Coffee has spent decades flipping between health food and health hazard. In 1991 the International Agency for Research on Cancer classed coffee as “possibly carcinogenic to humans”. In 2016 the same agency reviewed the evidence again and dropped it to Group 3, “not classifiable as to carcinogenicity”, because the human evidence had moved. That is one of the cleanest public examples of evidence changing a verdict. Since then, large reviews have generally trended toward benefit. The case for coffee is stronger than most people think. The case against is weaker than most people think. Neither is settled, and the honest answer has conditions attached.
What the large reviews actually found
The single most cited piece of evidence is a 2017 umbrella review in the BMJ by Poole and colleagues. It pooled 201 meta-analyses of observational research plus 17 interventional trials. Its headline findings: moderate coffee consumption, roughly three to four cups a day, was associated with lower risk of cardiovascular disease, type 2 diabetes, several cancers, and all-cause mortality, with the largest relative-risk reductions in the range of 5 to 25 percent depending on outcome. A 2024 umbrella review covering newer cohorts reached broadly similar conclusions for cardiovascular and metabolic outcomes. These are the reviews to cite in an argument, and it’s worth noting what they are: pooled observational data, where people self-report coffee habits over years. That design has known limits, which we will get to.
The strongest evidence: caffeine and hard outcomes
Solid interventional and large-cohort data exist for specific questions. A 2021 study in Circulation following over 380,000 people found that habitual coffee intake was associated with a 3 to 6 percent lower risk of developing atrial fibrillation, cutting directly against the folk belief that coffee causes heart-rhythm problems. Randomised trials have tested coffee against decaffeinated coffee in people with hypertension and found modest blood pressure effects at most. And the pregnancy question has a real ceiling: the American College of Obstetricians and Gynecologists recommends keeping caffeine under 200 mg per day during pregnancy, based on evidence linking higher intakes to miscarriage and low birth weight. That number is a genuine constraint, not a hedge.
The case against, stated properly
Three points survive scrutiny.
First, observational coffee research has a confounding problem. People who drink several cups a day differ from non-drinkers in income, shift work, smoking, and stress, and early studies that didn’t separate smokers from non-smokers produced inflated harm estimates. Later analyses that stratified by smoking status found the harm largely disappeared. But confounding cuts both ways: it inflated the harm, and healthier-user patterns may inflate the benefit. Neither direction of error has been fully eliminated.
Second, genetics. The CYP1A2 enzyme metabolises caffeine, and people carry different variants. A 2018 study of UK and Italian cohorts found slow caffeine metabolisers had higher risk of nonfatal heart attack at younger ages when they drank more coffee, while fast metabolisers didn’t. The sample sizes were moderate and the finding hasn’t been consistently replicated across populations, but it’s a real signal that “coffee is good for you” isn’t a species-wide statement.
Third, the cup isn’t just coffee. Acrylamide, a compound formed when coffee beans are roasted, appears on California’s Proposition 65 list, and coffee sellers in California faced labelling lawsuits over it. A 2019 ruling exempted coffee, on the basis that the acrylamide doses in coffee were far below those causing harm in animal studies and came alongside compounds with apparent protective effect. And what you add matters more than most headlines admit: the difference between black coffee and a large sweetened coffee drink is hundreds of calories and a large insulin hit, and studies that find “coffee associated with weight gain” are usually finding sugar.
The measured verdict
Weighing the evidence as a whole, on the same scale we use for every topic on this site:
- Coffee at three to four cups a day is associated with lower all-cause mortality in large cohorts, but the association design can’t fully exclude confounding. Confidence: moderate.
- For atrial fibrillation and routine blood pressure, coffee at normal intakes looks neutral to mildly favourable. Confidence: moderate.
- During pregnancy, cap caffeine at 200 mg per day. Confidence: high. This is the one recommendation with near-consensus backing.
- Slow caffeine metabolisers may face higher cardiovascular risk from heavy coffee intake. Confidence: low to moderate. Not actionable without a genetic test, and a genetic test isn’t justified by this alone.
- Unfiltered coffee (French press, Turkish, boiled) raises LDL cholesterol via cafestol and kahweol; paper filters remove most of the diterpenes. Confidence: moderate to high. If your LDL is high, this is the one preparation choice with a plausible mechanism behind it.
What this means for an actual decision
If you drink three cups a day and feel fine, the evidence doesn’t support quitting. If you’re pregnant, keep under 200 mg of caffeine, which is roughly one to two cups depending on size and strength. If your LDL cholesterol is high and you brew unfiltered coffee, switch to a paper filter before you switch to decaf. And if coffee makes you anxious or wrecks your sleep, no pooled relative risk overrides your own measured reaction; sleep loss has a larger measured health effect than any plausible coffee effect at these doses. For contrast, we covered a related pattern in our piece on vitamin D supplements, where a plausible mechanism also kept outliving its evidence, and in our look at ultra-processed foods, another case where the observational story is real but the mechanism is tangled. Coffee sits in the same drawer: a real signal, an incomplete mechanism, and a lot of confident people overstating both directions.