What a Biological Age Test Kit Measures

A biological age test kit takes a blood or saliva sample and measures DNA methylation at a few hundred to a few hundred thousand CpG sites. Those values feed into a linear model trained on chronological age or on mortality data, and the kit returns a single number in years. The number is real in the sense that it is computed from your molecules. It is also a point estimate from a model with a standard error of roughly two to five years, and most consumer kits do not tell you that error. If you want to use biological age as an instrument rather than a novelty, you need the raw methylation data and at least two timepoints. You also need a clear view of which clock you are running and what it was trained to predict.

This page covers how the clocks work and what accuracy means for them. It also covers which test to buy if you want one and why blood chemistry is often the better first measurement.

How an epigenetic clock computes a number

DNA methylation is the addition of a methyl group to a cytosine, usually where a cytosine sits next to a guanine (a CpG site). The methylation fraction at a given site, called a beta value, runs from 0 to 1 and shifts predictably with age across the genome. An epigenetic clock is a penalized linear regression over those beta values. Horvath’s 2013 multi-tissue clock uses 353 CpGs. Hannum’s uses 71. PhenoAge uses 513. GrimAge is built on plasma protein and smoking-pack-year surrogates rather than chronological age directly.

The training target matters more than the site count. First-generation clocks such as Horvath and Hannum were trained to predict chronological age, so the best possible clock of that kind is one that tells you nothing new. Second-generation clocks were trained on mortality or on a composite of clinical biomarkers, which is why they correlate better with disease outcomes than the age-trained clocks do ¹. When a kit reports “biological age,” ask which target was used. A DunedinPACE result (a rate of aging, in years per year) and a Horvath result (an age, in years) are not interchangeable quantities.

The measurement platform is usually an Illumina methylation array: the EPIC array covers about 850,000 CpGs, and its predecessor the 450K array covers about 450,000. Some groups have built targeted assays that measure only the CpGs a clock needs, using pyrosequencing or bisulfite amplicon sequencing, which is cheaper and can be more precise per site ²³. A simplified whole-blood assay of this kind can estimate age from a handful of sites with an error in the range of a few years ³. Whole-genome bisulfite sequencing is the most complete option and the most expensive, and at typical coverage its per-site noise is worse than an array’s.

What “accurate” means here, and why it is the wrong first question

Accuracy against chronological age is easy to state and mostly uninteresting. Good clocks correlate with chronological age at r above 0.9 with a median absolute error around three to four years in blood ⁴⁵. That error is the model’s, before you add technical noise.

Technical noise is the part that breaks personal tracking. The same blood sample split and run twice on an EPIC array can return epigenetic ages differing by several years, because individual beta values are noisy and the clock sums hundreds of them with fixed weights. Principal-component versions of the clocks were developed specifically to suppress this, and they substantially improve test-retest reliability. If a vendor will not tell you the technical replicate standard deviation of their pipeline, treat any year-over-year change smaller than five years as uninformative.

Two more sources of drift deserve attention. Blood cell composition changes with infection, exercise, and time of day. Because clocks are trained on bulk blood, a shift in the neutrophil-to-lymphocyte ratio moves the estimate. Good pipelines deconvolve cell fractions (Houseman or EpiDISH methods on the IDAT-derived betas) and report them alongside the age. Second, tissue matters: saliva and blood give different answers, and clocks trained in one tissue transfer imperfectly to another ⁶. Do not compare a saliva result from one vendor to a blood result from another.

The deeper question is what the number predicts. Epigenetic age acceleration (the residual of epigenetic age regressed on chronological age) associates with mortality and with organ-level outcomes across many cohorts ⁴. In transplanted kidneys, methylation-based and telomere-based age estimates both tracked donor biological aging, and the two measures did not agree closely with each other ⁷. The disagreement between them is the summary of the field: several defensible clocks exist, they are correlated but not equivalent, and each captures a partly different aspect of aging ¹.

If you buy a kit, buy raw data

Our recommendation is straightforward. Choose a provider that returns the raw IDAT files from the array. A PDF with one number is not enough. With IDATs you can run any clock you want and rerun them as better clocks are published.

A workable local pipeline in R:

library(sesame); library(minfi)
betas <- openSesame("idat_dir/", prep = "QCDPB")   # dye bias, detection p, background
library(methylclock)
ages <- DNAmAge(betas, clocks = c("Horvath", "Hannum", "Levine", "skinHorvath"))
library(EpiDISH)
cells <- epidish(betas, ref = centDHSbloodDMC.m, method = "RPC")$estF

Two practical failure modes. First, missing CpGs. If a clock’s required sites are absent from your array version, methylclock imputes them, and heavy imputation quietly degrades the estimate. Check the proportion imputed before you believe the output. Second, normalization choice changes the result by a year or more, so fix one pipeline (prep = "QCDPB" or funnorm via minfi::preprocessFunnorm) and never change it between timepoints.

Sampling discipline does more for interpretability than any modeling choice. Draw at the same time of day, fasted and before exercise. Do not draw within two weeks of an acute illness or vaccination. Bank two aliquots per draw so you can measure technical replication yourself.

Blood chemistry is the cheaper and better-understood alternative

If your goal is a number you can track quarterly and reason about, a clinical-chemistry age composite is a better starting point than an array. The BioAge R package implements Klemera-Doubal, PhenoAge, and homeostatic dysregulation from standard panel values. Those values include the following.⁸

• albumin
• creatinine
• glucose
• C-reactive protein
• lymphocyte percentage
• mean cell volume
• red cell distribution width
• alkaline phosphatase
• white blood cell count

Those analytes cost a fraction of an array, have well-characterized assay coefficients of variation, and are measured in every clinical lab.

The tradeoff is directness. Blood chemistry composites measure organ function downstream of the biology, so they respond to the same short-term perturbations that any clinical panel does. Methylation is upstream and more stable between draws, when the technical noise is controlled. Running both is informative precisely because they disagree in ways that point somewhere.

The broader direction of the field is multi-omic. Proteomic and transcriptomic aging models now match or exceed methylation clocks for some outcomes, and plasma proteomics supports organ-specific age estimates that a single blood methylation clock cannot resolve ⁹. Clocks that add functional data follow the same logic: DNAmFitAge incorporates fitness measures such as grip strength and VO2max into a methylation model and predicts mortality better than the methylation component alone ¹⁰. A biological age number is most useful as one channel in a longitudinal panel.

One boundary worth stating plainly. None of these numbers is a diagnosis and none of them indicates a treatment. If a panel shows an abnormal analyte such as a high C-reactive protein or a rising creatinine, that is a conversation with a physician. An abnormal analyte is not an input to a supplement decision.

Questions people also ask

How can I test my biological age? Order a DNA methylation test that returns raw IDAT files, or get a standard clinical chemistry panel and run the BioAge package on the values ⁸. The chemistry route costs less and is easier to repeat.

Are biological age tests accurate? Against chronological age, the better clocks reach a median absolute error of roughly three to four years ⁵. Test-retest noise on consumer array pipelines can be of similar magnitude, so a single result should be read as a range rather than a value.

How can I check my biological age at home? Fingerstick and saliva kits work for methylation arrays, and several validated targeted assays need only a small whole-blood volume ³. Collect at a consistent time of day and avoid sampling during an acute illness.

What is the most accurate biological age test available? There is no single winner. Second-generation clocks trained on mortality, such as PhenoAge and GrimAge, predict health outcomes better than age-trained clocks ¹. Principal-component versions of them are the most reliable for repeat measurement. For organ-level resolution, plasma proteomic models currently go further than any blood methylation clock ⁹.

Does a high epigenetic age mean something is wrong? Age acceleration is a population-level association with risk. It is not a finding about you specifically ⁴. Interpret it alongside cell-composition estimates and a clinical panel, and bring anything clinically abnormal to a physician.

Footnotes

1. Morgan E Levine. Assessment of Epigenetic Clocks as Biomarkers of Aging in Basic and Population Research. The Journals of Gerontology: Series A, 2020. https://doi.org/10.1093/gerona/glaa021 ↩ ↩² ↩³

2. Noémie Gensous, Claudia Sala, Chiara Pirazzini, et al. A Targeted Epigenetic Clock for the Prediction of Biological Age. Cells, 2022. https://doi.org/10.3390/cells11244044 ↩

3. Laura Vidal-Bralo, Yolanda Lopez-Golan, Antonio Gonzalez. Simplified Assay for Epigenetic Age Estimation in Whole Blood of Adults. Frontiers in Genetics, 2016. https://doi.org/10.3389/fgene.2016.00126 ↩ ↩² ↩³

4. Ran Duan, Qiaoyu Fu, Yu Sun, et al. Epigenetic clock: A promising biomarker and practical tool in aging. Ageing Research Reviews, 2022. https://doi.org/10.1016/j.arr.2022.101743 ↩ ↩² ↩³

5. Daniel J. Simpson, Tamir Chandra. Epigenetic age prediction. Aging Cell, 2021. https://doi.org/10.1111/acel.13452 ↩ ↩²

6. Calen P. Ryan. “Epigenetic clocks”: Theory and applications in human biology. American Journal of Human Biology, 2020. https://doi.org/10.1002/ajhb.23488 ↩

7. Sofia Pavanello, Manuela Campisi, Paolo Rigotti, et al. DNA Methylation - and Telomere - Based Biological Age Estimation as Markers of Biological Aging in Donors Kidneys. Frontiers in Medicine, 2022. https://doi.org/10.3389/fmed.2022.832411 ↩

8. Dayoon Kwon, Daniel W. Belsky. A toolkit for quantification of biological age from blood chemistry and organ function test data: BioAge. GeroScience, 2021. https://doi.org/10.1007/s11357-021-00480-5 ↩ ↩²

9. Jarod Rutledge, Hamilton Oh, Tony Wyss-Coray. Measuring biological age using omics data. Nature Reviews Genetics, 2022. https://doi.org/10.1038/s41576-022-00511-7 ↩ ↩²

10. Kristen M. McGreevy, Zsolt Radak, Ferenc Torma, et al. DNAmFitAge: biological age indicator incorporating physical fitness. Aging, 2023. https://doi.org/10.18632/aging.204538 ↩