For years, coffee research has revolved around quantity: none, one cup, three cups, perhaps too many cups. A 2025 study in the European Heart Journal added a more everyday question. Does it matter whether those cups arrive before lunch or keep appearing until evening?

The answer was intriguing, but it was not a set of instructions. This article explains one observational study, not settled consensus or medical advice, and the research did not show that moving an afternoon coffee to breakfast will prevent cardiovascular disease.

That boundary is especially important here because “31% lower” sounds like a simple count. It was actually a relative, adjusted estimate from a statistical model, and the raw event numbers tell a much messier story.

The study by Xuan Wang and colleagues analyzed 40,725 US adults in the National Health and Nutrition Examination Survey, or NHANES, from 1999 through 2018. People reported everything they had eaten and drunk during one or two 24-hour periods. That record included the type, amount and time of each coffee.

The researchers split the clock into three windows: morning from 4 a.m. through 11:59 a.m., afternoon from noon through 4:59 p.m., and evening from 5 p.m. through 3:59 a.m. A clustering analysis then found two broad habits among coffee drinkers. In the “morning type,” consumption was concentrated before noon and was scarce later. In the “all-day type,” it was spread across all three windows.

There were 14,643 morning-type drinkers, 6,489 all-day drinkers and 19,593 people who reported no coffee. The researchers also reproduced the two timing patterns in 1,463 participants from separate validation studies who kept much more detailed seven-day food and drink records. That supports the idea that the clusters captured recognizable habits rather than a quirk of one survey day.

During a median 9.8 years of follow-up, 4,295 participants died. Of those deaths, 1,268 were attributed to cardiovascular disease and 934 to cancer.

After adjusting for a long list of differences, the morning pattern had a cardiovascular mortality hazard ratio of 0.69 compared with drinking no coffee. Subtracting that figure from 1 gives the headline’s 31 percent relative reduction. The 95 percent confidence interval ran from 0.55 to 0.87, so the estimate was statistically significant under the study’s model.

The adjustments included age, sex, race and ethnicity, survey cycle, income, education, body mass index, diabetes, hypertension, high cholesterol, smoking, physical activity, overall diet quality, calories, caffeinated and decaffeinated coffee amounts, tea, caffeinated soda, short sleep and trouble sleeping. For death from any cause, the morning group’s adjusted hazard was 16 percent lower than that of non-drinkers.

The all-day group produced a cardiovascular hazard ratio of 0.82, with a confidence interval from 0.61 to 1.10. Its point estimate was below 1, but the uncertainty range included no difference. That is what “no significant reduction” means here.

Look only at the event counts and the headline seems to unravel. Cardiovascular deaths occurred in 536 of the 14,643 morning drinkers, 274 of the 6,489 all-day drinkers and 458 of the 19,593 non-drinkers. Those are crude proportions of about 3.7, 4.2 and 2.3 percent, respectively.

The obvious reason is age. At baseline, non-drinkers averaged 38.5 years old, while both coffee-drinking groups averaged just over 50. An older group has more time-proximate cardiovascular risk before coffee enters the picture. The hazard model tried to compare more similar people and account for follow-up time, which is why its adjusted estimate can point in a different direction from the raw totals.

This is a useful demonstration of both the value and the limit of adjustment. A good model can reduce unfair comparisons, but it cannot turn an observational survey into a randomized experiment. It can only adjust for factors the researchers measured, and only as accurately as those factors were recorded.

The authors specifically noted that they lacked information about shift work and usual waking time. A hospital worker drinking coffee at 6 p.m. before a night shift is not using caffeine in the same biological context as someone drinking it after dinner. Work schedules, illness, sleep habits and reasons for avoiding coffee could all leave residual differences between the groups.

It is tempting to read the paper as a contest that morning coffee won and all-day coffee lost. Statistics are less tidy.

The morning pattern was significantly associated with lower cardiovascular mortality when each group was compared with non-drinkers. The all-day pattern was not. But “significant in one group and not significant in another” does not automatically prove that the two groups differ significantly from each other. In the study’s analysis of coffee amount and cardiovascular mortality, the interaction with timing did not reach statistical significance.

So the evidence supports a cautious sentence: morning-concentrated coffee was associated with lower cardiovascular mortality than no coffee, while an association for the all-day pattern remained uncertain. It does not support saying that later coffee was harmful, that it erased a proven benefit, or that changing the clock alone will reproduce the estimate.

The researchers proposed two possible explanations. First, caffeine later in the day can push against the body’s circadian system. They cited an earlier trial in which heavy afternoon and evening coffee consumption was linked to a roughly 30 percent reduction in peak nighttime melatonin. In theory, repeated disruption of sleep and circadian timing could influence blood pressure, inflammation and other cardiovascular processes.

Second, coffee contains bioactive compounds with anti-inflammatory effects. Some inflammatory signals also rise and fall across the day, often peaking in the morning. The authors wondered whether delivering coffee’s compounds during that peak might matter.

Both ideas are mechanisms to test, not explanations the study demonstrated. In fact, the researchers found broadly similar timing patterns when they analyzed caffeinated and decaffeinated coffee separately. That makes a caffeine-only story too simple and leaves open the possibility that timing is partly a marker for some other feature of daily life.

“Coffee” is not a uniform exposure. An eight-ounce filter coffee, a large sweetened café drink, a double espresso and a mug of decaf can differ sharply in caffeine, calories and other compounds. Preparation matters too. A later review in the same journal explains that paper filtering removes much of the cafestol and kahweol found in unfiltered coffee, two compounds known to raise LDL cholesterol.

The timing study adjusted for how much caffeinated and decaffeinated coffee participants reported, but it still relied on one or two recalled days for the main cohort. It could identify population patterns; it could not capture every change in cup size, brewing method, additions or schedule across the following decade.

A stronger next step would randomly assign regular coffee drinkers to keep the same coffee and quantity while changing only the timing. Researchers could then track sleep, blood pressure, glucose regulation and inflammatory markers before asking the much harder, longer-term question about heart attacks, strokes and deaths. The European Society of Cardiology’s summary likewise notes that a clinical trial is needed to test whether changing coffee time changes risk.

Anyone considering caffeine in the context of pregnancy, breastfeeding, medication, a health condition, palpitations or persistent sleep trouble should ask a qualified health professional, not treat this cohort result as personalized guidance. The US Food and Drug Administration says sensitivity and the speed at which people clear caffeine vary widely, even though up to 400 milligrams a day is not generally associated with negative effects for most adults.

The genuinely interesting lesson is broader than breakfast coffee. Nutrition studies usually focus on what and how much. This one suggests that when may carry information too, while reminding us that an association on the clock is still several careful experiments away from becoming advice.