The Lifespan Blood Test: What Predicts Mortality, and How to Measure It

Oak
biomarkers mortality phenoage epigenetic-clocks lab-data
A feathered long-necked creature on a dark plinth, nine glowing nodes along its throat and a ring of light circling its neck.

There is no single blood test for lifespan. The closest thing that exists is a composite measure built from nine routine chemistry and hematology analytes plus chronological age. Those inputs are combined using Cox regression coefficients fit to 10-year mortality. That is the construction behind PhenoAge, which was derived from the NHANES III cohort1. Everything else worth ordering is an addition to that base. That includes ApoB for lipid exposure, hs-CRP and IL-6 for inflammatory tone, and cystatin C for filtration. It also includes NT-proBNP for cardiac wall stress and plasma neurofilament light (NfL) for neuroaxonal injury. The signal lives in the combination of markers and in the slope across repeated draws, rather than in any one value crossing a reference range.

The nine analytes that do most of the work

This section covers the core panel, what it costs, and the two ways people most often get it wrong. It matters because these nine measurements carry most of the predictive weight. They are also cheap enough that the real risk is a silent arithmetic error rather than an unavailable test.

PhenoAge uses ten inputs. They are albumin (g/L), creatinine (µmol/L), glucose (mmol/L) and log C-reactive protein (mg/dL). It also uses lymphocyte percent (%), mean cell volume (fL) and red cell distribution width (%). The remaining inputs are alkaline phosphatase (U/L), white blood cell count (10³/µL) and chronological age. The linear predictor from those inputs is mapped through a Gompertz mortality model to produce a “phenotypic age” expressed in years. Because the model was trained on mortality rather than on calendar age, it outperforms chronological age for mortality prediction and tracks healthspan outcomes as well1.

All nine can be ordered for under $100 as a comprehensive metabolic panel plus a complete blood count with differential plus hs-CRP. Each one is a routine test with a LOINC code, the standard identifier system for laboratory observations. That means any clinical lab already runs it.

The first failure mode is units. The published coefficients assume creatinine in µmol/L, while US labs report mg/dL, so multiply by 88.4. They assume glucose in mmol/L, so divide mg/dL by 18.016. Albumin enters in g/L, so multiply g/dL by 10. CRP enters in mg/dL as a natural log, and because US hs-CRP is reported in mg/L, you divide by 10 first. Get one of these wrong and the resulting phenotypic age can be off by a decade without any error being thrown. The defense is a unit-conversion layer with explicit source and target units, plus range assertions that run before the model sees anything.

The second failure mode is an hs-CRP result of 0 mg/L. Many assays report <0.3 below the limit of quantification, and log(0) evaluates to -Inf. Substitute the limit of quantification, or half of it, and record in the data which choice you made.

Red cell distribution width deserves a note of its own. It is the cheapest thing on the panel and one of the more consistent mortality associations in the hematology literature. Red cell indices more broadly carry aging signal that a CBC already gives you at no extra cost2.

What to add beyond the nine

Having established the base panel, the next question is what genuinely adds information on top of it. The multi-marker mortality literature is consistent on one point. Markers drawn from different physiological systems inform one another. Combinations of neuroendocrine and immune markers with metabolic and cardiovascular markers predict later-life mortality better than the markers of any single system alone3. The 10-year all-cause and disease-specific mortality signatures described in aging cohorts are similarly multi-system4. Proposed healthy-ageing panels converge on the same structure, sampling physical, cognitive, immune, endocrine, and metabolic domains together5.

With that logic in mind, here is what we would add and the reasoning behind each choice.

  • ApoB, measured by nephelometry and standardized to the WHO/IFCC SP3-07 reference material. It gives one particle count per lipoprotein, with no fasting requirement and no reliance on the Friedewald estimation of LDL. Lipid metabolism is deeply tied to aging biology and to age-related disease6.
  • Lp(a), once, reported in nmol/L from a particle-number assay rather than in mg/dL of mass. Levels are largely genetic, so a single lifetime measurement is enough unless the assay changes.
  • Cystatin C alongside creatinine. Creatinine tracks muscle mass, which drifts with age and with training. When eGFR estimated from cystatin C and from creatinine disagree by more than roughly 15 mL/min/1.73m², that disagreement is itself informative and worth a clinician’s read.
  • HbA1c plus fasting insulin. HbA1c is distorted by anything that changes red cell lifespan. That is precisely the population in which RDW is also moving, so the two should be read together.
  • IL-6 and hs-CRP. The two are correlated but not redundant.
  • NT-proBNP, which is age- and sex-dependent and carries a strong mortality signal in older cohorts.
  • Plasma NfL measured by single-molecule array, such as the Quanterix Simoa HD-X or equivalent. Plasma NfL rises with age and is associated with mortality in old age independent of clinical diagnosis7. The relationship also holds across species, which is unusual and suggests the assay is measuring something fundamental about neuroaxonal turnover rather than an artifact of one cohort8. Interpret it only against age-stratified reference data, and bear in mind that low eGFR and low BMI both raise it.

Epigenetic clocks, and when they are worth the money

DNA methylation clocks belong to a different measurement class from blood chemistry, and they are worth understanding before you pay for one. First-generation clocks were trained to predict chronological age from CpG beta values, the fraction of methylated molecules at a given site. They are remarkably accurate at that task9. Second-generation clocks such as DNAm PhenoAge were instead trained on the mortality-fit phenotypic age rather than on calendar age. That is why they carry mortality and healthspan signal that a chronological-age clock does not1.

In practical terms, order an Illumina EPIC array (v2, roughly 935k probes) on buffy coat or whole blood. Insist on receiving the raw IDAT files, meaning the green and red IDATs for each sample plus the sample sheet. Without them, all you have is a single number from a vendor pipeline you cannot audit.

Starting from the IDATs, the processing in R looks like this:

library(sesame)
betas <- openSesame("idats/", func = getBetas)   # includes noob + nonlinear dye bias
# then a clock package, e.g. dnaMethyAge or methylclock, against `betas`

There are two things to control for. The first is cell composition, because whole blood beta values move with the granulocyte-to-lymphocyte ratio. Estimate cell fractions using the FlowSorted.Blood.EPIC reference and retain them as covariates in any longitudinal comparison. The second is technical reliability. Single-CpG clocks have test-retest noise on the order of years, which swamps a change measured over six months. Principal-component versions of the clocks are substantially more reliable, and those are what we use for repeated measures. If you intend to sample only once, the clock is an expensive way to learn something the blood panel has already told you10.

Repeat measurement is the whole game

A single draw gives you a position. Three or more give you a slope, and the slope is what you and a clinician can act on. That makes the discipline around collection as important as the choice of assay, because uncontrolled pre-analytics turn a slope into noise.

The following practices keep repeated draws comparable.

  • Use the same lab, the same platform, and the same assay each time. A Roche hs-CRP and an Abbott hs-CRP are not interchangeable in the 0.5-2 mg/L range that matters most.
  • Fast for 10-12 hours before glucose and insulin measurements. Note the fasting hours in the record even for assays that do not require fasting.
  • Avoid strenuous exercise for 48 hours before a draw, since CK, AST, and CRP all move in response to it.
  • Account for posture, which has a real effect. Standing for 15 minutes before venipuncture concentrates albumin and hemoglobin by several percent compared with a supine draw.
  • Draw at the same time of day, within an hour.
  • Defer any draw for at least two weeks after an infection or a vaccination, because CRP and IL-6 will be uninterpretable.

Before calling a change real, compute the reference change value: RCV = 2.77 × sqrt(CVa² + CVi²). Here CVa is the lab’s analytical coefficient of variation, which you can ask for because they have it. CVi is the within-subject biological variation. The EFLM biological variation database provides both. For albumin the RCV is small, so a 10% drop is meaningful. For hs-CRP, within-person variation is wide enough that a single value near 3 mg/L tells you almost nothing. That is why we use the median of three draws.

How we would store it

Storage sounds like a detail, but it determines whether you can trust a number you computed six months ago. We recommend a single long-format table with one row per analyte per draw:

draw_id, collected_at, fasting_hours, analyte, loinc, value, unit, assay, instrument, lab, loq, flag

Keep values in the units the lab reported and convert at load time rather than in place. Store the original PDF or HL7 payload next to the parsed row. Version the scoring code and record which commit produced which phenotypic age. When you re-derive PhenoAge six months later with a unit bug fixed, you want to know exactly which historical numbers changed and why.

From there, compute the deltas that matter: PhenoAge minus chronological age, and the trajectory of that difference over time. Anything outside a reference range goes to a physician, as does any marker moving quickly in a bad direction. These are measurements, and interpretation for an individual is a clinical judgment. That applies with particular force to NT-proBNP, NfL, or a falling albumin.

Questions people also ask

What is the number one predictor of life expectancy? Chronological age, by a wide margin, which is why every useful biomarker is evaluated on what it adds beyond age. Among blood measures, the strongest single-analyte performers in older cohorts are albumin, RDW, and NfL. A multi-system composite beats any of them individually34.

What are the top 5 biomarkers for longevity? If you had to pick five from blood, we would choose albumin, RDW, hs-CRP, HbA1c, and ApoB. Add plasma NfL as a sixth if you are over 60 and can get a Simoa assay7.

What are the most important blood tests for seniors? That is a clinician’s call, driven by history and medications rather than by a generic panel. The commonly ordered set in older adults includes CBC with differential, CMP (albumin, creatinine, liver enzymes), cystatin C, TSH, B12, ferritin, HbA1c, and NT-proBNP. Age-adjusted interpretation matters here, because a “normal” creatinine in an 80-year-old with low muscle mass can hide meaningful loss of filtration.

What are the most important biomarkers for longevity? The ones trained on mortality rather than on disease diagnosis. In practice that means the PhenoAge composite1, second-generation methylation clocks910, and multi-system panels spanning the immune, metabolic, cardiac, and neuroaxonal domains5.

What increases longevity the most? We measure, and we do not prescribe. The interventions with the largest population-level effects are well documented in the public health literature, and they are not blood tests. Any change to your own regimen belongs in a conversation with a physician who has your full record.

Oak builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.

Footnotes

  1. Morgan E. Levine, Ake T. Lu, Austin Quach, et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging, 2018. https://doi.org/10.18632/aging.101414 ↩ ↩2 ↩3 ↩4

  2. Roula P. Kyriacou, Sapha Shibeeb. Red Blood Cells and Human Aging: Exploring Their Biomarker Potential. Diagnostics, 2025. https://doi.org/10.3390/diagnostics15161993 ↩

  3. Tara L. Gruenewald, Teresa E. Seeman, Carol D. Ryff, et al. Combinations of biomarkers predictive of later life mortality. Proceedings of the National Academy of Sciences, 2006. https://doi.org/10.1073/pnas.0606215103 ↩ ↩2

  4. Yanxia Lu, Gianni Monaco, Xavier Camous, et al. Biomarker Signatures Predicting 10-Year All-Cause and Disease-Specific Mortality. The Journals of Gerontology: Series A, 2018. https://doi.org/10.1093/gerona/gly138 ↩ ↩2

  5. Jose Lara, Rachel Cooper, Jack Nissan, et al. A proposed panel of biomarkers of healthy ageing. BMC Medicine, 2015. https://doi.org/10.1186/s12916-015-0470-9 ↩ ↩2

  6. Adiv A. Johnson, Alexandra Stolzing. The role of lipid metabolism in aging, lifespan regulation, and age‐related disease. Aging Cell, 2019. https://doi.org/10.1111/acel.13048 ↩

  7. Stephan A. Kaeser, Benoit Lehallier, Mikael Thinggaard, et al. A neuronal blood marker is associated with mortality in old age. Nature Aging, 2021. https://doi.org/10.1038/s43587-021-00028-4 ↩ ↩2

  8. Carina Bergmann, Lisa M. Häsler, Marius Lambert, et al. Neurofilament light chain may serve as a cross-species blood biomarker to assess aging and predict mortality. PLOS Biology, 2026. https://doi.org/10.1371/journal.pbio.3003606 ↩

  9. Steve Horvath, Kenneth Raj. DNA methylation-based biomarkers and the epigenetic clock theory of ageing. Nature Reviews Genetics, 2018. https://doi.org/10.1038/s41576-018-0004-3 ↩ ↩2

  10. Juulia Jylhävä, Nancy L. Pedersen, Sara Hägg. Biological Age Predictors. EBioMedicine, 2017. https://doi.org/10.1016/j.ebiom.2017.03.046 ↩ ↩2