CALERIE, dissected: two years of eating less, and what it actually bought

The most famous calorie-restriction trial ever run in humans did not test the thing everyone says it tested.

CALERIE prescribed a 25% cut. Its participants managed 11.9%. That is not a footnote — it is arguably the most useful number the trial produced, and it almost never survives the trip from the paper to the podcast. What CALERIE actually measured was what happens when healthy adults eat about 300 calories a day less than usual for two years, with a lot of professional help. The answer turns out to be genuinely interesting. It is also a good deal smaller, and a good deal more caveated, than “calorie restriction slows human aging.”

So let’s take it apart properly: what was run, what came out, what it cost, and which of it applies to you.

Why this trial carries the weight it does

Calorie restriction is the oldest claim in longevity science. Restrict intake without malnutrition and you extend lifespan in yeast, worms, flies, rodents — reproducibly, across decades of work. It is the closest thing the field has to a law.

The trouble is the jump to humans. You cannot randomise people to eat 25% less for forty years and count the funerals. So the evidence sits in two unsatisfying piles: animal experiments where lifespan was measured properly, and human observational data where the people eating less also tend to differ in twenty other ways.

CALERIE is the one bridge. Someone actually assigned the diet, in healthy humans, for two full years, with objective measurement of what people ate. Take CALERIE out and the human case for calorie restriction collapses back into mouse data plus inference. That’s why a single trial does so much lifting — and why it’s worth knowing what it does and doesn’t say.

The design: who, how many, how long

CALERIE Phase 2 ran at three US centres — Washington University, Pennington Biomedical Research Center and Tufts — with a coordinating centre at Duke. Of 238 people assessed as eligible, 220 were randomised — 145 to a prescribed 25% cut, 75 to eating freely, a 2:1 allocation in favour of restriction. The papers reporting the diet itself work from a slightly smaller number — the 218 who were assigned and started (143 restriction, 75 control), the two-person difference sitting in the restriction arm. Both figures are correct and they count different things, so it’s worth noting which one a given paper is quoting.

The eligibility criteria matter more than they look:

• Men aged 21–50; premenopausal women aged 21–47
• BMI 22.0–27.9 — normal weight to slightly overweight, nobody with obesity
• Healthy — this was not a trial in patients

The people who enrolled had a mean age of 38, and 70% were women. Hold onto that, because it’s the single biggest limit on who the results apply to.

Baseline calorie needs weren’t self-reported. Each participant’s energy requirement was measured with two two-week doubly-labelled-water periods — the gold standard, and unusually rigorous for a diet trial. The 25% target was set against that measured baseline, and doubly-labelled water was repeated at 6, 12, 18 and 24 months, so the achieved restriction is a measurement rather than a food diary.

The support was serious: 27 days of provided meals to teach portion sizes, then 24 group and individual counselling sessions across the first 24 weeks, weekly. Control participants carried on as normal with quarterly check-ins.

The gap: 25% prescribed, 12% delivered

Here is the number the headlines skip. Averaged across the two years, the restriction group achieved 11.9% restriction (standard error 0.7) — roughly half the prescribed dose. In the cardiometabolic paper, that is intake falling from about 2,467 to 2,170 calories a day. (According to PubMed: Kraus et al., The Lancet Diabetes & Endocrinology 2019, DOI.)

Do the subtraction and the famous longevity intervention is a deficit of about 297 calories a day. A pint and a packet of crisps. Two flat whites and a pastry. Something you could lose by not finishing dinner.

I want to be careful about what that gap means, because it can be read two ways and both are true.

The deflating read: nobody in this trial did the thing the mouse studies did. Rodent protocols run at 30–40% restriction for life. CALERIE ran at 12% for two years and then stopped. The extrapolation from one to the other is much longer than it looks.

The genuinely useful read: this was as supportive an environment as a calorie-restriction study will ever get — free food, dietitians, weekly counselling, objective measurement, and volunteers motivated enough to sign up for two years of it. If that averages 12%, then 12% is the honest planning number for sustained restriction in free-living adults, and a protocol that assumes everyone holds 25% is designing around the exception.

One important caveat on my own framing there: 11.9% is a group mean, not a cap. Individual doses varied a lot. A 20% cut sat at the 75th percentile of the restriction group at twelve months — so roughly a quarter were that deep — though by twenty-four months 20% had drifted up to the 87th percentile, meaning only about one in eight was still holding it. Plenty of people managed more than 12%; fewer managed it for the full two years. And as the dose–response results below show, the ones who cut deeper did get more out of it. The honest statement is that 12% is what the average person sustained, not the most anyone could.

The trial accidentally ran an adherence experiment alongside the aging experiment, and the adherence result may be the more actionable of the two.

Bit nerdy, but worth noting: the trial’s own analysts made the same adjustment. Rather than estimate an effect at the prescribed 25%, the 2023 methylation paper deliberately modelled a 20% dose instead — precisely because so few participants got to 25% that a number at the prescribed dose wouldn’t have meant much.

What two years of 12% actually did

The cardiometabolic results are the strong part

This is where CALERIE is most convincing, and least discussed. Over two years, the restriction group lost a sustained 7.5 kg against a 0.1 kg gain in controls, and essentially every conventional risk marker moved the right way: LDL cholesterol, total-to-HDL cholesterol ratio, and both systolic and diastolic blood pressure all improved significantly from baseline, alongside C-reactive protein, insulin sensitivity index, and metabolic syndrome score.

Two details raise it above ordinary weight-loss findings. First, these were people who were already healthy and not obese — there was no obvious pathology to fix. Second, the authors ran a sensitivity analysis controlling for relative weight change and the responses held up, which argues the benefit isn’t purely “they weighed less.”

Supporting evidence points the same way. Urinary F2-isoprostanes, a validated whole-body oxidative-stress marker, fell 17% at 12 months and 13% at 24 months in the restriction group relative to controls. (According to PubMed: Il’yasova et al., Aging Cell 2018, DOI.)

The aging-clock result is the famous part, and the weaker one

In 2023, Waziry and colleagues ran DNA-methylation assays on banked CALERIE blood and asked whether restriction slowed biological aging. This is the paper behind every “calorie restriction slows aging, proven in humans” headline. (According to PubMed: Waziry et al., Nature Aging 2023, DOI.)

Three measures, three different answers:

The DunedinPACE effect corresponds to a 2–3% slowing in the pace of aging, and it held from 12 months through 24. There was a dose–response: participants who achieved more than 10% restriction showed roughly d = −0.33 at both timepoints, against −0.19 and −0.14 in those below 10%. Modelled at a 20% dose, the effect roughly doubled to d ≈ −0.43.

That is a real result from a real randomised trial, and it’s the first time a geroscience intervention moved an aging biomarker under randomisation. It is also, in the authors’ own words, small. The paper states plainly that the two groups’ DunedinPACE trajectories show close to 90% overlap — pick a restriction participant and a control at random and you’d struggle to tell which was which.

The split between the three clocks isn’t noise, incidentally. It’s the most interesting mechanistic hint in the paper. PhenoAge and GrimAge were trained to predict mortality risk at a single point in time, so they encode accumulated damage — a stock. DunedinPACE was built from two decades of within-person physiological change, so it tracks a rate — a flow. Two years of eating slightly less plausibly changes your current rate without much denting a lifetime’s accumulation. If you’ve ever wondered why different biological age tests disagree, this is a textbook case: the clocks aren’t measuring the same thing, so they aren’t obliged to agree.

The caveats the headlines skipped

It cost bone

The trial’s safety paper is the one nobody cites next to the aging headline. At both 12 and 24 months, bone mineral density at the lumbar spine, total hip and femoral neck was significantly lower in the restriction group than in controls. Three participants were withdrawn on safety grounds. Within the restriction group, nervous-system, musculoskeletal and reproductive disorders were significantly more common in the normal-weight participants than the overweight ones — that is, the leaner you started, the worse you tolerated it. The authors’ conclusion is that two years of restriction at CALERIE’s achieved level was safe and well tolerated, with the explicit caveat that close monitoring for excessive bone loss and anaemia is important. (According to PubMed: Romashkan et al., Oncotarget 2016, DOI.)

About 2 kg of the loss wasn’t fat

Of the 7.5 kg lost, 5.3 kg was fat mass — 71%. Which means roughly 2.2 kg was not fat. For a 38-year-old that’s an acceptable trade. For someone in their late sixties, where muscle mass is one of the better predictors of how well you age, losing two kilos of lean tissue and measurable bone density in exchange for a 2–3% shift in a methylation score is a very different sum — and CALERIE studied nobody in that bracket.

The mortality figure is borrowed, not measured

You’ll see “a 10–15% reduction in mortality risk, comparable to smoking cessation” attached to CALERIE. That number is in the paper, but read where it comes from: it’s what observational cohorts show for people with slower DunedinPACE. It is an extrapolation from the marker to an outcome the trial never measured. Follow-up stopped when the intervention did. Whether any of this converts into fewer diseases or extra years is, in the authors’ phrasing, currently unknown.

The trial studied the wrong people for the question most people are asking

Mean age 38, BMI 22–28, healthy, 70% women, 77% white. The findings get quoted at people two or three decades older, which is precisely the group where sarcopenia and fracture risk stop being side effects and become the main event. That’s not a flaw in the trial — it’s a flaw in how it’s used.

The hunger was real, if survivable

One nice piece of honesty from the behavioural data: dietary restraint and self-efficacy went up, hunger spiked but was transient, and disinhibited eating increased. (According to PubMed: Dorling et al., Appetite 2019, DOI.) Against that, the quality-of-life results were genuinely reassuring: at 24 months the restriction group had better mood, less tension, better general health scores and improved sexual drive, with better sleep duration at 12 months. (According to PubMed: Martin et al., JAMA Internal Medicine 2016, DOI.) People eating less were not, on the whole, miserable. That surprised me.

What I take from it

CALERIE is a good trial that gets quoted badly. Stripped of the headline, here is what it supports:

1. A ~300 calorie daily deficit, sustained for two years, improves essentially every cardiometabolic marker in healthy non-obese adults — including in people who had nothing obviously wrong with them. That’s the strongest, least-hyped finding in the whole programme.
2. It slowed one aging measure by 2–3%, and did nothing to two others. Real, randomised, small, and not yet connected to any clinical outcome.
3. Twelve per cent is the realistic planning number. Not because 25% doesn’t work — the deeper cutters did better — but because that’s where the average landed in the most supported trial anyone will ever run, and only about one participant in eight was still at 20% or deeper after two years.
4. The costs are real and land hardest on the leanest people. Bone density, lean mass, and worse tolerance in normal-weight participants.

Which points at an unglamorous conclusion: if you want CALERIE’s benefits, the version worth copying is a modest deficit with the protein and resistance training to defend your muscle and bone — not the heroic 25%, which the trial’s own participants couldn’t manage either. Much the same lesson falls out of the fasting evidence, where the scheduling tools also turn out to work mainly by helping people eat a bit less, and where the lean-mass cost shows up again.

Slightly deflating, that. Also quite good, actually — a 300-calorie deficit is a thing a person can genuinely do, which is more than most of the longevity literature can say.