Age well

The short trials and the long follow-ups measure different things, weight on one side and disease on the other, so neither one checks the other. The long-term half only watched what people already drank, and people already at higher risk switching to diet drinks would produce the same pattern. This is a joint position from the Korean Diabetes Association and Korean Nutrition Society rather than a new analysis.

Every study pooled here watched people rather than assigning anything, so a flat curve may mark illness already underway rather than cause it. The slope was measured differently from study to study, from two saliva samples a day to six or more, which blurs the pooled figure.

This only tracks a pattern over time, so it cannot show cause. Just 32 of the deaths were from heart disease, which makes the estimate rough. Cortisol was measured on one ordinary weekday. Morning cortisol on its own showed no link at all.

This measures anticholinergic burden as a whole, not diphenhydramine and not antihistamines. Only 8 of the 25 studies found could be pooled at all. The scales used to score that burden agree poorly with one another, so what is being measured is not the same thing from study to study.

People carrying only one copy, and those with the milder H63D variant, showed no excess of anything — and they far outnumber the two-copy group, which is roughly 1 in 150 to 200 people of northern European ancestry. UK Biobank volunteers are healthier than the general population, so these percentages are not population rates. The one-in-ten lifetime figure is for men; the figure for women is substantially lower.

The larger figures come from a slice of the data examined after the fact rather than the review's main comparison, which makes them the weaker number even though they are the more striking one. The ten pooled reviews overlap, so some of the same original trials are counted more than once. And blood pressure is a stand-in here: none of these trials counted heart attacks or strokes.

The search closed in October 2025, so anything published since is missing. The human trials measured such different things that they could not be pooled, which makes the null conclusion a reading of the literature rather than a calculated number. The review also found no outcomes trial at all of intravenous NAD, which is sold commercially.

Between two and five trials per outcome, so a real effect could still be hiding in an evidence base this thin. What argues against that is how closely the trials agreed with each other: they scattered almost not at all around the null. The nicotinamide riboside results could not be pooled and ran in both directions, longer six-minute walks in peripheral artery disease against worse physical performance scores in mild cognitive impairment.

The over-60 systolic figure is a subgroup finding inside a 349-person pool, which generates a hypothesis rather than confirming one. The analysis assumed one true effect across trials spanning 250 to 2,000 mg a day and 14 days to 24 weeks, which narrows the confidence interval by assumption rather than by evidence. No trial in the pool measured a heart attack or a stroke; the reading on the cuff is the entire result.

Eight trials and 342 people in the metabolic pool, so a modest real effect could be missed. Participants were mostly healthy and not diabetic, which is the population least likely to show a metabolic improvement and leaves the question open in people who already have metabolic disease. Several pooled trials ran fourteen days, far too short for HbA1c, which reflects about three months of blood sugar, to move at all.

The longest trial in the pool ran 24 weeks. Ten trials contributed to the safety analysis, too few to catch a harm affecting one user in a thousand, and most were funded by manufacturers, who historically under-report adverse events. The four-week study establishes the highest dose formally tested, not a dose any regulator has certified. No trial measured a cardiovascular event.

The trial cannot separate how much of the benefit came from the weight loss and how much from the drug's other effects, since the treated arm got both. Adverse events forced 16.6% off the drug permanently, against 8.2% on placebo, in a group with strong reason to persist. Novo Nordisk funded it and employed several of the authors.

Participants were picked for being unusually healthy for their age, with no major health event in five years and good mobility, so a frailer group might respond differently. Testing six outcomes at once, the authors demanded a p value below 0.01, a strict bar that would hide a modest real effect. Three years is short for both fractures and cognition.

Both trials gave very large intermittent doses, so nothing here suggests a daily tablet carries the same risk. The falls trial enrolled 200 people and had no placebo group, comparing one dose against another rather than against nothing. The annual-dose trial's fracture interval starts at exactly no effect, and its explanation for the timing, a spike in blood levels just after each dose, is a hypothesis it raises rather than one it tests.

Neither trial screened people in for low vitamin D, so both test adding more to people who mostly had enough rather than correcting a shortfall. The Australian participants averaged 77 nanomoles per litre before they started, in a country with a great deal of sun. Five years may be short for cancers that take longer than that to surface.

The over-75 recommendation rests on a very small reduction in deaths, graded as high certainty but never translated into a number of lives. Most of the guideline's positions are worded as suggestions rather than recommendations, the wording used when certainty is low. The trials behind it used widely varying doses and often let participants keep taking their own vitamin D, so the right dose is left unresolved.

Baseline protein intake was not consistently controlled across the trials, so some of these people may already have been eating enough for a supplement to add nothing. The handgrip analysis was highly inconsistent between studies. Appendicular muscle mass did improve, but on two trials that disagreed sharply with each other. Older adults only, so this says nothing about young trained people.

Both tests soaked teeth in the drink for a day or a week at a stretch, with no saliva rinsing between sips. In one of them the damage did not track a drink's measured acidity at all, so phosphoric acid's own share never separates from the rest of cola. The seven-day test used bovine enamel, a stand-in for human enamel rather than the thing itself. Twelve of the nineteen studies in the review are lab work too.

The group was small and elderly, and the published summary never says how many people were in it. In humans this is a correlation; the evidence that cortisol itself damages the structure comes from the rat work the same paper cites. Ageing or illness could equally shrink that structure and raise cortisol at once, which would draw this same pattern with no link between them.

One hospital, 250 people. Facing major surgery can itself distort cortisol patterns, so this may not reflect ordinary life. A flatter drop may also mark someone who is already sicker, rather than adding risk of its own.

Everyone in these trials already had one of those conditions, so a healthy adult has no reason to expect the same shift. The trials ran between one and six months, which says nothing about a year on it. And 543 people across 11 trials is a modest base for a figure this precise.

Blood is not tissue. The one trial that biopsied muscle to check delivery found NAD-related metabolites raised there, including the products of clearing the compound away, while the muscle's mitochondria performed no differently. In mice, oral doses of both compounds are taken apart into plain nicotinamide in the liver; the same compounds injected arrive intact.

Open-label, fourteen days, healthy adults, so this establishes a route rather than an outcome. The bacterial step was shown in fermentation vessels seeded with human gut microbes rather than watched in the participants themselves. Nine of the twelve authors work for Nestle, whose health science arm sells NAD-related products.

A narrative appraisal rather than a systematic review, so its selection of evidence cannot be reproduced by anyone else. Its authors include the group running the pharmaceutical NMN trials, which makes this the field appraising itself. Its phrase about raising NAD in blood and some tissues is deliberately narrower than the marketing and should not be read as delivery throughout the body.

Fourteen days and open-label, so the comparison is of absorption, not of benefit, and neither arm measured anything a person would feel. Nine of the twelve authors work for Nestle, whose health science arm sells NAD-related products, which is worth noting mainly because the result cuts against both branded products rather than favouring either.

This was one result among many in a study built to map how fast NAD is made and consumed across tissues, not a comparison it was powered to make. It tracks where the molecule goes, not whether anything follows from its arriving. No equivalent measurement has been made in people.

Thirty-six people is small for a vascular endpoint, so this cannot rule out a modest real effect. The paper's own closing line says NMN showed potential in alleviating arterial stiffness, which is a firmer reading than its numbers support. The work was done in the laboratories of a Japanese supplement company, though the authors declare no competing interests.

Mice, at 100 to 300 mg of NMN per kilogram of body weight a day. A 70 kg adult swallowing a standard 250 mg capsule takes about 4 mg per kilogram, and body-surface scaling narrows that gap without closing it. More was not better even in the mice: the higher-dose group moved around less and reared less than the lower-dose one. The senior author holds NMN patent-licensing interests.

Fourteen men, one dose each, and sixty minutes of blood sampling, so this describes the first hour and nothing about daily use or any outcome. The authors could not say where the extra 2PY and 4PY came from. All three work for Meiji Holdings, which sells NMN.

Funded by the company that sells the ingredient, with two of the three authors on its staff and the third the inventor of its licensed patents. Eight weeks, in healthy overweight adults. Every number reported is a blood or safety measurement, so the curve describes how much of the pill turns up, not how much it does.

The outcome is a marker in spit, not a symptom: nothing was reported about breathlessness, flare-ups, lung function or quality of life. The confidence interval runs from a 76 percent reduction to a 7.6 percent one, so the size of the effect is barely pinned down. Inflammation in the blood did not change, so the effect did not travel beyond the airway. Two authors are employees and shareholders of the company selling the compound.

Twenty people over four weeks, in a phase one safety trial. Its apparent improvement in Parkinson's symptom scores is not an efficacy finding: the authors traced it to participants on the compound having taken their levodopa more recently before assessment. Four weeks establishes the highest dose tested, not a dose anyone has certified for long-term use.

Men only, all obese and insulin-resistant, so this does not describe healthy or older users, and twelve weeks is not years. The mitochondrial half is a sub-study of the same forty men rather than an independent replication, and the two count once. The trial did not report whether NAD actually rose in these participants, so a failure of delivery cannot be fully separated from a failure of the mechanism.

Twelve men over 21 days is a mechanistic study, not an efficacy trial. What rose was the NAD metabolome, the family of related compounds including clearance products; the paper does not report that muscle NAD itself rose in proportion. The gene-expression changes ran downward in the energy and mitochondrial pathways, the opposite direction to the one the rationale predicts. One author invented the licensed patents behind the product and holds stock in the company selling it.

Expressed per kilo of whole body mass, the widely quoted 0.40 against 0.24, the same comparison reached only p = 0.055. This pools data from studies run earlier for other purposes rather than a trial built for the question. Men only, in both age groups. It measures a building rate over hours, not muscle held onto over years.

That upper limit was set on the risk of high blood calcium with an allowance for uncertainty, not on any observed pattern of harm at 4,000 international units. Reaching a toxic blood level takes very large doses sustained over a long period. Safety at 2,000 international units daily says nothing about the very large intermittent doses that were followed by more falls in other trials.

Nobody was assigned to drink coffee, so whatever else separates a three-cup habit from none, in diet, smoking, exercise or income, is folded into that number. Coffee carries hundreds of compounds besides caffeine, and nothing here tells them apart. The size of the association differed by region, running stronger in Europe and Asia than in the US.

That coffee drinkers die less often is not in doubt; whether anything in the cup causes it is. Nobody was assigned to drink coffee, and nothing separates caffeine from the hundreds of other compounds in it.

Both pools followed what people already chose to drink, so these are links rather than proof that either drink causes anything. Someone told to watch their weight or blood sugar tends to switch to diet, which would draw this same pattern with no effect from the drink at all. The pooled studies varied widely in their results for several of these outcomes. And these are drink categories, not Coca-Cola and Coca-Cola Zero specifically.

The cohorts were stable cardiac patients, not supplement users, so the absolute risk does not transfer to anyone taking a capsule. The human half is observational, though 4PY at physiological levels switched on the vascular inflammation machinery in mice and a genetic analysis across 106,000 people pointed the same way. Every trial of these supplements that measured the metabolites ran weeks to months, far too short for a heart attack to be an endpoint.

Twenty-five postmenopausal women with prediabetes who were overweight or obese, at a single site, for ten weeks. The endpoint is insulin sensitivity, a metabolic measure, not an ageing or longevity one. The senior authors hold NMN patent-licensing interests through their university. Later pooled analyses of the wider NMN literature found no metabolic effect, in populations that were mostly not metabolically ill.

Whether NMN improved insulin handling in a 25-person trial is disputed: the women given it began with less than half the liver fat of those on placebo, and liver fat is something NMN is meant to act on. A trial that size cannot settle it.

Twenty people per arm across four arms, analysed per protocol rather than by intention to treat, in healthy adults over sixty days. The first author works for an NMN company and two co-authors for NMN raw-material producers. The trial's biological-age outcome comes from a free online calculator rather than a validated clinical instrument.

Whether NMN helps people walk further is unsettled. The one trial that found it was run by companies selling NMN and tested healthy middle-aged adults; a later pooling in older adults did not repeat it.

This is diet recall at a single point in time, so daily cola drinkers may differ from occasional ones in ways nobody recorded, including what they were drinking instead. The difference showed at the hip and not the spine. Daily cola drinkers were not taking in more phosphorus overall, which makes phosphoric acid the authors' inference rather than something they measured.

This followed what people already consumed rather than assigning it, so the arrow could run the other way: someone watching their blood sugar has good reason to switch to sweeteners well before a diagnosis arrives. Intake came from participants' own repeated diet diaries. And these are the same participants and the same diaries behind the cancer finding, so the two results are not independent of each other.

The English null is wide: a real 20 to 30% increase sits inside its range. The Taiwanese comparison group is small, 36,081 non-users against 677,971 users, and people with hay fever who take nothing are unlikely to resemble those who do. Neither design separates the drug from the reason it was taken.

Whether antihistamines raise dementia risk is unsettled. English prescription records show no significant link, Taiwanese dispensing records a rise with dose, and English records miss most shop-bought use, so neither can correct the other.

The conditions these drugs treat, insomnia and low mood among them, are early signs of dementia themselves, so the drug may be marking the illness rather than causing it. Dropping the last twelve months of use blunts that without removing it. One health system in Seattle, shop-bought use only partly captured.

A pattern shared by drugs that do and do not reach the brain points somewhere other than the drug. More medicine dispensed means more medical contact and more chances for a diagnosis to be recorded, and heavier users are sicker to begin with. The authors call for trials to confirm any association.

A secondary outcome is exploratory unless it was named in advance and corrected for testing many things at once, and this paper reports no such correction. Three of the four authors work for Meiji Holdings, which sells NMN. The authors highlight that walking-time changes tracked changes in blood NAD, 2PY and 4PY, which is an association inside one trial rather than evidence that raising NAD caused the walking.

Thirteen people is very small even for a crossover design, which makes the two positive results the fragile half and the long list of nulls the sturdy one. The fat-free mass change was one of many outcomes tested, exactly the setup in which a chance finding appears. Six weeks, in healthy overweight and obese volunteers.

Twenty-four people in a crossover design, six weeks per arm, all of them healthy, so this is an early tolerability and target-engagement study rather than a test of anything. NAD was measured in blood, which shows the pill is absorbed and metabolised, not that it reached the tissues where ageing is supposed to slow. The trial was investigator-initiated at a university with no declared competing interests, which sets it apart from most of this literature.

The cohort recorded what people chose to consume rather than assigning it, and its own authors flag selection effects, unmeasured differences between users and non-users, and the possibility that illness came first. The percentage rises are real, but the absolute difference in risk between users and non-users is small. The rodent work is the original researchers re-examining their own contested findings, not an independent repeat. The review pointing the other way came from a consultancy regularly retained by regulated industries.

None of the three studies on sweeteners and cancer could settle it: a cohort of people's own choices, a team re-analysing its own rodent work, an industry consultancy's review. The study that could has not been run.