# How Biological Age Is Actually Calculated: The 2026 Field Guide
> Updated: 2026-05-27 · Source: https://dorsi.ai/blog/how-biological-age-is-calculated
Three generations of DNA-methylation clocks (Horvath, PhenoAge, GrimAge, DunedinPACE), the blood-only PhenoAge formula you can run yourself, what InsideTracker / TruDiagnostic / GrimAge actually do under the hood, and whether an Apple Watch is enough to estimate biological age without a blood draw.
Key Takeaways
- "Biological age" is not one number — it's a family of estimators, split into three generations. First-gen clocks (Horvath 2013) predict your chronological age from DNA methylation; they're very accurate at guessing age but only modestly predict mortality. Second-gen (PhenoAge, GrimAge) trains directly on mortality data. Third-gen (DunedinPACE) measures the rate at which you're aging right now, which is what intervention studies actually move.
- The gold standard is GrimAge for mortality risk and DunedinPACE for tracking interventions, both DNA-methylation tests. Cost: $200–$500, requires a clinic visit. TruDiagnostic, NOVOS, and Elysium sell these. Bryan Johnson's Blueprint "Speed of Aging" test is TruDiagnostic underneath.
- You can run PhenoAge yourself from a standard blood panel for $30–80. The formula is fully published (Levine 2018) and takes 9 markers from a CBC + CMP + hs-CRP. It's not as strong as GrimAge but predicts all-cause mortality far better than chronological age alone.
- Apple Watch + a few self-report inputs can give a defensible educational estimate without any blood test. VO₂max (a single number from Apple Watch's Cardio Fitness) is one of the strongest mortality predictors in the literature (Mandsager 2018: 5× mortality difference between elite and low fitness quintiles). Sleep regularity, daily steps, and resting heart rate add the rest. We built a free calculator that does exactly this.
- Most consumer "biological age" scores are marketing. Function Health concedes its score "is not a clinical measure." WHOOP Age is a composite of behavior signals. Oura's Cardiovascular Age is a wrist-PPG estimate of arterial stiffness — useful but not what most users think they're getting. The honest-naming award goes to Garmin: their "Fitness Age" is literally just VO₂max relative to age-sex norms, and they don't pretend otherwise.
## What "biological age" actually means
If you ask ten longevity researchers to define biological age, you get ten subtly different answers — but they all agree on the framing: it's an estimate of how old your body looks at the molecular, physiological, or functional level, independent of how many birthdays you've had.
Three terms to keep straight:
- **Chronological age** — what your driver's license says.
- **Biological age** — the estimate. A 42-year-old whose biology "looks like" a typical 36-year-old has a biological age of 36.
- **Age acceleration** — biological age minus chronological age. Positive means you're aging faster than average; negative means slower.
- **Pace of aging** — separate concept introduced by the DunedinPACE clock. Tells you how many biological years you're racking up per chronological year. 1.0 = average; 1.2 = aging 20% faster. This is the version that interventions actually move.
The whole field exists because chronological age is a terrible predictor of *anything you care about*. Two 60-year-olds can have wildly different 10-year mortality risks; biological-age clocks try to put a number on that difference.
## The three generations of clocks
Biological-age estimators sort cleanly into three generations based on what they were trained to predict.
### First generation: predict chronological age
The original DNA-methylation clocks — **Horvath 2013** (353 CpG sites, multi-tissue) and **Hannum 2013** (71 CpG sites, blood only) — were trained on one target: predict the donor's chronological age from methylation patterns. Both achieved correlations of ~0.96 with age, with median errors of 3–5 years.
Conceptually they're elegant: aging leaves a chemical fingerprint on DNA, and you can read the fingerprint. But predicting chronological age is the *wrong target* if your goal is predicting health outcomes. By construction, a first-gen clock can't exceed chronological age as a mortality predictor — because that's what it was trained to be.
First-gen clocks still have research uses (cross-tissue comparisons, MyDNAge sells one), but the field has moved on. If a consumer product is selling you a "biological age" based on Horvath or Hannum alone, you're paying for the cheaper version of what's possible.
### Second generation: predict mortality
The big break came with **DNAm PhenoAge** (Morgan Levine, 2018) and **GrimAge** (Ake Lu, 2019). Both train the methylation predictor on outcomes that matter:
- **PhenoAge** is a two-step construction. Step 1 takes 9 blood biomarkers and trains a "Phenotypic Age" against all-cause mortality in NHANES III (n ≈ 9,926, ~21 years of follow-up). Step 2 then trains a 513-CpG methylation predictor against that Phenotypic Age target. The result strongly predicts cancer, cardiovascular events, and dementia incidence.
- **GrimAge** goes further: it builds DNAm surrogates for 7 plasma proteins and a methylation-based estimator of smoking pack-years, then combines them via Cox regression directly against time-to-death in the Framingham cohort. It's the best single-timepoint mortality predictor in the published literature. GrimAge v2 (2022) improved performance in non-white cohorts.
Second-gen clocks finally made biological age clinically useful. They're what TruDiagnostic and NOVOS sell.
### Third generation: pace of aging
**DunedinPACE** (Daniel Belsky, 2022) does something different. Instead of estimating an accumulated "age," it estimates the *rate* at which you're aging right now — expressed as biological years per chronological year.
The training data is special: the Dunedin Study cohort (New Zealand, born 1972–73) has had 19 biomarkers measured longitudinally at ages 26, 32, 38, and 45. DunedinPACE was trained to predict the *slope* of decline across those 19 markers from a single methylation snapshot.
Why this matters for intervention: an accumulated-age clock (Horvath, PhenoAge, GrimAge) is like an odometer — it summarizes total wear and changes slowly. A pace-of-aging measure is like a speedometer — it tells you whether what you're doing right now is accelerating or slowing your aging. The TAME-metformin trial and most current longevity studies use DunedinPACE as their outcome metric for exactly this reason.
DunedinPACE was also engineered for high test-retest reliability (ICC > 0.9), which the older clocks didn't always achieve.
## The blood-only version you can actually run
Skip the methylation array for a moment. The **blood-chemistry version of PhenoAge** is the most underrated estimator in this space — it uses 9 markers from a standard panel and gives you a defensible biological age for $30–80.
The 9 inputs:
| Marker | Source panel |
|---|---|
| Albumin | CMP |
| Creatinine | CMP |
| Glucose | CMP |
| C-reactive protein (CRP, ln-transformed) | hs-CRP |
| Lymphocyte percentage | CBC |
| MCV (mean cell volume) | CBC |
| RDW (red cell distribution width) | CBC |
| Alkaline phosphatase | CMP |
| White blood cell count | CBC |
Plus your chronological age. The Levine 2018 formula is fully published:
```
xb = -19.9067
- 0.0336 × albumin
+ 0.0095 × creatinine
+ 0.1953 × glucose
+ 0.0954 × ln(CRP)
- 0.0120 × lymphocyte%
+ 0.0268 × MCV
+ 0.3306 × RDW
+ 0.00188 × ALP
+ 0.0554 × WBC
+ 0.0804 × age
M = 1 - exp( -exp(xb) × (exp(120 × 0.0076927) - 1) / 0.0076927 )
PhenoAge = 141.50225 + ln(-0.00553 × ln(1 - M)) / 0.09165
```
Validation: in NHANES IV, every 5 years of PhenoAge advance roughly doubles all-cause mortality odds. Replicated in UK Biobank and the Health and Retirement Study.
Costs: $30–80 for the lab panel at LabCorp / Quest, free if you upload existing labs your doctor already ordered. The `BioAge` R package (Kwon & Belsky 2021) implements this and a couple of related composites.
## What the consumer apps actually do
Here's what's under the hood at the products you've heard of.
| Product | Substrate | Underlying method | Price | Honest? |
|---|---|---|---|---|
| **TruDiagnostic TruAge** | Blood DNAm | PhenoAge + GrimAge + DunedinPACE + OMICmAge + SYMPHONYAge | $229–499 | Yes — names every clock and links the papers |
| **NOVOS Age** | Blood DNAm | DunedinPACE + Organ Age + telomere | ~$499 | Yes |
| **Elysium Index** | Saliva DNAm | Custom clock developed with Morgan Levine | $299–499 | Yes — peer-reviewed |
| **MyDNAge** | Blood/urine DNAm | Horvath 2013 (first-gen) | $299 | Honest, but selling first-gen tech |
| **InsideTracker InnerAge 2.0** | Blood (17 markers) | Klemera-Doubal method + proprietary | $249–589 | Above-average — names the academic basis |
| **Bryan Johnson's Speed of Aging** | Blood DNAm | TruDiagnostic underneath (DunedinPACE + SYMPHONYAge) | ~$299 | Underlying tech is gold standard; Johnson's personal interpretation ("I reversed aging by 31 years") has been critiqued by aging researchers |
| **GlycanAge** | Finger-prick (IgG glycans) | 29-glycan proprietary; ~140 peer-reviewed papers | $348 | Yes — differentiated substrate |
| **Function Health** | Blood (100+ markers) | Proprietary formula. Function itself states the age score "is not a clinical measure." | $365/yr | Low — sold as a clinical measure de facto |
| **Lifeforce** | Blood (40-50 markers) | No explicit clock — sells biomarker dashboard + clinician access | $129/mo + setup | N/A as a clock |
| **WHOOP Age** | Wearable behavior | 9-metric composite (sleep, HR zones, steps, RHR, VO₂max, lean mass) | $199–359/yr | Medium — it's really a "behavioral age" |
| **Oura Cardiovascular Age** | Finger PPG | Pulse-wave-velocity proxy from overnight PPG morphology | Ring $349+ + $5.99/mo | Medium — labels it Cardiovascular Age (not Biological Age), which is honest, but the PPG-to-PWV mapping isn't independently validated |
| **Apple Watch Cardio Fitness** | Wrist HR + GPS | VO₂max estimate via NN + physiological ODE | Free with Watch | High — they correctly don't call it biological age |
| **Garmin Fitness Age** | Wearable | Literally VO₂max repackaged against age/sex norms | Free with device | Most honest naming in the category |
The pattern: the more honest the product is about what it's actually measuring, the less marketing-friendly the number sounds. "Cardiovascular Age" and "Fitness Age" don't generate the same wow factor as "Biological Age" — but they're what those products are really measuring.
## Can a wearable plus a questionnaire get you close?
The short answer: **yes, for an educational estimate**. Not a clinical-grade one.
Here's why. The strongest single mortality predictor in the published literature is **cardiorespiratory fitness**. Mandsager et al. (2018, *JAMA Network Open*, n=122,000) tracked patients undergoing exercise treadmill testing and found the lowest-fitness quintile had ~5× the mortality rate of the elite quintile. Apple Watch's Cardio Fitness gives you a VO₂max estimate within about ±1.2 mL/kg/min of a lab cardiopulmonary exercise test (per Apple's own validation paper).
Stack a few more high-signal inputs on top:
- **Sleep regularity** — Windred et al. (2024, *SLEEP*) found regularity is a stronger mortality predictor than duration itself. Self-rateable in one question.
- **Sleep duration** — Yin 2017 *J Am Heart Assoc*: U-shape with optimum at 7–8 hours.
- **Resting heart rate** — Aune 2017 *BMC Medicine*: each 10-bpm increase above ~65 is associated with ~17% higher all-cause mortality.
- **Daily steps** — Paluch 2022 *Lancet Public Health*: clear dose-response through ~10,000 steps.
- **Strength training frequency** — Saeidifard 2019 meta-analysis: 2–3 sessions/week reduce all-cause mortality by ~21%.
- **BMI + waist circumference** — Aune 2016 *BMJ* (BMI U-shape) and Janssen 2004 *AJCN* (waist independent of BMI).
- **The lifestyle Big Four**: smoking, alcohol, diet quality, stress.
That's 18 inputs total. None require a blood draw. All are self-reportable in two minutes, and most are already in Apple HealthKit if the user owns a Watch.
The honest framing: the *absolute* number will be ±5 years off because of self-report bias. But the *relative ranking* across dimensions — which factor is dragging you down most, which one is buying you the most years back — is robust, because it's reading directly off the cohort-level evidence.
## So what should you actually do?
A practical decision tree, depending on how seriously you want to engage with this:
**If you're curious but not paying $200+ for a test.** Use a free self-report estimator that's grounded in published research. We built one — 18 inputs, fully cited, 2 minutes, no signup, no blood test, the per-dimension breakdown shows you exactly which factor is your single biggest lever.
**If you have recent bloodwork from a doctor's visit.** Run the Levine PhenoAge formula on it. Free tools exist online; the math is published. You'll get a number that's actually validated against ~20-year mortality outcomes in NHANES III.
**If you're tracking an intervention (caloric restriction, GLP-1, an exercise program) and want a real outcome metric.** Get a TruDiagnostic TruAge (DunedinPACE) test. The pace-of-aging clock is the only one in the literature that moves meaningfully on a 6–12-month timescale. Retest every 3–6 months.
**If you want the single strongest mortality-prediction number with one blood draw.** GrimAge via TruDiagnostic or NOVOS. It's the heaviest hitter for single-snapshot mortality prediction in the published literature.
**If your insurance covers it.** Some clinics now offer biological-age testing under wellness benefits. Ask.
## What this whole field is missing
A few caveats no one in the consumer space talks about much:
1. **Ethnicity bias.** Almost every clock was trained on predominantly European-ancestry cohorts. Recent work (e.g. 2023 *eBioMedicine* refitting KDM and PhenoAge in Chinese adults) finds meaningfully different coefficients. If you're East Asian, South Asian, or African-ancestry, expect ±3–5 years of systematic error on top of everything else.
2. **Test-retest reliability.** First-gen clocks (Horvath, Hannum) can vary 3–9 years between technical replicates of the same blood sample. That number alone should make you suspicious of any "I dropped 4 years!" anecdote that wasn't done on DunedinPACE (ICC > 0.9) or similar high-reliability second/third-gen clocks.
3. **Intervention sensitivity is modest.** Even successful interventions (caloric restriction in CALERIE-2, exercise programs, GLP-1 agonists) typically move DunedinPACE by 0.03–0.05 over 6–12 months. Real, replicable, but not the "10 years younger" headlines suggest.
4. **"Biological age" has no FDA-recognized clinical use.** All consumer claims have to be wellness-framed. This is also why every product has a "not medical advice" disclaimer.
## Bottom line
Biological-age estimation is real science with a real evidence base. It's also a marketing category, and the gap between those two is where most consumers get lost.
If you want the gold-standard number: DNA-methylation, GrimAge or DunedinPACE, $200–$500, via TruDiagnostic or NOVOS.
If you have recent blood work: Levine PhenoAge from your CBC + CMP. Free if you have the values.
If you want a 2-minute estimate with no blood draw: a self-report tool grounded in cohort studies. Try ours, free; the underlying scoring engine is open source ([github.com/YusongCui/biological-age-calculator](https://github.com/YusongCui/biological-age-calculator)) so you can read every coefficient and the paper behind it.
The most useful number isn't the headline age. It's the per-dimension breakdown: which lifestyle factor is your single biggest lever right now. That's what intervention studies actually move.
---
*For full citations and the underlying scoring algorithm, see the open-source skill at github.com/YusongCui/biological-age-calculator.*