What the Yale biological age calculator computes, and how to run it on your own labs

The “Yale biological age calculator” that the consumer sites wrap in a form is PhenoAge, developed by Morgan Levine and colleagues while she was at Yale School of Medicine. It takes nine routine clinical chemistry and hematology values plus your chronological age, runs them through a Gompertz proportional-hazards model of ten-year all-cause mortality fit in NHANES III, and then inverts that mortality risk back onto the age scale. The output answers one question: what is the chronological age at which the average person in the reference population carries your ten-year mortality risk. It is a risk score wearing the units of years, and that framing matters for how you should read a result.

Below is the full specification, the places implementations go wrong, and what we would do with the number once you have it.

The formula, with units

There are two stages. First a linear predictor, then a rescaling. The linear predictor is a weighted sum of ten terms plus an intercept. Units are not optional here, because the coefficients were fit in specific ones.

Term	Unit	Coefficient
Albumin	g/L	−0.0336
Creatinine	µmol/L	0.0095
Glucose, serum	mmol/L	0.1953
ln(C-reactive protein)	mg/dL	0.0954
Lymphocyte percent	%	−0.0120
Mean cell volume	fL	0.0268
Red cell distribution width	%	0.3306
Alkaline phosphatase	U/L	0.00188
White blood cell count	10³ cells/µL	0.0554
Chronological age	years	0.0804
Intercept	—	−19.9067

Call that sum xb. The second stage converts it to a ten-year mortality probability and then to an age:

import math

GAMMA = 0.0076927

def phenoage(xb):
    m = 1 - math.exp(-math.exp(xb) * (math.exp(120 * GAMMA) - 1) / GAMMA)
    return 141.50225 + math.log(-0.00553 * math.log(1 - m)) / 0.090165

The 120 is months, giving a ten-year horizon. The final line is the inverse of a mortality-to-age mapping, which is why PhenoAge can exceed 120 or go negative for extreme inputs. If you see a result outside roughly 20 to 110, suspect a unit error rather than an interesting biology finding.

The unit conversions that break most implementations

US labs report several of these analytes in units the model does not use. Convert before you sum.

• Creatinine mg/dL to µmol/L: multiply by 88.4.
• Glucose mg/dL to mmol/L: divide by 18.016 (or multiply by 0.0555).
• Albumin g/dL to g/L: multiply by 10.
• CRP mg/L to mg/dL: divide by 10. Then take the natural log.

The CRP term has two further traps. High-sensitivity CRP assays often report “<0.3 mg/L” or similar at the low end, and ln(0) is undefined. We substitute half the detection limit and record that we did so. Second, do not confuse hs-CRP with standard CRP panels that floor at 3 or 5 mg/L. The standard assay will systematically inflate your score.

Lymphocyte percent is the percentage of white cells. The absolute count in 10³/µL is not the same quantity. If your CBC reports only absolutes, compute percent as lymphocyte absolute divided by total WBC, times 100.

Which inputs move the number, and by how much

Dividing each coefficient by the age coefficient of 0.0804 gives an intuition for sensitivity in “years of chronological age equivalent” inside the linear predictor. These are approximations near the middle of the range, since the second stage is nonlinear.

• Red cell distribution width, +1 percentage point: about 4.1 years.
• Glucose, +1 mmol/L (about 18 mg/dL): about 2.4 years.
• Creatinine, +10 µmol/L (about 0.11 mg/dL): about 1.2 years.
• White blood cell count, +1 × 10³/µL: about 0.7 years.
• Albumin, −1 g/L: about 0.4 years.
• Mean cell volume, +1 fL: about 0.3 years.
• Alkaline phosphatase, +10 U/L: about 0.2 years.
• CRP, one doubling: about 0.8 years.
• Lymphocyte percent, −1 point: about 0.15 years.

RDW dominates. That is worth sitting with, because RDW is a measure of variability in red cell size and is sensitive to iron status, B12, folate and recent blood loss. A single draw during a mild deficiency can shift your PhenoAge by several years without anything having changed about your rate of aging. CRP and WBC are the other volatile inputs. Both respond to an infection, a vaccination, a poor night of sleep or a hard training session in the preceding days.

How we would run it

Do not treat a single draw as a measurement of your biology. Treat it as one noisy sample from a distribution and estimate the distribution.

Standardize the pre-analytic conditions. Draw fasting in the morning, hydrated, at the same lab with the same instrument platform. Have no acute illness in the prior two weeks and no strenuous exercise in the prior 48 hours. Instrument-to-instrument differences in RDW and MCV are real, so switching labs mid-series introduces a step change you will misread as signal.

Then compute PhenoAge on every draw and fit a trend rather than reading points. Three to four draws a year gives you a usable slope within two years. We look at the slope of PhenoAge minus chronological age over time, which is dimensionless and is the quantity that would change if something about the underlying trajectory changed. We also keep the nine inputs visible alongside the composite, because a two-year move driven entirely by CRP is a different observation from one driven by glucose and creatinine together.

Any individual value outside the reference range belongs with a clinician, independent of what the composite says. Elevated creatinine, a rising alkaline phosphatase or a persistently high RDW are clinical findings. PhenoAge was not built to work them up and will happily average them into a single reassuring number.

If you would rather not hand-code the arithmetic, the BioAge R package implements PhenoAge along with Klemera-Doubal biological age and handles the NHANES-based standardization. Reading its source is also the fastest way to check your own implementation against a reference.

Blood-based PhenoAge versus the epigenetic clocks

PhenoAge is the reason DNAm PhenoAge exists: Levine’s group trained a DNA methylation predictor to reproduce the blood-based score, and that methylation clock is what most longevity studies now report. If you are choosing where to spend attention, the tradeoff is concrete.

The blood version is cheap, repeatable and built from analytes your physician already understands. Its weakness is that several inputs are acute-phase reactants, so week-to-week variance is large relative to the annual trend you care about.

The methylation versions are further from clinical interpretation and carry their own reliability problem. First-generation clocks show substantial technical variance between replicate assays of the same sample. Principal-component reconstructions of the clocks were developed specifically to suppress that noise for longitudinal tracking and trial use.¹ Careful decomposition shows that even after that correction, the split between biological and technical reliability differs by clock. That changes how much of a measured shift you should believe.² The field’s own consensus recommendations emphasize validation and standardization before these measures are used to guide decisions in individuals.³

There is also a direction-of-causation question that neither approach settles. Methylation changes track age closely, and whether they drive aging or record it remains open.⁴ Intervention data are beginning to arrive, and the responsiveness of epigenetic measures to human longevity interventions is now being tested directly rather than assumed.⁵ Results can also run counter to intuition. In low-income adolescents, higher self-control was associated with faster epigenetic aging, which is a useful reminder that these scores are not a straightforward virtue meter.⁶

Our view: compute PhenoAge because the inputs are nearly free and the series is informative, and read it as a slope over years. Add a methylation measure if you want a second, partly independent axis, and prefer the principal-component or otherwise reliability-corrected versions when you do. Cell-type-resolved approaches are where the interesting work is heading, since single-cell immune clocks show that different immune compartments age at different rates within the same person.⁷

Questions people also ask

Is the Yale biological age calculator the same as PhenoAge? Yes. The Yale attribution refers to Morgan Levine’s work at Yale School of Medicine, and the calculator implements the phenotypic age equation from her 2018 paper. The various consumer sites are all running the same nine-biomarker formula with different front ends.

Which blood test do I need? A comprehensive metabolic panel gives albumin, creatinine, glucose and alkaline phosphatase. A CBC with differential gives white blood cell count, lymphocyte percent, mean cell volume and red cell distribution width. The one addition is high-sensitivity CRP, which is often not bundled and must be ordered specifically.

How accurate is it for one person? It was validated as a population-level mortality predictor. It was not validated as a per-person measurement instrument. Most people land within a few years of their chronological age, and a single result should be read with an uncertainty band of several years given the day-to-day variability in CRP, WBC and RDW.

Can I lower my PhenoAge? The score will move when its inputs move, and inputs like glucose and CRP respond to many things. Whether a change in the score reflects a change in underlying aging rate is the open question the intervention literature is currently testing.⁵³ We would not treat the number as a target to optimize directly.

Footnotes

1. Albert T. Higgins-Chen, Kyra L. Thrush, Yunzhang Wang, et al. A computational solution for bolstering reliability of epigenetic clocks: implications for clinical trials and longitudinal tracking. Nature Aging, 2022. https://doi.org/10.1038/s43587-022-00248-2 ↩

2. Raghav Sehgal, Daniel S. Borrus, John Gonzalez, et al. Biological Versus Technical Reliability of Epigenetic Clocks and Implications for Disease Prognosis and Intervention Response. Aging Cell, 2026. https://doi.org/10.1111/acel.70635 ↩

3. Biomarkers of Aging Consortium, Chiara M. S. Herzog, Ludger J. E. Goeminne, et al. Challenges and recommendations for the translation of biomarkers of aging. Nature Aging, 2024. https://doi.org/10.1038/s43587-024-00683-3 ↩ ↩²

4. Morgan E Levine, Sara Hagg. DNA METHYLATION: CAUSE OR CONSEQUENCE OF AGING?. Innovation in Aging, 2019. https://doi.org/10.1093/geroni/igz038.127 ↩

5. Raghav Sehgal, Daniel Borrus, Jenel F. Armstrong, et al. Responsiveness of epigenetic aging biomarkers to longevity interventions in humans. Nature Medicine, 2026. https://doi.org/10.1038/s41591-026-04562-9 ↩ ↩²

6. Hyungkyung Kim, Edith Chen, Gregory E. Miller, et al. Does High Self-Control Accelerate Epigenetic Aging in Low-Income Adolescents?. Journal of Adolescent Health, 2025. https://doi.org/10.1016/j.jadohealth.2024.10.012 ↩

7. Wenchao Li, Zhenhua Zhang, Saumya Kumar, et al. Single-cell immune aging clocks reveal inter-individual heterogeneity during infection and vaccination. Nature Aging, 2025. https://doi.org/10.1038/s43587-025-00819-z ↩