How Accurate Is Function Health's Biological Age?
Function Health’s biological age is a composite score derived from its blood panel, and its accuracy should be read as “directionally informative, poorly reproducible at the individual level.” The underlying biomarkers are real and mortality-associated. The single number is a weighted summary of inflammation, metabolic, renal, liver and hematology markers, most of which vary substantially within one person from week to week. So a result telling a 46-year-old that they are 36 is a low-precision estimate of a quantity that has no gold-standard reference value. The result is not a measurement error in the usual sense. Function does not publish the exact model or its coefficients, which means you cannot audit it directly. You can still reproduce the published clocks that models like it are built on, using the same lab values Function gives you.
What a blood-based biological age is computed from
Blood biological age scores almost all descend from two methods. The first is the Klemera-Doubal approach, which fits each biomarker against chronological age in a reference population and then inverts the regressions to estimate an age from a new person’s values. The second is PhenoAge, trained on mortality rather than on age itself. PhenoAge uses nine markers plus chronological age.
- albumin
- creatinine
- glucose
- log C-reactive protein
- lymphocyte percent
- mean cell volume
- red cell distribution width
- alkaline phosphatase
- white blood cell count
Two properties of that construction matter for interpretation. Chronological age is itself an input, so the output is anchored to your real age and can only deviate from it by so much. And the deviation is driven mostly by the two noisiest inputs, C-reactive protein and glucose. If a consumer panel reports a number well below your chronological age, the most likely explanation is a low CRP and a favorable metabolic profile on the morning you drew blood.
Why the number moves when nothing about you has changed
Precision here has two components. Analytical variability is the assay’s own imprecision, typically 2 to 5 percent coefficient of variation for routine chemistries on a modern analyzer. Within-subject biological variability is the day-to-day swing in your true value, and it is much larger for the inflammatory and metabolic markers. High-sensitivity CRP has within-person variation on the order of tens of percent, and a minor infection or a hard training block can move it several-fold. Fasting glucose runs roughly 5 to 7 percent within-person. Albumin and MCV are stable, closer to 2 to 4 percent.
Because CRP enters PhenoAge as a log term with a large coefficient, that instability propagates. Repeating the same panel two weeks apart with no change in your health can shift a blood-derived biological age by a few years in either direction. That is the precision envelope. Practically, this means a single result is not interpretable as a level. Three or four results across a year, with consistent fasting and consistent draw time, are interpretable as a trend.
There is also a population-fit problem. These models were trained largely on NHANES-style cohorts, which are older and less healthy on average than the people buying consumer panels. Model-based age estimates regress toward the training population’s mean, so healthy people in their forties are systematically told they are younger than they are. Recent work developing and validating methylation-based age and biomarker models on Japanese data makes the same point from the other direction: coefficients fitted in one population need revalidation before they are applied to another.1
How this compares with methylation clocks
Blood-chemistry clocks and DNA methylation clocks are different instruments measuring loosely related things. Methylation clocks read the fraction of methylated cytosines at specific CpG sites, and the first generation of them predicted chronological age with a median absolute error of a few years across tissues.2 Second-generation clocks trained on mortality and on pace-of-change outperform them for disease association. Epigenetic age acceleration has been linked to cardiovascular risk stratification beyond chronological age, and to cognitively healthy longevity in large prospective cohorts.34 Newer interpretable models map their CpGs onto specific aging pathways rather than returning an opaque scalar.5
Methylation clocks are not automatically more reliable. A careful decomposition of technical versus biological reliability found that many widely used clocks have poor test-retest performance, and that principal-component and pace-of-aging formulations are the ones stable enough to use for tracking an intervention in one person.6 Reviews of clinical application reach a similar conclusion, treating these measures as research instruments rather than diagnostics.7 So the choice turns on whether a given implementation has published within-person reliability, and most consumer scores do not. Blood versus methylation is not the choice that matters.
How to check the number yourself
You can recompute published clocks from the values Function already returns, which is the most useful audit available. Export the panel as CSV and convert units first, since the published coefficients are unit-specific. PhenoAge expects albumin in g/L, creatinine in µmol/L and glucose in mmol/L. It expects CRP in mg/dL before the log, alkaline phosphatase in U/L and MCV in fL. It also expects WBC in 1000 cells/µL, RDW in percent, and lymphocyte percent. US labs report albumin in g/dL, creatinine in mg/dL and glucose in mg/dL. Multiply those by 10, 88.4 and 0.0555 respectively.
For the computation itself, use biolearn (Python) which implements PhenoAge, Klemera-Doubal and the major methylation clocks against a common data model. In R, methylclock and dnaMethyAge cover the methylation side. If you have array data, process IDAT files with sesame rather than raw minfi defaults, and check probe coverage before trusting any clock output. Newer EPIC array versions dropped or replaced some of the CpGs used by the original clocks, and implementations silently impute the missing ones, which degrades the estimate in ways the output does not flag.
Then do the thing that matters more than the model choice. Plot each input biomarker as a time series and look at its own variability before you interpret the composite. If your biological age dropped four years and your CRP went from 2.1 to 0.4 mg/L, you have learned that your CRP moved. Environmental exposures also register in these measures. Ambient pollution has been associated with accelerated epigenetic aging, which is a reminder that these scores are partly reading your environment rather than an intrinsic rate.8
What we would track instead of one age number
We would keep the panel and discard the scalar. The individual trajectories carry the information. Track fasting insulin and HOMA-IR, hsCRP, ApoB and HbA1c alongside continuous glucose data. Track eGFR, ALT, and a full CBC with RDW as well. Each has a known reference interval and a known within-person variability, so a change is interpretable. A composite age collapses all of that into one figure whose confidence interval is never shown.
If you want an aging measure specifically, use a clock with published test-retest reliability and run it more than once.6 Tissue-specific and pathway-level clocks are where the field is moving, and they produce interpretable output rather than a single year count.95 Associations with depression severity and cognitive deficits illustrate both the promise and the confounding: these measures track many things at once.10
None of this is a diagnosis. If a component marker sits outside its reference interval, that is a finding for a physician to evaluate with your history in hand. This applies in particular to a persistent elevation in CRP, liver enzymes, or glucose. A biological age of 36 does not offset an abnormal individual result, and a biological age of 55 is not itself a condition.
Questions people also ask
How does Function Health calculate biological age? Function has not published the model or its coefficients, and describes the metric as derived from its biomarker panel. The method space is small, so it is very likely a Klemera-Doubal or PhenoAge-style weighted combination of inflammation, metabolic, renal, liver and hematology markers, anchored to chronological age.
Is a biological age 10 years below my real age meaningful? Treat it as weak evidence. Blood clocks systematically return younger estimates for healthier-than-average people because of regression toward the training population mean. Confirm with a repeat panel under the same conditions before concluding anything.
Is a methylation test more accurate? More accurate for chronological age, and better validated against disease outcomes.23 Reliability varies sharply by clock, and only the principal-component and pace-of-aging variants have demonstrated the within-person stability needed to track change in one individual.6
Can I lower my biological age? You can move the inputs. Whether that reflects a change in an underlying aging rate is unresolved, and current reviews treat these clocks as research tools rather than clinical endpoints.7
Footnotes
-
Tatsuma Shoji, Yui Tomo, Ryo Nakaki. Prediction of biological age and blood biomarkers from DNA methylation profiles measured by the methylation screening array: development and validation of models on Japanese data. Epigenetics Reports, 2026. https://doi.org/10.1080/28361512.2026.2715974 ↩
-
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
-
Ramzi Ibrahim, Balaji Tamarappoo, Kwan Lee, et al. Assessing Differences Between Epigenetic versus Chronologic Age for Cardiovascular Risk Assessments in the United States. Circulation: Population Health and Outcomes, 2026. https://doi.org/10.1161/circoutcomes.126.013600 ↩ ↩2
-
Andrea Z LaCroix, Bowei Zhang, Steve Nguyen, et al. Epigenetic Clocks of Biological Aging and Cognitively Healthy Longevity: The Women’s Health Initiative Memory Study. The Journals of Gerontology, Series A: Biological Sciences and Medical Sciences, 2026. https://doi.org/10.1093/gerona/glag215 ↩
-
Rajarshi Mandal, Ning Xie, Gil Alterovitz. Interpretable epigenetic clock links aging pathways and disease-specific methylation profiles. npj Aging, 2026. https://doi.org/10.1038/s41514-026-00476-5 ↩ ↩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 ↩ ↩2 ↩3
-
A. Baki Yildirim, Duygu T. Yildirim, Hilal Akalin, et al. Epigenetic Clocks: Biological Aging, Disease, and Clinical Applications. The EuroBiotech Journal, 2026. https://doi.org/10.2478/ebtj-2026-0014 ↩ ↩2
-
Ya-Qian Xu, Linghui Cai, Yaowen Nuo, et al. How Ambient Pollutants Accelerate Aging: Bridging Epigenetics, Age-Stratified Effects, and Public Health Interventions. Environment & Health, 2025. https://doi.org/10.1021/envhealth.5c00542 ↩
-
Jinxiao Liao, Zhongting Huang, Yuxin Liu, et al. An epithelial-derived epigenetic clock reveals pathway-level aging mechanisms in chronic lung disease. npj Aging, 2026. https://doi.org/10.1038/s41514-026-00487-2 ↩
-
Epigenetic Clocks Identify Harmful Epigenetic Aging Linked to Depression Severity and Cognitive Deficits in Major Depressive Disorder. aging and disease, 2025. https://doi.org/10.14336/ad.2025.0781 ↩
Pothos builds longitudinal molecular profiles of individuals: whole-genome sequencing, single-cell RNA sequencing, and plasma proteomics, delivered as a detailed report, every data file, and stored cells. Build your profile.