How to Actually Lower Your Biological Age: What the Trials Show

Key Takeaways

The phrase "reverse your biological age" has a specific evidentiary footprint, and it is small. Between the first clock in 2013 and today, fewer than ten randomised trials in humans have measured an epigenetic clock before and after an intervention. Most found nothing between groups. The ones that found something found months. The results that circulate as headlines — two and a half years of reversal, three years in eight weeks — come from the studies with the weakest designs.

This review walks through every verified trial, then the null results, then the large cohort associations that tell you where the real levers are. It is honest to the point of being deflating, and then it explains why that is not actually bad news. If you have not yet read what the clocks measure and how noisy they are, that guide comes first.

The Randomised Evidence, Ranked

DO-HEALTH: omega-3, vitamin D and exercise over three years

This is the best evidence that anything you can buy slows a clock. DO-HEALTH randomised 2,157 adults aged 70 and over, across five European countries, to vitamin D 2,000 IU/day, marine omega-3 1 g/day, a simple home strength programme, or any combination, for three years. In 2025 the investigators reported methylation clocks in the 777 participants with stored blood.

Omega-3 alone slowed PhenoAge, GrimAge2 and DunedinPACE. All three interventions together had an additive effect on PhenoAge that none achieved alone. The standardised effects were 0.16 to 0.32 — in plain terms, 2.9 to 3.8 months of biological ageing over three years. Vitamin D alone did not slow the clocks. First-generation GrimAge was not among the measures omega-3 moved.

The strengths: large, randomised, long, second-generation clocks, a population old enough for the clocks to be meaningful. The weakness: it is a post hoc analysis, and the effect is small enough that no individual could detect it in their own results. It is the backbone of the Biological Age Optimization Protocol precisely because it is the only backbone available.

CALERIE: two years of caloric restriction

CALERIE randomised 220 non-obese adults to 25% caloric restriction or their usual diet for two years — the most rigorous test of restriction ever run in humans. The 2023 clock analysis found that restriction slowed DunedinPACE and produced no significant change in PhenoAge, GrimAge or any other methylation clock. The authors described the effect sizes as small.

Two readings of this are possible, and both are correct. First: the intervention that extends lifespan in nearly every species tested barely registers on the clocks that predict human mortality. Second: DunedinPACE — the clock built to measure rate rather than age — detected a change that age-estimating clocks could not. That is an argument for DunedinPACE as the tracking tool, and an argument against expecting any clock to move far in two years.

Diet trials: mostly null between arms

DIRECT PLUS randomised 256 adults to a healthy diet, a Mediterranean diet or a polyphenol-rich "green" Mediterranean diet for 18 months and measured seven clocks plus DunedinPACE. There were no between-arm differences in clock change. Only within the green-Mediterranean group did greater adherence correlate with attenuated ageing on two first-generation clocks, with a Mediterranean-style pattern associated with about nine months of favourable difference on one of them.

NU-AGE gave 120 elderly Italians and Poles a Mediterranean-style diet for a year. Overall, a trend toward rejuvenation on the Horvath clock; significance only in Polish women and in participants who were epigenetically older at baseline.

TwiNS randomised identical twins to a vegan or a healthy omnivorous diet for eight weeks. The vegan arm showed significant decreases in overall epigenetic age acceleration on principal-component clocks. The authors note the vegan twins also consumed fewer calories, so the result cannot separate diet composition from restriction.

DAMA randomised 219 postmenopausal women to diet and physical-activity interventions in a factorial design for two years. The dietary arm slowed GrimAge. The activity arm reduced stochastic epigenetic mutations but was not reported to change a clock.

MACRO compared low-carbohydrate and low-fat weight-loss diets in 144 obese adults for a year. Clock changes were not associated with changes in cardiometabolic biomarkers and did not mediate the benefits of weight loss. The authors' conclusion is one the field should take seriously: caution in treating epigenetic age as a modifiable intervention target.

The pattern across all five: modest intake and a plant-rich, polyphenol-rich composition track slower ageing in adherence analyses; named diets do not beat each other when randomised.

The famous reversals, examined

TRIIM. A one-year protocol of growth hormone, DHEA and metformin aimed at regenerating the thymus. The 2019 paper reported mean epigenetic age about 1.5 years below baseline after a year — 2.5 years against expectation — with GrimAge two years lower and the effect persisting six months after stopping. This is the most-cited reversal in the field. It was uncontrolled, and the number of participants, nine, is in the full text rather than the abstract. It generated a hypothesis and a larger follow-up trial. It did not establish a treatment.

The eight-week lifestyle trial. Forty-three healthy men aged 50 to 72 were randomised to eight weeks of diet, sleep, exercise and relaxation guidance plus a probiotic and a phytonutrient powder, or to no intervention. The headline was −3.23 years on the Horvath clock versus controls. The details: the sample was saliva, the clock was first-generation, the within-group change in the treatment arm was −1.96 years with p = 0.066 — not significant — and the paper carries two errata. A 2023 follow-up applied the same programme to six women without a control group and reported −4.6 years. These are the studies behind most "methylation-support" supplement programmes, and they are the weakest randomised evidence in this review.

Surgery and weight

Bariatric surgery in 40 severely obese patients reduced Horvath age acceleration by 0.92 years at 12 months, with half of the methylation changes explained by BMI change. But a 2014 study across tissues found obesity-related epigenetic ageing concentrated in the liver — 3.3 years per 10 BMI units — and not reversible in the short term after surgical weight loss. The clocks respond to weight, slowly and partially.

What Has Not Moved a Clock

Metformin. In 4,575 people across two cohorts, metformin use was not associated with slower epigenetic ageing. In a 24-week trial in virally suppressed people with HIV, metformin reduced principal-component PhenoAge in monocytes by 3.5 years — and did nothing in CD8 T cells in the companion single-arm study of eight. Cell-type specific, small, and not generalisable.

Rapamycin. The PEARL trial of weekly rapamycin for 48 weeks reported on body composition and safety; no epigenetic clock endpoint appears in its abstract. No human clock data for rapamycin were verifiable for this review. The rapamycin profile covers what the drug does have evidence for.

Multi-nutrient supplements. Eighty older adults took a combination supplement for 12 weeks. No change in Horvath or Hannum age overall. A reduction appeared only in the subgroup with baseline age acceleration of two years or more — the textbook signature of regression to the mean, not of an effect.

NMN, spermidine, senolytics, peptides. No randomised human trial with a clock endpoint. The epigenetic-clock article on peptides covers what is known mechanistically; none of it has been tested against a clock in people.

The Cohort Evidence: Where the Big Numbers Are

Randomised trials measure what researchers chose to test. The cohorts measure what people actually do, and the associations there are an order of magnitude larger than any trial effect.

A 2021 systematic review of 156 publications confirmed the direction: higher BMI, male sex and HIV infection consistently accelerate the clocks; acceleration consistently predicts mortality, cardiovascular disease, cancer and diabetes. Two cautions. First, GrimAge contains a methylation signature of smoking by construction, so its sensitivity to tobacco is partly built in — though first-generation clocks, tellingly, did not respond to smoking at all in one large German cohort. Second, the physical-activity evidence is almost entirely cross-sectional, and in working-age Finns neither insomnia nor shift-work years survived adjustment; only sleep-apnoea symptoms did.

What This Means in Practice

The honest ranking of levers, by the size and reliability of their clock signal:

1. Do not smoke anything. The largest effect in the literature, on the best clock. Cannabis smoking also accelerated clocks in one cohort, and cessation normalised them.
2. Keep alcohol low. Consistent across three cohorts, three clocks and two age groups, with dose-response.
3. Be aerobically fit. The clock signal is small and mostly cross-sectional; the mortality benefit is enormous regardless.
4. Eat modestly, plant-rich, polyphenol-rich. The adherence analyses point there; no named diet wins when randomised.
5. Fix sleep apnoea and avoid chronic night work. The sleep signals that survive adjustment.
6. Take omega-3 1 g/day, and vitamin D if you are low. Three months over three years, randomised, in older adults. Cheap, safe, real, small.
7. Do not take metformin, rapamycin or growth hormone to move a clock. The evidence does not support it and the risks are not zero.

Notice that items one through five are the standard advice for not dying early, independent of any clock. That is the deflating conclusion and also the reassuring one: the clocks, when they move, move in response to the things already known to matter. They have not revealed a shortcut. Their value is as a research instrument that confirms the slow levers are real, and — used carefully, as the protocol sets out — as a way of checking, once a year, that you are still pulling them.

Frequently Asked Questions

Is a 3-month slowing over 3 years worth anything? At the population level, a shift of that size in a mortality-predicting biomarker is meaningful. At the individual level it is undetectable against the noise of any current test. Take omega-3 for its cardiovascular evidence; the clock effect is a bonus you will never see.

Why did CALERIE move DunedinPACE but not GrimAge? DunedinPACE measures the current rate of ageing, built from longitudinal change; GrimAge estimates accumulated age. A two-year intervention changes the rate before it changes the accumulation. It is also why DunedinPACE is the better clock for tracking an intervention.

The eight-week trial showed three years. Why dismiss it? Not dismissed — downgraded. Saliva, a first-generation clock, a non-significant within-group change and two errata make it hypothesis-generating. If the effect were real at that size, DO-HEALTH and CALERIE would not have struggled to find months.

Does weight loss lower biological age? Slowly and partially. Bariatric surgery moved the Horvath clock by under a year at 12 months, and liver epigenetic age did not reverse in the short term. In the MACRO trial, clock changes did not track the metabolic improvements of dieting.

What about the methylation-support supplements — folate, B12, TMG? They are the logic behind the eight-week programme, and the mechanistic case for TMG and methyl donors is sound. No trial has shown them to change a clock independently of the lifestyle programme they were bundled with.

What should I do first? Read the testing guide so you do not misread a result, then run the protocol, which starts with the behaviours in this table and tests once a year.

Related Research

• Biological Age Tests Explained: GrimAge, DunedinPACE and What They Can't Tell You
• The Biological Age Optimization Protocol: Test, Intervene, Retest
• The Biological Age Optimization Protocol
• The Epigenetic Clock: How Peptides and Interventions Change DNA Methylation Age
• A Data-Driven Longevity Blueprint: What Extreme Optimizers Actually Do