How Long Should You Fast? The Honest Evidence on Hours, Days, and Autophagy

August 4, 2026 · 27 min read

Software Engineer

An empty plate, a glass of water and a clock on a bare table, suggesting the passing hours of a fast

Fasting promises a lot: fat loss, cellular renewal, a longer life. Here is what the best evidence actually supports — how many hours, how many days, how many times a year, and where the popular claims outrun the science.

note

This is general health information, not personalized medical or dietary advice. Fasting is not safe for everyone — talk to a doctor before changing how you eat, especially if you take medication or have a health condition.

Almost every fasting claim is really a claim about dose. Sixteen hours a day, one day a week, five days a month, a week once a year — these are wildly different interventions that happen to share a name, and lumping them together is why the topic feels so confusing. They deserve three separate questions, with three separate answers and three very different grades of evidence:

How many hours a day should the eating window be?
How many days can a fast usefully last?
How many times a year is a longer fast worth doing?

The honest summary before any of the detail: nearly everything fasting reliably does for a healthy adult, it does by making them eat less. The effects on top of that are real but modest, they depend more on when you eat than on how long you abstain, and the single most exciting mechanism in the whole field — autophagy — has barely ever been measured in a living human being.

The fuel timeline: what actually happens between meals

Before judging any protocol, it helps to know what the body is doing hour by hour, because every fasting claim hangs off one of these phases.

The body runs on two fuels: glucose it can store as glycogen in the liver, and fat it can break down into fatty acids and ketones. A meal tops up the first. A long enough gap forces a handover to the second. That handover is the whole physiological story.

The first four hours or so are the fed state. Insulin is high, the meal's glucose is being used and stored, and fat burning is suppressed.
From roughly four to twelve hours, the liver takes over. It releases stored glycogen to keep blood glucose steady. Nothing dramatic is happening; this is just the ordinary state between dinner and breakfast.
Somewhere between twelve and thirty-six hours, liver glycogen runs out and the metabolism flips. Fat mobilization accelerates, and fatty acids and the ketones made from them become the dominant fuel. This is what researchers call the metabolic switch^[1]. The window is that wide because the timing depends on how carbohydrate-rich your last meal was and how much you moved during the fast — a hard morning run flips it far sooner than a day on the sofa.
After about a day, ketones are supplying a serious share of the brain's energy, and the body starts protecting muscle protein rather than burning it for glucose.
Around the third day, something bigger happens. When researchers tracked roughly 3,000 blood proteins in twelve volunteers through a seven-day water-only fast, the substantial systemic changes — proteins linked to the brain, the immune system, and the body's connective scaffolding — appeared only after three days of no calories^[2]. The volunteers lost 5.7 kg on average; the fat loss largely persisted after refeeding, while the lean tissue came back.

That last point is the one to carry forward. The interesting biology starts around day three — which is well past anything a daily eating window can reach. Keep that in mind through everything that follows, because the popular case for skipping breakfast is usually borrowed from research on fasts an order of magnitude longer.

Autophagy: the most oversold word in nutrition

Autophagy — from the Greek for "self-eating" — is the process by which a cell wraps up damaged proteins and worn-out components in a membrane, ships them to the lysosome, and digests them back into reusable parts. It is genuine, fundamental cell biology: Yoshinori Ohsumi won the 2016 Nobel Prize in Physiology or Medicine for working out its machinery in baker's yeast^[3].

Nutrient scarcity is one of its known triggers, and in animals this part is not in dispute. A day without food produces a dramatic rise in autophagy in the neurons of mice — including in the brain, long assumed to be shielded from it^[4]. Nor is the link to lifespan merely suggestive: in C. elegans, the roundworm, dietary restriction stops extending life at all once you disable the autophagy genes^[5], and mice engineered to run more autophagy live around 17% longer than their littermates^[6]. In animals, fasting-induced autophagy and its longevity payoff are results, not conjecture.

The internet's version takes those animal findings and applies them to humans, at a specific hour count, in every tissue: sixteen hours without food "switches on" autophagy and cleans out your cells. It is that last step — the translation — that has essentially never been demonstrated.

The first problem is that nobody can reliably measure autophagy in a living person. It's a dynamic flux, not a quantity, and a 2021 review in Trends in Molecular Medicine named this measurement gap outright as an underappreciated barrier to translating the field: several of its central concepts, nutritional modulation included, remain unvalidated in humans^[7]. There is no blood test for it. Reading autophagy requires biopsies or specialized cell preparations, which is why the human literature is so thin.

The second problem is that when researchers have looked, the results have been unhelpful. In a study that took muscle biopsies at 2, 12, 24 and 36 hours of fasting, the autophagy markers in skeletal muscle fell rather than rose, and only in untrained subjects — trained ones barely responded at all^[8]. A 2022 study that ran the comparison across species found the same split: intermittent fasting raised autophagy markers in mouse liver, but not in mouse muscle, and in humans the muscle markers went down^[9].

The best-designed human attempt to date is encouraging and very hedged. An exploratory analysis of a six-month randomized controlled trial (a study where participants are assigned to groups at random, the strongest common design for testing an intervention) measured autophagic flux in blood cells and found it higher in the intermittent-fasting group than in standard care. But the difference was borderline, there was no significant increase from the fasting group's own baseline, and the authors note the effect may partly reflect autophagy declining in the comparison group^[10]. Their own verdict is "may increase" and "further studies are required."

Three corrections follow from this, and they matter more than any protocol choice.

No hour threshold has ever been established in humans. The confident numbers circulating online — 16 hours, 18 hours, 24 hours — are not measurements. They are extrapolations from rodent studies, dressed up with a precision the underlying data does not have.

Autophagy is not a light switch. It runs continuously in every cell as basic housekeeping. The question was never whether fasting turns it on, but whether fasting raises it enough, in the right tissues, for long enough to matter — and that is exactly what nobody has yet shown in a person.

More autophagy is not automatically good. It's a survival mechanism that cells use under stress, including cancer cells; the biology is context-dependent rather than a dial you want turned to maximum.

None of this means autophagy is fake or that fasting does nothing. It means the human half of the cellular-renewal story is a hypothesis being tested, not a benefit you can bank — and it should not be the reason you skip breakfast.

The daily question: how many hours?

So set the cell biology aside and look at the outcomes that have been measured properly in humans. Time-restricted eating — confining all food to a window of typically 8 to 10 hours, written as 16:8 when 16 hours of fasting bracket an 8-hour window — is the most-studied daily protocol, and the trials are remarkably consistent.

Trial	Protocol	Design	Beat its control?	What it found
TREAT, 2020^[11]	16:8, no calorie target	Randomized trial, n=116, 12 weeks	No	0.26 kg difference, not significant; the fasting group lost more lean mass
Guangzhou, 2022^[12]	8-hour window plus a calorie limit	Randomized trial, n=139, 12 months	No	8.0 kg vs 6.3 kg on the calorie limit alone — a 1.8 kg gap that was not significant
Early window, 2022^[13]	07:00–15:00 window plus a calorie deficit	Randomized trial, n=90, 14 weeks	Partly	2.3 kg more weight lost, but the fat-mass difference missed significance
Controlled feeding, 2018^[14]	6-hour early window, weight deliberately held constant	Crossover trial, n=8	Mechanism only	Insulin sensitivity improved and blood pressure fell about 11/10 mmHg with no weight change

Read down that "beat its control" column and the pattern is hard to miss. When a trial matches the calories between groups, the eating window stops mattering. The one trial that clearly won on weight gave both groups the same calorie-deficit counselling and still missed its own co-primary endpoint on fat mass. The one that found the cleanest metabolic effect achieved it by feeding participants everything they needed and clamping their weight — proof that timing has some independent effect, from a study of eight men.

The two best syntheses land in the same place. A Cochrane review of 22 randomized trials covering 1,995 people found intermittent fasting produced no meaningful extra weight loss over ordinary dietary advice, and its lead author put it bluntly: the evidence doesn't justify the enthusiasm on social media^[15]. A network meta-analysis in The BMJ pooling 99 trials and 6,582 participants reached the same verdict for time-restricted eating specifically — no advantage over continuous calorie restriction — and found that even the modest advantages that did show up were confined to trials shorter than 24 weeks^[16].

The one timing effect that does hold up is circadian, not duration-based. A meta-analysis of 18 studies covering 1,169 people found that early time-restricted eating significantly lowered fasting glucose, while time-restricted eating overall did not^[17]. Eating your food earlier in the day appears to work with your circadian rhythm rather than against it, and that is a better-supported reason to move dinner earlier than anything about hour counts. Even so, keep the size in perspective: when the Salk Institute ran its own controlled trial of a personalized 8-to-10-hour window in people with metabolic syndrome, the improvement in glycated haemoglobin (HbA1c) — the three-month average blood-sugar marker — was 0.10 percentage points^[18], statistically real and clinically negligible.

There is still a practical case for a shorter eating window, and it's an honest one: it's a rule you can follow without counting anything. In a six-month trial in people with type 2 diabetes, an 8-hour window with no calorie counting produced 4.5 kg of weight loss while explicit calorie restriction produced 2.6 kg and missed significance — and the mechanism was visible in the adherence data, where 87% stuck to the window but only 68% hit their calorie targets^[19]. That is a behavioural win, not a metabolic one. If closing the kitchen at 20:00 is easier for you than logging meals, it is a perfectly good tool. Just don't expect it to do anything your calorie intake isn't already doing.

The weekly question: whole fasting days

Alternate-day fasting and its gentler cousin 4:3 — eating very little on three non-consecutive days a week — are the one family of protocols with a measurable edge over ordinary dieting, and it is small.

The BMJ network meta-analysis found alternate-day fasting was the only fasting pattern to beat continuous calorie restriction, by 1.29 kg^[16]. The strongest single trial is more encouraging: 165 adults randomized to 4:3 fasting or daily calorie restriction for a full year, with the same weekly calorie deficit prescribed in both arms, and 4:3 came out 2.89 kg ahead with better adherence^[20]. Note the fine print, though — both groups received intensive group behavioural support and a 300-minute weekly activity prescription, so this is not a comparison of bare diets.

Against that sits the cost. In a twelve-month trial of alternate-day fasting against daily calorie restriction, the two were indistinguishable on weight at both six and twelve months, alternate-day fasting had the highest dropout rate of any group at 38%, and it raised low-density lipoprotein (LDL) cholesterol — the "bad" cholesterol — by 11.5 mg/dL, roughly 0.3 mmol/L, relative to daily dieting^[21]. Whole fasting days create a real deficit, which is why they sometimes win, and they are the hardest pattern to keep up, which is why the win rarely survives a year.

The yearly question: multi-day fasts and how often

This is where the biology gets genuinely interesting — day three, remember, is where the systemic changes start — and where the evidence gets weakest. Two research traditions dominate.

Supervised prolonged fasting. The largest safety dataset comes from a German clinic that has run the Buchinger method for decades: 1,422 people fasting for 4 to 21 days on 200–250 kcal a day of juice, broth and honey, with medical supervision throughout. Adverse effects appeared in under 1% of participants, weight fell from 3.2 kg over five days to 8.6 kg over twenty, and blood pressure dropped substantially^[22]. It is genuinely reassuring on safety — but it is an observational study, with no control group, in self-selected and largely affluent participants, at one clinic, with unblinded data collection. The authors say so themselves.

The fasting-mimicking diet. Valter Longo's group designed a five-day, low-calorie, low-protein, high-unsaturated-fat regimen intended to produce fasting's metabolic signature while people still eat something. In a randomized trial of 100 people, three monthly cycles reduced body weight, trunk and total body fat, blood pressure, and insulin-like growth factor 1 (IGF-1), with no serious adverse effects^[23]. A 2024 re-analysis of trial blood samples reported reduced insulin resistance, lower liver fat, and a 2.5-year drop in a computed biological-age score, independent of weight loss^[24]. Read that last one with the appropriate caution: it is a secondary and exploratory analysis of biomarkers rather than a trial of health outcomes, "biological age" is a model estimate and not a measured quantity, and the diet is a commercial product sold by a company the lead researchers are connected to.

On "how many times a year," the honest answer is that nobody has tested it. Every cadence you will see quoted — five days monthly for three months, then once or twice a year; a week-long fast annually — comes from how a particular study was designed, or from clinical tradition, not from any trial that compared frequencies against each other. There is no dose-finding study. There is no long-term randomized trial of periodic fasting against not fasting with hard outcomes like heart attacks, cancer, or death. Anyone giving you a confident annual number is extrapolating.

Fasting dose	Typical protocol	What the human evidence supports	Evidence strength
Overnight, 12–14 h	Stop eating after dinner, eat breakfast normally	A sane baseline; no special claims needed	Well established as safe
Daily window, 8–10 h	16:8, ideally with the window early	Modest weight loss, entirely via eating less; a small glucose benefit if early	Multiple randomized trials, small effects
Weekly fast days	Alternate-day or 4:3	Roughly 1–3 kg more than daily dieting, mostly in shorter trials	Mixed, poor adherence
Multi-day, 2–7 days	Water-only or very-low-calorie, supervised	Large short-term weight and blood-pressure changes; safety data is observational	Uncontrolled studies
Periodic 5-day cycles	Fasting-mimicking diet, monthly for three cycles	Improved risk markers in randomized trials; the cadence itself is untested	Biomarkers only

strong good mixed weak

One firm practical rule does emerge from this literature: a fast beyond about 72 hours belongs under medical supervision. That is not caution for its own sake. Coming out of a prolonged fast carries a specific danger called refeeding syndrome, in which reintroducing carbohydrate causes a sudden insulin surge that drives phosphate, potassium and magnesium out of the bloodstream and into cells, with consequences up to cardiac arrest. Clinical guidance flags anyone with little or no food intake for more than five days as at risk, and recommends restarting nutrition at a deliberately low rate with thiamine and vitamin supplementation^[25]. Every multi-day protocol in the research above is run by clinicians for exactly this reason.

What it costs

A fair account of fasting has to price the downsides at the same rigour as the benefits.

You may lose more muscle. The TREAT trial's most cited result isn't the null weight finding — it's that the 16:8 group lost significantly more lean mass in the arms and legs than the control group^[11]. Compressing your eating window tends to compress your protein intake too, and low protein plus a calorie deficit plus no resistance training is a recipe for losing the tissue you least want to lose.

Your cholesterol may move the wrong way. Alternate-day fasting raised LDL cholesterol over twelve months^[21]; the controlled-feeding early-window study raised triglycerides and total cholesterol^[14]; the BMJ analysis found time-restricted eating increased total and LDL cholesterol relative to whole-day fasting^[16]. There is no clean lipid-improvement story here.

Rapid weight loss promotes gallstones. When the body mobilizes fat quickly, the liver secretes extra cholesterol into bile and the gallbladder empties less often — the exact conditions in which stones form. Very-low-calorie dieting is a recognized risk factor, per the US National Institute of Diabetes and Digestive and Kidney Diseases^[26].

Fasting is entangled with disordered eating. In a study of 2,762 Canadian adolescents and young adults, intermittent fasting in the previous year was associated with every eating-disorder behaviour measured in women — overeating, loss of control, binge eating, vomiting, laxative use, compulsive exercise — and with higher eating-disorder psychopathology scores in men and gender-diverse participants too^[27]. That is a correlation, and people already struggling may be drawn to fasting rather than harmed by it. But a protocol whose core skill is overriding hunger signals is not a neutral practice for everyone, and the researchers' own recommendation is that clinicians be cautious about prescribing it to young people.

If you take glucose-lowering medication, fasting is a dosing problem, not a diet. Insulin and sulfonylureas lower blood glucose whether or not you have eaten, so skipping meals on an unchanged dose risks hypoglycaemia. This is well-charted territory thanks to Ramadan: the International Diabetes Federation and the Diabetes and Ramadan International Alliance publish guidelines that stratify patients into low, moderate and high risk six to eight weeks in advance, adjust medication accordingly, and advise the high-risk group not to fast at all^[28].

Group	Guidance
Pregnant or breastfeeding	Do not fast
Children and adolescents	Do not fast
History of an eating disorder	Do not fast
Type 1 diabetes	Not recommended — untested
Frail or underweight older adults	Do not fast
Taking insulin or sulfonylureas	Only with medication adjustment
Any fast beyond about 72 hours	Medical supervision required
Healthy adult, modest eating window	Reasonable to try

Finally, a worked example in how to read fasting headlines. In March 2024, an abstract presented at an American Heart Association conference reported that people with an eating window under 8 hours had a 91% higher risk of cardiovascular death, and it went around the world^[29]. The findings came from an observational analysis in which years of eating behaviour were inferred from two days of self-reported dietary recall; the abstract was never peer-reviewed or published as a full paper; no hazard ratio with a confidence interval was released; and the obvious alternative explanation — that people who are already ill eat less and over a shorter span — was not excluded. The association's own release said the work was preliminary and did not show causation^[30]. Treat it as a lesson in headline scepticism rather than as evidence of harm — and apply the same scepticism to the glowing headlines, which come from the same conference circuit.

What this all adds up to

Line the three questions up next to their answers and the picture is much simpler than the discourse around it.

Fasting is a legitimate and generally safe way to structure eating, and a poor substitute for the things that actually move health outcomes. If a shorter eating window makes it easier for you to eat less without counting anything, that is a real benefit and a good enough reason to do it — that convenience, not autophagy, is the honest case. Shift the window earlier if your life allows, keep protein up and keep lifting something heavy so the weight you lose is fat rather than muscle, and stop well short of any multi-day fast unless a clinician is involved.

And be precise about what is unproven, because the gap is narrower and more specific than "fasting is hype". In animals the case is made: block autophagy and dietary restriction stops extending life, boost it and mice live longer, and in rhesus monkeys eating less from adulthood onward improves both health and survival^[31]. What has never been shown is that any particular human protocol reproduces any of it. Nobody has demonstrated that a 16:8 window raises autophagy in a person, and no trial has ever tested whether a fasting schedule adds a single day to a human life. The uncertainty isn't in the underlying biology — it sits in the two translations: from short-lived animals to us, and from "no food" to a schedule you can actually keep. Until someone closes those, the honest case for a shorter eating window is the convenient one, not the cellular one.

The fuel timeline: what actually happens between meals​

Autophagy: the most oversold word in nutrition​

The daily question: how many hours?​

The weekly question: whole fasting days​

The yearly question: multi-day fasts and how often​

What it costs​

What this all adds up to​

References​