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

How to read coffee health news

A sourced checklist for judging a coffee-and-health headline – the type of study behind it, relative versus absolute risk, what a “cup” means, confounding, missing studies, funding and press-release spin.

Across the chain: Science

Source-checkedby Coffearium verification pass· Last source-checked 21 Oct 2026· Prepared with AI assistance· 20 sources (20 tier A)· Updated 21 Oct 2026
Reading depth
Contents
  • In the GRADE system for rating evidence, randomised trials start as high-quality evidence and observational studies as low-quality evidence.1
  • Most coffee-and-health evidence is observational; a 2017 umbrella review rated about 25% of its outcomes low and 75% “very low” quality.8
  • A relative risk says nothing about absolute numbers. Our illustration – a hypothetical 50% lower risk applied to a lifetime risk of 1.1% – gives about 5.5 fewer cases per 1,000 people.10
  • “One cup of coffee” is not a reproducible measure; café espressos held 51–322 mg of caffeine each.11,12
  • In a study of UK university press releases, 33% exaggerated causal claims; news based on exaggerated releases was far more likely to exaggerate too.3

What kind of study is behind the headline#

Health headlines about coffee rest on studies of very different strength. The GRADE system, used to rate evidence in health research, starts randomised trials as high-quality evidence and observational studies as low-quality evidence.1 In a randomised trial, researchers assign coffee or no coffee at random. In an observational study, people choose for themselves, and researchers record what happens.1,8

The figure arranges the main study types in a rough order. It is a guide, not a rule: the Oxford Centre for Evidence-Based Medicine notes that hierarchies of evidence have been used inflexibly and criticised for decades.2 A well-run study lower down can be more informative than a poor one higher up.1

Laboratory and animal studiesCase-control studiesCohort studiesMendelian randomisationRandomised trialsSystematic reviews, meta-analysesless open to biasmore open to bias123456
  1. Laboratory and animal studies. Show possible mechanisms. In a study of UK university press releases, 36% exaggerated what animal research meant for humans.3
  2. Case-control studies. Compare patients with people without the disease and ask about coffee drunk years before diagnosis.4 Observational: they show association, not cause.1
  3. Cohort studies. Follow people forward and count who falls ill. IARC gave cohort studies the greatest weight in its 2016 coffee review.5
  4. Mendelian randomisation. Uses genetic variants as a natural experiment; less prone to confounding and reverse causation, but depends on three assumptions.6
  5. Randomised trials. Coffee is assigned at random, so they can test cause.1,8 Coffee trials have been short.7,8
  6. Systematic reviews and meta-analyses. Pool many studies; an umbrella review pools meta-analyses. They inherit the weaknesses of the studies they pool.8
A rough ladder of study types in health research, with what each can and cannot show, as described in the sources cited. Higher steps are less open to bias, but quality within each step varies.1,2 Schematic, not to scale.

Coffee research sits mostly on the lower and middle steps. A 2017 umbrella review of coffee and health covered 201 meta-analyses of observational research and only 17 of interventional studies.8 Its authors concluded that most of the evidence “provides only low or very low quality evidence” and that robust randomised trials are needed to tell whether the associations are causal.8 What that review found, outcome by outcome, is covered in Coffee and long-term health.

Relative and absolute risk#

Headlines usually report relative risk (RR): how much higher or lower a risk is in one group than in another. An RR of 0.82 means 18% lower. It does not say how common the disease is to begin with. Absolute risk does: how many people out of 1,000 are affected.

One figure from the 2017 umbrella review and one hypothetical headline show why the difference matters. High versus low coffee intake was associated with an 18% lower risk of any cancer (RR 0.82).8 For a rare outcome, take a hypothetical headline reporting a 50% lower risk of liver cancer (RR 0.50; an illustration, not a result quoted from the review). In the United States, about 39.2% of people are diagnosed with some cancer during their lifetime9, and about 1.1% with liver cancer (data 2021–2023).10

Our calculation applies each RR to its baseline. For all cancers: 392 per 1,000 × 0.82 ≈ 321 per 1,000, about 71 fewer. For the hypothetical liver-cancer headline: 11 per 1,000 × 0.50 ≈ 5.5 per 1,000, about 5.5 fewer.8,9,10 The larger relative change gives the smaller absolute change, because liver cancer is rare.

0100200300400per 1,000 peopleAny cancer · US baseline392 (39.2%)9 Any cancer · RR 0.82applied≈321 · 18% lower8,9 Liver cancer · US baseline11 (1.1%)10 Liver cancer ·hypothetical RR 0.50≈5.5 · 50% lower (hypothetical)10
Data table
per 1,000 peopleSource
Any cancer · US baseline392 (39.2%)9
Any cancer · RR 0.82 applied≈321 · 18% lower8, 9
Liver cancer · US baseline11 (1.1%)10
Liver cancer · hypothetical RR 0.50≈5.5 · 50% lower (hypothetical)10
Lifetime risk per 1,000 people in the United States (SEER, 2021–2023)9,10, and the same baseline with a relative risk applied: for any cancer the umbrella review’s high-versus-low estimate8, for liver cancer a hypothetical halving. Our calculation, for illustration only: the relative risks are observational associations, and the baselines include coffee drinkers.

This is an exercise in arithmetic, not a prediction. The relative risks come from observational studies, which cannot show that coffee caused the difference.8 The baselines are for a whole population, drinkers included, and in another country.9,10 The point is the method: always look for the starting risk before judging a percentage.

What counts as a cup#

Most coffee studies measure intake in cups a day. A 2014 study of espressos bought in Scotland, Italy and Spain concluded that “one cup of coffee” is not a reproducible measurement.11 Outlet averages in Scotland ranged from 72 to 212 mg of caffeine per espresso, and serving volumes varied widely.11 An earlier survey of 20 cafés in Glasgow found 51–322 mg per espresso, in cups of 23–70 mL.12

The 2017 umbrella review notes that there is no universally recognised standard cup size.8 When a headline says “three cups a day”, it helps to know what kind of cup the study counted. How much caffeine a serving holds is covered in Caffeine.

Confounding, reverse causation and missing studies#

Confounding. A confounder is a third factor linked both to coffee and to the disease. For coffee, smoking is a well-documented example.8,13 In 2016 an IARC working group concluded that positive associations with bladder cancer in some early studies may have been due to other risk factors, such as tobacco smoking.13 In the 2017 umbrella review, harmful associations were largely nullified by adequate adjustment for smoking, except in pregnancy.8

Reverse causation. Sometimes the disease changes the coffee habit, not the other way round. The umbrella review gives the example of symptoms leading people to drink less coffee before a diagnosis.8 Low coffee intake then looks linked to disease without causing it.

Missing studies. Not every study gets published. A Cochrane methodology review found that trials with positive findings are published more often, and more quickly, than trials with negative findings.14 A meta-analysis can only pool what was published.

Who paid for the study#

Funding is associated with results. A Cochrane review found that drug and device studies sponsored by manufacturers more often reported favourable efficacy results (RR 1.27) and favourable conclusions (RR 1.34) than studies sponsored by other sources.15 In nutrition, a 2007 analysis of 206 articles on soft drinks, juice and milk found conclusions favourable to the sponsor far more often when all funding came from industry than when none did.16

Member journals of the International Committee of Medical Journal Editors (ICMJE) require authors to declare financial relationships on a standard form, and the committee encourages other journals to adopt it.17 Articles should also state what role the funders played in study design, data analysis and interpretation.17 Articles should be published with these disclosure statements.17 Coffearium uses information sites funded by the coffee industry only as pointers to primary papers, never as evidence.

From paper to headline#

Exaggeration often starts before the newsroom. In a 2014 study of 462 press releases from 20 UK universities, 40% contained exaggerated advice and 33% exaggerated causal claims.3 When a press release exaggerated a causal claim, 81% of the news stories based on it did too, against 18% when it did not.3

The paper itself can be the source. A 2012 study of randomised trials found “spin”, a misleading presentation of results, in 40% of abstract conclusions and 47% of press releases.18 Spin in the abstract was the only factor linked to spin in the press release.18

Questions to ask of a coffee headline#

Use these steps when you read a new coffee-and-health story:

  1. Find the paper. Look for the journal name or a DOI in the article or press release; read at least the abstract.
  2. Name the study type. Laboratory, animal, case-control, cohort, Mendelian randomisation, randomised trial or review? Only trials test cause directly.1
  3. Name the people. Who was studied, how many, in which country and for how long?
  4. Find the dose. Cups a day or milligrams of caffeine? A “cup” varies widely.11
  5. Look for absolute numbers. A percentage change needs a starting risk.
  6. Check the wording. “Associated with” in the paper should not become “causes” or “prevents” in the headline.3
  7. Check confounders. Were smoking and other habits taken into account?8
  8. Check funding and disclosures. Read the statements on funding and conflicts of interest.17
  9. Compare with reviews and authorities. One new study rarely overturns a body of evidence; see Coffee and long-term health.

How evidence is graded#

GRADE rates the quality of evidence, starting from the study design. Randomised trials start high and observational studies start low.1 Quality may be downgraded for limitations in design or conduct, imprecise estimates (wide confidence intervals), inconsistent results, indirect evidence or publication bias.1 It may be upgraded for a large effect, a dose-response gradient, or when all plausible biases would reduce the apparent effect.1

In the Oxford levels, the top level for questions about treatment benefit is “either N-of-1 randomized trials or systematic reviews of randomized trials”.2 For coffee and long-term disease, few such trials exist: the umbrella review notes that coffee has been given as an intervention only for short durations.8 A 2023 randomised trial shows what is feasible: 100 adults wore continuous monitors while coffee and no-coffee days were assigned at random, and the outcomes were measured over days, not decades.7

Mendelian randomisation fills part of this gap. Genetic variants are fixed at conception, so they are less likely to be affected by confounding or reverse causation.6 A variant is a valid instrument only if three assumptions hold: relevance, independence and the exclusion restriction; horizontal pleiotropy, where a variant acts through another pathway, violates them.6 The estimates describe effects of a risk factor over a lifetime, not of a change at a specific time.6 Results for coffee are discussed in Coffee and long-term health.

Worked example: pancreatic cancer#

Early reports linked coffee with pancreatic cancer.4 In 1986 a research group published further data from its multi-year case-control study. It compared coffee intake 10 years before diagnosis in patients and controls, and concluded that its findings did not support earlier reports of a link.4

Thirty years later, IARC reviewed 20 cohort studies and 22 case-control studies that controlled for smoking.5 Cohort studies and population-based case-control studies showed no overall association, and a pooled analysis of cohort studies found none, including in non-smokers.5 IARC concluded that there was no evidence of an association.5 It listed pancreatic cancer among cancers that coffee is unlikely to cause.13

The lesson: a first association, from a design open to bias, is a question rather than an answer. For the same review, IARC rated drinking coffee overall as not classifiable as to its carcinogenicity (Group 3). That means the data do not allow a conclusion either way; it is not a statement that coffee is safe.13

Worked example: the California warning label#

Cancer warnings on coffee in California came out of a court case, CERT v. Starbucks. In its first phase, the trial court found that the coffee companies had failed to establish a legal defence to the requirement to post warnings for acrylamide in coffee, because they had not performed a quantitative risk assessment.20 The Phase 2 statement of decision followed in May 2018.20 Acrylamide forms during roasting; see Roasting contaminants.

The state agency then changed the rule. On 3 June 2019, California’s Office of Administrative Law approved a regulation adopted by the Office of Environmental Health Hazard Assessment (OEHHA). It states that chemicals in coffee created by and inherent in roasting or brewing “do not pose a significant risk of cancer”.19 It took effect on 1 October 2019.19 OEHHA wrote that the regulation is strongly supported by the research summarised by IARC, which placed coffee in Group 3, and expressed no view on whether the court had reached the right result.20

The ruling turned on what the companies had to prove under one state law, not on a new study of coffee.20 News of a “cancer warning” is therefore a legal fact, not a measure of risk.

Reference values#

Values as reported in the sources.

Parameter Value reported Source
GRADE starting levels Randomised trials high; observational studies low Guyatt et al.1
Coffee umbrella review, scope 201 meta-analyses of observational research (67 outcomes); 17 of interventional studies Poole et al.8
Coffee umbrella review, evidence quality ≈25% of outcomes “low”, ≈75% “very low” Poole et al.8
Any cancer, high vs low intake RR 0.82 (0.74–0.89) Poole et al.8
Lifetime risk, any cancer, US 39.2% (2021–2023 data) SEER9
Lifetime risk, liver and intrahepatic bile duct cancer, US 1.1% (2021–2023 data) SEER10
Absolute difference, any cancer (our calculation) 392 → ≈321 per 1,000 from8,9
Absolute difference, liver cancer, hypothetical RR 0.50 (our illustration) 11 → ≈5.5 per 1,000 from10
Caffeine per espresso, 20 Glasgow cafés 51–322 mg (cups 23–70 mL) Crozier et al.12
Caffeine per espresso, outlet means, Scotland 72–212 mg Ludwig et al.11
Press releases with exaggerated advice · causal claims · animal-to-human inference 40% · 33% · 36% (462 releases, 20 UK universities, 2011) Sumner et al.3
News exaggeration when the press release exaggerated vs not Advice 58 vs 17% · causal claims 81 vs 18% · animal-to-human 86 vs 10% Sumner et al.3
Spin in randomised-trial reports 40% of abstract conclusions; 47% of press releases (70 trials) Yavchitz et al.18
Spin in press release linked to spin in abstract RR 5.6 (2.8–11.1) Yavchitz et al.18
Industry-sponsored drug and device studies, favourable results · conclusions RR 1.27 (1.17–1.37) · RR 1.34 (1.19–1.51) Lundh et al.15
Nutrition articles, all-industry vs no-industry funding, favourable conclusion Odds ratio 7.61 (1.27–45.73); 206 articles Lesser et al.16
IARC, drinking coffee Group 3, not classifiable (2016) IARC13
California regulation on chemicals in coffee Approved 3 June 2019; effective 1 October 2019 OEHHA19

Ranges in brackets are 95% confidence intervals. The Lundh review covers drug and device studies, not food; the Lesser analysis covers soft drinks, juice and milk, not coffee.15,16 No study opened for this entry measured the effect of funding on coffee research specifically. The umbrella review’s authors found only one article that declared support from an organisation linked to the coffee industry.8

Limits of the absolute-risk example#

The worked example uses the simplest possible arithmetic. Three things make it rough:

  • Comparison groups. “High versus low” compares the top and bottom intake groups of each meta-analysis; the categories differ between studies.8
  • Baseline. The SEER lifetime risks describe the whole US population, coffee drinkers included.9,10 The risk among non-drinkers would differ.
  • Cause. Both relative risks come from observational data and may reflect confounding or reverse causation.8

A trial or review that reports absolute numbers directly (for example, events per 1,000 people in each group) avoids these steps.

Open questions#

  • Cup size in older cohorts. Long-running cohorts recorded cups, not milligrams.11 No study opened for this entry estimated how much this measurement variation changes the published risk estimates for coffee.
  • Trials for long-term outcomes. Randomised trials of coffee have been short8; trials long enough to test cancer or mortality have not been reported in the sources opened here.
  • Spin in nutrition news. The press-release studies covered biomedical research and drug trials in general3,18; no study opened for this entry measured exaggeration in coffee news specifically.

Key studies#

Sumner et al., 2014 – Exaggeration starts in the press release. Compared 462 health-related press releases from 20 UK universities with the papers they described and 668 resulting news stories. When press releases exaggerated advice, causal claims or animal-to-human inference, news stories were far more likely to exaggerate too, with little sign that exaggeration earned more coverage.

BMJ 349, g7015 · doi:10.1136/bmj.g7015

Yavchitz et al., 2012 – Spin in trial reports and news. Followed 70 randomised trials from abstract to press release to news. Spin was common in abstract conclusions and in press releases, and spin in the abstract was the only factor linked to spin in the press release.

PLoS Medicine 9(9), e1001308 · doi:10.1371/journal.pmed.1001308

Lesser et al., 2007 – Funding and nutrition conclusions. Classified 206 articles on soft drinks, juice and milk by funding source and by conclusion. Articles funded entirely by industry were much more likely to reach conclusions favourable to the sponsor than those with no industry funding.

PLoS Medicine 4(1), e5 · doi:10.1371/journal.pmed.0040005

Ludwig et al., 2014 – What is a cup? Measured caffeine and chlorogenic acids in espressos bought in Scotland, Italy and Spain. Content and volume varied widely between outlets, and the authors concluded that “one cup of coffee” is not a reproducible unit for epidemiology.

Food & Function 5, 1718–1726 · doi:10.1039/c4fo00290c

Poole et al., 2017 – Umbrella review of coffee and health. Gathered 201 observational and 17 interventional meta-analyses. Associations were mostly with lower risk, but the evidence was rated low or “very low” quality, and harmful associations largely disappeared with adequate adjustment for smoking.

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

General information, not medical advice.

Share or cite this entry

Change log

  • – The ICMJE disclosure-form requirement is now attributed to ICMJE member journals (other journals are encouraged to adopt it); the unsourced statement that disclosures are often headed 'Funding' and 'Competing interests' was removed. Reason: The ICMJE page says member journals require the form and other journals are encouraged to adopt it; it names no such headings.
  • – The California regulation is no longer said to be based on the IARC review; OEHHA's own wording is used (strongly supported by the research summarised by IARC). Reason: OEHHA's Final Statement of Reasons rests the regulation on the weight of the evidence, with IARC as a major but not sole basis.

Sources

  1. AGuyatt, G. H.; Oxman, A. D.; Kunz, R.; Vist, G. E.; Falck-Ytter, Y.; Schünemann, H. J.; GRADE Working Group (2008). What is “quality of evidence” and why is it important to clinicians?. BMJ 336(7651), 995–998. doi:10.1136/bmj.39490.551019.BE · other copies
  2. AOCEBM Levels of Evidence Working Group (2011). OCEBM Levels of Evidence (The Oxford Levels of Evidence 2). Oxford Centre for Evidence-Based Medicine, University of Oxford.
  3. ASumner, P.; Vivian-Griffiths, S.; Boivin, J.; Williams, A.; Venetis, C. A.; Davies, A.; Ogden, J.; Whelan, L.; Hughes, B.; Dalton, B.; Boy, F.; Chambers, C. D. (2014). The association between exaggeration in health related science news and academic press releases: retrospective observational study. BMJ 349, g7015. doi:10.1136/bmj.g7015 · other copies
  4. AHsieh, C. C.; MacMahon, B.; Yen, S.; Trichopoulos, D.; Warren, K.; Nardi, G. (1986). Coffee and pancreatic cancer (chapter 2). New England Journal of Medicine 315(9), 587–589 (abstract via AGRIS, FAO).
  5. AIARC Working Group on the Evaluation of Carcinogenic Risks to Humans (2018). Summary of data reported (Section 5). In: 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; pp. 414–415.
  6. ADavies, N. M.; Holmes, M. V.; Davey Smith, G. (2018). Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians. BMJ 362, k601. doi:10.1136/bmj.k601 · other copies
  7. 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
  8. 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
  9. ANational Cancer Institute, Surveillance, Epidemiology, and End Results Program (SEER) (2026). Cancer of Any Site – Cancer Stat Facts. National Cancer Institute (US), data 2021–2023.
  10. ANational Cancer Institute, Surveillance, Epidemiology, and End Results Program (SEER) (2026). Liver and Intrahepatic Bile Duct Cancer – Cancer Stat Facts. National Cancer Institute (US), data 2021–2023.
  11. ALudwig, I. A.; Mena, P.; Calani, L.; Cid, C.; Del Rio, D.; Lean, M. E. J.; Crozier, A. (2014). Variations in caffeine and chlorogenic acid contents of coffees: what are we drinking?. Food & Function 5, 1718–1726. doi:10.1039/c4fo00290c · other copies
  12. ACrozier, T. W. M.; Stalmach, A.; Lean, M. E. J.; Crozier, A. (2012). Espresso coffees, caffeine and chlorogenic acid intake: potential health implications. Food & Function 3, 30–33. doi:10.1039/C1FO10240K · other copies
  13. AInternational Agency for Research on Cancer (IARC) (2016). Debunking the myths of the Monographs: six common misconceptions dispelled. International Agency for Research on Cancer (WHO), Lyon, 15 June 2016.
  14. AHopewell, S.; Loudon, K.; Clarke, M. J.; Oxman, A. D.; Dickersin, K. (2009). Publication bias in clinical trials due to statistical significance or direction of trial results. Cochrane Database of Systematic Reviews, Issue 1, MR000006. doi:10.1002/14651858.MR000006.pub3 · other copies
  15. ALundh, A.; Lexchin, J.; Mintzes, B.; Schroll, J. B.; Bero, L. (2017). Industry sponsorship and research outcome. Cochrane Database of Systematic Reviews, MR000033. doi:10.1002/14651858.MR000033.pub3 · other copies
  16. ALesser, L. I.; Ebbeling, C. B.; Goozner, M.; Wypij, D.; Ludwig, D. S. (2007). Relationship between funding source and conclusion among nutrition-related scientific articles. PLoS Medicine 4(1), e5. doi:10.1371/journal.pmed.0040005 · other copies
  17. AInternational Committee of Medical Journal Editors (ICMJE) Disclosure of Financial and Non-Financial Relationships and Activities, and Conflicts of Interest (Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly Work in Medical Journals). International Committee of Medical Journal Editors.
  18. AYavchitz, A.; Boutron, I.; Bafeta, A.; Marroun, I.; Charles, P.; Mantz, J.; Ravaud, P. (2012). Misrepresentation of Randomized Controlled Trials in Press Releases and News Coverage: A Cohort Study. PLoS Medicine 9(9), e1001308. doi:10.1371/journal.pmed.1001308 · other copies
  19. ACalifornia Office of Environmental Health Hazard Assessment (OEHHA) (2019). Notice of Adoption – Section 25704 Exposures to Listed Chemicals in Coffee Posing No Significant Risk. OEHHA, 7 June 2019.
  20. ACalifornia Office of Environmental Health Hazard Assessment (OEHHA) (2019). Final Statement of Reasons – Title 27, California Code of Regulations, Section 25704: Exposures to Listed Chemicals in Coffee Posing No Significant Risk. OEHHA.

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 How to read coffee health news. Tell us what you would change and why. A source that supports it helps most.