Biological age tests, read carefully
A kit arrives, you spit or prick a finger, and a few weeks later a lab tells you that you are 41 on the inside. The number comes from a formula trained on someone else's blood, and the same sample can come back years apart. We explain which clocks mean something and how to read yours.
Clocks That Measure Your Aging
30 minutes. A narrated research briefing from the Bio Clock Test editors, with every source checked.
Read what the episode covers
Consumer Biological Age Tests and Supplement Marketing
Imagine you spit into a tube, mail it to a lab, and a few weeks later receive a report telling you that your body is nine years older than your birth certificate. No symptoms. No diagnosis. Just a number, clinical and authoritative, sitting at the top of a glossy PDF next to a suggested shopping cart full of supplements.
This is not a hypothetical. A case study published in the AMA Journal of Ethics describes a healthy 56-year-old man — call him AA — who did exactly that. He took a direct-to-consumer epigenetic age test and received the news that he was, biologically speaking, 65. He was not sick. He felt fine. But the number said otherwise, and so he did what any reasonable person might do upon being told his cells were nine years behind on their mortgage: he bought the supplements the company recommended. Brain health. Cellular repair. The whole lineup.
The AMA commentary is not entirely sympathetic to his predicament. It raises the uncomfortable possibility that "biological age" is, in certain respects, a fiction — a number manufactured by an algorithm and delivered in the language of clinical medicine. And yet the same commentary acknowledges that the underlying science is real. There is a genuine predictive signal in the data. These tests measure something. The question, which the field has been arguing about since the first clock was published in 2013, is exactly what.
To follow that argument, you need to know a little about what these tests are actually reading. Every cell in your body contains the same DNA sequence — the same three billion base pairs, the same instruction manual. What makes a liver cell different from a brain cell is not the sequence but the annotation: a system of chemical tags that sit on top of the DNA and tell the cell which genes to switch on and which to ignore. The most studied of these tags is DNA methylation. A methyl group — one carbon, three hydrogens — gets attached to a cytosine molecule at specific locations in the genome called CpG dinucleotides. CpG just means a cytosine followed by a guanine on the same strand. The methyl group does not change the underlying DNA sequence. It changes whether that section of DNA gets read. It is a dimmer switch on a light you cannot physically remove from the wall.
In mammals, roughly 75% of CpG dinucleotides carry a methyl group in normal somatic cells. That is a lot of tags, and the locations where they are present or absent define the epigenome — which is not static. It drifts systematically over a lifetime. Some CpG sites accumulate methyl groups as you age; others lose them. The pattern is consistent enough across individuals that a machine learning algorithm, trained on thousands of samples from people with known ages, can read a snapshot of your methylation landscape and estimate how old you are. That algorithm is an epigenetic clock. The DNA sequence is your hardware. The methylation pattern is the software log, and it keeps a record of everything.
The first researcher to do this at scale was Steve Horvath, a UCLA biostatistician and geneticist who in 2013 published a clock that read 353 specific CpG sites and predicted chronological age across diverse tissue types with a mean absolute error of about 3.6 years. That is a genuinely tight result for a biological measurement — and it immediately raised an obvious follow-up question. If the clock can estimate your age, and your clock says you are older than your birth certificate claims, what does that gap mean for you, specifically?
Lifespan Versus Healthspan: What Biological Age Predicts
Scientists studying aging separate two outcomes that biological age might predict: lifespan, meaning total years alive, and healthspan, meaning years lived in reasonably good health. Those are not the same number, and closing the gap between them is arguably more important than extending the first one. Biological age, in theory, tells you not just how many years have passed but how much wear the machinery has accumulated — where you sit on a trajectory that ends the same way for everyone, just at different speeds.
The AMA commentary pushes back on this framing. Calling the result a biological age implies a clinical certainty the test has not earned. The 56-year-old told he is biologically 65 is also told, in effect, that his cells are behind schedule. Whether that is true for him specifically is a question the test cannot answer. Epigenetic drift does track with increased risk for diseases like cancer, heart disease, and lung disease, and that association is consistent across large population studies. But population risk and individual fate are not the same calculation, and consumer test marketing rarely pauses to explain the difference. The science arrived, it was real and striking, and the marketplace followed roughly forty-five minutes later with a checkout button and a supplement bundle. That gap — between what the clocks can actually tell you and what the industry says they can — is what the rest of this piece is about.
Those 353 CpG sites did not emerge from biological theory. Horvath assembled publicly available methylation datasets covering 51 different non-cancerous tissue and cell types — blood, brain, liver, saliva, kidney, and about forty-six other things your body maintains without consulting you — and fed them into an elastic-net regression model. Elastic-net regression is a mathematical technique that simultaneously selects and shrinks predictors, finding the minimum set of variables that still do the job without collapsing into noise. Out of hundreds of thousands of candidate CpG sites, the model converged on 353. The striking result was not the count but what those 353 sites could do: they tracked chronological age with correlations above 0.9 across all 51 tissue types. A liver cell and a brain cell do not share many ambitions, yet the same methylation fingerprint predicted age in both. The clock worked in saliva, in blood, in brain tissue, and in prenatal samples. It covered people from birth to 100, a wider range than most biological measurements stay coherent across. That tissue universality is what made the clock a landmark rather than a clever parlor trick.
The Hannum clock, published the same year, came from a narrower question. Gregory Hannum and colleagues profiled methylation across more than 450,000 CpG markers in a whole-blood cohort of 482 individuals ranging in age from 19 to 101. Same mathematical technique — elastic net — but trained exclusively on blood. The model settled on 71 CpG sites and hit a mean absolute error of 3 to 4 years in blood samples. The Hannum study also found accelerated aging in tumor tissue, found that methylome aging rates differ by sex and genetic background, and found that individual variation in how fast the methylome ages shows up in the transcriptome. The model was validated in human tissues beyond blood, though it was always most accurate where it learned to walk.
Why Clocks Trained on Age Miss Disease Risk
Both clocks were trained on one objective: guessing how old you are. That is a coherent and achievable goal. The problem is that knowing someone is 58 does not tell you whether they will have a heart attack at 60 or spend their eighties being aggressively opinionated at dinner parties. Chronological age correlates with health outcomes because aging and health deterioration tend to travel in the same direction. But the clocks were not designed to separate the people-aging-fast-and-getting-sick from the people-aging-fast-and-just-getting-worse-to-be-around. They picked up the mortality signal as a side effect of the chronological training objective. That ceiling is not a bug that later researchers fixed with better code; it is baked into what the first-generation clocks were asking.
What researchers actually analyze, when using these clocks for science rather than sales copy, are the residuals. Epigenetic Age Acceleration — AgeAccel — is what remains after you regress epigenetic age on chronological age. A positive value means you are epigenetically older than expected. From the Horvath clock, this residual adjusted for blood cell counts is called Intrinsic Epigenetic Age Acceleration, or IEAA. IEAA accounts for age-related shifts in immune cell populations — specifically the decline in naïve CD8-plus T cells and the accumulation of exhausted ones. From the Hannum clock, a version that upweights blood cell composition before doing the regression is called Extrinsic Epigenetic Age Acceleration, or EEAA. EEAA is more sensitive to immunosenescence: the gradual degradation of the immune system that makes elderly people worse at fending off things that barely register in younger ones.
Both first-generation clocks hold up well for tissue-specific comparisons and for benchmarking sequencing platforms — if you run the same sample twice and the epigenetic ages come back wildly different, you have a technical problem, not a finding. For that kind of quality-control work, these clocks are solid. As instruments for predicting what will actually happen to a specific human being, they were built with a structural limitation that no amount of statistical sophistication could obscure: they were trained to predict your age, and your age is already on your driver's license.
The problem with training a clock to predict your age is that your age is the one thing you already know. This observation took the field until 2018 to fully act on.
Morgan Levine, a biogerontologist then at Yale, asked a different question. Instead of teaching a model to guess a birthday, what if you trained it on something that actually mattered, like whether you were going to die? This sounds morbid, but science has a long tradition of cutting straight to the point.
Levine's approach had two steps, and the first step did not involve methylation at all. She started with the National Health and Nutrition Examination Survey, known as NHANES IV, which is a large American health study that tracks people over time and records, among other things, when they die. From that dataset, she identified nine standard blood chemistry markers that together predicted mortality. The list includes creatinine, blood glucose, C-reactive protein, and lymphocyte percentage, among others. None of these are exotic. Creatinine tells you how well your kidneys are clearing waste. C-reactive protein tracks inflammation. Lymphocyte percentage reflects immune function. Your doctor has ordered these at some routine checkup and handed you the printout without once suggesting you should be alarmed.
Building PhenoAge from Physiological Mortality Signals
From those nine values, Levine constructed a composite score she called phenotypic age. Then, in the second step, she went looking for the methylation signature that best matched it. She was not asking which CpG sites correlate with being 47 years old. She was asking which CpG sites correlate with looking, physiologically, like you are on a faster path toward the outcomes that kill people. The clock that emerged from that process is PhenoAge, published in 2018.
In the NHANES IV validation sample of 11,432 adults between the ages of 20 and 84, PhenoAge was significantly associated with all-cause mortality after adjusting for chronological age and sex. That association held across strata of age, race and ethnicity, education level, disease count, and health behaviors. Even among people defined as disease-free with a normal BMI, PhenoAge still predicted mortality. The C-index for mortality prediction was 0.750, which in clinical prediction terms means the model ranked a future decedent above a survivor about 75 percent of the time. For comparison, a coin flip gets you to 0.5. PhenoAge sits meaningfully above chance and meaningfully below omniscience, which is about as good as medicine usually manages.
GrimAge, published the following year by Ake Lu and colleagues with Steve Horvath among the co-authors, pushed the mortality-optimization logic further. Where PhenoAge mapped methylation to a composite of blood chemistry values, GrimAge mapped methylation to a set of plasma protein surrogates and smoking exposure, then fed those surrogates into a mortality model trained on time-to-death data. The plasma proteins include PAI-1, a clotting factor linked to cardiovascular risk, and GDF15, which rises with inflammation, cancer, and various forms of metabolic distress. Smoking exposure was included because smoking leaves a distinct and durable methylation signature even after someone quits. Your genome holds a grudge in a way your lungs technically cannot.
In population cohort studies, a five-year acceleration in GrimAge has been reported to carry a hazard ratio of around 1.50 for all-cause mortality, outperforming PhenoAge, Hannum, and Horvath on that specific measure. That figure is widely cited and should be read as an estimate with a range around it, not a precise physical constant. The direction is consistent across studies, though: GrimAge is currently the strongest single-clock predictor of mortality among the established clocks.
One practical difference separating both second-generation clocks from their predecessors is what they pick up in population data. Lifestyle factors — smoking, physical activity, socioeconomic circumstances — show up more consistently in PhenoAge and GrimAge than in the Horvath or Hannum clocks. Living in a lower-income census tract was associated with older PhenoAge and GrimAge scores, and the GrimAge association survived after adjusting for smoking status. The first-generation clocks were largely indifferent to that signal. The second-generation clocks are capturing the methylation consequences of how you actually live. That is a more complete picture of biological aging. It is also a more efficient mechanism for selling supplements to people who are already under financial stress, and the industry did not take long to notice.
Second-Generation Epigenetic Clocks as Frozen Snapshots
The second-generation clocks were a genuine step forward. But they still had a structural limitation that nobody highlighted prominently on the landing page of their consumer test: they were taking a photograph. They looked at your methylation pattern at one moment in time and said, based on this snapshot, here is approximately where your biology sits on the age spectrum. That is real information. It is also information that tells you where you are, not how fast you are moving. If what you actually want to know is whether your morning jogs and your vitamin D supplements are doing anything useful, a photograph is the wrong instrument. You want a speedometer.
Daniel Belsky and colleagues built the speedometer. The dataset they used was the Dunedin Study, a 1972-to-1973 birth cohort from Dunedin, New Zealand, in which a group of people born in the same year have been followed and tested at regular intervals for their entire lives. Everyone in the cohort is the same age. There are no generational confounds, no older people who grew up eating different chemicals in different decades skewing your comparisons. Just the same people, returning every few years to donate blood and submit to testing, which says something admirable about New Zealanders and also something practical about the advantages of small island nations where there is nowhere to hide from your researchers.
Belsky's team measured 19 biomarkers of organ-system integrity — kidneys, liver, lungs, heart and blood vessels, immune function, metabolic processing — at ages 26, 32, 38, and 45. For each person, they calculated the rate at which those 19 indicators were declining over roughly two decades of follow-up. That rate is the Pace of Aging: a single number expressing how many years of physiological wear a person accumulates per chronological year. The methylation clock trained on that longitudinal decline is DunedinPACE, published in eLife in 2022. The training dataset contained roughly 69,715 data points drawn from those 19 biomarkers across four time points, and the entire clock was built on change over time rather than the cross-sectional snapshots every previous clock had used.
The output of DunedinPACE is not a biological age. It is a pace. A score of 1.0 means you are aging at the population average. A score of 1.2 means you are aging 20% faster than average. A score of 0.8 means you are aging 20% slower, which is excellent news that you should share sparingly at social gatherings because your host may be sitting on a 1.4 and has access to sharp kitchen implements.
The clock correlates at r equals 0.78 with the actual 20-year longitudinal Pace of Aging measured in the Dunedin cohort. That is strong. It is not complete — a correlation of 0.78 accounts for about 60% of the variance in true aging pace, and the remaining 40% is elsewhere. The training process deliberately excluded CpG probes with poor test-retest reliability, filtering the input data before the elastic-net regression ran, which pushed the clock's own reliability above 0.90. This is worth pausing on because most epigenetic clocks do not do this and it matters, as later sections of this story will illustrate with some precision.
DunedinPACE Measures the Longitudinal Speed of Aging
A predecessor called DunedinPoAm, published in eLife in 2020, tracked 18 biomarkers across a shorter window from ages 26 to 38 — about 12 years rather than the full 20. DunedinPACE extended the follow-up to four measurement occasions, added a nineteenth biomarker, and is computed from standard Illumina methylation array data. The upgrade is not cosmetic. Longer longitudinal windows give you a more accurate picture of the slope, and the additional measurement occasion contributed substantially to the clock's correlation with actual physiological decline.
Among the co-authors on the 2022 paper were Avshalom Caspi and Terrie Moffitt, two researchers who built their careers studying how childhood environments shape adult health and behavior. Their involvement is a clue about what DunedinPACE is sensitive to. The clock has been shown to reflect faster aging in adults who experienced childhood adversity, socioeconomic hardship, and other early-life stressors — the kinds of exposures that compound slowly and show up in the cells of a 45-year-old who may not connect them to their current biology at all.
DunedinPACE has been validated in more than 65 large cohorts across more than 17 countries and in more than 300 published studies, which is a remarkable rate of adoption for a clock that is only a few years old. It adds incremental prediction of morbidity, disability, and mortality on top of what GrimAge already captures, and across intervention studies it moves more in response to behavioral and clinical changes than any other clock in the current toolkit. That last property is the one that matters most for what comes next, because the central question the entire industry is circling around is not what your number is today. It is whether you can make it go down.
So far, every clock we have covered requires methylation profiling, a specialized assay that reads chemical tags across hundreds of thousands of genomic positions. That is not what your doctor ordered last Tuesday. What your doctor ordered was a basic metabolic panel, maybe a CBC, possibly a lipid screen. The relevant detail is that PhenoAge, the mortality-optimized clock from the previous section, is built from nine standard clinical blood values: creatinine, blood glucose, C-reactive protein, albumin, alkaline phosphatase, lymphocyte percentage, red cell distribution width, mean corpuscular volume, and white blood cell count. Every one of those appears on a routine comprehensive metabolic panel or complete blood count. The epigenetic version of PhenoAge is a proxy for the blood chemistry version, not the other way around. The blood test came first, and the methylation model was trained to mimic it.
Commercial biological age companies noticed. InsideTracker's InnerAge 2.0 computes a biological age from blood markers including glucose and GGT, a liver enzyme. Blood glucose carries more weight because it correlates more strongly with age in InsideTracker's proprietary dataset. That dataset is not published, which is a detail that does a certain amount of work all on its own.
UK Biobank Blood Chemistry Clocks Outperform PhenoAge
Researchers working with data that can actually be audited have built blood chemistry clocks from the UK Biobank. A 25-biomarker clock derived from 306,116 participants using elastic-net Cox regression on 60 candidate biomarkers achieved a C-index of 0.778 for mortality prediction, a step above PhenoAge's 0.750. Biological age estimates from that clock ranged between 20 years younger and 20 years older than chronological age, which is a wide enough spread to make some people very happy and others considerably less so. Neural network and ensemble models trained on NHANES data pushed prediction accuracy to a mean absolute error of 2.98 years, with all-cause mortality hazard ratios of 1.059 per unit of biological age acceleration. The three strongest predictive features in those NHANES models, identified by SHAP analysis, were prescription medication use, hepatitis B surface antibody status, and vigorous physical activity. Biological age clocks capture every lifestyle and environmental variable that happened to correlate with age in the training data. That is both their strength and their limitation, and the two are largely the same thing.
Then there is proteomics. Rather than methylation tags or a handful of blood values, proteomic clocks read the full landscape of circulating proteins in plasma. A Proteomic Aging Clock built from the UK Biobank Pharma Proteomics Project measured 2,923 plasma proteins using the Olink Explore platform across 53,021 participants aged 39 to 70, training on 45,441 of them. The clock correlated at r equals 0.94 with chronological age in the discovery cohort, validated at r equals 0.92 in a Chinese cohort of 3,977 participants and r equals 0.94 in a Finnish cohort of 1,990. Over a mean follow-up of 13.3 years, 10.9% of study participants died. The clock outperformed PhenoAge and leukocyte telomere length in predicting both all-cause mortality and 18 incident diseases including heart, liver, kidney, and lung disease, diabetes, neurodegeneration, and cancer. The 204 proteins contributing to it are enriched in immunoinflammatory processes, cellular senescence, extracellular matrix remodeling, and vascular biology, plus hormone regulation, neuronal structure, and developmental signaling. The single plasma protein most strongly correlated with age was fibrinogen beta chain, at a Spearman correlation of 0.590. SERPINA3 came second at 0.560, which is an impressive performance for a protein whose name sounds like a mid-tier Bond villain with a biochemistry degree.
Those are strong numbers. They should also be read alongside a methodological cautionary tale that deserves considerably more attention than it gets in promotional materials.
One mass-spectrometry-based proteomic clock study analyzed plasma from 350 adults processed across 27 batches. The order in which samples ran through the mass spectrometer correlated with chronological age at a Spearman correlation of 0.336. That correlation has nothing to do with aging. Acquisition order alone predicted age with a mean absolute error of 3.98 years in a negative control model. When complete batches were held out during cross-validation, aggregate mean absolute error jumped to 10.56 years. Age and sex distributions differed strongly across those 27 batches, which meant the model had learned when in the run schedule a sample appeared and reported it as aging. A 50-feature gradient boosting approach on that same data yielded a mean absolute error of 8.90 years under rigorous cross-validation, a result that sits considerably further from clinical utility than the within-batch figures would suggest.
Platform Bias and Variability in Proteomic Clocks
The lesson is not that proteomic clocks are broken. The scale and design of the UK Biobank PAC study put it in a different category from a single-site mass spectrometry run. The lesson is that proteomic aging clocks vary widely in platform, population, and modeling strategy, and a number that looks like biological truth can sometimes be your mass spectrometer keeping a record of what time it ran your sample. The field's most ambitious instrument has a confound baked into the hardware. That is a useful thing to know before the industry builds a checkout button around it.
All of which brings us to the question the entire industry is built around, even if most consumer-facing materials do not ask it this directly: can you actually move the number?
Two trials get cited more than any others when this question comes up. The first is the CALERIE trial, which stands for the Comprehensive Assessment of Long-term Effects of Reducing Intake of Energy, a name so bureaucratically thorough it could only have been assembled by a federal grant committee. CALERIE randomized 220 healthy, non-obese adults — men and premenopausal women, ages 21 to 50, BMIs between 22 and 27.9 — to either 25% caloric restriction or eating whatever they wanted, for two years. The randomization was 2:1, meaning roughly twice as many people were assigned to eat less than to eat freely, which suggests the researchers understood that telling people to cut a quarter of their food intake for two years would generate some attrition and built accordingly.
The post-hoc analysis of that trial found that caloric restriction slowed DunedinPACE by approximately 3%. It did not produce statistically significant changes in PhenoAge or GrimAge. The treatment effect sizes were small. The fact that DunedinPACE moved and the other clocks did not is not an accident — it is exactly what you would expect from a clock specifically designed to detect change over time rather than characterize a static position. A 3% slowdown in pace-of-aging sounds modest, and in the context of a two-year trial it is. Extrapolated across decades, using independent observational data from older adult cohorts, a 3% slower DunedinPACE has been associated with approximately 15% lower risk of death. That extrapolation is disputed, and you should hold it loosely. It is not a guarantee. It is a signal. But CALERIE is still the first properly randomized controlled trial to demonstrate that a dietary intervention can measurably shift a biological pace-of-aging measure. That is not nothing, and the field had been waiting for that result for a long time.
The second trial is TRIIM, which stands for Thymus Regeneration, Immunorestoration, and Insulin Mitigation — a name that sounds like the agenda items of a very optimistic medical conference. The TRIIM trial was conducted at Stanford, published in Aging Cell in 2019, and enrolled nine volunteers: putatively healthy men aged 51 to 65. Nine. The protocol combined recombinant human growth hormone, DHEA, and metformin. Growth hormone was chosen for evidence that it has thymotrophic effects — meaning it can stimulate thymus regeneration and immune reconstitution. DHEA and metformin were added specifically to offset growth hormone's tendency to induce hyperinsulinemia, which is a polite clinical term for making your pancreas extremely unhappy.
TRIIM Trial Claims Epigenetic Age Reversal
The reported results were that mean epigenetic age was approximately 1.5 years below baseline after one year of treatment, as measured by the Horvath clock. The GrimAge predictor reportedly showed a 2-year decrease. And the rate of epigenetic age reversal reportedly accelerated to negative 6.5 years per year during months nine through twelve of the trial. The paper claimed to be the first report of an increase in predicted human lifespan via a currently accessible aging intervention. That is a large claim for a study with no control group and nine participants.
To be direct about what n equals 9 with no control means: you cannot separate the drug effects from regression to the mean, from seasonal variation in methylation, from the placebo effect, or from the simple statistical reality that nine people is not enough to generalize to a species of eight billion. The rate acceleration in months nine through twelve is the kind of finding that, in a larger controlled study, would either replicate and be genuinely important, or disappear entirely and be quietly filed under Things That Happened Once. The follow-on trial, TRIIM-X, registered as NCT04375657 and run as a Phase 2 study by Intervene Immune, attempts to generate more controlled data with personalized dosing across male and female volunteers aged 40 to 80, over a 12-month treatment duration. That trial is the appropriate venue to find out whether any of this holds up. The original TRIIM result catalyzed a wave of research and investor attention. Whether it catalyzed a scientific finding remains an open question.
The 2026 systematic review of 26 geroscience intervention studies, covering 13 comparative and 13 single-arm pre-post studies, delivered a conclusion that should accompany every consumer test report as a mandatory disclaimer: no epigenetic clock currently meets the criteria for a validated surrogate endpoint. Across all assessed bodies of evidence, certainty of evidence was rated very low. Of the 26 studies, 15 were classified as mixed, 6 as indeterminate, and 4 as jointly inconclusive for concordance between clock changes and clinical outcome changes. One study was uniformly concordant. One. No intervention-clock-clinical outcome combination met the criteria for meta-analysis. The field has produced clocks that are real, predictive in population data, and sensitive to the kinds of things that are genuinely bad for you. It has not yet produced clocks that reliably tell you whether a specific intervention is working for a specific person.
That is where the science actually sits. Not at the checkout button, not in the glossy PDF with the supplement bundle, but somewhere considerably more honest and considerably less convenient: biological age tests measure something real, the clocks have improved dramatically across three generations, and the gap between what they can tell a population and what they can tell you specifically remains wide enough that you should read your result with curiosity rather than alarm, and with a healthy suspicion toward anyone who charges extra for the action plan.
Start here
-
Biological Age Tests: What They Measure and What They Don't
A biological age test turns blood into a number in years. Learn the four main types, why one sample can return different ages, and what the result can and cannot say.
-
Biological Age, Defined Carefully
Biological age is an estimate in years, not a measurement. See how clocks trained on birthdays or on health risk reach different numbers and why labs disagree.
-
Biological Age vs Chronological Age: What a Gap Means
Biological age vs chronological age explained: how a clock's residual becomes 'age acceleration', why the comparison group matters, and how test noise can exceed the gap.
Everything else we have written
- Bryan Johnson's Biological Age and What One Person Can Measure
Bryan Johnson biological age searches want one number, but the evidence shows a single reading can swing by years. Here is how far one person's tests can be trusted.
- Tally Health's Cheek-Swab Age Test
Tally Health sells a cheek-swab age test, but most epigenetic clocks are trained on blood. Here is what that means for accuracy, repeatability and trust.
- TruAge and TruDiagnostic: What the Test Reports
Truage is TruDiagnostic's epigenetic age test. Learn what its three clocks and 64 biomarkers report, and what published validation does and does not show.
- Do Ageing Clocks Predict Disease and Death?
Epigenetic clock mortality evidence from blood cohorts shows small but real links to death and disease, yet no validated surrogate for trials or saliva tests.
- The TRIIM Trial and the Claim of Reversing Epigenetic Age
A triim trial explainer: nine volunteers, three drugs and a 2.5-year epigenetic drop, plus why clock noise and the missing control group limit the result.
- The CALERIE Trial: Calorie Restriction Slowed a Clock
The CALERIE trial put 220 healthy adults on 25% less food for two years. One aging clock slowed by about 3%, two did not, and what that means is unsettled.
- How Reliable Are Consumer Biological Age Tests?
Epigenetic age test accuracy depends on the clock, the lab and the kind of reliability. One tube can return ages up to nine years apart, and no standard exists.
- Telomeres: The Ageing Marker That Underperformed
Why telomeres lose to newer ageing measures: noisy assays, one averaged number against thousands of inputs, and a calorie restriction trial whose result shifts by timeframe.
- Proteomic Clocks and Organ Ages
How a proteomic aging clock turns blood proteins into an age estimate, why it trains on birthdays, what the gap predicts, and where the evidence is still thin.
- Biological Age Calculators From Your Blood Work
How a biological age calculator turns routine blood work into one number, how the weights are chosen, and why results from different calculators are hard to compare.
- DunedinPACE: Measuring the Speed of Ageing
DunedinPACE reads aging speed from one blood draw, not an age. Learn how the Dunedin cohort built it, what it predicts, and where the evidence is still unsettled.
- GrimAge: The Clock Trained on Death
Grimage explained: GrimAge is an epigenetic clock trained to predict mortality, not birthdays. Learn what it measures and why lowering it is unproven.
- PhenoAge: A Clock Trained on Blood Tests
PhenoAge is a weighted blend of nine routine blood values, and its DNA methylation twin is trained to estimate that same blend. Here is how they relate and where they differ.
- The Horvath Clock: The Pan-Tissue Clock That Started It
The horvath clock reads 353 DNA sites to guess age within about 3.6 years. Learn how it was built, what its age gap means, and where it still gets used.
- 'Methylation Test' Means Three Different Things
What a methylation test actually means: an epigenetic age clock, an MTHFR gene test or a homocysteine blood level, and why the clock results rarely agree.
- DNA Methylation: The Chemical Tags Clocks Read
How dna methylation works, from a methyl tag on a CpG site to an epigenetic clock age, plus where blood, saliva and read depth make the number shaky.
- How an Epigenetic Clock Is Built
How an epigenetic clock is built: elastic net regression picks a few hundred CpG sites to guess age, and why the clock's errors are the useful part.
| Question to ask | Why it matters |
|---|---|
| Which clock does the test use? | Horvath, PhenoAge, GrimAge and DunedinPACE were trained to predict different things. |
| How much does a repeat test move? | Some clocks give the same sample results years apart. Newer versions were built to cut that noise. |
| What has it been checked against? | A clock that predicts disease and death tells you more than one that only predicts your birthday. |
The full guide has twelve questions to ask before you pay for a biological age test, and how to read the result without overreacting.
The Biological Age Test Buyer's Guide
What to ask before you pay for a biological age test: which clock it uses, its repeat-test noise, what it has been validated against, and how to read your result without overreacting. Twelve questions, one page.