Is Melatonin Safe? What the Heart Failure Study Actually Found

The melatonin headline came around twice.

It landed first in November, presented at a cardiology conference, and went everywhere in about a day: melatonin, the thing sitting in half the nightstands in America, linked to a 90% higher risk of heart failure. Then it went quiet. This week it came back — same study, same number, a fresh wave of coverage, nine months later.

I keep melatonin in the house. I always have one handy.

I used to take it daily — that was its own era, and at the time I'd have told you it was working. Now I can't tell you when I last took one. Not roughly. At all. The nightly habit stopped somewhere, without a decision I can point to, and the bottle stayed in the drawer.

So the headline didn't arrive as someone else's problem. It arrived aimed at the version of me who took one every night — which is close to exactly the group this study was looking at.

Which is a worse feeling than being implicated, and a much better reason to read carefully. So I read the thing instead of the coverage — the abstract, the methods, the numbers. And the methods section turned out to be the entire story. There is one word in the definition of who counted as a melatonin user, and once you see it you cannot unsee it. It doesn't make the finding go away. It changes what the finding is about.

What the Study Actually Reported

Start with the design, because the design is most of the story.

This was a retrospective cohort analysis run on the TriNetX Global Research Network — a pool of de-identified electronic health records from healthcare organisations across multiple countries. Nobody was assigned to take melatonin. Researchers went looking in records that already existed.

They built two groups of adults with insomnia: 65,414 with documented long-term melatonin use, and 65,414 without. Anyone with existing heart failure was excluded, as was anyone on other prescription sleep medication. The groups were then matched one-to-one on demographics, fifteen comorbidities, cardiometabolic medications, labs, vitals and healthcare utilisation — and matched well, every standardised mean difference landing under 0.02. That's careful work, and I want to give it its due before I take it apart.

Over five years, heart failure showed up in 3,021 of the melatonin group and 1,797 of the controls. In percentages: 4.6% versus 2.7%. The hazard ratio was 1.89, 95% confidence interval 1.78 to 2.00 — a tight interval around a large number. The authors also reported roughly three times the rate of heart-failure hospitalisation and roughly double the all-cause mortality.

That 1.89 became "90% higher risk" in the coverage, a fair rounding of an 89% relative increase. It is also the number that did all the travelling, mostly detached from the 4.6-versus-2.7 that produced it.

One more thing about status, because it matters more than anything else here. Abstracts presented at American Heart Association meetings are not peer reviewed — the AHA says so itself, in the release. Findings stay preliminary until they appear as a full manuscript. This one was presented in November 2025, and no manuscript has appeared since. The study behind two international news cycles is still, formally, a conference abstract that has never passed review.

The study that generated two international news cycles is still, formally, a conference abstract that has never passed review.

The Word Is "Prescription"

Here is the line that reframes everything. To count as exposed, a person needed at least one melatonin prescription and at least 365 days of exposure.

A prescription. Not a purchase. Not a bottle in a bathroom cabinet. A prescription, written by a clinician, sitting in a medical record.

In the United States, melatonin is a dietary supplement under the 1994 legislation governing supplements: no prescription required, no dose cap, any strength, any shop. It almost never generates a prescription record, because there is nothing to prescribe. In the United Kingdom, Australia, Switzerland, Japan and much of the European Union, melatonin is a prescription medicine — France caps supplement melatonin below 2 mg a day, several EU states at 1 mg.

Now put that against the exposure definition. In a global records network, requiring a prescription selects hard for people in places where a doctor looked at their insomnia and decided it warranted a year or more of a prescribed hormone. Those are not people who grabbed gummies because they'd been scrolling too late. Those are people whose sleep was broken badly enough to reach a clinic and stay there.

Epidemiologists have a name for this: confounding by indication. The drug and the disease arrive together, and the records cannot tell you which one you're looking at.

We can put a number on it, because insomnia and heart failure have been studied directly. A meta-analysis of seven prospective cohorts covering 177,008 people found insomnia associated with a 26% higher risk of heart failure — pooled hazard ratio 1.26, confidence interval 1.09 to 1.45. More telling, the risk scales with severity: in a cohort of more than 12,700 middle-aged and older adults, incident heart failure risk climbed from 22% higher with one insomnia symptom to 80% higher with four.

Read that next to the study design. Both arms had insomnia, so the 26% is not a number you subtract from the 89%. But severity within insomnia is where the risk lives — and severity is precisely what the matching could not touch. You can match on fifteen comorbidities, on labs, on vitals. There is no field in a health record for how badly this person sleeps. The researchers balanced everything they could see, and the thing that most separated the two groups was the thing nobody can see.

The Second Clue Is the Size of the Signal

There's a second tell in the results, and it comes from a habit I picked up somewhere else entirely. When a platform incident degrades one service, you go looking for the change that touched it. When every dashboard goes yellow at once — latency, error rate, queue depth, throughput, all together — you stop hunting for a subtle bug in one component. A signal that broad usually means something upstream of everything moved.

The abstract reports elevated heart failure, elevated heart failure hospitalisation, and roughly doubled all-cause mortality. All-cause. Every way a person can die, over five years, at twice the rate. Ask what mechanism would do that. Melatonin is a hormone with real physiological effects, and I won't pretend it's inert — but there is no proposed pathway by which a nightly dose doubles your chance of dying of anything at all within five years. That isn't a cardiac signal. That's the signature of one group being sicker than the other at the start, which is exactly what the prescription requirement would produce.

I want to be careful not to overplay this. Confounding by indication does not automatically account for a hazard ratio of 1.89, and waving a study away with the word "confounded" is the same lazy move as putting "90%" in a headline. A gap that large, in a sample that large, with matching that tight, deserves to be taken seriously. It's a reason for a proper study. It isn't, on its own, a reason to change what you do tonight.

When every dashboard goes yellow at once, you stop looking for a subtle bug in one component. Something upstream of everything moved.

What This Evidence Cannot Tell You

It cannot tell you melatonin causes heart failure. The authors said so themselves: "our study cannot prove a direct cause-and-effect relationship." This is an association found in records. Nobody randomised anyone.

It cannot tell you the dose mattered, because the dose wasn't recorded. One milligram of prolonged-release and ten milligrams of a gummy are the same data point here. If there were a dose-response relationship — the most persuasive thing an observational study can show — this design could not have found it.

It cannot tell you the control group wasn't taking melatonin. Controls were defined by the absence of a record. In every country where melatonin sits on a shelf, that absence proves nothing, and an unknown share of the "unexposed" were almost certainly taking it. That kind of mixing usually shrinks a gap rather than inventing one, so it doesn't explain the finding away — but the comparison is not melatonin against no melatonin. It's documented melatonin against undocumented anything.

It cannot tell you this applies to occasional use. Exposure required a year or more. Nothing here speaks to the person who takes melatonin four times a year crossing time zones, which is most people who own a bottle.

And it cannot be checked. No manuscript, no peer review, no full methods, no supplementary tables, nine months on. I can read a summary of what was done. I cannot read the sensitivity analyses — and neither can any cardiologist quoted in any of the coverage.

Now the part that cuts the other way, because leaving it out would make my own case tidier than it is.

Nothing in the randomised literature predicted a cardiac harm signal, and some points the other way. A meta-analysis of seven randomised trials of melatonin and nocturnal blood pressure found no significant effect overall — but in the controlled-release subgroup, nocturnal systolic pressure fell 6.1 mmHg and diastolic 3.5 mmHg, which is if anything cardioprotective. And a systematic review of 79 studies in 3,861 adults, at doses of 10 mg or more, found no detectable increase in serious adverse events among the higher-quality trials — though it did find more drowsiness, headache and dizziness, and noted that 37% of those studies never mentioned adverse events at all.

So the randomised evidence says melatonin looks tolerable. It also can't rebut the abstract, because none of those trials ran five years — and five years is the window in question. Short trials cannot see slow harm. Records cannot see cause.

Short trials cannot see slow harm. Records cannot see cause. Each body of evidence has a hole exactly where the other one's answer would go.

Seven Minutes

Here's the question the scare skipped past, and the one I'd want answered before deciding anything: how well does melatonin work?

The best answer we have is a meta-analysis of 19 randomised controlled trials covering 1,683 people with primary sleep disorders. Melatonin beat placebo on all three measures, and every effect is small.

• Time to fall asleep: 7.06 minutes faster than placebo (95% CI 4.37 to 9.75).
• Total sleep time: 8.25 minutes longer (95% CI 1.74 to 14.75).
• Sleep quality: significantly improved versus placebo.

Seven minutes. Eight minutes.

Those are real effects — the confidence intervals clear zero, and the benefit didn't fade with continued use, which is more than you can say for most sleep medication. They're also nowhere near what the person buying the bottle thinks they're buying. Nobody takes melatonin hoping to fall asleep seven minutes sooner.

The guidelines reflect that gap. The American Academy of Sleep Medicine's 2017 guideline recommends that clinicians not use melatonin for sleep-onset or sleep-maintenance insomnia in adults — a recommendation the AASM itself grades as weak, reflecting genuine uncertainty rather than proven uselessness. The American College of Physicians recommends cognitive behavioural therapy for insomnia as the initial treatment for every adult with chronic insomnia, and grades that one strong.

The intervention with the strong recommendation behind it is the one nobody markets, because there's nothing to sell. The one in the checkout aisle carries a weak recommendation against it.

You Don't Know What's in the Bottle

This is the part of the melatonin story with genuinely solid evidence behind it, and it has never once led a news cycle.

In 2017, researchers analysed 31 melatonin products bought from ordinary shops and pharmacies. More than 71% missed their label claim by more than 10%, with actual content ranging from 83% below the stated dose to 478% above it — and within a single product, one lot differed from another by as much as 465%. A quarter of them contained unlabelled serotonin.

In 2023, a team did the same to melatonin gummies and published it in JAMA. Of 25 products, 22 were inaccurately labelled — only three landed within 10% of the stated dose. Actual melatonin ran from 74% to 347% of the declared amount. One product contained no detectable melatonin at all, and 31.3 mg of CBD.

Sit with that. If you take a 5 mg gummy nightly, the honest description of your dose is: somewhere between roughly 4 and 17 mg, varying by bottle, and you will never know which.

It's the same structural problem I wrote about in why your supplements may not be working the way you think — and melatonin is its sharpest example, because melatonin is not a nutrient. It's a hormone. A bottle at 478% of its label isn't a slightly stronger vitamin. It's a hormone dose nobody chose.

A bottle that's 478% of its label isn't a slightly stronger vitamin. It's a hormone dose nobody chose — least of all you.

The Dose Nearly Everyone Gets Wrong

Even setting the label problem aside, the dose most people take has an odd history: nobody established it. It's what got sold.

In 2001, a controlled trial gave adults over 50 with measured reductions in sleep efficiency three doses of melatonin — 0.1, 0.3 and 3.0 mg — half an hour before bed, against placebo. The finding is, to me, the most useful thing in this entire body of work.

0.3 mg restored sleep efficiency and brought night-time blood melatonin back to the levels typical of a young adult. That's what a physiological dose means: enough to reproduce the signal your own body used to send.

3.0 mg also improved sleep — but it induced hypothermia, and left melatonin circulating in the blood into the daylight hours.

Ten times the dose did not produce a better night. It produced a night plus a morning, with the sleep signal still switched on while the body was trying to run its day. And the shelf sells 5 and 10 mg — fifteen to thirty times the dose that restored a normal signal in that trial, before you account for a label that might be off by 478%.

Melatonin is not a sedative. It's a timing signal: the message that the dark part of the cycle has started. More of a message doesn't make it truer. It makes it louder, and later, and blurred into the part of the day it was never meant to reach. Whatever you make of the heart failure abstract, this is the part I'd fix first, because it rests on direct measurement rather than inference.

Which is where the Train pillar walks into an article about a supplement. The strongest melatonin signal available to you is free, and it's the one your own body makes: bright light early, physical work during the day, darkness arriving at a consistent hour. I lift three to five days a week and do low-intensity work most days. The sleep effect was never why I started. It's the compound interest nobody advertises, because it isn't for sale.

I'd like to tell you that's why the bottle went quiet in my drawer. I can't. Claiming it would be inventing a mechanism I can't demonstrate, in an article about people doing exactly that. What I can tell you is that I stopped reaching for it, and I never decided to — which is its own kind of answer, just not the kind you can put in a study.

What I Actually Take From This

I got this wrong on the first pass, and the way I got it wrong is more useful than the study.

My mistake wasn't believing the headline. It was the reflex underneath it — reading a finding by asking does this indict me? instead of is this true? Alarm and smug dismissal are the same failure wearing different clothes. Both put you at the centre of a study that isn't about you, and both let you skip the methods.

The discipline that fixes it is one I keep coming back to: the difference between correlation and causation is not a debating trick. It's the habit of asking who ended up in which group, and why, before asking what happened to them. Here the answer was one word — prescription — and it turned a story about a supplement into a story about who gets prescribed one.

What I'd want to know is what this study couldn't measure. Does the risk track with dose? Does it appear at 0.3 mg as well as 10? Does it hold in the United States, where exposure would be a purchase rather than a prescription? Until someone answers that, the honest position is that a large signal sits in a dataset that cannot explain it — and the reading most consistent with the data is that people taking melatonin for a year were, from the start, the people sleeping worst.

Which is not reassuring. Badly broken sleep at 54 is a real cardiovascular concern in its own right, separate from anything in a bottle. If this study nudges anyone toward treating chronic insomnia as a serious medical problem rather than a nuisance to medicate over the counter, it will have done more good than its headline ever did.