Biomarkers · Explainer

What an epigenetic clock can—and cannot—tell you

DNA-methylation clocks can estimate age-related patterns, pace or risk—but different clocks answer different questions. A result is not a diagnosis, and a lower score is not proof of longer life.

Reviewed under standing publication authorization

Source, correction-status, evidence, conflict and prose reviews were completed July 31, 2026.

Evidence box

Established research tools; unsettled clinical meaning

Evidence type
Algorithm development, cohort validation, reliability and intervention analyses
Studied in
Human tissues and longitudinal cohorts
Sample
Varies by clock and validation dataset
Publication status
Peer reviewed; featured CALERIE analysis corrected
Outcome type
DNA-methylation biomarker estimates and associations
Development stage
Research use; clinical utility not generally established
Conflicts / funding
Some clock technology is licensed commercially; disclosures differ by paper
Our assessment
Useful when the construct and uncertainty are explicit
01

The bottom line

An epigenetic clock is an algorithm that combines DNA-methylation measurements into an estimate. Early clocks were trained mainly to predict chronological age; later tools target mortality-related risk, physiological decline or pace of aging. Because their targets differ, two clocks can give different answers without either being a literal measurement of how old a person “really” is.

02

Why this matters

Waiting decades for lifespan outcomes makes aging trials difficult. A trustworthy surrogate biomarker could help researchers test interventions sooner. But a biomarker can correlate with an outcome without lying on the causal pathway to it. Treating every clock as an interchangeable endpoint can turn a useful research signal into an unsupported consumer promise.

03

How clocks are built

DNA methylation is a chemical modification associated with gene regulation. Horvath’s 2013 multi-tissue clock selected methylation sites that predicted chronological age across tissues. DunedinPACE instead used repeated physiological data from a longitudinal birth cohort to train a blood measure intended to estimate pace of aging. Other clocks optimize for different combinations of age, proteins, smoking or mortality. Training target, tissue, laboratory platform and reference population all shape the output.

04

What researchers have found

Clock measures can associate with disease and mortality in cohorts, and some respond in interventions. In CALERIE, two years of calorie restriction produced a small change in DunedinPACE but not significant changes in PhenoAge or GrimAge. That disagreement is informative: the intervention did not turn one universal “biological age” dial. The analysis also has an author correction, linked below.

05

Reliability is not validity

A measure can be repeatable without measuring the intended construct, or valid on average without being precise enough for one person. A 2022 reliability study found that technical noise could shift outputs from six prominent clocks by as much as nine years between replicates and proposed principal-component versions that reduced variation. The paper’s authors also disclosed technology licensing and consulting relationships. Better repeatability helps, but it does not by itself prove that a score is a surrogate for clinical benefit.

06

What a result does not show

A clock result does not diagnose a disease, identify a treatment, reveal an exact biological age, or state how many years someone will live. A before-and-after change can reflect measurement noise, cell-composition shifts, regression to the mean or a genuine biomarker response. Even a genuine response does not prove reduced disability, disease or mortality unless that clock has been validated for the specific use.

07

Commercial tests, conflicts and privacy

Consumers should ask which clock was used, what it was trained to predict, the test’s repeatability, the expected variation, and whether the result changes an evidence-based decision. Some clock methods or derivatives are commercially licensed, including technology discussed in the reliability paper. A methylation test also creates sensitive biological data; storage, secondary use and deletion policies matter independently of scientific validity.

08

How to read the next clock headline

Look for a prespecified clock, a randomized comparison, repeated measurements, quality control, correction for multiple analyses, absolute effect size and a clinically relevant outcome. Ask whether the result replicated with another clock and whether a correction exists. Expert consensus still describes biomarker validation and clinical utility as unfinished work—not a settled diagnostic standard.

Primary sources

Sources, roles and limits

  1. Horvath, multi-tissue age predictor
    Identifier
    PMID:24138928 · PMCID:PMC4015143 · DOI:10.1186/gb-2013-14-10-r115
    Role
    Foundational chronological-age clock
  2. Belsky et al., DunedinPACE
    Identifier
    PMID:35029144 · PMCID:PMC8853656 · DOI:10.7554/eLife.73420
    Role
    Pace-of-aging measure development
  3. Higgins-Chen et al., clock reliability
    Identifier
    PMID:36277076 · PMCID:PMC9586209 · DOI:10.1038/s43587-022-00248-2
    Limitation
    Methodological reliability study; commercial relationships disclosed.
  4. CALERIE methylation analysis
    Identifier
    PMID:37118425 · DOI:10.1038/s43587-022-00357-y
    Role
    Randomized intervention response; corrected
  5. Author correction
    Identifier
    PMID:37161091
    Role
    Correction record
  6. 2025 expert consensus on aging biomarkers
    Identifier
    PMID:39708300 · DOI:10.1093/gerona/glae297
    Role
    Expert consensus and remaining validation needs

Connected topics

Continue through the knowledge system

Biomarker guideEpigenetic changeCALERIE briefResearch methods

Disclosures and history

  • July 31, 2026 — three research passes and separate evidence/prose reviews completed.
  • July 31, 2026 — correction and commercial-conflict context verified.