Aging Blood Tests: What They Measure and How to Compute Your Own

Oak
biological-age biomarkers epigenetics blood-tests longitudinal
A tall laboratory instrument turns a carousel of glowing blood vials beneath three glass columns of different suspended media merging into one beam.

There is no single aging blood test. What exists is three separate families of measurement that a vendor may sell you under one label: (1) an algorithm over standard clinical chemistry (albumin, CRP, creatinine, glucose, white cell indices) fit to predict mortality, (2) a DNA methylation clock computed from an array run on blood leukocytes, and (3) plasma protein panels, including organ-specific and neurodegeneration markers. The first is the one we would start with, because you can compute it yourself from a $100 blood draw, recompute it every quarter, and inspect every input. The methylation clock adds a genuinely different signal but costs more per measurement and carries more technical noise than its two-decimal output implies.

The three families, and what each one is fit to predict

Blood-chemistry clocks. The canonical construction takes nine routine analytes plus chronological age and fits them to a mortality hazard: albumin, creatinine, glucose, log CRP, lymphocyte percent, mean cell volume, red cell distribution width, alkaline phosphatase, and white blood cell count. A second approach, Klemera-Doubal, regresses each biomarker on age separately and combines the residuals into a single age estimate. Both are implemented in the BioAge R package, trained on NHANES, with published validation against mortality and functional decline 1. Deep networks trained on tens of thousands of blood panels converge on similar inputs: albumin, glucose, and alkaline phosphatase consistently rank among the most informative single markers 2. Larger biobank models built on circulating biomarkers reproduce the same basic result, that a linear or lightly nonlinear combination of common analytes predicts mortality better than chronological age alone 3.

Methylation clocks. These run on an Illumina EPIC array (or a targeted sequencing assay) over blood DNA and produce beta values at hundreds of thousands of CpG sites. First-generation clocks (Horvath, Hannum) were trained to predict chronological age and are therefore, by construction, poor at measuring deviation from it. Second-generation clocks trained on mortality or on longitudinal rate of decline are more useful. DunedinPACE is the one we would use: trained on the rate of change across 19 organ-system markers in a birth cohort followed from age 26 to 45, it reports a pace of aging in years of biological change per calendar year, with reported test-retest ICC around 0.96, far better than earlier clocks 4.

Plasma proteins. Immunoassay panels (Simoa, Olink, SomaScan) measure specific proteins rather than an aggregate index. The neurodegeneration set (p-tau181, p-tau217, GFAP, NfL) is the most developed: in a midlife community cohort, plasma Alzheimer’s neuropathology markers tracked cognition decades before typical clinical presentation 5. These are single analytes with real biology behind them, which makes them more interpretable than any composite and more easily confounded by acute events. NfL and GFAP move after surgery, head injury, and intercurrent illness 6, so a single elevated value with no baseline tells you little.

The panel we would run

Quarterly, from one venous draw:

  • CMP (albumin, glucose, creatinine, ALP, ALT, AST, electrolytes)
  • CBC with differential (WBC, lymphocyte %, MCV, RDW)
  • hs-CRP
  • Lipid panel with ApoB and Lp(a) (Lp(a) once, it is largely genetic)
  • HbA1c and fasting insulin
  • Cystatin C (a creatinine-independent kidney filtration estimate, useful if you carry a lot of muscle)
  • IGF-1, and sex hormones if relevant to you

Annually: an EPIC-array methylation run with DunedinPACE and a mortality-trained clock, and a plasma neuro panel if you have family history that makes you want a baseline.

That covers every input to PhenoAge and KDM, plus the markers that reviews of functional and molecular aging biomarkers converge on 7. Total cost is well under the price of most direct-to-consumer “biological age” products, and you keep the raw values.

Computing the number yourself

Blood chemistry first. Get your results as a CSV with the analyte, value, and unit. Units matter more than anything else here: albumin in g/dL vs g/L, CRP in mg/L vs mg/dL, and creatinine in mg/dL vs µmol/L will silently move your result by a decade.

# install.packages("devtools"); devtools::install_github("dayoonkwon/BioAge")
library(BioAge)

me <- data.frame(
  age = 41, gender = 1,
  albumin = 46,          # g/L
  lymph = 30,            # %
  mcv = 89,              # fL
  glucose = 4.9,         # mmol/L
  rdw = 12.6,            # %
  creat = 77,            # umol/L
  crp = 0.9,             # mg/L  (package log-transforms)
  alp = 58,              # U/L
  wbc = 5.2              # 1000 cells/uL
)

phenoage_calc(data = me, biomarkers = c("albumin","lymph","mcv","glucose",
  "rdw","creat","crp","alp","wbc"))

Run it on every historical panel you have. The absolute value is less informative than the slope across eight quarters.

Methylation next. Insist on the raw IDATs (_Grn.idat, _Red.idat) from whoever runs the array. Vendors that return only a PDF with a number are selling you an opaque function call.

library(sesame)
sdfs  <- openSesame("idats/", func = NULL)     # QC-aware preprocessing
betas <- openSesame(sdfs)                       # matrix: CpGs x samples

# DunedinPACE from the authors' package
library(DunedinPACE)
PACEProjector(betas)

Check detection p-values and the number of probes passing QC before you trust anything downstream. A sample with 8% failed probes will still produce a confident-looking clock output. Also run a leukocyte deconvolution (FlowSorted.Blood.EPIC) and record the cell fractions alongside the clock value: a shift in neutrophil-to-lymphocyte ratio from an infection two weeks before your draw will move first-generation clock outputs on its own.

Reading the number without fooling yourself

Three rules we hold to.

Compare yourself to yourself. Population percentiles depend on the training cohort’s demographics and assay platform. Your own four-point trend on a fixed assay is a cleaner instrument than your percentile against NHANES.

Treat the composite as a summary, not a finding. Biomarker signature work in long-lived families shows that people cluster into distinct patterns, and those patterns carry different morbidity and mortality risk even at similar ages 8. A PhenoAge that moved two years is a prompt to look at which analytes moved, not a result in itself. If it was driven by CRP alone, you measured an inflammatory episode.

Subjective sense of aging is not a calibrated instrument. Self-rated health maps onto measured biomarkers differently across socioeconomic groups, so “I feel fine” and “I feel old” carry different biological content depending on who is saying it 9. That is the reason to measure.

On responsiveness: methylation clocks do move with intervention. A small clinical study of therapeutic plasma exchange reported reductions in epigenetic age measures 10. We read that as evidence the assays are not inert, not as a reason to pursue any particular intervention. Interpretation of an out-of-range clinical analyte, and any decision that follows from it, belongs with a physician.

Where the reader needs a clinician

Anything outside the reference interval on a standard panel is a clinical question, not a data question: a creatinine trend up, ferritin high, ALP high, a plasma p-tau217 above assay threshold. Bring the numbers and the trend to a physician. We are describing measurement and interpretation of your own data, nothing more.

Questions people also ask

What is the biggest single indicator of longevity? Among routine blood analytes, serum albumin and markers of chronic inflammation (hs-CRP, white cell indices) carry disproportionate weight in mortality-trained models 2 3. No single value beats a multi-marker composite tracked over time.

What is the best blood test for aging markers? For value per dollar, a standard CMP plus CBC with differential plus hs-CRP, run into PhenoAge and KDM via the BioAge package 1. Add DunedinPACE annually if you want a rate-of-change measure rather than a level 4.

What are the four tests to see if you’re aging well? We would use: a blood-chemistry biological age composite, a methylation pace-of-aging measure, function tests you can do at home (grip strength, VO2max estimate, gait speed), and a fasting metabolic set (HbA1c, fasting insulin, ApoB). Functional and molecular measures track partially independent processes 7.

Is a genetic age test the same as a biological age test? No. Your genome does not change with age. Methylation clocks read chemical marks layered on the genome, which do change, and they require a different assay from whole-genome sequencing.

At what age do you age most rapidly? Rate varies by person and by organ system. DunedinPACE was built precisely to quantify that rate individually rather than assume a population curve 4.

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. Dayoon Kwon, Daniel W. Belsky. A toolkit for quantification of biological age from blood chemistry and organ function test data: BioAge. GeroScience, 2021. https://doi.org/10.1007/s11357-021-00480-5 ↩ ↩2

  2. Evgeny Putin, Polina Mamoshina, Alexander Aliper, et al. Deep biomarkers of human aging: Application of deep neural networks to biomarker development. Aging, 2016. https://doi.org/10.18632/aging.100968 ↩ ↩2

  3. Jordan Bortz, Andrea Guariglia, Lucija Klaric, et al. Biological age estimation using circulating blood biomarkers. Communications Biology, 2023. https://doi.org/10.1038/s42003-023-05456-z ↩ ↩2

  4. Daniel W Belsky, Avshalom Caspi, David L Corcoran, et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife, 2022. https://doi.org/10.7554/elife.73420 ↩ ↩2 ↩3

  5. Xiaqing Jiang, Tina D Hoang, Leslie M Shaw, et al. Alzheimer’s disease neuropathology plasma biomarkers and cognition in midlife: a community-based cohort study. The Lancet, 2026. https://doi.org/10.1016/s0140-6736(26)00515-5 ↩

  6. Anahita Amirpour, Lina Bergman, Gabriela Markovic, et al. Understanding neurocognitive recovery in older adults after total hip arthroplasty—neurocognitive assessment, blood biomarkers and patient experiences: a mixed-methods study. BMJ Open, 2025. https://doi.org/10.1136/bmjopen-2024-093872 ↩

  7. Karl-Heinz Wagner, David Cameron-Smith, Barbara Wessner, et al. Biomarkers of Aging: From Function to Molecular Biology. Nutrients, 2016. https://doi.org/10.3390/nu8060338 ↩ ↩2

  8. Paola Sebastiani, Bharat Thyagarajan, Fangui Sun, et al. Biomarker signatures of aging. Aging Cell, 2017. https://doi.org/10.1111/acel.12557 ↩

  9. Jennifer Beam Dowd, Anna Zajacova. Does Self-Rated Health Mean the Same Thing Across Socioeconomic Groups? Evidence From Biomarker Data. Annals of Epidemiology, 2010. https://doi.org/10.1016/j.annepidem.2010.06.007 ↩

  10. Daehwan Kim, Dobri D. Kiprov, Connor Luellen, et al. Old plasma dilution reduces human biological age: a clinical study. GeroScience, 2022. https://doi.org/10.1007/s11357-022-00645-w ↩