Epigenetic age test explained

An epigenetic age test is an algorithm applied to methylation data

DNA methylation is a chemical mark on DNA, commonly measured at cytosine–phosphate–guanine (CpG) sites. A commercial DNA methylation age test takes a blood or saliva sample, measures methylation across many CpGs, and applies one or more published weighted algorithms. It does not sequence your genome, diagnose a disease, or look at every possible epigenetic mark. It calculates a score from a selected set of methylation sites. The original multi-tissue Horvath clock, for example, combined 353 CpGs with an elastic-net model to predict age. Horvath’s 2013 paper

The label “biological age” can make the output sound more literal than it is. For an age clock, the useful derived quantity is often age acceleration: whether the methylation-predicted age is above or below the value expected for someone of the same chronological age. It is a statistical comparison with a reference model, not a countdown of years remaining. Even the GrimAge developers explicitly caution that an age-acceleration result expressed in years is not equivalent to a change in life expectancy. GrimAge validation paper

That distinction matters because different clocks are built to answer different questions. A clock can predict calendar age extremely well and still be a poor tool for detecting a short-term lifestyle effect. Conversely, a score trained against mortality-related signals can be informative for population-level risk associations without being a personal diagnosis. Treat the name of the clock and its training target as part of the result.

First generation, second generation, and pace are different instruments

First-generation clocks were trained mainly to predict chronological age from methylation patterns. The Horvath clock was developed across 51 healthy tissues and cell types; the Hannum model was developed from whole-blood methylation data from 656 people aged 19 to 101. They are useful demonstrations that methylation tracks age, but their primary target was calendar age. Horvath study Hannum study

Second-generation clocks changed the target. DNAm PhenoAge was trained to estimate a phenotypic-age construct built from chronological age and nine clinical biomarkers, then distilled into a 513-CpG methylation predictor. DNAm GrimAge used methylation surrogates for seven plasma proteins and smoking pack-years, together with age and sex, to predict time-to-death. Those design choices explain why a “good” result on one clock can disagree with another: they are not measuring the same target. PhenoAge paper GrimAge paper

Comparison: what each type is for

DunedinPACE is a rate, not an age

DunedinPACE stands for Pace of Aging Calculated from the Epigenome. Its training target was a 20-year longitudinal Pace of Aging measured from changes in 19 indicators of organ-system integrity in the Dunedin Study, from ages 26 to 45. The final blood-methylation algorithm uses 173 CpGs selected after a reliability screen. Its reference value is 1.0: one biological year of aging per calendar year in the study’s reference cohort. DunedinPACE paper

So a pace score of 0.67 means an estimated pace of 0.67 biological years per one calendar year—about 33% below the reference rate. It does not mean that someone has a biological age of 0.67 years, will live 33% longer, or has permanently changed their aging trajectory. The researchers’ implementation notes likewise describe 1.0 as the normative midlife reference and caution that age affects the expected value. DunedinPACE documentation

The Rejuvenation Olympics leaderboard is a public ranking built around pace results: its page says the average pace uses the best three tests over two years, requires at least three tests, and accepts tests less than four years old. That makes a rank a comparison among voluntary entries under those rules, not proof of clinical superiority or a representative population benchmark. The leaderboard itself does not publish an uncertainty range for an individual’s rank.

Reliable measurement is not the same as a reliable story about change

“Accurate” needs a qualifier. A clock can have a strong population-level association with age or later outcomes and still give a noisy estimate when one person retests. DunedinPACE performed well in technical duplicates in its development paper: an intraclass correlation of 0.96 in 36 replicated adult-blood samples on the 450K array, and 0.87 across 450K-to-EPIC array replicates. Those statistics describe repeated measurement of essentially the same material or samples, not certainty that a small month-to-month shift is a durable biological change. DunedinPACE reliability analysis

Pre-analytic conditions such as time of day and fasting status, plus acute stress, menstrual cycles, and time of year, can complicate repeat interpretation. A 2024 analysis of multiple longitudinal datasets found that first-generation clocks often showed isolated intervention effects that were not supported by higher-reliability versions or other clocks, while stronger agreement appeared after age-accelerating events. Its practical message is appropriately conservative: do not announce “reversal” from one isolated clock change. Longitudinal reliability analysis

What actually moves a result, then? The immediate answer is: methylation measurements move. Over longer periods that signal may reflect aging, cumulative exposures, disease-related stress, immune-cell shifts, behavior, and technical variation. GrimAge includes a methylation surrogate for smoking pack-years by design, so smoking exposure is not merely an incidental correlation in that result. The same longitudinal analysis found concordant increases across several newer clocks after intensive surgery or radiotherapy/chemotherapy, and more consistent decreases across newer or principal-component clocks in a two-year dietary intervention. These are useful observations, not a menu of guaranteed score-lowering tactics or proof that every observed shift is causal. GrimAge construction Intervention analysis

Use a clock as a prompt to investigate, not an instruction to chase a number

Start by recording the exact clock, provider, specimen, collection conditions, and report version. Retest only when you can keep the provider and conditions reasonably comparable, and give a result enough time to become a trend. A consumer provider may present DunedinPACE as feedback on diet, exercise, and sleep, but its own documentation does not supply an individual threshold for a meaningful change. TruDiagnostic’s DunedinPACE documentation

Then put the score beside measurements that answer more direct questions. Your bloodwork and biomarker panel can show what is happening with ApoB, fasting insulin, and inflammatory markers; Abbott Lingo CGM curves show glucose responses in daily life; body composition and wearable trends add a view of behavior, training, recovery, and sleep. Read the clock as one high-level layer across those data, not as a substitute for them. If a clock result and the rest of the picture disagree, that is a reason to check context and discuss clinically relevant findings with an appropriate clinician—not a reason to discard the direct measurements.

For a useful result, ask: Which clock was used? What was it trained to predict? Was the sample taken under comparable conditions? Is the shift larger than the likely noise and repeated on another test? What do my standard measurements and real-world habits say? That framework is more valuable than trying to become younger on a dashboard.

From health data to a written Personal Longevity Blueprint

The point of measuring is to make better decisions. In The Longevity Blueprint, students work through 18 live weekly sessions across 18 weeks, Mondays from 7:00–7:50pm ET. By week 18, they combine their own bloodwork, ApoB, fasting insulin, inflammatory markers, body composition, CGM curves, and wearable data into a written Personal Longevity Blueprint. You finish with a written protocol built on your numbers. It stays yours.

That process is designed to keep an epigenetic clock in proportion: a useful signal alongside more direct markers and practical habits, not a stand-alone target. The instructors who teach are Natalie Blackbourne, Courtney Donofrio, Amy Jamieson (Senior Lecturer, UC Santa Barbara), and Jordan Lattimore. Julie Gibson Clark is founding faculty and ranks #2 on the Rejuvenation Olympics leaderboard for slowest measured aging pace; she does not teach the course.

See how The Longevity Blueprint works

Frequently asked questions

What is an epigenetic clock?

An epigenetic clock is a weighted algorithm that estimates an age-related measure from methylation at selected CpG sites. Its meaning depends on its training target: calendar age, a mortality-related construct, or pace of aging.

Is a DNA methylation age test accurate?

It can be accurate for the population-level target it was trained to estimate, but that does not make a small change in one person’s repeated results definitive. Check the clock, method, conditions, and trend before treating a shift as meaningful.

What does a DunedinPACE of 0.67 mean?

It estimates a pace of 0.67 biological years per calendar year, relative to a 1.0 reference rate. It is a rate estimate, not a biological age, personal life-expectancy forecast, or guarantee of slower aging in the future.

Can diet, exercise, or supplements reverse epigenetic age?

Some studies report clock changes after interventions, but clock responses differ by algorithm and study, and technical or biological variation can contribute. A single score change is not proof that an intervention reversed biological aging.

Should I use an epigenetic age test instead of bloodwork?

No. A clock is best treated as a complementary, high-level signal. Use standard bloodwork and relevant clinical care for direct information about lipids, glucose regulation, inflammation, and other health questions.