Beverages
Coffee has the most consistent good news of any drink; the old claim that a drink a day protects the heart has collapsed nearly everywhere it was tested; and tea sits between them with a smaller, less certain positive signal. Water matters more than its folklore suggests, both for how fast you seem to age and for what is in the bottle, and the one habit that cuts across coffee and tea is letting a hot drink cool.
If you enjoy coffee, 2–4 cups a day is a sound habit, and 2–3 cups of tea likewise; let any hot drink cool below about 60 °C, which for a fresh pour means about six minutes; drink water steadily and to thirst; and treat alcohol as a cost, with the lowest-risk amount at or near zero. Almost all of the evidence here is observational — large cohorts, plus genetic studies that use inherited gene variants as a natural experiment (Mendelian randomisation) — so it shows associations; the randomised trials — on drinking more water, coffee and heart rhythm, caffeine and sleep — are named where they appear.
Water: high-normal blood sodium is the warning sign
People whose blood sodium sits at the high end of the normal range look biologically older and get more chronic disease, and habitually drinking too little is the most likely driver, though the studies cannot prove it. In a US cohort of 15,752 adults followed for 25 years, sodium above 142 mmol/L went with about 10–15% higher odds of being biologically older than chronological age, rising to roughly 50% only above 144 mmol/L, and with roughly 39% higher risk of a combined measure of heart failure, dementia, chronic lung disease and stroke; above 144, mortality ran about 21% higher, an estimate that only just clears no effect.[1] Nearly all of this sits inside 135–145 mmol/L, the range a laboratory reports as normal.
The same research group found the pattern again in Israel. In 407,187 adults healthy at baseline, sodium of 140–142 mmol/L carried about 13% higher risk of later high blood pressure, rising to about 29% above 143, with heart-failure risk about 20% higher in that top band.[2] Pooling an Israeli population with the US cohort, dementia risk started climbing around 141 mmol/L, running 1.5- to 3-fold higher at the top of the range and above it.[3] That corroborates rather than independently replicates, and the two cohorts put the turning point in different places, so read the low 140s as a grey zone rather than a safe plateau. Sodium is a proxy for habitual intake, not a readout of it, and the two trials that tested whether drinking more helps an already-diseased kidney were both null.
Trials of simply drinking more have found little. Of 18 randomised trials, only those on kidney stones and weight loss had more than one trial pointing the same way, recurrent urinary-tract infections have one strong positive trial (against one null four-day study), and eight of the 18 were negative overall. Migraine rests on two trials, the larger leaving headache days unchanged, and the one blood-pressure trial found pressure rose, not fell.[4]
Kidney stones: probably helps if you already form them. Randomising 199 first-time stone formers to a high-water target halved five-year recurrence, 12% against 27%, and delayed the stones that did recur by about a year.[5] The largest trial since, coaching 1,658 stone formers with low urine output to drink more, found no difference over two years — about one in five had a new stone either way — though intake rose in both arms, diluting the contrast.[6] There is no randomised evidence that drinking more prevents a first stone.[7]
Drink steadily, and let thirst and urine colour be the check. Spread drinks through the day until your urine stays pale and you rarely feel thirsty; for most healthy adults under 65 that comes to roughly 2–3 L of total fluid a day, food included, and the guide is those signals rather than a target volume. Past 65, both become unreliable, so drink on a schedule instead. During endurance exercise the rule flips to drinking to thirst, not ahead of it: overdrinking dilutes blood sodium dangerously and has been linked to at least 14 athlete deaths since 1981; because it is caused by fluid volume rather than salt loss, sports drinks do not prevent it. The international consensus — a strong recommendation on evidence its panel grades low — is that thirst prevents both dehydration and over-hydration.[8] Sipping may beat chugging: in a heat-chamber trial of eight people, metered sipping kept about 75% of the fluid against about 55% for the same volume drunk as a bolus.[9] A sensible habit; the numbers are loose.
What's in the water, and which filter
Microplastics and the smaller nanoplastics are in practically all tap and bottled water. Improved detection puts bottled water at about 240,000 plastic particles per litre, around 90% of them nanoplastics, though a published critique of the control blanks means the absolute number is best read as directional.[10] Particles have been reported in human tissue, including placenta; the concerns — crossing the gut lining, local inflammation, the hormone-disrupting plasticisers they carry — come from animal and cell studies, and harm at everyday doses is unproven in people.
PFAS (per- and polyfluoroalkyl substances), near-indestructible carbon-fluorine "forever chemicals", accumulate in groundwater and human blood, with links to immunosuppression, thyroid disruption, abnormal blood lipids and reproductive harm.[11] In late 2023 the World Health Organization's cancer agency, the International Agency for Research on Cancer (IARC), classified perfluorooctanoic acid (PFOA) as a Group 1 human carcinogen, its highest tier — a hazard classification, which grades what the chemical can do rather than the risk at everyday drinking-water levels.[12]
Choose a filter by what your water report shows. Reverse osmosis removes most of both, but also strips the minerals; add a remineralisation stage, because mineral-stripped water works against the stroke signal linked to magnesium in drinking water.
Water covers the sodium evidence in full, the older contaminants (arsenic, nitrate, chlorination byproducts), filter choices, electrolytes, functional waters, and dehydration past 65.
Coffee: the most consistent good news, still mostly observational
Coffee drinkers die a little less often, and the pattern is unusually consistent. Pooling 40 cohort studies of roughly 3.85 million adults, mortality was lowest at 3.5 cups a day, about 15% lower;[13] an umbrella review of 201 meta-analyses put the low point at 3 cups.[14] In UK Biobank, among 449,000 adults, the greatest reduction sat at 2–3 cups for ground (27% lower), instant (11%) and decaffeinated (14%) coffee,[15] and in 171,616 adults unsweetened coffee went with up to about 29% lower mortality, largest between roughly 2.5 and 4.5 cups.[16] Genetic studies often fail to back this up, but decaf shares most of it, which suggests polyphenols such as chlorogenic acids, not just caffeine, do some of the work.
The disease-specific associations point the same way. Across 1.28 million adults, cardiovascular risk was lowest at 3–5 cups a day.[17] Across 28 prospective studies of over 1.1 million adults, coffee tracked with lower risk of type 2 diabetes; per daily cup, in the studies that separated the two, about 9% lower for caffeinated and about 6% for decaf.[18] Each extra 2 cups a day tracked with about 35% lower risk of hepatocellular carcinoma, the main form of primary liver cancer,[19] and the signal extends to cirrhosis and to scarring in fatty liver disease. Parkinson's disease has the most consistent inverse association, roughly 28% lower risk at around 3 cups a day in a dose-response meta-analysis of about 900,000 people,[20] but genetic studies find no causal effect on risk, though they may be underpowered.[21]
The brain is where caffeine itself seems to matter, and where the evidence is most contested. In two large US health-professional cohorts, 2–3 cups a day of caffeinated coffee went with 18% lower dementia risk, while decaf showed no association.[22] A pooled analysis of other cohorts found no protective effect of moderate coffee, so treat the dementia link as unsettled.
What goes in the cup matters. In UK Biobank a modest spoonful of sugar, about 4 g per cup, tracked with much the same mortality benefit as unsweetened coffee, though for diabetes even added sugar roughly halved the favourable link; it is the sugary café drink, at 200–500 kcal, that likely cancels the favourable signal. Paper filtration removes the coffee oils, chiefly cafestol, that raise LDL (low-density lipoprotein) cholesterol slightly, which matters if yours is already high. Two consumer worries can be set down: ochratoxin A from coffee runs at roughly 2% of the tolerable threshold even at 4 cups a day, and the acrylamide formed in roasting causes cancer in animals but has not been shown to in people, most recently in a 2022 meta-analysis of 1.15 million.
Coffee and heart rhythm: the one place with real trials
The advice to give up coffee if you have atrial fibrillation is not supported by the one randomised test of it. In the DECAF trial, 200 current or recent coffee drinkers whose heart rhythm had just been reset electrically either kept drinking at least a cup a day of caffeinated coffee or gave up caffeine. Recurrence was 47% against 64%, about a 39% lower rate (hazard ratio 0.61, 95% confidence interval 0.42–0.89 — the range where the true effect most plausibly lies, here entirely below 1.0, so chance alone is an unlikely explanation).[23] It was a single open-label trial, and trials this small tend to overstate effects, so trust the direction more than the size: quitting does not appear to help an established drinker, which is not the same as coffee treating the arrhythmia. The 2023 American heart-rhythm guideline had already rated caffeine abstention of no benefit for preventing recurrence.[24] If you reliably notice palpitations after coffee, trust that.
The second trial is the counterweight. In 100 adults wearing continuous monitors who drank or avoided coffee on randomly assigned days, extra beats from the upper chambers were essentially unchanged, but extra beats from the lower chambers (ventricles) rose by about half — a secondary result, concentrated above one cup a day — and people slept about 36 minutes less on coffee days.[25]
Keep the caffeine under 400 mg a day, and early. Europe's food-safety authority and the US Food and Drug Administration both put the safe limit for healthy adults at 400 mg; in pregnancy the European figure is 200 mg, and the FDA sets none and defers to your clinician. The cut-off before bed depends on the dose: 400 mg, a full day's allowance in one go, disturbed sleep when taken within 12 hours of bedtime and fragmented it within 8, while 100 mg, about one cup, had no measurable effect even 4 hours before bed.[26] Don't trust your own sense of it: in an earlier trial, 400 mg six hours before bed cut sleep by more than an hour without the subjects noticing.[27]
Coffee covers the dose-response by outcome, blood pressure, cancer, brewing methods, timing, and who should be careful, including pregnancy, medicines and slow metabolisers.
Tea: a smaller signal in the same direction
Tea is probably mildly good for you, with smaller effects than coffee's and a thinner evidence base, almost all observational. In UK Biobank, about 498,000 adults mostly drinking black tea, 2 or more cups a day went with 9–13% lower all-cause mortality, the largest gap at 2–3 cups. The link held regardless of milk, tea temperature or caffeine-metabolism genotype, a hint that polyphenols rather than caffeine do the work; whether added sugar matters is disputed.[28]
Dementia is tea's best result. In the same two US cohorts, the third who drank the most tea, at a median of about one cup a day, had about 14% lower dementia risk than the third who drank least (hazard ratio 0.86, 95% confidence interval 0.83–0.90, entirely below 1.0); the larger 18% figure often quoted belongs to caffeinated coffee.[29] For coronary heart disease the two teas run different curves: black tea's association keeps strengthening across the range studied, while green tea's is best at two to three large cups a day and fades beyond that. Genetic studies of tea and cardiovascular disease conflict with one another, so some of the signal may be confounding.
The rest is plausible rather than shown. A 2025 meta-analysis covering roughly 560,000 adults found tea drinkers had about 25% lower odds of osteoporosis, shrinking to about 13% in the stronger prospective studies, and coffee drinkers scored much the same — a hint that the drinkers rather than the drink do some of the work; no trial has measured fractures.[30] In women aged 65 and over, tea drinkers' hip bone density was higher, but only slightly.[31] The idea that green tea feeds the gut bacterium Akkermansia muciniphila, linked to metabolic health, rests on animal and observational data. And the claim that green-tea drinkers age "more successfully" comes from a one-time snapshot of about 3,300 Greek adults over 50, in which the green-tea drinkers were also more active and less hypertensive, and neither difference was adjusted for.[32]
Concentrated green tea extract is the exception. The pills have caused rare liver injury serious enough to need a transplant; brewed tea is generally safe, with only rare reactions.
Tea covers green versus black, L-theanine, blood pressure and cholesterol, cancer, medicine interactions, fluoride and the plastics in tea bags.
Iron: tea or coffee before the meal, not after
Both drinks inhibit absorption of non-heme iron, the plant-derived iron that makes up most of the diet, tea substantially more than coffee. The timing fix runs in one direction only, which is where most advice goes wrong: drink before the meal, not after. Tea drunk an hour after a meal cut inhibition from about 37% to 18% — it roughly halves the effect rather than abolishing it.[33] The effect is local to the gut, not a drain on stored iron, so it matters only for people whose iron is marginal: menstruating women, plant-based eaters and anyone on iron therapy.
Very hot drinks: the risk is the temperature
The one cancer caution that applies to coffee and tea alike is about heat, not the drink. In a cohort of about 50,000 adults whose tea temperature was measured rather than self-reported, drinking at 60 °C or hotter carried 41% higher risk of oesophageal squamous-cell cancer, rising to roughly 90% higher at high daily volumes.[34] The World Health Organization's cancer agency draws its line a little higher, classifying the drinking of beverages above 65 °C as probably carcinogenic, on limited human evidence.[35] That evidence comes mostly from regions where this cancer is far more common than in Europe or the US, but typical Western drinking temperatures land around 56–60 °C, at the boundary rather than safely below it. Letting a fresh pour stand about six minutes is the cheapest risk reduction on this page. Very hot drinks covers the evidence.
Alcohol: the heart benefit has largely collapsed
For 30 years, cohorts showed light drinkers outliving abstainers — a J-shaped curve credited to the heart. The flaw was a comparison group that included many former drinkers who had quit because they were already ill: the sick-quitter effect. Bias-corrected pooling of 107 cohort studies, 4.8 million people, reproduced the J-curve in its raw numbers and lost it once former drinkers were separated from lifelong abstainers: low-volume drinkers showed no mortality advantage, and risk rose significantly from about 45 g a day in men and 25 g in women.[36]
Genetic studies point the same way. In the China Kadoorie Biobank, which enrolled 512,715 adults and genotyped 161,498 men, any genetically predicted increase in drinking among those men raised blood pressure and stroke incidence linearly, with no protective threshold — though it showed no significant effect on heart attack, the endpoint the J-curve was built on.[37] A 94-variant analysis found about 27% higher stroke risk and roughly triple the risk of peripheral artery disease in people genetically inclined to drink substantially more.[38] In UK Biobank, each additional daily drink was causally linked to 2.65 mmHg higher systolic blood pressure, roughly double the risk of bleeding stroke and about 26% higher atrial-fibrillation risk.[39] Among US veterans of African, European and Hispanic ancestry, the U-shape appeared in the ordinary analysis and vanished in the genetic one — no benefit and no harm, the signature of confounding.[40]
The argument is not fully over. A 2025 US National Academies review pooled eight studies and found moderate drinkers had about 16% lower all-cause mortality than never-drinkers, graded moderate certainty.[41] Every study in it was observational, and its never-drinkers misreport their own history and differ from drinkers in ways no cohort fully adjusts away. Heart attack is the one endpoint where genetic studies still fail to show harm, though no large effect either way survives there, and almost none has tested for a J-shape. Enough to keep the argument alive; not enough to make drinking a health behaviour.
Alcohol and cancer: harm from the first drink
Alcohol is a Group 1 carcinogen, and the cancer evidence is the clearest harm. A meta-analysis of 572 studies and 486,538 cancer cases found risk climbing steadily with dose and no protective dip at the bottom: breast, mouth and oesophageal cancer risk was already raised at about one drink a day, with bowel and liver cancer rising as intake climbs.[42] Alcohol accounts for about 4% of cancers worldwide, roughly 741,000 cases in 2020; about 103,000 of them arose in people drinking under 20 g a day, under two drinks, so roughly one in seven occurs at intakes most people would not call heavy drinking.[43]
The main culprit is acetaldehyde, the first product of breaking down alcohol. An East Asian enzyme variant that slows its clearance, and makes the face flush, is close to a natural experiment: carriers who drink develop far more mouth, throat and oesophageal cancer at the same intake.[44] If you flush when you drink, that is a warning sign.
The US Surgeon General's January 2025 advisory named alcohol the third leading preventable cause of cancer in the US, after tobacco and obesity.[45] The World Health Organization's January 2023 headline was that no level of drinking is safe;[46] the peer-reviewed statement behind it says, more carefully, that no safe amount for cancers and health "can be established."[47] Even moderate drinking is associated with lower grey-matter volume on brain scans — a reason not to feel reassured rather than a diagnosis, since what a few per cent of volume means for how you think is not established.[48] In pregnancy no safe level is established, harm is dose-dependent and susceptibility varies too widely to set a threshold; every major authority recommends complete avoidance, ideally from the point of trying to conceive.
Guidance has not followed the evidence. In January 2026 the US 2025–2030 Dietary Guidelines dropped their numeric limits, and the separate limits for men and women, for unquantified advice to drink less; the American Association for the Study of Liver Diseases objected, and The Lancet Gastroenterology & Hepatology called it a step backwards.[49] Canada's 2023 guidance puts the lower-risk threshold at two drinks a week or fewer.
Alcohol covers the genetic data and where it stops short, cancer by site, the liver, brain and sleep, standard-drink sizes and national guidelines, and who should be most careful.
Practical beverage guidance
- Coffee, 2–4 cups a day if you already drink it. The effect doesn't justify starting a habit you don't enjoy. Use a paper filter if your LDL cholesterol is high.
- Let hot drinks cool below about 60 °C — about six minutes for a fresh cup of coffee or tea.
- Keep the caffeine early. One afternoon cup is fine for most people; a heavy total wants 10–12 hours before bed.
- Skip the sugary café drink, not the teaspoon — though for diabetes risk, less sugar is better.
- Alcohol: the minimum-risk dose is at or near zero. Don't start for "heart benefits"; if you drink, less is better. Keep the last drink well before bed, since even a moderate dose that has cleared by bedtime still costs sleep, and none at all in pregnancy.
- Drink water steadily, not in boluses, guided by thirst and pale urine; past 65, on a schedule. Filter for what your supply contains, and remineralise reverse-osmosis water.
- Tea, green or black, 2–3 cups a day, where most cohort signals converge; the dementia signal sits at 1–2. As with coffee, don't take it up purely for health.
- Drink tea or coffee an hour or more before iron-rich meals or iron supplements, not with or after them, if your iron status matters.
- Sugar-sweetened drinks are off the list. They sit on the negative side of the ledger with alcohol; see Sugar substitutes and Foods to limit.
What's overrated
- "Wine is healthy because of resveratrol." A glass holds roughly 0.05–0.3 mg; supplement trials use 100–1,500 mg. No special protective effect of wine over other alcohol survives rigorous methods.
- The "J-curve" for moderate drinking. See the alcohol section above.
- Blanket coffee avoidance for atrial fibrillation. Not supported by the one trial (above), though coffee is not proven to prevent it either.
- "Bulletproof" butter coffee. Calorie-dense, no demonstrated benefit over plain coffee, and a meaningful dose of saturated fat.
- Alkaline, "structured" and marine-plasma water. No clinical evidence of benefit over plain mineralised water; "structured" water is physicochemical pseudoscience. Hydrogen water has plausible biology and small trials in metabolic syndrome, but is still preliminary.
- Daily electrolyte powders for sedentary adults. Among desert ultramarathoners, how much supplemental sodium runners took did not predict who developed blood-sodium disturbances; the rule of thumb puts any benefit past roughly an hour of sustained exertion or heavy heat, and many powders add sugar and unneeded sodium.
- Vitamin-fortified "wellness" waters. The vitamin doses are negligible; the sugar and sweetener loads usually are not.
For the full ordered action list drawn from every pillar, see the healthspan long list.