The phrase "biological age" gets attached to four very different measurements that answer four different questions. PhenoAge estimates how old your bloodwork looks relative to a reference cohort. GrimAge predicts how likely you are to die in the next decade. DunedinPACE measures how fast you are aging per chronological year. Horvath's original clock estimates cellular age from DNA methylation. They are not interchangeable, they do not replace each other, and which one you should care about depends on what you are trying to learn. This article walks through each, with the original methodology papers, validation work, current cost, and the practical case for use.
Why "biological age" is plural
Chronological age is one number. Biological age, in the most useful framing, is a family of measurements that each estimate different aspects of the same underlying process: cumulative wear, system dysfunction, and time-to-disease. The first biological age proxies came from clinical chemistry — a person whose bloodwork looked like a typical 50-year-old's was given a biological age of 50, regardless of their birth year. The next generation came from DNA methylation patterns, which turn out to drift with age in highly predictable ways. The most recent generation moves further toward predicting outcomes directly: not "how old does this person look biologically" but "what is this person's mortality risk and rate of aging right now."
The four methods below are the ones with the most validation and the most clinical traction. There are dozens of other candidates, including telomere length panels, glycan clocks, transcriptomic clocks, and proteomic predictors. None of them have yet accumulated the evidence base of the four covered here.
PhenoAge (Levine, 2018)
Levine and colleagues (Aging, 2018, training cohort n=9,926 NHANES participants) developed PhenoAge by fitting a Cox proportional hazards model to predict 10-year all-cause mortality from nine standard clinical chemistry markers plus chronological age. The nine inputs are albumin, creatinine, glucose, C-reactive protein, lymphocyte percentage, mean cell volume, red cell distribution width, alkaline phosphatase, and white blood cell count.
The output is expressed as a "phenotypic age" in years. A 50-year-old whose PhenoAge calculation returns 44 has bloodwork that resembles the average 44-year-old in the reference cohort, and is associated with the mortality risk of that age. The headline validation result: PhenoAge in the NHANES test cohort predicted all-cause mortality with a hazard ratio of about 1.05 per year of PhenoAge above chronological age, after adjusting for chronological age itself.
The case for PhenoAge as a starting point: it uses tests that any standard lab performs, costs under $100 in most countries, and is easy to track over time. The case against: it is a function of the inputs, which means it is sensitive to acute states (inflammation from a cold, dehydration affecting creatinine) and does not capture cellular aging directly. It is a useful proxy, not a ground truth.
The PhenoAge calculator is open-source and the methodology is fully documented in the original Aging paper. Several consumer services (InsideTracker, SiPhox, others) report PhenoAge as part of their panels.
GrimAge (Lu, 2019)
GrimAge (Lu et al., Aging, 2019) takes a fundamentally different approach. Rather than fit a methylation clock to chronological age, Lu and colleagues fit methylation patterns to seven plasma protein levels that themselves predict mortality (including GDF15, adrenomedullin, beta-2 microglobulin, cystatin C, leptin, PAI-1, and TIMP-1), plus smoking pack-years. The resulting "GrimAge" is a methylation-derived estimate of these mortality-associated proteins, calibrated to predict time to death.
In the Framingham Heart Study training cohort, GrimAge outperformed all prior epigenetic clocks for predicting all-cause mortality, time to coronary heart disease, time to cancer, and time to age-related conditions. The hazard ratio per year of GrimAge above chronological age was approximately 1.10 for all-cause mortality, roughly twice the magnitude of PhenoAge in the same cohort.
GrimAge requires DNA methylation profiling, typically on the Illumina EPIC array, from a blood sample. Cost is currently $300 to $500 through commercial providers (TruDiagnostic, Elysium Index, others). Turnaround is two to six weeks. A second-generation GrimAge V2 (Lu et al., Aging, 2022) improved precision and is the current standard.
The case for GrimAge: it is the strongest single predictor of mortality among widely available tests. The case against: cost, turnaround time, and the fact that it is fundamentally an outcome prediction rather than a direct measurement of aging biology. A person who quits smoking will see GrimAge drop substantially, which is biologically meaningful but does not mean their cellular age has actually reversed.
DunedinPACE (Belsky, 2022)
DunedinPACE is the conceptual outlier. Belsky and colleagues (eLife, 2022) developed it from the Dunedin Multidisciplinary Health and Development Study, a longitudinal cohort followed since birth in 1972-1973. By comparing methylation patterns at age 26, 32, 38, and 45 in the same individuals, the team could measure the actual rate of biological change per chronological year, rather than a cross-sectional age estimate. DunedinPACE is a methylation-based estimate of that rate.
A DunedinPACE of 1.0 means you are aging at one biological year per chronological year (the cohort average). A score of 0.9 means you are aging 10 percent slower than the average person of your age. A score of 1.15 means you are aging 15 percent faster.
This is the metric that fits most cleanly into the question longevity practitioners care about: am I bending the curve right now, this year? PhenoAge and GrimAge produce a static estimate that changes slowly. DunedinPACE produces an estimate of velocity, which can in principle respond to interventions on a months-not-years timescale, although the long-term test-retest data is still limited.
Validation work includes Belsky et al.'s original eLife paper showing that DunedinPACE predicts all-cause mortality, disability incidence, and incident chronic disease independently of chronological age and prior methylation clocks. Higher DunedinPACE in the CALERIE caloric restriction trial cohort correlated with frailty progression (Waziry et al., Nature Aging, 2023). Several consumer services now report DunedinPACE (TruDiagnostic was first to commercialize it), with current cost in the $200 to $400 range.
The case for DunedinPACE: it answers the velocity question better than any other available metric. The case against: as a single-cohort-derived measure, it may not generalize perfectly to populations of different ancestry, and the response to interventions is still being characterized.
Horvath (2013)
Horvath's original multi-tissue clock (Genome Biology, 2013, training across 8,000 samples from 51 tissues) was the breakthrough that established DNA methylation as a usable biological age estimator. The clock uses 353 CpG sites and predicts chronological age across most human tissues with a median error of 3.6 years.
It is important to understand what the Horvath clock does and does not do. It estimates chronological age very well. Its deviation from chronological age (the "epigenetic age acceleration") is associated with mortality, but the effect size is modest compared to PhenoAge or GrimAge, which were explicitly built to predict mortality rather than chronological age.
The Horvath clock remains valuable as a research tool and as a cross-tissue methylation clock. Most consumer epigenetic age tests today use PhenoAge, GrimAge, or DunedinPACE rather than the original Horvath clock for outcome prediction.
A second Horvath clock, the Skin and Blood Clock (Horvath et al., Aging, 2018), is more precise for blood specifically and is sometimes reported alongside the others.
What Bryan Johnson's data tells us
Bryan Johnson's Blueprint protocol is the most public single-subject longitudinal data set on these clocks. Across roughly three years of self-reported data, he has documented PhenoAge five to seven years below chronological, GrimAge several years below chronological, and DunedinPACE consistently around 0.69 to 0.78, meaning his measured rate of aging has been roughly 25 to 30 percent slower than the population mean. These data are self-reported and not peer-reviewed, but they are useful as an existence proof that meaningful clock movement is achievable in a real human, not just in mice. The interventions are well-documented (sleep regularity, structured exercise, hyper-controlled diet, modest pharmacology) and broadly track the directions the longevity literature would predict.
Practical guidance: which test, when
If you want to start with one test today, PhenoAge is the most accessible. It uses standard bloodwork, costs the least, and gives you a tracking baseline you can re-test every six months.
If you want the strongest single mortality predictor and can afford it, GrimAge V2 once a year is the highest-information annual test. It is what you want if your question is "where am I on the long-arc curve."
If you are running interventions and want to see whether they are working on a months-not-years timescale, DunedinPACE is the metric built for that question. Pair it with PhenoAge for the static estimate and you have most of the information available from current consumer testing.
If your goal is research or you are comparing across tissue types, the original Horvath clock and its Skin and Blood derivative are the established references but are not the most actionable consumer tests.
A final note on test-retest noise. All methylation clocks have some assay-level variability. Higgins-Chen et al. (Nature Aging, 2022) developed PC-clocks (principal-component-based clocks) that substantially reduce this noise, and most reputable commercial providers now use PC-corrected versions. If you are comparing two readings six months apart, look for confidence intervals or repeat-test variation reporting. A drop of 0.3 years in PhenoAge between tests may be inside the assay noise and not a real change.
Key takeaways
- "Biological age" refers to a family of measurements, not a single number. PhenoAge, GrimAge, DunedinPACE, and Horvath answer different questions.
- PhenoAge: cheap, blood-based, useful as a tracking baseline. PhenoAge above chronological age associates with mortality risk but is sensitive to acute states.
- GrimAge V2: methylation-based, strongest single predictor of all-cause mortality, $300-500, slower turnaround.
- DunedinPACE: methylation-based rate-of-aging metric, best suited to track whether interventions are working over months rather than years.
- Horvath: foundational chronological age estimator, less actionable than PhenoAge or GrimAge for individual mortality prediction.
- Look for PC-corrected versions and confidence intervals when comparing successive readings.
Sources
1. Levine ME, et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging. 2018;10(4):573-591. https://doi.org/10.18632/aging.101414 2. Lu AT, et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging. 2019;11(2):303-327. https://doi.org/10.18632/aging.101684 3. Lu AT, et al. DNA methylation GrimAge version 2. Aging. 2022;14(23):9484-9549. 4. Belsky DW, et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife. 2022;11:e73420. https://doi.org/10.7554/eLife.73420 5. Horvath S. DNA methylation age of human tissues and cell types. Genome Biology. 2013;14:R115. https://doi.org/10.1186/gb-2013-14-10-r115 6. Horvath S, et al. Epigenetic clock for skin and blood cells applied to Hutchinson Gilford Progeria Syndrome and ex vivo studies. Aging. 2018;10(7):1758-1775. 7. Higgins-Chen AT, et al. A computational solution for bolstering reliability of epigenetic clocks. Nature Aging. 2022;2:644-661. 8. Waziry R, et al. Effect of long-term caloric restriction on DNA methylation measures of biological aging in healthy adults from the CALERIE trial. Nature Aging. 2023;3:248-257.
---
Want VITA to do this for you automatically? Join the waitlist