← Longevity & Supplement Guides
Coffee and longevity: a permission slip, not a prescription
Somewhere between the second sip and the bottom of the cup, a headline arrives: coffee adds two years to your life. It is a good headline. It has been a good headline for about fifteen years now, in rotation, with the number nudged up or down depending on whose cohort is talking.
Coffee is the most-studied object in most people's kitchens. Nearly half a million adults have been followed for a decade with their coffee habits on file. And after all of it, the literature still cannot answer the only question you were actually asking, which is whether the cup in your hand is doing anything for you.
What it can do — and this turns out to be more useful than it sounds — is tell you to stop asking.
Half a million people, and the curve barely bends back up
In 2018, Erikka Loftfield and colleagues published the coffee study that most others still get measured against: 498,134 UK Biobank participants, 3.4 million person-years, 14,225 deaths. Compared with non-drinkers, coffee drinkers died at lower rates at every level of intake — a hazard ratio of 0.92 at one cup a day, 0.88 at two to three, 0.84 at six to seven, and 0.86 among people drinking eight or more.
That last number is the one that gets attention. Eight cups a day is not moderation by anybody's definition, and the association declined to punish it.
A US cohort later converted the same shape into years. Using NHANES data on 43,114 adults followed for a median of 8.7 years, researchers estimated life expectancy at age 50 at 30.1 years for non-drinkers and 32.1 years for people drinking one to two cups a day. That gap is the two-year figure the headlines use.
Both findings are real, in the sense that the arithmetic is sound and the samples are enormous. Neither one says what a reader assumes it says.
Then the same paper quietly dismantled its own headline
Loftfield's study did something most coffee papers never bother with. It went looking for the mechanism — and failed to find it.
Two results, buried in the tables. First: decaffeinated coffee showed inverse associations too. Second, and more damaging: the researchers scored every participant on the genetic variants that govern how quickly they clear caffeine, and the score did not modify the association at all (P = 0.17 for heterogeneity). Slow metabolisers and fast metabolisers — people who get meaningfully different caffeine exposure out of an identical cup — showed comparable results.
If caffeine were doing the work, that is precisely the test it should have failed. It didn't fail it. It came back blank.
There are two honest ways to read a blank. Either the active ingredients are among the hundreds of non-caffeine compounds in coffee — the chlorogenic acids and their relatives, which decaf keeps. Or nothing in the cup is doing the work at all, and what we are looking at is the people, not the drink.
What happens when you let the genes randomise for you
Nutrition cannot randomise people to fifteen years of coffee. Genetics runs a version of that trial by accident. The variants that nudge someone toward drinking more coffee are dealt at conception — before income, before employment, before illness — so they arrive unentangled from the things that usually confound a diet study. Comparing outcomes across those variants is called Mendelian randomisation, and it is the closest thing nutrition science has to an unrigged coin.
Coffee has been through it repeatedly. A 2022 review in the European Journal of Nutrition lined the two literatures up side by side: observational studies put hazard ratios for all-cause and cardiovascular mortality at 0.85 to 0.90 in high consumers, while the Mendelian randomisation studies offered no support for causality. The same split held across the cardiometabolic outcomes where the observational coffee findings look strongest.
Mendelian randomisation is not a verdict machine. The genetic instruments for coffee intake are weak, they model a lifelong tendency rather than what you drank last year, and a null result can conceal a small effect. But when a signal this large in observation collapses this completely under randomisation, the honest word is not "proven safe" and not "proven useless." It is unconvicted.
The dullest explanation is usually the right one
Here is the sentence nobody puts in a headline. People stop drinking coffee when they feel unwell.
Reflux, palpitations, a stomach that turned, a night that wouldn't come, a pregnancy, a diagnosis, a new medication, an appetite that quietly went. Every one of those routes a person out of the coffee column and into the non-drinker column — and drags their subsequent mortality along with them. The reference group in every coffee study is quietly enriched with people who were already becoming ill. Epidemiologists call it reverse causation, and it flatters coffee for free.
Then the other half of it. Drinking coffee tracks having a morning: a job, colleagues, a routine, somewhere to be at nine. Loftfield's cohort had a participation rate of roughly 5.5% — the people who volunteer for a decade-long biobank are not a random slice of Britain. Adjusting for smoking and household income does not unmake that.
The wellness economy has a standing appetite for a permission slip, and coffee is the most agreeable one on the shelf: a habit almost everyone already has, attached to a literature that keeps voting yes. Which is exactly the condition under which it is worth checking who is counting the votes.
The evidence
One association, four tests
Large cohorts
498,134 adults, 14,225 deaths, 3.4 million person-years. Lower mortality at every intake level, including 8 or more cups a day.
Decaffeinated coffee
Decaf carried inverse associations too. If caffeine were the active ingredient, it should not have.
Caffeine-metabolism genotype
Slow and fast metabolisers showed comparable associations (P = 0.17 for heterogeneity).
Mendelian randomisation
Genetically predicted coffee intake gives no support for a causal effect on mortality.
The timing study is the same trap wearing a watch
Coffee epidemiology's most recent hit is about when rather than how much. In the European Heart Journal in 2025, a group analysed 40,725 US adults over a median 9.8 years and identified two drinking patterns: a morning type (36% of participants) and an all-day type (14%). Against non-drinkers, the morning type carried a hazard ratio of 0.84 for all-cause mortality and 0.69 for cardiovascular mortality. The all-day type carried neither. The dose-response — more coffee, lower risk — appeared only in the morning group.
It is a genuinely interesting result, and it was reported as a rule: drink your coffee before noon.
Months later the same journal ran a critical perspective on it. The central objection is the one you can now see coming: morning-only coffee drinking is a marker of a regular life. It travels with fixed wake times, structured days, and people who are not drinking coffee at nine at night in order to stay upright. The commentary also noted that the study never examined caffeine-metabolism genes, and that adjusting for household income does not capture socioeconomic position.
What survives that critique is not a rule about coffee. It is that regular timing tracks better outcomes than scattered timing — a finding we already hold with far more direct evidence in sleep regularity and in the science of circadian rhythm.
What is in the cup is the bigger variable
The NHANES life-expectancy analysis split its results by added sugar, and the split is stark. Among people who sweetened their coffee, no significant association appeared at any level of intake. Among people who did not, one to two cups a day carried the two-year estimate.
Resist the obvious conclusion. This is not evidence that black coffee extends life; it is the same observational data with the same confounding, sliced thinner. What it does hint at is where the nutritionally interesting variable actually sits. Coffee itself is a near-zero-calorie infusion. The syrup, the sugar and the cream are food — consumed daily, often entirely unaccounted for. If any part of a coffee habit deserves attention, it is more likely the part you add than the part you brew, which is the same lesson as glucose spikes and ultra-processed food arriving through a different door.
The one part of this that is not observational
Everything above is association. One thing about coffee is not.
Caffeine is a drug with a half-life of roughly five hours, a known action at adenosine receptors, and a randomised-trial evidence base. In the EU the authorised claim is narrow and accurate: caffeine helps increase alertness and helps improve concentration, at 75 mg — about one espresso. That is essentially the whole list of what caffeine is permitted to say about itself in Europe, and it is the one part built on experiments rather than cohorts.
The flip side is equally causal and much more personal. Caffeine's effect on sleep has been measured in controlled trials, it depends on both dose and timing, and it is where a coffee habit can genuinely cost you something — which we cover in caffeine and sleep. The epidemiology cannot see this at all. It is averaged away. And if you are pregnant, taking medication, or have been told to watch your caffeine for any reason at all, that is a conversation with your doctor rather than with a cohort study.
Run the experiment epidemiology cannot run for you
Cohorts answer the question "what happened to a population." You are asking "what is this doing to me." Those are different questions, and only one of them is answerable this month.
Hold everything constant for two weeks — same coffee, same amount, same type — and change one variable: move your last cup to at least six hours before you intend to be asleep. Then read three things against your own baseline rather than against a population: how long it takes you to fall asleep, your overnight resting heart rate, and the direction of your HRV trend.
The honest caveat, because it matters: you cannot blind yourself, expectation moves all three of those numbers, and a single bad night proves nothing. Read the two-week trend, not Tuesday. Trends are what the Agen Band is for, and it is worth knowing what resting heart rate and HRV actually mean — and what a wearable can and cannot measure — before you interpret either.
If nothing moves, you have your answer, and it is a good one: your coffee is not costing you your nights. If your time-to-sleep drops by twenty minutes and your overnight heart rate follows it down, you have learned something about yourself that 498,134 strangers could not tell you.
The bottom line
Coffee has survived more attempts to convict it than almost any other item in the human diet — across half a million people, eight cups a day, and every genotype anyone thought to test. That is a genuinely reassuring result, and it deserves to be said plainly: if you like coffee, the evidence gives you no reason to stop.
It also gives you no reason to start. The association is large, consistent, and unsupported by every method built to test causation — which is the signature of a habit that marks healthy people rather than one that makes them. It is a permission slip, not a prescription, and the two get filed in different drawers.
So drink it because you like it, watch what it does to your sleep, and spend your attention on the parts of a longevity routine that survived randomisation. There are fewer of them than the headlines imply. That, in the end, is the useful finding.
