Across several very large cohort studies, coffee drinkers were recorded dying at somewhat lower rates than non-drinkers. That finding is real, replicated and easy to overstate. The curve is not a straight line, decaffeinated coffee showed much the same association as caffeinated, and the whole result nearly went the other way until researchers adjusted for smoking. This article covers what the cohorts found, where the dose-response turns, what decaf implies about mechanism, and the two places where the evidence gives a genuine limit: pregnancy and sleep.
What the large cohorts found — and the confounder that nearly reversed them
The NIH-AARP Diet and Health Study followed 402,260 US adults aged 50 to 71 from 1995 to 2008. In the age-adjusted analysis, coffee drinkers died at a higher rate than non-drinkers. Coffee drinkers also smoked considerably more. After adjustment for smoking and other confounders, the association reversed and became inverse: at six or more cups a day, the hazard ratio for death was 0.90 in men and 0.85 in women.
That reversal is worth sitting with. The direction of the headline finding depended on a statistical adjustment. It is a good result and a warning label at the same time, and the general problem is set out in our overview of diet and chronic disease.
Two UK Biobank analyses followed. Loftfield and colleagues, in JAMA Internal Medicine in 2018, followed 498,134 participants over about ten years and recorded 14,225 deaths. A 2022 analysis in the European Journal of Preventive Cardiology followed 468,629 participants for a median 11 years and recorded a hazard ratio of 0.88 (95% CI 0.83–0.92) for all-cause mortality at half a cup to three cups a day, compared with none.
Where the dose-response curve turns
The old version of this article carried the headline “as coffee drinking increases, risk of death decreases,” and told readers that coffee “didn’t stop being beneficial until beyond ten cups.” That is not what the data show, and we are withdrawing it.
The relationship is a shallow U, not a downward slope. In the JAMA Internal Medicine analysis the hazard ratios by daily intake were:
| Cups per day | Hazard ratio for death (95% CI) |
|---|---|
| Less than 1 | 0.94 (0.88–1.01) |
| 1 | 0.92 (0.87–0.97) |
| 2 to 3 | 0.88 (0.84–0.93) |
| 4 to 5 | 0.88 (0.83–0.93) |
| 6 to 7 | 0.84 (0.77–0.92) |
| 8 or more | 0.86 (0.77–0.95) |
The curve flattens somewhere around two to four cups and stops improving. The 2022 European analysis found no significant association at all above three cups a day after full adjustment. Reading these as “more is better” is the one conclusion the numbers do not support.
Timing may matter too. A 2025 analysis in the European Heart Journal of 40,725 NHANES participants recorded a hazard ratio of 0.84 for all-cause mortality in morning-only drinkers and 0.96 — not statistically distinguishable from non-drinkers — in people who drank coffee throughout the day. That is one observational study, and people who stop drinking coffee at noon may differ in other ways.
Decaf, and what it says about mechanism
The most informative finding in this literature gets the least attention. In the JAMA Internal Medicine analysis, decaffeinated and instant coffee showed inverse associations with mortality comparable to ground caffeinated coffee. The 2022 European analysis reported a hazard ratio of 0.84 for decaffeinated coffee at half a cup to three cups a day.
If caffeine were doing the work, decaf should not track the same way. Two readings remain open. Either the polyphenols and other non-caffeine compounds in the bean account for most of the association, or a good deal of what is being measured is the kind of person who drinks coffee rather than the coffee. Both explanations are consistent with the data, and neither has been tested in a randomized trial. No trial has randomized anyone to coffee and counted deaths, and none is likely to.
Caffeine sensitivity and CYP1A2
Caffeine is cleared mainly by the liver enzyme CYP1A2, and common variants in the CYP1A2 gene, along with smoking and oral contraceptive use, shift how fast an individual metabolizes it. Half-life in a typical healthy adult is around five hours, but the range across individuals runs from roughly two hours to more than twelve.
It would be reasonable to expect fast metabolizers to show a different mortality pattern from slow ones. They did not. The UK Biobank analysis built a genetic score for caffeine metabolism and found no evidence of effect modification across strata (P = 0.17 for heterogeneity) — the association looked similar regardless of genotype. That is another point against caffeine being the active ingredient, and it also means consumer gene tests marketed as telling you your “optimal” coffee dose are selling more precision than the evidence contains.
Subjective sensitivity is a different matter and is worth taking seriously. If two cups produce palpitations, anxiety or reflux, that is information about you, and no cohort study overrides it.
Pregnancy, and other real limits
The American College of Obstetricians and Gynecologists, in advice reaffirmed in 2023, states that moderate caffeine consumption — under 200 mg a day — does not appear to be a major contributing factor in miscarriage or preterm birth, and that a definitive conclusion cannot be drawn about higher intakes. Two hundred milligrams is roughly one 12-ounce brewed coffee, though café servings vary widely.
For other healthy adults the FDA cites 400 mg a day — about four or five cups of home-brewed coffee — as an amount not generally associated with negative effects. People taking certain medications, and anyone with an arrhythmia or an anxiety disorder, should raise it with their clinician rather than with a cohort study.
Two further limits. Unfiltered coffee — French press, boiled, some espresso preparations — retains diterpenes that filtered coffee removes, and a 2025 analysis in Nutrition, Metabolism and Cardiovascular Diseases measured markedly higher concentrations of these compounds in unfiltered and machine coffee than in paper-filtered brews. Diterpene intake has been associated with higher LDL cholesterol in controlled feeding studies. And what goes into the cup counts: a large sweetened coffee drink is a different food from black coffee, closer in composition to the products discussed in our piece on caffeinated energy drinks.
Sleep, half-life, and a sensible cut-off
Caffeine’s five-hour half-life is the useful number. A 200 mg afternoon coffee still has roughly 100 mg circulating five hours later and 50 mg after ten.
A 2025 randomized crossover trial in SLEEP put actual figures on this. Twenty-three healthy men received 100 mg or 400 mg of caffeine at 12, 8 or 4 hours before bed, with polysomnography. At 100 mg there were no significant effects on sleep at any timing. At 400 mg the effects were substantial: taken four hours before bed it cut total sleep time by about 51 minutes and deep sleep by about 30 minutes; taken eight hours before, it still increased wakefulness after sleep onset by 26 minutes; even at 12 hours it delayed sleep onset by about 15 minutes.
The authors’ guidance was that 100 mg can be taken up to four hours before bed, while 400 mg should not be taken within 12 hours of bedtime. In practice that means the size of the dose matters more than the hour on the clock — a single small coffee mid-afternoon is a different proposition from a large one. The trial ran in 23 young men, so the numbers are indicative rather than universal. For alternatives, see energy without stimulants.
A claim from this blog we are retracting
The 2013 version of this post ended by pointing readers toward a green coffee bean extract weight-loss supplement. The trial that launched that entire category was retracted in 2014, and the Federal Trade Commission obtained a $3.5 million settlement from its sponsor, Applied Food Sciences. The FTC found the study’s lead investigator had repeatedly altered subject weights and measurements, changed the trial length mid-study, and misreported which subjects received placebo. The agency described the trial as so flawed that no reliable conclusions could be drawn from it.
We should not have recommended it. Green coffee bean extract is a different substance from coffee, and nothing in the mortality literature above applies to it.
Common questions
How many cups a day is the sweet spot?
The cohort data flatten around two to four cups, and the 2022 UK Biobank analysis found no additional association above three. Nobody has run a trial, so “sweet spot” is a stronger word than the evidence deserves. If you drink one cup and feel fine, there is no basis for drinking more.
Is decaf pointless?
The opposite. Decaffeinated coffee showed inverse associations with mortality comparable to caffeinated coffee in the large UK Biobank analyses. For anyone sensitive to caffeine, or drinking coffee late, that is useful news.
Does adding sugar cancel the association?
A 2022 UK Biobank analysis in Annals of Internal Medicine separated unsweetened, sugar-sweetened and artificially sweetened coffee, and recorded inverse associations with mortality for both unsweetened and modestly sugar-sweetened coffee. What that will not survive is a 400-calorie flavored drink, which belongs in a different category of food entirely.
What time should I stop drinking coffee?
Work backward from dose. In the crossover trial, 100 mg four hours before bed did nothing measurable, while 400 mg still disturbed sleep at eight hours. A small coffee in the early afternoon is defensible; a large one after mid-afternoon is where the trial data start to object. Caffeine also acts as a mild diuretic, which is worth factoring in alongside ordinary hydration.
Where this leaves the cup
Coffee has been associated with lower recorded mortality in several very large cohorts, the association flattens rather than continuing to improve, and decaf tracks it closely enough to cast doubt on caffeine as the mechanism. None of that makes coffee a health intervention. It makes it a drink that the evidence does not ask anyone to give up.
The gap in most people’s week is vegetables, not coffee. If a concentrated greens powder is a practical way to close some of that, Greens Plus has been making them since 1993.
About the Author
