GrimAge, PhenoAge and DunedinPACE: Next-Generation Aging Clocks
Three clocks, three different questions. What each one was trained to predict, how they differ in practice, and which to use when evaluating an intervention.
The Short Answer
PhenoAge, GrimAge and DunedinPACE are the three most consequential aging clocks developed after the first generation, and they differ in a way that determines how each should be used. PhenoAge was trained to predict a mortality-weighted composite of nine clinical biomarkers. GrimAge was trained to predict time to death, using methylation-based surrogates for plasma proteins and smoking history, and is the strongest outcome predictor of the three. DunedinPACE was trained on two decades of repeated multi-organ measurement in a birth cohort and estimates the current rate of aging rather than accumulated age, which makes it the appropriate choice for judging whether an intervention has changed a trajectory.
Why the First Generation Was Not Enough
The first epigenetic clocks were trained to predict chronological age, and they succeeded. That success was also the problem: a model that predicts chronological age perfectly contains no information beyond the calendar, and the deviation people care about is, from the model's perspective, error to be minimised.
The insight behind the second generation was to change the target. Instead of asking a model to predict how many years someone has lived, ask it to predict something that matters: how likely they are to die, or what their clinical biomarkers look like. The methylation sites selected under that objective are different, and the resulting scores correlate less with chronological age and more with outcomes.
PhenoAge
Construction. A two-stage design. First, a phenotypic age was built from nine clinical measures, albumin, creatinine, glucose, C-reactive protein, lymphocyte percentage, mean cell volume, red cell distribution width, alkaline phosphatase and white cell count, plus chronological age, trained against mortality. Then a methylation model was trained to predict that phenotypic age, yielding a methylation clock with a clinically grounded target.
Behaviour. Correlates with chronological age less tightly than first-generation clocks. Age acceleration on PhenoAge associates with mortality, cardiovascular events, and physical and cognitive functional decline across multiple cohorts.
Notable property. Because its target is built from clinical chemistry, PhenoAge is conceptually close to a blood biomarker composite, and the non-methylation version can be calculated from a routine blood panel at almost no cost. For many practical purposes this is the most accessible member of the family.
Best used for. An outcome-weighted status assessment that connects cleanly to actionable clinical markers.
GrimAge
Construction. A three-stage design. Methylation surrogates were developed for seven plasma proteins associated with mortality and for smoking pack-years. Those surrogates, plus age and sex, were then combined to predict time to death. The output is expressed in years.
Behaviour. The strongest mortality and morbidity predictor among widely used clocks. Associations have been reported with cancer incidence, cardiovascular events, cognitive decline, physical function and multimorbidity, generally exceeding those of PhenoAge and greatly exceeding first-generation clocks.
Important caveat. The smoking pack-years surrogate is a substantial contributor to its performance, which is unsurprising given that smoking is the largest single modifiable factor in this space. For a lifelong non-smoker, that component contributes less discriminating information, and the reading should be interpreted with that in mind. This is not a flaw, it is how the model was built, but a GrimAge result is not a uniformly weighted read across all of aging biology.
Best used for. The strongest available single-shot epigenetic estimate of outcome-relevant biological age.
DunedinPACE
Construction. Fundamentally different. Rather than a cross-sectional target, it was built from a birth cohort followed from birth into midlife, with nineteen biomarkers across multiple organ systems measured repeatedly at several ages. The rate of change across those markers was calculated per individual, and a methylation model was trained to predict that rate.
Output. A ratio rather than a number of years. A value of 1.0 means biological aging at the population-typical rate of one year of biological change per calendar year. A value of 0.85 means aging fifteen percent slower; 1.2 means twenty percent faster.
Why it is different in kind. Accumulated-age clocks answer where you are. A pace measure answers how fast you are moving. For evaluating an intervention, the second question is the right one, because a person's accumulated damage does not undo itself quickly but their rate of accumulation can plausibly change.
Validation. Higher pace associates with faster functional decline, earlier morbidity onset and higher mortality across several cohorts. It has also been used in intervention analysis, most notably in the secondary analysis of a randomised caloric restriction trial in humans, where the intervention group showed a slower pace of aging, while accumulated-age clocks showed no significant change in the same participants. That divergence is the clearest empirical demonstration of why the distinction matters.
Best used for. Judging whether an intervention has changed the trajectory.
Side-by-Side
| PhenoAge | GrimAge | DunedinPACE | |
|---|---|---|---|
| Generation | Second | Second | Third |
| Trained to predict | Clinical-biomarker phenotypic age | Time to death | Rate of multi-organ change |
| Output | Years | Years | Ratio (1.0 = typical) |
| Outcome association | Strong | Strongest | Strong |
| Sensitive to intervention | Moderate | Moderate | Highest |
| Main limitation | Reflects current clinical state | Smoking surrogate weighting | Single-cohort derivation |
| Non-methylation version | Yes, from a blood panel | No | No |
How to Choose, and How to Read a Result
The practical guidance is short. If the question is "where do I stand", use a second-generation clock, and prefer a principal-component version if the provider offers one, since technical noise is lower. If the question is "is what I am doing working", use a pace-of-aging measure, because the accumulated-age clocks are less responsive and can miss a genuine trajectory change, as the caloric restriction analysis demonstrated.
Reading a result well means holding three things at once. The measurement has real technical variability, so a small change is not a result. The clocks disagree with each other by design, so a discrepancy is not an error. And any of them can be moved by transient states, so conditions at sampling matter.
What none of them provides is a cause. A high GrimAge acceleration does not indicate which system is driving it, and no clock is a substitute for looking at inflammatory markers, metabolic markers, fitness and sleep. The clock says something is off. The panel says what.
The AEONNN Perspective
The divergence between accumulated-age clocks and pace-of-aging measures in the caloric restriction analysis is the single most important result for anyone building a longevity platform, and it shaped how AEONNN thinks about evaluation. An intervention can change a trajectory without changing an accumulated-damage estimate on any useful timescale, which means a system that judges interventions by accumulated-age readings will systematically conclude that nothing works.
AEONNN Age is therefore constructed to make trajectory legible rather than to report a static score, and it is recomputed as a member's context and data change. It is a composite, not a diagnosis, and it is not any of the clocks described here. Where a member brings a laboratory clock result into Synched Mode, it is weighted according to what that specific clock generation can support.
This is also why continuity is the product rather than the feature. A single reading from the best clock available is worth less than a consistent sequence of composite readings interpreted alongside a member's changing context, and building the second requires a system that persists over years rather than a test that arrives once.
Pillar Matrix mapping
Database Matrix layers
- Evidence Layer (PubMed, Cochrane, ClinicalTrials.gov)
- Population Layer (UK Biobank, NHANES)
- Meta / Consensus Layer (JAMA, BMJ, specialty society positions)
- Mechanistic Layer (KEGG, Reactome, UniProt)
- Innovation Layer (bioRxiv preprints, patent filings)
Frequently Asked
Which clock is the best predictor of mortality?
GrimAge has the strongest reported associations with mortality and morbidity among widely used clocks, exceeding PhenoAge and greatly exceeding first-generation clocks. Its performance depends substantially on a methylation surrogate for smoking pack-years.
What does a DunedinPACE score of 0.9 mean?
It indicates aging at roughly ten percent slower than the population-typical rate, meaning about 0.9 years of biological change per calendar year on the multi-organ markers the measure was trained on.
Why did caloric restriction change DunedinPACE but not other clocks?
Because they measure different things. Accumulated-age clocks estimate damage already present, which does not reverse quickly. A pace measure estimates the current rate of change, which an intervention can plausibly alter. The divergence is the clearest demonstration of why the distinction matters.
Can PhenoAge be calculated without a methylation test?
Yes. The original phenotypic age was built from nine routine clinical markers plus chronological age, and can be calculated from a standard blood panel using the published formula. This is the most accessible member of the family.
Should I test more than one clock?
Testing a second-generation clock alongside a pace-of-aging measure answers two different and both useful questions. Testing several accumulated-age clocks mostly produces disagreement without additional insight.
Does GrimAge penalise former smokers permanently?
GrimAge includes a methylation surrogate for smoking pack-years, and methylation marks associated with smoking do change after cessation, though not all of them return fully to never-smoker patterns. A former smoker should expect that component to reflect history to some degree.
How much can these clocks change with lifestyle change?
Reported intervention effects are typically in the range of one to three years for accumulated-age clocks and a few percent for pace measures, over months to a couple of years. Changes smaller than the assay test-retest variability should not be read as real.
Evidence and review
Any dosage ranges cited here reflect the ranges used in published human trials, not personal recommendations. Evidence in this field moves, so this article is reviewed quarterly and carries its last-updated date above. Nothing here is intended as medical advice, and supplementation should be discussed with a qualified clinician, particularly alongside prescribed medication or an existing condition.