Why biological-age tests disagree and what I use them for

I took five biological age tests in a year. My age ranged from 15 years younger to 3 years older than my chronological age.

Different biology, different models, and measurement error all play a part. I wanted to know what I could actually learn from the disagreement.

Biological-age tests promise a number that feels easy to act on. Before buying one, I want to know what it measures, how well it predicts an outcome, and what the result can change.

I compared the results from TruDiagnostic, GlycanAge, Generation Lab, a PhenoAge calculation, and DunedinPACE. Some of these are platforms; others are scores that the platforms report.

Start with the health questions your usual tests can answer. An aging score is optional extra information, and a change in it does not prove a treatment slowed aging.

IN THIS ISSUE:

📊 Your bio-age test gave you a number. Here's what it actually means (and doesn't).

🧬 Why 14 epigenetic clocks can't agree on your age, and what that tells us about the science.

Why routine clinical markers remain important when comparing cardiovascular prediction.

🔬 My testing stack: what I run, what I skip, and what changed my clinical plan.

What is "biological age" anyway?

Chronological age is time since birth. A biological-age estimate is a model of selected features associated with age, disease, or mortality.

There is no single agreed biological-age measurement. The result depends on the measured biology, model, training population, sample, and technical variation.

Epigenetic clocks are the most well-known category. These measure chemical tags on your DNA called methyl groups, which shift in predictable patterns as you age. Scientists trained algorithms on thousands of people to estimate "how old your DNA looks."

Some commercial reports assign organ-system age estimates. Those may generate questions, but their clinical value depends on whether an estimate reliably changes care. An inferred lung or brain age is not a direct test of that organ’s function.

GlycanAge measures glycans on IgG antibodies. These patterns relate to age and inflammatory states, but the report does not diagnose inflammation or tell us its source.

Clinical chemistry models such as PhenoAge use routine lab values. LifeClock is a separate model with its own inputs. A PhenoAge calculation needs chronological age, values from CBC/CMP, and CRP; CBC and CMP alone are not enough.

The point: when someone says they "tested their biological age," you have to ask which kind. A GlycanAge result and a DunedinPACE score and a SystemAge respiratory reading are measuring completely different dimensions of your biology.

Three generations of clocks

Not all clocks ask the same question. And the question matters more than the answer.

Horvath and Hannum clocks were trained to predict chronological age. A comparison of fourteen clocks found different associations with later disease. A forensic age estimate or one benchmark ranking does not make a clock a validated tool for clinical decisions.

A clock trained to predict how old you are will be good at predicting how old you are. It won't necessarily tell you how sick you're getting.

Later models used clinical risk and mortality-related features. Clinical PhenoAge combines nine lab values with chronological age; DNAm PhenoAge estimates that phenotype from methylation. GrimAge uses methylation proxies for proteins and smoking history. They are related approaches, not interchangeable tests.

Big improvement. But both PhenoAge and GrimAge still produce a static number: your estimated biological age at one moment. A snapshot.

DunedinPACE estimates a pace-related score using methylation patterns. It was trained against repeated organ-system measures across years. It is not a direct observation of your present speed of aging.

Age estimates and pace-related scores

DunedinPACE came from a New Zealand birth cohort with repeated measures of nineteen organ-system indicators over two decades. The score is scaled around one. A value of 0.8 is a lower modeled pace, not proof that a calendar year costs one person only 9.6 months of biological life.

A marker associated with faster aging may help research. Whether changing it improves a person’s future health needs separate evidence.

A study published earlier this month validated this in the strongest terms yet. Vetter et al. followed 1,083 older adults in the Berlin Aging Study II (BASE-II) for 7.4 years, comparing 14 expert-consented aging biomarkers including IGF-1, GDF-15, CRP, IL-6, muscle mass, grip strength, gait speed, balance, cognition, and multiple epigenetic clocks (Biomarker Research 2026).

DunedinPACE was the strongest individual predictor of all-cause mortality.

In BASE-II, DunedinPACE had the strongest mortality association among the fourteen markers studied. A selected model using it with muscle mass and balance had a C-index of 0.63, compared with 0.65 for the full model. That describes ranking ability in this cohort, not a personal lifespan forecast or demonstrated benefit from ordering the tests.

That is a useful research finding. It still needs validation before becoming a standard clinical testing package.

OMICmAge and the multi-modal future

OMICmAge research combined clinical and multiomics information during model development, then used methylation proxies to estimate those features.

It performed well against selected clocks in the study’s benchmarks. That doesn’t establish superiority over every existing clock for every patient or use.

The resulting estimate is informed by several types of data. It doesn’t directly measure every layer in each person’s sample.

Before paying for a test, ask what it measures, how the company checked its estimates, and what you would do with the result.

Its performance adds to measurement research. It doesn’t itself prove that a reported organ age identifies a treatable problem.

A landmark Science paper this month mapped the first atlas of organ-specific chromatin accessibility across the mouse lifespan. Twenty-one tissues, three age groups, both sexes. The finding: about 25% of organ-specific cell types show significant age-related population shifts. Kidney and liver showed faster epigenomic aging. Brain showed partial protection. Forty percent of aging dynamics were sex-dependent.

Animal organ differences help researchers ask questions. They don’t tell us which treatment a human organ-age report requires.

But does any of this really matter?

The 2024 CARDIA analysis of 913 adults compared accelerated-aging measures with a conventional cardiovascular risk score. Clinical-marker age acceleration and the Framingham score had stronger associations with incident cardiovascular disease than the epigenetic measures studied.

That supports taking established cardiovascular risk factors seriously. It doesn’t show that any $200 lab panel equals every $500 epigenetic test.

A 2025 UK Biobank analysis confirmed that Phenotypic Age acceleration does track with cardiovascular events. But so do cholesterol and blood pressure. The question isn't whether biological age testing captures something real. It's whether it captures enough beyond what you already have.

A February 2026 preprint also compared clocks and traditional risks. It had not completed peer review when this issue appeared, so it should not settle the comparison.

Clock associations with dementia, mortality, and disability are worth studying. Prediction can support future uses, but it does not yet show that an individual’s clock-guided treatment improves those outcomes.

So are these tests worth your money? The honest answer requires understanding what each test is actually measuring, and what it can't see.

What's actually new (2025-2026)

The field is moving fast. A few things worth knowing:

Longitudinal clock research examines whether changes improve prediction. Serial measurements can add information, but illness, treatment, technical variation, and regression to the mean can also change a score. A rise is not automatically a danger signal.

In CALERIE, a two-year randomized calorie-restriction intervention produced a small average change in DunedinPACE; some other clocks did not change. The target was 25% restriction, but participants did not all sustain that amount. A clock response is not proof of longer life.

A small eight-week randomized pilot reported a difference in Horvath age between intervention and control groups. It tested several changes together and did not show a three-year gain in lifespan or whole-body rejuvenation.

The COSMOS substudy found a modest slowing on PhenoAge and GrimAge over two years. It did not significantly slow all five clocks or prove the result came from correcting deficiencies. That possible mechanism and the clinical meaning need further study.

A study of aging anxiety in women reported some associations with age scores. The cumulative-anxiety association weakened after health and behavior adjustment. It doesn’t show that anxiety writes faster aging into your DNA.

New clock types beyond DNA methylation. In 2025, researchers built epigenetic clocks based on histone modifications rather than DNA methylation, a new class of aging biomarker. And EpiAge, published in Nature Aging, is a pan-tissue clock that works across 19 tissue types.

A unified benchmarking framework. Ying et al. in Nature Aging (2025) published the first systematic framework for evaluating and comparing aging biomarkers.

Technical-variation research shows why replication matters. Error differs by clock and platform. Using the same lab and method can help comparison, but it cannot remove biological variation or prove that a trend represents slower aging.

THE CEO'S PROTOCOL: How to use biological age tests

My Results

I've tested my biological age on five different platforms. Here's what came back.

These are my reported results from different dates and methods. They are not interchangeable ages or verified rates of whole-body aging.

Five of the six results looked favorable against their reference values. Reassuring, but five favorable scores are not five independent proofs that I’m younger.

The number itself is not the takeaway. What surprised me was what I learned from the disagreements between tests.

Five tests, five different answers. Here's why.

These tests don't disagree because some are right and others wrong. They disagree because they're measuring fundamentally different things.

Clinical PhenoAge was trained using population data and routine markers. A favorable result can reflect favorable lab values; it is not guaranteed for every healthy person.

Epigenetic clocks (OMICm, SYMPHONY, Generation Lab) measure DNA methylation patterns at different CpG sites, with different algorithms, trained on different cohorts. SYMPHONY maps 133 clinical biomarkers via methylation; Generation Lab uses 460+ CpG sites across 19 organ systems. Same patient, different data layers, different numbers.

IgG glycosylation and methylation capture different features. Both can change, and we should not assign each one a universal weeks-versus-years biological window.

My DunedinPACE result was 0.75. I find the pace concept useful for discussion, but it does not show that each calendar year costs me exactly nine months of biological time.

Three things a bio-age test can actually tell you (and one thing it can't)

After running all of these on myself, I think a bio-age test has three legitimate uses:

Cross-platform questions: Generation Lab reported a respiratory age of 40.2 and TruDiagnostic a lung-capacity estimate at the forty-seventh percentile. I have borderline asthma and several prior COVID infections, so the pattern interested me. Two inferred scores do not diagnose the cause; respiratory symptoms and direct lung testing still matter.

Tracking: My GlycanAge changed from thirty-two to thirty-one over five months, and the report’s Glycan Shield percentile rose from eighty-three to ninety-six. I suspected recovery after COVID contributed because I had not made another major change. I can’t tell from that score alone whether the inflammation resolved.

3. Motivation with guardrails. Concrete numbers change behavior more than general advice. But set expectations before you test. Ask: "What clinical decision would this result change?"

What it can't tell you: Your actual risk of a specific disease. A low bio-age does not mean you don't have a ticking lipid problem, a brewing autoimmune condition, or an undiagnosed cancer.

Before buying an aging test, choose routine labs around your actual preventive or clinical question. Blood pressure, lipid measures including ApoB and Lp(a) when useful, glycemic assessment, and other indicated tests often change care more directly than an age estimate.

Start with clinical questions and routine markers

These are the traditional markers that held their own against bio-age scores in the CARDIA study.

Clinical PhenoAge needs albumin, creatinine, glucose, CRP, lymphocyte percentage, MCV, RDW, alkaline phosphatase, white-cell count, and chronological age. CRP is not part of a CBC or CMP. Use the calculator’s required units; a unit error can produce a misleading result.

hsCRP: My value was 0.5 mg/L. This is one clue to systemic inflammation and cardiovascular-risk context, not a diagnosis. An illness can raise it, and lab reference ranges and preventive-risk thresholds serve different purposes.

Glucose assessment: My fasting insulin was 3.1 µIU/mL and HbA1c 5.3%. HbA1c, fasting glucose, and insulin-based estimates answer related but different questions. A fasting insulin above seven or eight alone does not diagnose prediabetes, and there is no clinical “elite” category.

Lipids: Before treatment, my LDL-C was 192 mg/dL, ApoB 128 mg/dL, and LDL particle number 2,291 nmol/L. After Repatha and ezetimibe, the reported values were 37, 45, and 911, respectively. That is my treatment history. Targets depend on cardiovascular risk; a good particle-size report does not erase elevated ApoB or LDL-C.

No short set of “optimal” numbers guarantees a sound metabolic foundation. Read the results with history, blood pressure, medicines, and the clinical question.

Optional aging measures

I track DunedinPACE, but it is not a direct measurement of the last two or three months of aging. Research calibration and a person’s retest interval are different. The study does not establish 0.75 or 0.85 as clinical treatment targets.

I review exercise, sleep, food, stress, and metabolic health because they matter even without an aging score. We don’t have proof that a specific weekly exercise or 16:8 fasting schedule reliably lowers every individual’s DunedinPACE.

I also find glycan trends interesting. A faster response is not automatically a more useful or more accurate clinical test.

An organ-age outlier can prompt a question, but it does not localize a diagnosis. Testing for liver disease or exposures needs clinical evidence beyond the inferred age.

Tier 3: Where two tests agree (~$500+, 1-2x/year)

TruDiagnostic SYMPHONY breaks aging into 11 organ systems. Every system was aging slowly except immune and hormone, which lagged. I suspect COVID did a number on my immune metrics. I'll recheck to see if the pattern is persistent.

Generation Lab flagged my respiratory system aging fastest (40.2 vs calendar age 35). Not surprising given the COVID and asthma. I've started breath training and will track whether respiratory age shifts.

I have seen favorable methylation scores alongside higher glycan ages. That disagreement can prompt review, but it does not establish gut permeability, autoimmunity, or a particular inflammatory cause.

A research model that combines three measures

BASE-II selected DunedinPACE, muscle mass, and standing balance as a smaller predictive set. This was model selection within a cohort, not a trial showing that the package improves care.

Specifically:

• DunedinPACE: a pace-related methylation score, with no universal intervene/monitor target.
• Appendicular lean-mass index: arm and leg lean mass divided by height squared; interpret with strength, function, and the relevant diagnostic criteria.
• Balance: use a defined test with safe support nearby. Eyes-open and eyes-closed results are different tests and should not share one cutoff.

The cohort’s smaller model had a C-index of 0.63 versus 0.65 for the full model. Those values describe discrimination, not a guaranteed ten-minute assessment or an individual life expectancy.

If You Want to Test on Your Own

If you want to order directly, check the assay, reported scores, current price, evidence, and access to interpretation. These were examples I discussed in March 2026, not a current ranking or shopping list.

TruDiagnostic: I used its reports for several clocks and inferred organ-system measures. Check the current package and price; a longer report does not prove more clinical value.

NOVOS Age: An option discussed for pace-related tracking. Verify the current included assay and retest terms before buying.

Elysium Index: Check its current methodology, assay, availability, and price rather than relying on the old quoted range.

Tally Health: The cheek-swab approach differs from blood testing. Convenience alone does not establish clinical utility; check current membership and assay details.

Thorne: This issue discussed a clinical-marker age report. Check the current product and full cost, including any lab draw.

myDNAge: This issue discussed a Horvath-based reference estimate. Confirm the current assay and specimen options; predicting chronological age does not demonstrate treatment guidance.

GlycanAge: It uses IgG glycosylation, a different signal from methylation. That difference does not make it a diagnostic test for inflammation.

Clinical PhenoAge: The calculation can be free when you already have every required lab value, including CRP, with correct units. It is a model result, not a diagnosis.

For repeat testing, use the same method when possible and compare results with changes in symptoms, fitness, and routine markers. A lower score may reflect biology, error, or both; it does not prove your intervention slowed aging.

How I Counsel Patients

When a DunedinPACE result is high relative to its reference, I explain that it is a modeled estimate. A value such as 1.12 does not prove that this patient is aging exactly twelve percent faster or identify what needs treatment.

I then review sleep, activity, metabolic trends, stress, alcohol, and symptoms. The test can start a conversation, but a finding in that review is not proof the clock detected its cause.

Exposure tests need an exposure history or clinical reason. A persistent elevated clock alone does not justify PFAS or heavy-metal testing.

A favorable result does not remove existing health risks. I encourage patients to keep the habits that support their health and judge them by useful outcomes, not only an age score.

A lower retest can feel encouraging. It is evidence that the measured score changed, not proof that choices reversed biological aging. I also want to see how the patient feels and functions.

Before ordering, I ask what the result can change. Standard preventive or metabolic evaluation depends on the patient’s risks and goals. HbA1c and fasting insulin are not interchangeable, and an aging test does not make homocysteine or vitamin D testing necessary for everyone.

I also set expectations about precision. I tell every patient: "This number will have noise. We're not looking for your age to the decimal. We're looking for your trend over 6 to 12 months. One number means almost nothing. Two or three data points start telling a story."

The Honest Cost

I spend roughly $1,500 a year on these tests because understanding them is part of my work. For most people, I would first address the clinical risks we already know how to treat.

The test is only as good as what you do with the result.

My Retesting Schedule

Links & Codes

I have clinic partnerships with several of these companies, which means I can offer special pricing to my readers and patients.

• My longevity practice offers testing and interpretation.
• TruDiagnostic: ask the clinic about current options.
• GlycanAge: an IgG-glycosylation age estimate; check current CARECORE terms.
• Generation Lab: inferred organ-system estimates; provider options can change.
• Clinical PhenoAge: needs all nine biomarkers and chronological age, including CRP.

And you can book a consult with me to go over any of the above if helpful.

KEY TAKEAWAYS

📌 Clinical risk factors and routine markers remain useful. Their value depends on the outcome and decision; no one panel equals every age test.

📌 DunedinPACE showed the strongest mortality association among fourteen markers in BASE-II. That is not a universal clinical ranking or a treatment target.

📌 Different tests can disagree because of biology, model design, and error. Disagreement does not tell us which factor caused it.

📌 First-generation clocks (Horvath, Hannum) are research tools, not clinical tools. They predict chronological age, not disease or death. Don't spend money on these for personal health tracking.

📌 A repeated pattern or changing score can raise a question. It doesn’t diagnose a cause or prove an intervention works.

📌 Consistent methods help serial comparison. A trend still needs context; more data points do not automatically separate treatment effects from noise.

📌 Before buying a test, ask: "What clinical decision would this change?" If you can't answer that, you're not ready for the test.

LONGEVITY QUICK HITS

🧠 Gut–brain research in mice: Researchers studied a bacterial pathway and memory in aged mice. It identifies a possible target, not a proven treatment for human age-related memory loss.

🍊 Vitamin C and aging models: Animal research on ACSL4 and ferroptosis reported molecular and clock changes. It doesn’t prove that vitamin C reverses aging or extends life in people.

⚠️ Before you try senolytics, read this. The popular dasatinib + quercetin combo caused profound myelin loss in mice by pushing oligodendrocytes into a dysfunctional state. Damage was actually worse in younger animals. Preclinical, but a real caution signal for anyone taking D+Q prophylactically (PNAS).

🍄 Psilocybin beats nicotine patch for smoking cessation. First head-to-head RCT: 40.5% quit rate vs. ~10% for the patch. The psychedelic-as-medicine evidence keeps building.

💊 GLP-1s and muscle: Research combining experimental and clinical data distinguishes fat, lean tissue, and function. It doesn’t make muscle loss impossible or settle every patient’s nutrition and training needs.

💉 BELIEVE: The bimagrumab–semaglutide trial studied weight and body composition. Those results do not make the experimental combination an approved muscle-sparing treatment.

🏆 Wegovy just got a lot stronger. FDA approved Wegovy HD at 7.2mg, the highest-dose semaglutide injection ever. STEP UP trial showed ~21% weight loss.

👁️ Semaglutide and optic neuropathy: An analysis of FDA adverse-event reports found a stronger reporting signal for Wegovy than Ozempic. It cannot show that dose caused the difference or tell us a person’s risk. Sudden vision loss needs urgent assessment; review medication decisions with a clinician.

💉 PCSK9 base editing: Early human research reported LDL-C reductions. Follow-up cannot yet establish permanent benefit, long-term safety, or replacement of standard lipid treatment.

🔥 PEP and inflammation: Mechanistic research links phosphoenolpyruvate with cGAS–STING signaling. It doesn’t establish a human anti-aging treatment.

💊 GLP-1s and substance-use outcomes: A large veterans analysis found associations with lower recorded substance-use-disorder risks. It was not a randomized addiction-treatment trial, and veterans may differ from other populations.

😴 Napping and aging measures: A genetic analysis examined napping, age measures, and frailty. Its assumptions and outcomes do not establish that an afternoon nap makes one reader age faster.

FROM THE LONGEVITY SHOW

In this early podcast video, I discuss biological-aging measurements and the evidence behind attempts to change them. A lower measurement should not be read as proof of rejuvenation.

Keep learning,

Hillary Lin, MD

Was this useful? Forward it to someone who's been thinking about testing their biological age.

Follow me: YouTube | LinkedIn | Instagram | X | TikTok

Want to turn your wellness brand into a full-service health destination? Learn about Curator's platform or Get Started Here

Events listed when this issue was published

👱🏻‍♀️👩🏻‍🦰👩🏻👧🏽👧🏾 Livelong Women’s Health Summit: April 17-18, 2026, SF, CA. Delighted to join 50 other thought leaders in speaking on women’s longevity! (Code HILLARY50 for $50 off)

Offers listed when this issue was published

This issue included a NeuroAgeTx affiliate offer. Check current services and prices before buying.

This issue included a Timeline offer. Urolithin A marker studies do not establish longer human life. Check current terms.

GlycanAge reports an IgG-glycosylation age estimate. It doesn’t diagnose or tell you how to treat inflammation.

I have liked ZBiotics products. That preference is not proof that they extend life or offset alcohol’s health risks. Check current offer terms.