#386 - Aging clocks—what they measure, how they work, and their clinical and real-world relevance

Summary of #386 - Aging clocks—what they measure, how they work, and their clinical and real-world relevance

by Peter Attia, MD

42m•April 6, 2026
Listen to Episode

Overview of #386 - Aging clocks—what they measure, how they work, and their clinical and real-world relevance

Host Peter Attia walks through what “aging clocks” are, how they’re built (especially epigenetic/DNA‑methylation clocks), what they may actually measure, and whether they have clinical or individual utility today. He reviews two recent/illustrative studies—a 3‑year randomized trial (DO‑HEALTH) testing simple interventions against four epigenetic clocks, and a brain‑MRI based pace‑of‑aging study—to show both the promise and the major limitations of using clocks as surrogate endpoints.

Key takeaways

  • Aging clocks are prediction models built from biological data (most commonly DNA methylation) intended to estimate biological age or pace of aging, not chronological age.
  • Different clocks are trained on different outcomes (chronological age, clinical biomarkers, mortality risk, or longitudinal rate of decline) and thus can report different results on the same person/intervention.
  • In the DO‑HEALTH randomized trial (≈800 healthy seniors, 3 years), 1 g/day omega‑3 produced small but consistent shifts in several clocks; vitamin D (2,000 IU) and a modest home exercise add‑on did not show consistent effects. The omega‑3 effect translated roughly to ~3 months less “aging” over 3 years (very small).
  • A brain‑MRI based clock can estimate pace of aging and associates with cognitive decline, frailty, and mortality—illustrating other modalities beyond methylation can be useful.
  • Major unresolved question: does changing a clock meaningfully change the clinical outcomes we care about (dementia, cancer, heart disease, disability, mortality)? We currently don’t have definitive answers.
  • Practical advise: clocks are best seen as experimental research tools today. For individual decision‑making, prioritize established, clinically validated measures (blood pressure, glucose, lipids, smoking status, fitness, body composition).

How aging clocks work

DNA methylation basics

  • DNA methylation: epigenetic modification (methyl groups on cytosine‑guanine sites, “CpGs”) that can influence gene expression without changing DNA sequence.
  • Many CpG methylation patterns change predictably with age, making them useful inputs for statistical models that estimate age‑related biology.

Generations/types of clocks (examples)

  • First‑generation clocks (e.g., Horvath): trained to predict chronological age from cross‑sectional methylation data. Good at estimating calendar age but not necessarily clinical risk.
  • Second‑generation / outcome‑trained clocks:
    • PhenoAge: ~500 CpG sites; trained to reproduce a multi‑biomarker score linked to mortality (albumin, glucose, CRP, etc.).
    • GrimAge: ~1,000 CpG sites; estimates methylation‑derived plasma proteins and smoking exposure; predicts time‑to‑death.
    • GrimAge2: updated biomarker set (including CRP, A1c refinements).
  • Pace‑of‑aging clocks:
    • DunedinPACE: ~173 CpG sites; trained on longitudinal physiological decline data to estimate current rate of aging (a first derivative rather than a state).
  • Other approaches: structural/functional imaging (e.g., single MRI) can produce imaging‑based pace estimates analogous to epigenetic clocks.

Studies reviewed

DO‑HEALTH randomized trial (Europe; mean age ~75; n ≈ 800; 3 years; 2×2×2 factorial)

  • Interventions: 2,000 IU vitamin D daily; 1 g/day omega‑3 (330 mg EPA + 660 mg DHA from marine algae); home‑based strength exercise 30 min × 3/week (added on top of baseline activity).
  • Clocks measured at baseline and 3 years: PhenoAge, GrimAge, GrimAge2, DunedinPACE.
  • Findings:
    • Omega‑3: the most consistent effect—significant small shifts in 3 of 4 clocks (PhenoAge, GrimAge2, DunedinPACE), not GrimAge (first gen). Magnitude ≈ ~3 months less aging over 3 years.
    • Vitamin D: no clear effect (dose/ baseline insufficiency issues may matter).
    • Exercise: no independent effect detected—likely limited by participants being already physically active at baseline and modest added dose.
  • Interpretation: small molecular shifts are detectable but effect sizes were tiny and interpretation depends on which clock you trust.

Brain MRI / Dunedin‑PACE imaging study (summary)

  • Structural MRI features from a single scan were analyzed to estimate an individual’s pace of aging.
  • Imaging‑based pace estimates associated with cognitive decline, frailty measures, and mortality—showing non‑methylation modalities can capture aging‑related biology.
  • Suggests multi‑modality clocks (epigenetics, proteomics, imaging) may each capture different aging dimensions.

Main limitations and caveats

  • Different clocks measure different biology: clocks trained on differing endpoints (age vs biomarkers vs mortality vs pace) will disagree.
  • Measurement/technical noise:
    • Assay variability (sample handling, extraction, conversion efficiency, batch effects).
    • Heterogeneity of blood cell mixtures influences methylation signals.
    • Collapsing hundreds of thousands of CpGs to a single score introduces uncertainty.
  • Biological (transient) noise:
    • Short‑term events (infection, intense exercise, inflammation) may transiently shift clocks without long‑term significance.
  • Many clocks are trained on cross‑sectional data rather than longitudinal individual trajectories—limits inference about within‑person change.
  • Lack of proven causal link: showing a clock shifts does not prove that disease risk, disability, or mortality will change.
  • Commercialization risk: direct‑to‑consumer tests are marketed aggressively; if companies sell interventions alongside tests, conflict and overinterpretation are likely.
  • Established metrics still outperform clocks for actionable decisions (decades of outcome data link BP, glucose, lipids, smoking, fitness to clinical outcomes). Life insurers, who predict mortality precisely, do not currently use epigenetic clocks.

Practical recommendations / action items

  • Treat aging clocks as experimental research tools—useful for hypothesis generation and potentially for prioritizing interventions in trials, but not yet definitive guides for individual care.
  • For personal health decisions, prioritize proven measures:
    • Control blood pressure, glucose (A1c), lipids.
    • Stop smoking.
    • Maintain physical fitness and healthy body composition.
    • Get adequate sleep; follow evidence‑based nutrition and exercise.
  • If you use a clock:
    • Interpret small changes cautiously (minutes/ months of “aging” are not the same as clinical outcomes).
    • Consider repeated measures and multiple clocks to reduce chance findings.
    • Ask labs/companies about technical reproducibility, batch effects, and whether the clock was validated longitudinally.
  • In trials: pre‑specify clocks, report multiple clocks (as DO‑HEALTH did), and estimate technical noise/power for plausible small effects.

Notable quote

  • “All models are wrong, some are useful.” — invoked to emphasize clocks are models with potential but important limitations.

Bottom line

Aging clocks are scientifically promising tools that can compress complex biology into a single metric and may speed early phase aging research. However, they are heterogeneous, technically noisy, and currently unproven as surrogate endpoints for the clinical outcomes that matter most (disease, disability, mortality). For individuals, focus on well‑validated risk factors and lifestyle measures; view epigenetic and imaging clocks as experimental data points rather than definitive guidance.