Oct 2026
Reading depth
Contents
  • A 2017 umbrella review of 201 observational meta-analyses found coffee more often associated with lower than higher risk; most evidence was low or very low quality.1
  • In that review, the largest relative risk reductions were at about three to four cups a day versus none; all-cause mortality was RR 0.83 at three cups.1
  • In genetic (Mendelian randomisation) studies, genetically predicted coffee intake showed no association with cardiovascular disease and probable evidence of higher type 2 diabetes risk.3,4
  • The International Agency for Research on Cancer (IARC) rates drinking coffee as not classifiable as to its carcinogenicity (Group 3).6,7
  • In the umbrella review, harmful associations remained in pregnancy after adjustment for smoking; fracture risk in women was also higher.1

What the evidence is made of#

Almost everything known about coffee and long-term health comes from observational studies. These follow large groups of people, record how much coffee they drink and count who later develops a disease.1 They show associations: coffee drinkers may differ from non-drinkers in many other ways.1,5 Randomised trials, which could show cause, have given coffee only for short periods.1

A wide summary is a 2017 umbrella review, a review of meta-analyses. It covered 201 meta-analyses of observational research on 67 health outcomes and 17 meta-analyses of trials, in adults in any country.1 About a quarter of the evidence was rated low quality and three quarters very low quality, on the GRADE scale used to rate evidence.1 No authors of the review reported support from any organisation for the work.1

Outcome by outcome#

In the umbrella review, coffee drinking was more often associated with lower risk than with higher risk.1 For several outcomes the largest relative risk reductions were at about three to four cups a day, compared with none.1 A relative risk (RR) of 0.83 means a 17% lower risk in that group.1 It is not proof that coffee caused the difference.

  • All-cause mortality: RR 0.83 at three cups a day versus none.1
  • Heart and blood vessels: cardiovascular mortality RR 0.81 at three cups a day versus none; incident cardiovascular disease RR 0.85, with the largest benefit at three to five cups a day.1
  • Type 2 diabetes: RR 0.70 for high versus low intake.1
  • Liver: lower risk of chronic liver disease, cirrhosis and liver cancer.1
  • Nervous system: consistent associations with lower risk of Parkinson’s disease and of depression.1
  • Bones: in women, high versus low intake was associated with a higher risk of fracture (RR 1.14).1
0.20.40.60.811.21.4Relative risk, high vs low intake (1 = no difference)Dot = summary relative risk; bar = 95% confidence interval. Observational data.Type 2 diabetesRR 0.70 (0.65–0.75)1 Chronic liver diseaseRR 0.35 (0.22–0.56)1 CirrhosisRR 0.69 (0.44–1.07)1 Cancer, all sitesRR 0.82 (0.74–0.89)1 Metabolic syndromeRR 0.91 (0.86–0.95)1 Fracture, womenRR 1.14 (1.05–1.24)1
Data table
Relative risk, high vs low intake (1 = no difference)Source
Type 2 diabetesRR 0.70 (0.65–0.75)1
Chronic liver diseaseRR 0.35 (0.22–0.56)1
CirrhosisRR 0.69 (0.44–1.07)1
Cancer, all sitesRR 0.82 (0.74–0.89)1
Metabolic syndromeRR 0.91 (0.86–0.95)1
Fracture, womenRR 1.14 (1.05–1.24)1
Summary relative risks with 95% confidence intervals for high versus low coffee intake, six outcomes, as reported in one umbrella review of observational meta-analyses.1 Values below 1 mean lower risk among high consumers; intervals that cross 1 are compatible with no difference. Associations, not proof of cause.

Pregnancy is the main exception. There, harmful associations remained after adjustment for smoking.1 The European Food Safety Authority (EFSA) intake statement for pregnancy, up to 200 mg of caffeine a day, is covered in Caffeine.10

How much, and which coffee#

A UK cohort shows what the shape of the association looks like. The UK Biobank followed 498,134 adults aged 38–73 for about 10 years.2 Compared with non-drinkers, every intake from one cup to eight or more cups a day was associated with lower all-cause mortality.2 The hazard ratios differed little between these intake levels, from 0.84 to 0.92.2

0.70.80.911.1Hazard ratio vs non-drinkers (1 = no difference)Dot = hazard ratio; bar = 95% confidence interval. Observational data.Less than 1 cup/dayHR 0.94 (0.88–1.01)2 1 cup/dayHR 0.92 (0.87–0.97)2 2–3 cups/dayHR 0.88 (0.84–0.93)2 4–5 cups/dayHR 0.88 (0.83–0.93)2 6–7 cups/dayHR 0.84 (0.77–0.92)2 8 or more cups/dayHR 0.86 (0.77–0.95)2
Data table
Hazard ratio vs non-drinkers (1 = no difference)Source
Less than 1 cup/dayHR 0.94 (0.88–1.01)2
1 cup/dayHR 0.92 (0.87–0.97)2
2–3 cups/dayHR 0.88 (0.84–0.93)2
4–5 cups/dayHR 0.88 (0.83–0.93)2
6–7 cups/dayHR 0.84 (0.77–0.92)2
8 or more cups/dayHR 0.86 (0.77–0.95)2
All-cause mortality by cups per day, compared with non-drinkers, in one UK cohort of 498,134 adults (hazard ratios with 95% confidence intervals).2 Observational data: association, not proof of cause.

In the same cohort, similar associations appeared for instant, ground and decaffeinated coffee.2 They also held for people whose genes predict slower or faster caffeine breakdown.2 The umbrella review found no convincing harmful association for decaffeinated coffee with any outcome.1 Which coffee compounds might matter is covered in Bioactive compounds.

Brewing method matters for one known effect. Unfiltered coffee, such as boiled coffee, raises blood cholesterol through the diterpenes cafestol and kahweol; paper filters hold them back.11 See Coffee oil for the details.

Why association is not cause#

Four problems recur in coffee research.

  • Confounding. People who drink more coffee have often smoked more. Many harmful associations largely disappeared with adequate adjustment for smoking.1 A 2020 narrative review of 36 cohort samples found that many did not adjust adequately for smoking.5
  • Reverse causation. Illness can change habits. In case-control studies, symptoms of disease may have led people to drink less coffee; other observational studies share this risk.1
  • Measurement. There is no standard cup size, so “cups” in different studies can mean different amounts.1
  • Short trials. Randomised trials have given coffee only for short durations and measured few outcomes, such as blood pressure, not long-term disease.1

How to judge a single new headline about coffee is the subject of How to read coffee health news.

What health authorities say#

Several bodies have assessed the evidence. Their statements, as published:

  • IARC (World Health Organization), 2016: drinking coffee is “not classifiable as to its carcinogenicity to humans (Group 3)”. In 1991 it had been rated “possibly carcinogenic” (Group 2B).6,7
  • World Cancer Research Fund and American Institute for Cancer Research (WCRF/AICR), 2018: coffee “probably protects against liver cancer and endometrial cancer”.9
  • US Dietary Guidelines Advisory Committee, 2015: it graded as strong the evidence on moderate intake in healthy adults. By its conclusion, 3–5 cups a day (up to 400 mg caffeine) is not associated with higher risk of major chronic diseases or premature death.8
  • EFSA, 2015: habitual caffeine intake up to 400 mg a day raises no safety concerns for healthy non-pregnant adults; it assessed caffeine, not coffee.10

Very hot drinks are a separate matter. The IARC classifies drinking beverages above 65 °C as probably carcinogenic (Group 2A); see Serving temperature.6 Roasting by-products such as acrylamide are covered in Roasting contaminants.

Reference values#

Values as reported. They describe associations in observational data unless marked as a trial or a genetic study. RR = relative risk; HR = hazard ratio; OR = odds ratio; CI = 95% confidence interval.

Outcome Comparison Estimate (95% CI) Study type Source
All-cause mortality 3 cups/day vs none RR 0.83 (0.83–0.88)* Umbrella review of observational meta-analyses Poole et al.1
Cardiovascular mortality 3 cups/day vs none RR 0.81 (0.72–0.90) Same Poole et al.1
Cardiovascular disease Non-linear dose–response; largest benefit at 3–5 cups/day RR 0.85 (0.80–0.90) Same Poole et al.1
Type 2 diabetes High vs low RR 0.70 (0.65–0.75) Same Poole et al.1
Chronic liver disease High vs low RR 0.35 (0.22–0.56) Same Poole et al.1
Cirrhosis High vs low RR 0.69 (0.44–1.07) Same Poole et al.1
Cirrhosis Per extra cup/day RR 0.83 (0.78–0.88) Same Poole et al.1
Non-alcoholic fatty liver disease Any vs none RR 0.71 (0.60–0.85) Same Poole et al.1
Cancer, all sites High vs low RR 0.82 (0.74–0.89) Same Poole et al.1
Metabolic syndrome High vs low RR 0.91 (0.86–0.95) Same Poole et al.1
Fracture, women High vs low RR 1.14 (1.05–1.24) Same Poole et al.1
Liver cancer Per cup/day RR 0.86 (0.81–0.90) Dose–response meta-analysis of observational studies WCRF/AICR9
All-cause mortality, UK adults 38–73 < 1 cup/day vs none HR 0.94 (0.88–1.01) Cohort, 498,134 people Loftfield et al.2
Same 1 cup/day HR 0.92 (0.87–0.97) Same Loftfield et al.2
Same 2–3 cups/day HR 0.88 (0.84–0.93) Same Loftfield et al.2
Same 4–5 cups/day HR 0.88 (0.83–0.93) Same Loftfield et al.2
Same 6–7 cups/day HR 0.84 (0.77–0.92) Same Loftfield et al.2
Same ≥ 8 cups/day HR 0.86 (0.77–0.95) Same Loftfield et al.2
15 cardiovascular outcomes Per 50% higher genetically predicted intake No association (univariable) Mendelian randomisation Yuan et al.3

* The interval is printed as 0.83 to 0.88 in both the abstract and the results of the review, so the point estimate equals the lower limit; this is probably a printing slip.1 The review takes the value from an earlier meta-analysis that was not opened for this entry. We report it as printed and leave it out of the chart.

The observational evidence in detail#

The umbrella review re-analysed each meta-analysis and graded it.1 It found evidence of a non-linear association for several outcomes, with the largest relative risk reductions at three to four cups a day versus none.1 Beyond the outcomes above, it reported lower risk of several specific cancers and of neurological, metabolic and liver conditions.1 For Parkinson’s disease and depression it describes consistent associations but gives no single summary estimate in its text.1

The 2015 US advisory committee reached similar conclusions from observational data. It graded as moderate the evidence that moderate coffee intake is associated with reduced risk of type 2 diabetes and cardiovascular disease, and that regular intake is associated with reduced risk of liver and endometrial cancer.8 It also graded as moderate the evidence for an inverse association between caffeine intake and Parkinson’s disease.8

For cancer, the WCRF/AICR dose–response meta-analysis found RR 0.86 for liver cancer per cup a day.9 The report names smoking and hepatitis B and C viruses as possible confounders or effect modifiers.9 It graded the evidence for a lower risk of cancers of the mouth, pharynx and larynx, and of some skin cancers, as limited–suggestive.9 IARC found evidence suggesting lack of carcinogenicity for cancers of the female breast, uterine endometrium, prostate, pancreas and liver.7

What genetic studies add#

Mendelian randomisation (MR) uses gene variants linked to coffee intake as a stand-in for intake itself.3,4 The umbrella review’s authors suggested it as a way to test whether residual confounding explains the observational findings.1 They also note one limit: the approach relies on an assumption of linearity.1

  • Cardiovascular disease. A 2021 MR study used 12 variants and data from the UK Biobank (up to 35,979 cases) and the FinnGen consortium in Finland (up to 17,325 cases).3 In the main (univariable) analysis, genetically predicted intake was not associated with any of 15 cardiovascular outcomes. The authors concluded that earlier observational studies may have been confounded.3
  • Many outcomes. A 2025 systematic review covered 59 MR studies and 160 disease or biomarker associations.4 It found probable evidence for a lower risk of hepatocellular carcinoma (the main liver cancer), ovarian cancer, migraine, kidney disease and gallstone disease.4
  • Opposite directions. The same review found probable evidence for a higher risk of type 2 diabetes, oesophageal and digestive cancers and Alzheimer’s disease, among other conditions, and for higher total cholesterol, LDL cholesterol and apolipoprotein B.4 It reports these as associations.4

The liver findings point the same way in observational and genetic studies.1,4 For cardiovascular disease and type 2 diabetes they do not.1,3,4 The higher cholesterol in MR is in line with the trial evidence for unfiltered coffee.4,11 The 2025 review calls for more MR studies in non-European cohorts.4

1Observational studiesCohorts, case-controlShowsAssociations in largepopulations 1 LimitsConfounding, e.g. smoking 1,5 Reverse causation 1 No standard cup size 1 Mostly low or very lowquality (GRADE) 1 2Mendelian randomisationGene variants as proxiesShowsWhether geneticallypredicted intake tracksdisease 3,4 LimitsAssumes linearity 1 Non-European cohortsstill needed 4 3Randomised trialsCoffee assigned at randomShowsCause, for what theymeasure 1,12 LimitsShort durations 1 Blood pressure, lipids,heart rhythm, sleep –not long-term disease 1,12 Schematic. Not a ranking of individual studies.
Three kinds of evidence on coffee and long-term health, what each can show and its main limits, as described in the sources.1,3,4,5,12 Schematic.

Randomised trials#

Trials randomly assign who drinks coffee, so they reduce the risk of confounding.1 In the umbrella review, 17 meta-analyses of trials covered nine outcomes: blood pressure, blood lipids and three pregnancy outcomes.1 Coffee was given for short durations only.1 The review found coffee consistently associated with unfavourable changes in the lipid profile, and brewing method is an important factor there.1,11

One 2023 randomised case-crossover trial, funded by a university and the US National Institutes of Health, shows what short trials can measure.12 One hundred adults wore heart, movement and glucose monitors for 14 days, and daily text messages told them at random to drink caffeinated coffee or avoid caffeine.12 Premature atrial contractions did not rise significantly (58 versus 53 a day). Premature ventricular contractions rose from 102 to 154 a day, daily steps rose from 9,665 to 10,646, and nightly sleep fell from 432 to 397 minutes.12 These are short-term effects in one study, not long-term outcomes.

Biases in detail#

  • Smoking. In the umbrella review, harmful associations were largely nullified by adequate adjustment for smoking, and residual confounding by smoking, despite some adjustment, is named as a likely explanation for an apparent harm.1 A 2020 narrative review of cohort studies reported that 1–5 cups a day was related to lower mortality among never smokers and in studies that adjusted for pack-years of smoking.5 It recommends separate analyses for never smokers.5
  • Diet and sugar. The same review notes that possible effects of sugar added to coffee have not been adequately investigated.5
  • Reverse causality. Bias from reverse causality can occur in observational studies, not only in case-control designs.1
  • Exposure. Cup sizes differ, so studies comparing cups may compare ranges of exposure.1 Brewing method changes what is in the cup.1,11
  • Small-study bias. Egger’s test for small-study effects could be run on only 40% of the meta-analyses re-analysed in the umbrella review.1
  • Decaf drinkers. People who drink decaffeinated coffee might differ from those who drink caffeinated coffee.1

Open questions#

  • Cause or confounding for the heart. Observational data show lower cardiovascular risk; MR data show none.1,3 No long-term trial opened for this entry has tested cardiovascular events.
  • Type 2 diabetes. Observational and MR results point in opposite directions.1,4
  • Fracture in women. The umbrella review suggests that women at increased risk of fracture should possibly be excluded from any trial of coffee.1
  • Added sugar. Possible effects of sugar added to coffee have not been adequately investigated.5
  • Populations. The 2025 MR review calls for more studies in non-European cohorts.4

Key studies#

Poole et al., 2017 – Umbrella review of coffee and health. Re-analysed 201 meta-analyses of observational studies and 17 of trials covering 67 outcomes. Coffee was more often associated with benefit than harm, with the largest risk reductions at three to four cups a day; most evidence was of low or very low quality, and pregnancy and fracture in women were exceptions.

BMJ 359, j5024 · doi:10.1136/bmj.j5024

Loftfield et al., 2018 – Coffee and mortality in the UK Biobank. Followed 498,134 UK adults for about 10 years. Every intake level from one to eight or more cups a day was associated with lower all-cause mortality than no coffee, for instant, ground and decaffeinated coffee alike and regardless of genetic caffeine-metabolism score.

JAMA Internal Medicine, 2018 · doi:10.1001/jamainternmed.2018.2425

Yuan et al., 2021 – Mendelian randomisation and cardiovascular disease. Used 12 genetic variants for coffee intake in UK and Finnish data. Genetically predicted intake was not associated with any of 15 cardiovascular outcomes, which the authors read as a sign that earlier observational findings may have been confounded.

Nutrients 13(7), 2218 · doi:10.3390/nu13072218

Pham et al., 2025 – Systematic review of Mendelian randomisation studies. Reviewed 59 studies covering 160 disease or biomarker associations. It found probable evidence linking genetically predicted intake with lower risk of liver cancer and kidney disease but higher risk of type 2 diabetes and higher LDL cholesterol, partly against the observational picture.

Nutrition Research Reviews 38(2), 1006–1026 · doi:10.1017/s0954422425100206

Marcus et al., 2023 – Randomised coffee days. In this case-crossover trial, one hundred adults were told at random each day for 14 days to drink caffeinated coffee or avoid caffeine, while monitors recorded heart rhythm, movement and sleep. Coffee days brought more ventricular extra beats and steps and less sleep, but no significant rise in atrial extra beats.

New England Journal of Medicine, 2023 · doi:10.1056/NEJMoa2204737

General information, not medical advice.

Contested
  • Type 2 diabetes: observational meta-analyses link high coffee intake with lower risk (RR 0.70 for high versus low intake)1, and a 2015 US advisory committee graded the evidence for a reduced risk as moderate8; a 2025 systematic review of Mendelian randomisation studies found probable evidence for an increased risk with genetically predicted higher coffee intake.4 The two methods answer different questions and carry different biases.
  • Cardiovascular disease: observational meta-analyses link coffee non-linearly with lower risk of incident cardiovascular disease (RR 0.85), with the largest benefit at three to five cups a day1; a 2021 Mendelian randomisation study found no association with any of 15 cardiovascular outcomes in its main analysis and concluded that earlier observational studies may have been confounded.3
  • Liver cancer: IARC found evidence suggesting lack of carcinogenicity for liver cancer, and its press release noted reduced risks6,7; WCRF/AICR judged that coffee probably protects against it.9 Both describe the same direction with different strength of wording; neither rests on randomised trials.
Share or cite this entry

Change log

  • – All-cause mortality (RR 0.83) and cardiovascular mortality (RR 0.81) now given at three cups a day versus none, and incident cardiovascular disease (RR 0.85) at a largest benefit of three to five cups a day, instead of all three at three to four cups. Reason: Verification pass; the results section of source 1 states these intakes.
  • – Trial evidence in the umbrella review described as nine outcomes (blood pressure, blood lipids, three pregnancy outcomes) instead of "limited to systolic and diastolic blood pressure"; trials "reduce the risk of" rather than "remove" confounding. Reason: Verification pass; source 1 lists these trial outcomes and uses this wording.
  • – WCRF/AICR limited–suggestive evidence now stated as evidence for a lower risk, and for some skin cancers only. Reason: Verification pass; source 9 states the direction and the skin cancer types.

Sources

  1. APoole, R.; Kennedy, O. J.; Roderick, P.; Fallowfield, J. A.; Hayes, P. C.; Parkes, J. (2017). Coffee consumption and health: umbrella review of meta-analyses of multiple health outcomes. BMJ 359, j5024. doi:10.1136/bmj.j5024 · other copies
  2. ALoftfield, E.; Cornelis, M. C.; Caporaso, N.; Yu, K.; Sinha, R.; Freedman, N. D. (2018). Association of Coffee Drinking With Mortality by Genetic Variation in Caffeine Metabolism: Findings From the UK Biobank. JAMA Internal Medicine 178(8), 1086 (abstract via UK Biobank publications page). doi:10.1001/jamainternmed.2018.2425 · other copies
  3. AYuan, S.; Carter, P.; Mason, A. M.; Burgess, S.; Larsson, S. C. (2021). Coffee Consumption and Cardiovascular Diseases: A Mendelian Randomization Study. Nutrients 13(7), 2218 (abstract via UK Biobank publications page). doi:10.3390/nu13072218 · other copies
  4. APham, K.; Assefa Kassaw, N.; Mulugeta, A.; Zhou, A.; Hyppönen, E. (2025). Coffee and health outcomes: a systematic review of Mendelian randomisation studies. Nutrition Research Reviews 38(2), 1006–1026. doi:10.1017/s0954422425100206 · other copies
  5. AThomas, D. R.; Hodges, I. D. (2020). Dietary research on coffee: Improving adjustment for confounding. Current Developments in Nutrition 4(1), nzz142. doi:10.1093/cdn/nzz142 · other copies
  6. AInternational Agency for Research on Cancer (2016). IARC Monographs evaluate drinking coffee, maté, and very hot beverages (Press release N° 244). International Agency for Research on Cancer (WHO), Lyon, 15 June 2016.
  7. AIARC Working Group on the Evaluation of Carcinogenic Risks to Humans (2018). Drinking Coffee, Mate, and Very Hot Beverages (IARC Monographs on the Evaluation of Carcinogenic Risks to Humans, Vol. 116). International Agency for Research on Cancer (WHO), Lyon.
  8. A2015 Dietary Guidelines Advisory Committee (2015). Appendix E-2.39a: Usual Caffeine Consumption and Health Evidence Portfolio (Scientific Report of the 2015 Dietary Guidelines Advisory Committee). U.S. Department of Health and Human Services; U.S. Department of Agriculture.
  9. AWorld Cancer Research Fund; American Institute for Cancer Research (2018). Non-alcoholic drinks and the risk of cancer (Continuous Update Project Expert Report 2018). World Cancer Research Fund International.
  10. AEFSA Panel on Dietetic Products, Nutrition and Allergies (NDA) (2015). Scientific Opinion on the safety of caffeine. EFSA Journal 13(5), 4102. doi:10.2903/j.efsa.2015.4102 · other copies
  11. AUrgert, R.; Katan, M. B. (1997). The cholesterol-raising factor from coffee beans. Annual Review of Nutrition 17, 305–324. doi:10.1146/annurev.nutr.17.1.305 · other copies
  12. AMarcus, G. M.; Rosenthal, D. G.; Nah, G.; Vittinghoff, E.; Olgin, J. E.; et al. (2023). Acute Effects of Coffee Consumption on Health among Ambulatory Adults. New England Journal of Medicine (abstract via eScholarship, University of California). doi:10.1056/NEJMoa2204737 · other copies

Tier A · authoritative · B · trade and expert · C · general. Supporting quotes for every source are in the data version of this entry. If a publisher’s site refuses access, try the doi link or “other copies”.

Suggest a change

On Coffee and long-term health. Tell us what you would change and why. A source that supports it helps most.