Chronological age — the number of years since you were born — is a remarkably poor predictor of how long you'll live. Two 60-year-olds can have biological systems that look like they belong to 45 and 75 respectively. The question is: how do you measure which one you are, and can you move that number?
The answer came from an unexpected direction: DNA methylation. In 2013, Steve Horvath at UCLA discovered that the methylation pattern across hundreds of specific CpG sites in the genome changes with age in a highly predictable way — predictable enough to create a molecular clock that estimates chronological age from a blood or tissue sample with remarkable accuracy. That was the Horvath clock, the first generation of epigenetic age estimators, and it opened a field that has since produced half a dozen distinct clocks — each measuring different aspects of biological aging with different predictive power for different outcomes.
The challenge for consumers is that these clocks are not interchangeable. A test that tells you your "epigenetic age" may be reporting any of them — and the gap between what each measures is substantial enough that choosing the wrong test for your purpose means getting meaningless data.
DNA Methylation: The Technology Behind the Clocks
DNA methylation is an epigenetic mark — a chemical modification (addition of a methyl group to cytosine bases at CpG dinucleotides) that regulates gene expression without changing the underlying DNA sequence. Methylation patterns are cell-type specific, environmentally responsive, and — critically — change systematically with age in a way that can be measured and modeled.
Modern methylation arrays (the Illumina EPIC array is most commonly used commercially) can measure methylation at 850,000 CpG sites across the genome simultaneously from a blood sample. The different epigenetic clocks use different subsets of these sites, weighted by algorithms trained on specific outcomes — some trained on chronological age, others on mortality risk, others on disease risk, others on the pace of aging itself.
This means a single blood sample processed on the same array can generate results from multiple clocks simultaneously — which is why most commercial testing services report several clocks from one sample.
Horvath Clock Validation
The original Horvath 2013 clock used 353 CpG sites and was trained on chronological age across 51 tissue types. It achieves mean absolute error of approximately 3.6 years from chronological age and shows pan-tissue consistency remarkable for a single algorithm. First-generation clocks like Horvath correlate with mortality but less strongly than second-generation clocks trained directly on mortality outcomes.
GrimAge — Mortality-Trained Clock
Lu et al. 2019 trained GrimAge not on chronological age but on time-to-death, making it the strongest predictor of all-cause mortality among the established clocks. Each 1-year increase in GrimAge over chronological age is associated with a 12–19% increase in all-cause mortality risk in prospective studies. GrimAge acceleration is also associated with smoking, BMI, inflammation, and cardiovascular risk factors.
DunedinPACE — Pace of Aging
Belsky et al. 2022 trained DunedinPACE on longitudinal changes in 19 biomarkers of organ system aging across ~1000 individuals tracked for decades in the Dunedin birth cohort. Unlike other clocks, DunedinPACE reports the current speed of aging (faster or slower than 1 year per calendar year) rather than an estimated biological age. It's the most sensitive to lifestyle interventions in short-term trials.
PhenoAge — Disease Burden Clock
Levine et al. 2018 trained PhenoAge using clinical phenotypic markers associated with morbidity (albumin, creatinine, glucose, CRP, lymphocyte percentage, mean cell volume, red blood cell distribution width, alkaline phosphatase, white blood cell count) and then built a methylation clock to predict this phenotypic age. PhenoAge acceleration correlates strongly with cardiometabolic disease risk.
The Clock Comparison: What Each One Measures
Horvath Clock (2013)
Trained on: chronological age, 51 tissue types, 353 CpGs.
Strengths: pan-tissue consistency, robust technical validation across labs.
Weaknesses: relatively weak predictor of mortality compared to later clocks; calibrated to chronological age rather than health outcomes.
Best for: understanding basic epigenetic age; tissue-to-tissue comparisons; research applications.
Verdict: Foundational but superseded for clinical decision-making by GrimAge and DunedinPACE.
Hannum Clock (2013)
Trained on: chronological age in blood cells only, 71 CpGs.
Strengths: blood-specific, technically clean.
Weaknesses: blood-only (not tissue-generalizable); correlated with Horvath but less predictive of health outcomes.
Best for: historical comparison; rarely used as a primary metric in modern commercial tests.
Verdict: Largely superseded; still reported by some services for completeness.
PhenoAge (2018)
Trained on: clinical biomarker composite of disease risk phenotype.
Strengths: directly tied to cardiometabolic disease outcomes; PhenoAge acceleration strongly predicts diabetes, CVD, and all-cause mortality.
Weaknesses: trained on a single phenotypic composite; may miss aging pathways not captured in those biomarkers.
Best for: understanding metabolic health trajectory; correlates well with CRP, glucose, albumin changes.
Verdict: Highly actionable for individuals focused on metabolic health outcomes.
GrimAge (2019)
Trained on: time-to-death in prospective cohorts.
Strengths: strongest single predictor of all-cause mortality; GrimAge acceleration robustly predicts cardiac events, cancer mortality, and overall survival in multiple validation cohorts.
Weaknesses: less responsive to short-term lifestyle interventions than DunedinPACE.
Best for: overall mortality risk assessment; the "headline" biological age metric most clinicians focus on.
Verdict: The current gold standard for mortality-predictive biological age.
DunedinPACE (2022)
Trained on: longitudinal rate of change across 19 organ system biomarkers.
Strengths: most sensitive to short-term interventions (diet, exercise, caloric restriction); reports pace of aging (1.0 = average; <1.0 = aging slower) rather than static age estimate; used in caloric restriction trials (CALERIE) and shown to decelerate with interventions.
Weaknesses: single birth cohort training data; less validated for absolute mortality prediction.
Best for: tracking the impact of lifestyle interventions; the most practical clock for n=1 optimization.
Verdict: Best choice for measuring whether your interventions are actually working.
TruAge (TruDiagnostic) / Elysium Index
These commercial clocks incorporate multiple established clocks (Horvath, GrimAge, PhenoAge, DunedinPACE) into proprietary composite scores alongside additional algorithms trained on proprietary datasets. TruDiagnostic's TruAge PACE is specifically designed for intervention tracking. Elysium Health's Index uses a validated subset of CpG sites and reports a single composite score.
Strengths: combine multiple validated clocks; consumer-friendly reporting with trend tracking.
Weaknesses: proprietary weighting not fully transparent; more expensive per test than running individual clocks.
Verdict: Good all-in-one options; TruDiagnostic is the leader for intervention-focused users.
Commercial Testing Services Compared
| Service | Clocks Reported | Sample Type | Cost | Best For |
|---|---|---|---|---|
| TruDiagnostic (TruAge) |
Horvath, GrimAge, PhenoAge, DunedinPACE, proprietary composite | Blood (at-home fingerprick) | $299–$399 | Comprehensive baseline + intervention tracking |
| Elysium Health (Index) |
Proprietary composite based on validated CpGs | Saliva (at-home) | $299 (single) $499/year subscription |
Ongoing monitoring with trend reporting |
| Iollo (Metabolomic age) |
Metabolomic age (not methylation-based) | Blood (at-home) | $299 | Metabolic aging specifically; different signal |
| Epigenomics (via physicians) |
All major clocks reportable | Blood draw (clinic) | $150–$300 | Clinician-ordered, most data |
| Chronomics | Multiple clocks + gut microbiome | Blood + stool | $399–$499 | Combined epigenetic + microbiome profiling |
What Actually Moves the Clocks
The most valuable question for anyone who has tested their biological age is: what interventions are proven to reduce it? The evidence is cleaner for some clocks than others.
DunedinPACE is the most responsive to interventions in controlled trials. The CALERIE-2 trial of 25% caloric restriction showed a meaningful deceleration of DunedinPACE after 2 years. Exercise training, particularly aerobic exercise and resistance training combined, consistently reduces DunedinPACE in intervention studies. The Mediterranean diet and other anti-inflammatory dietary patterns are associated with lower DunedinPACE in cohort studies.
GrimAge is strongly associated with smoking — smoking cessation causes measurable GrimAge deceleration over 1–3 years. BMI reduction reduces GrimAge. Sleep improvement (particularly addressing sleep apnea) reduces GrimAge. The lifestyle interventions with the largest GrimAge effects are smoking cessation, weight loss, and regular exercise.
PhenoAge responds to dietary interventions targeting its training biomarkers: reducing CRP (anti-inflammatory diet), improving albumin (adequate protein intake), controlling blood glucose (low glycemic diet, reduced processed carbohydrates), and managing red blood cell volume (B12, folate, iron adequacy).
Several supplements have shown effects on epigenetic clocks in small trials: rapamycin (GrimAge reduction in a small open-label study), spermidine, vitamin D (in deficient individuals), omega-3 fatty acids, and certain senolytic combinations. These are preliminary findings requiring replication.
Testing Protocol for Maximum Signal
- Choose TruDiagnostic or Elysium for consumer-grade testing — both report multiple clocks
- Test at baseline before any major intervention change
- For intervention tracking, use DunedinPACE as primary metric — it's the most intervention-sensitive
- For mortality risk context, prioritize GrimAge as your headline number
- Retest after minimum 6 months of consistent intervention — clocks have some test-retest variability
- Control confounders: same time of day, similar recent activity, no acute illness at time of draw
- Best value: order 2–3 tests per year if actively tracking interventions
- Interpret within context: no single clock should drive major medical decisions alone
The Test-Retest Reliability Problem
A practical concern for anyone using epigenetic clocks to track interventions is test-retest variability. Epigenetic clocks are not perfectly reproducible — repeat measurements on the same individual within a short period show variability of approximately 1–3 years depending on the clock and testing service. This means a 1-year apparent reduction in biological age after an intervention might be partially attributable to measurement noise rather than a true change.
This is why DunedinPACE is preferable for intervention tracking — its continuous (rather than absolute) output and its design specifically for detecting change makes it less prone to false-positive "improvements" from measurement variability. It's also why TruDiagnostic's approach of testing multiple clocks simultaneously and reporting confidence intervals around the estimates is more informative than a single clock single number.
Rule of thumb: a 2+ year improvement in GrimAge or a 0.05+ reduction in DunedinPACE represents a likely real signal; smaller changes may be noise. Serial testing over 2–3 time points provides much more confidence than a single before/after comparison.
Biological Age vs. Other Aging Biomarkers
Epigenetic clocks are not the only biological age estimates available. Understanding their place in the landscape helps calibrate expectations:
Telomere length testing (offered by companies like LifeLength and TeloYears) predicts mortality independently of epigenetic age but with lower precision. Telomere length has high inter-individual variability and relatively high test-retest noise. It's less responsive to short-term interventions than epigenetic clocks.
Proteomics-based biological age (SomaScan platform, used by Levels and others) measures circulating protein levels that change with age. A 2023 paper in Nature Aging by Lehallier et al. showed that plasma proteomic aging clocks predict cognitive decline and physical function better than DNA methylation clocks in some contexts — suggesting complementary information rather than redundancy.
Blood test composites (such as PhenoAge calculated from standard blood panel data) are free if you have a comprehensive metabolic panel — a useful first step before committing to methylation testing.
Interpreting Your Results: A Practical Framework
Getting your first epigenetic age test back and seeing a number that's significantly higher than your chronological age is confronting. Here's how to contextualize it usefully rather than catastrophize or dismiss it:
First: which clock are you looking at? A high Horvath age and a normal GrimAge suggests your cells show some age-related methylation drift but your mortality risk profile is normal. A high GrimAge with high DunedinPACE is the combination that warrants the most attention and the most aggressive lifestyle intervention.
Second: what's your uncertainty range? A single test with no prior baseline provides minimal actionable information beyond a rough quartile sense of where you stand. The real value comes from repeat testing after 6–12 months of a defined intervention protocol.
Third: what's your lifestyle backdrop? If you're sedentary, overweight, sleeping poorly, and eating a processed diet, a high epigenetic age is expected and actionable — address those fundamentals before investing in supplement protocols. If your lifestyle is already optimized, the test may identify specific areas (metabolic function, inflammation, etc.) where targeted supplements or further investigation make sense.
The promise of biological age testing is genuinely exciting: for the first time in history, we have tools that can tell us whether the interventions we're making are actually slowing aging at a molecular level, rather than relying on indirect proxies like cholesterol numbers or VO2max. The technology is real. Using it well requires understanding what each number means — and what it doesn't.