PIEZAS.

Science · Health

Three and a half cups a day: 15% lower mortality (with two caveats)

The big studies of recent years all point the same way: people who drink coffee in moderation die less over the years the studies follow them. Lower mortality, which is not living longer: none of them measures that. With two caveats.

For decades coffee had the reputation of a vice: bad for your heart, bad for your blood pressure, better give it up. The big studies say the opposite, and not timidly: people who drink around three and a half cups a day have, on average, slightly lower mortality over the years the studies follow them. Less of what? Above all, of what kills us most: the evidence links it to less cardiovascular disease (heart and stroke), less type 2 diabetes and less liver disease and liver cancer. The problem is not the coffee. It is what you put in it, and how far you take it.

This is an association, not a miracle: these studies follow a lot of people for years, they are not experiments that “prove” coffee extends life. But the signal turns up in large follow-up studies from the United States, Europe and Asia, with a warning from the meta-analysis itself: region is the one factor where the association is not consistent. The most cited is the meta-analysis by Kim, Je and Giovannucci (2019), which pooled 40 cohort studies and 3,852,651 people; for type 2 diabetes, Ding et al. (2014), with 28 studies and 1,109,272 people, and a later and larger one, by Carlström and Larsson (2018), with 30 studies and 1,185,210 people. What that science says, and how to tell whether your coffee is working for you or against you.

What the studies say

The sweet spot is around three and a half cups a day. The exact figure is worth giving, because the study draws a distinction: the meta-analysis by Kim, Je and Giovannucci puts the lowest risk of all-cause mortality at 3.5 cups a day (15% lower than for someone who drinks no coffee) and the lowest risk of cardiovascular mortality a little earlier, at 2.5 cups (17% lower). For type 2 diabetes, Ding et al. put the fall at 9% for each cup of caffeinated coffee and 6% for decaf; both falls are statistically significant, and the difference between them is not. And liver cancer risk also falls as intake rises.

Risk is not life expectancy. What Kim measures is the relative risk of dying during follow-up, not how long you live: that meta-analysis does not contain a single year or a single month. Turning that 15% into “you live X years longer” is a calculation the study does not allow.

The first caveat: the curve flattens out

More is not better. What the meta-analysis found is that, past that sweet spot, drinking more coffee no longer lowers the risk any further: the benefit stops growing. Watch out for the easy leap, which is where half the internet goes wrong: the benefit running out does not mean coffee starts doing harm, and the study sets no threshold beyond which the balance tips the other way. What does show up, through clinical experience rather than through this meta-analysis, are the known effects of too much caffeine (sleep, anxiety, blood pressure). And there are people the advice does not apply to: pregnant women and people with severe high blood pressure. The conclusion is not “drink coffee to live longer”; it is “if you already drink it in moderation, relax”.

Chart 1

How far the risk falls by cups of coffee a day

The drop in risk grows up to around 3.5 cups a day; beyond that, drinking more adds nothing.

Drop in mortality risk by cups of coffee a day The drop in risk grows from 0 cups to a peak at around 3.5 cups a day (optimal zone highlighted) and from there the line is flat: drinking more does not lower the risk any further. The meta-analysis does not find that the benefit is lost beyond that point, nor that coffee goes on to increase the risk. 0 2 4 6 8 SWEET SPOT CUPS OF COFFEE A DAY − risk same as no coffee

Source: based on Kim, Je and Giovannucci, European Journal of Epidemiology, 2019 (40 cohorts, 3,852,651 people, 450,256 deaths). A hand-drawn sketch of the shape of the association, not measured values; the risk minimum sits at 3.5 cups for all-cause mortality and 2.5 for cardiovascular mortality. From that minimum the curve runs flat, not falling: the meta-analysis says that beyond it “additional intakes were not associated with further lower mortality” —it stops growing, it is not lost— and its abstract gives no value above the minimum with which to support a fall.

The second caveat: what you put in it

This is where half the story falls apart. The benefit is measured on coffee, not on liquid desserts. But little has been measured about what goes in, and what has been measured does not agree: in Zhou’s US cohort (2025, 46,222 adults) the link with lower mortality shows up only in black coffee and in coffee with little added sugar and little saturated fat; in the UK Biobank (Liu, 2022, 171,616 people) coffee with a moderate amount of sugar is also linked to lower mortality. In Zhou’s study, according to a note from Tufts, the lead author’s university, syrup and honey count as added sugar, and milk and cream as added saturated fat; the fat threshold is set at 1 gram per eight-ounce cup — about five tablespoons of 2% milk or one of light cream, by the same note’s count — and the association with lower mortality shows up only below it. A black coffee is not a caramel frappuccino, but how much that changes the risk nobody has measured.

Chart 2

Fill your cup

Default: black coffee, regular size — coffee exactly as the studies measure it. Change the base, the milk, the extras and the size, and see what has been measured about each one and what has not.

Regular

Base
Milk
Extras
Size

Black coffee

Black coffee, no added sugar and no added fat: this is how the benefit is measured in the studies.

This cup does not score. No source measures how much each addition takes away, so there is no score here: only what has been measured about each thing and what has not.

An educational tool, not a medical diagnosis. The point is to understand what you are drinking, not to drink more or less of it.

Neither Kim 2019 nor Ding 2014 stratifies by what you add or by format, and Ding says so among its own limitations: “none of the studies assessed the amount of sugar and dairy added to coffee”. The notes on this cup record only what has actually been measured: Zhou 2025 (added sugar and saturated fat, with its threshold of one gram of added saturated fat per eight-ounce cup), Liu 2022 (sweetened coffee) and Tverdal 2020 (filtered versus unfiltered). On capsules and the size of the cup there is no study in this list that measures it. This is not medical advice.

Coffee stopped being a suspect without the coffee changing: what changed was what we knew how to measure. The useful question is no longer “is it good or bad?”, but “what am I putting in it?”.

Correction · 4 November 2026

Arrhythmias. Removed from the list of people the advice does not apply to. In the DECAF randomised trial (JAMA, January 2026), patients with atrial fibrillation who drank coffee had fewer recurrences than those who did not: 47% against 64%, in 200 patients, without blinding.

High blood pressure. It said “uncontrolled”. Teramoto et al. (2023), now in the source list, write about severe hypertension, grade 2 or 3, with two or more cups a day.

Shape of the curve. The subheading and the metadata said “inverted U”. Kim et al. (2019) describe a non-linear inverse association: the risk falls to the minimum and then stops falling.

Chart 1. The dots on the curve have been removed. It is a hand-drawn sketch: the abstract of Kim et al. gives no intermediate values.

Diabetes. It said the risk falls by “up to 9%” per cup, and the sentence on decaf could be read as saying its 6% was not significant. It falls by 9% per cup of caffeinated coffee and by 6% with decaf, and both falls are significant; what is not significant is the difference between them (P = 0.17).

Diabetes, reference. Ding et al. (2014) is no longer the newest or the largest meta-analysis. Carlström and Larsson (2018) has been added.

Syrup. It said there was no study. In Zhou’s study (2025), syrup and honey count as added sugar.

Tablespoons. The equivalence in tablespoons of milk and cream was attributed to Zhou’s paper. It comes from the Tufts University note on the study.

“The most recent data”. The Kim meta-analysis closed its search on 8 March 2019. It now reads “the big studies”.

“Recommended”. None of the sources in this piece recommends drinking coffee: EFSA sets caffeine limits and Poole et al. (2017) call for trials. Removed.

Format. The source list said Tverdal (2020) was the only thing found on format. Loftfield et al. (2018) and Chieng et al. (2022) also separate ground, instant and decaf coffee.

Cup. The buttons “Good coffee” and “Cheap capsule” become “Coffee” and “Capsule”: no study in the source list measures quality or price.

Description and standfirst. They said these were caveats “nobody tells you”. Tufts University reported them in June 2025. Removed.

Sources. Removed the entry for the Spanish Heart Foundation and specialist press, which linked to no document, and added the primary sources for pregnancy, high blood pressure and caffeine.

Cup notes. They said a capsule is filtered coffee, that milk had not been measured, that there was no study on syrup, that sugar was the only addition with studies and that decaf is measured by “the two sources in this piece”. No study in the source list measures capsules; in Zhou (2025), milk counts as added saturated fat and syrup as added sugar. The decaf note now names Kim (2019) and Ding (2014).

Sources for this correction, consulted on 4 November 2026: the PubMed abstracts of Kim et al. (2019), Ding et al. (2014), Carlström and Larsson (2018), Poole et al. (2017), Teramoto et al. (2023), Zhou et al. (2025), Loftfield et al. (2018), Chieng et al. (2022) and Wong et al. (2026, DECAF trial, DOI 10.1001/jama.2025.21056); the Tufts University note of 16 June 2025 on Zhou’s study; and EFSA’s page on caffeine.

Sources
  1. Kim Y, Je Y, Giovannucci E. “Coffee consumption and all-cause and cause-specific mortality: a meta-analysis by potential modifiers”. European Journal of Epidemiology, 2019; 34(8):731–752. DOI: 10.1007/s10654-019-00524-3 — primary source (40 cohort studies, 3,852,651 people).
  2. Ding M, et al. “Caffeinated and Decaffeinated Coffee Consumption and Risk of Type 2 Diabetes: A Systematic Review and a Dose-Response Meta-analysis”. Diabetes Care, 2014; 37(2):569–586. DOI: 10.2337/dc13-1203 — primary source (28 studies, 1,109,272 people).
  3. Carlström M, Larsson SC. “Coffee consumption and reduced risk of developing type 2 diabetes: a systematic review with meta-analysis”. Nutrition Reviews, 2018; 76(6):395–417. DOI: 10.1093/nutrit/nuy014 — primary source (30 studies, 1,185,210 people).
  4. Zhou B, Ruan M, Pan Y, Wang L, Zhang FF. “Coffee Consumption and Mortality among United States Adults: A Prospective Cohort Study”. The Journal of Nutrition, 2025; 155(7):2312–2321. DOI: 10.1016/j.tjnut.2025.05.004 — primary source (46,222 adults, NHANES 1999-2018). The only source found that measures the sugar and saturated fat added to coffee.
  5. Liu D, Li ZH, Shen D, et al. “Association of Sugar-Sweetened, Artificially Sweetened, and Unsweetened Coffee Consumption With All-Cause and Cause-Specific Mortality”. Annals of Internal Medicine, 2022; 175(7):909–917. DOI: 10.7326/M21-2977 — primary source (UK Biobank, 171,616 people). It points the opposite way to Zhou on sugar; both are cited.
  6. Tverdal A, Selmer R, Cohen JM, Thelle DS. “Coffee consumption and mortality from cardiovascular diseases and total mortality: Does the brewing method matter?”. European Journal of Preventive Cardiology, 2020; 27(18):1986–1993. DOI: 10.1177/2047487320914443 — primary source (508,747 people). On format: filtered versus unfiltered.
  7. Poole R, Kennedy OJ, Roderick P, Fallowfield JA, Hayes PC, Parkes J. “Coffee consumption and health: umbrella review of meta-analyses of multiple health outcomes”. BMJ, 2017; 359:j5024. DOI: 10.1136/bmj.j5024 — primary source (review of 201 observational meta-analyses). Source for the risk in pregnancy.
  8. Teramoto M, Yamagishi K, Muraki I, Tamakoshi A, Iso H. “Coffee and Green Tea Consumption and Cardiovascular Disease Mortality Among People With and Without Hypertension”. Journal of the American Heart Association, 2023; 12(2):e026477. DOI: 10.1161/JAHA.122.026477 — primary source (Japanese JACC cohort). Source for severe high blood pressure: two or more cups a day with grade 2 or 3 hypertension.
  9. EFSA, Panel on Dietetic Products, Nutrition and Allergies. “Scientific Opinion on the safety of caffeine”. EFSA Journal, 2015; 13(5):4102. DOI: 10.2903/j.efsa.2015.4102 — primary source. Caffeine limits for adults and in pregnancy.