All analyses
5 min read

Last updated:

Biological age: the real value of the measures, and their blind spots

A client arrives with a number: 47 years of biological age for 52 years of civil age. Before doing anything with it, you need to know where that number comes from, what it was trained to predict, and how far it moves if the test is repeated the next day. The answers vary widely by family of measure.

10 studies citedsee the sources

Benjamin Sant

Co-founder of the School of Longevity

Practitioner comparing two functional assessments of the same person
Practitioner comparing two functional assessments of the same person

Three families of measures that do not measure the same thing

The term biological age covers methods that have nothing to do with one another. A 2017 review lists four classic statistical approaches: multiple linear regression, principal component analysis, the Hochschild method, and the Klemera and Doubal method. What separates them is not the calculation but the role given to real age. In the first two, real age serves as the selection criterion for markers. In the other two, it becomes a variable in its own right inside the model.

That detail decides everything. The founding paper by Klemera and Doubal, in 2006, shows that a large number of estimates published with multiple linear regression are unusable, because civil age predicts better than the model meant to outperform it. A number presented as a biological age can therefore be less informative than the date of birth.

Alongside these methods drawn from blood biology, two other families coexist. The first relies on DNA methylation: these are the epigenetic clocks, from DNAm PhenoAge in 2018 to GrimAge in 2019 and DunedinPACE in 2022. The second looks at no molecule at all and simply measures what the person does with their body, with walking speed and grip strength at the top of the list.

A biological age is only worth its predictive value

The right question is not whether the number looks correct, but whether it predicts something better than real age. A 2013 study compared five algorithms in 9,389 people aged 30 to 75 from the NHANES III survey, followed for eighteen years, with 1,843 deaths recorded.

The Klemera and Doubal method came out ahead. It predicts mortality better than chronological age, and above all, when both are placed in the same model, chronological age stops being associated with mortality. That is the only argument that counts: the measure carries information the date of birth does not contain.

Second generation clocks were built on the same requirement. DNAm PhenoAge was trained on composite clinical measures rather than on age, and predicts all cause mortality, cancers, physical function and Alzheimer's disease. GrimAge was presented in 2019 as a strong predictor of lifespan and healthspan. DunedinPACE, published in 2022, does not give an age but a pace of aging, modeled on the decline of nineteen organ integrity indicators tracked at four points over two decades in the New Zealand Dunedin cohort.

Measurement noise often exceeds the gap being looked for

A paper published in Nature Aging in 2022 ran six of the most cited epigenetic clocks on duplicate samples. The same draw, analyzed twice, produced gaps of up to nine years.

Nine years is more than the gap a practitioner hopes to see after a year of coaching. In other words, part of the variation announced to clients comes from the technical noise of the array, not from their biology. The authors propose a principal component correction that brings agreement between replicates under one and a half years for most samples.

That correction is not applied by default in consumer offerings. Before comparing two results from the same person, you therefore need to know whether the laboratory uses a corrected version, and refuse to comment on a gap when you do not know whether it exceeds measurement error.

Clocks do not behave the same way against the same criterion

A study published in Neurology in 2022 tested three generations of algorithms against cognitive impairment in 649 ADNI cohort participants, with replication in the offspring of the Framingham cohort.

The result separates the generations clearly. First generation clocks, Horvath and Hannum, and second generation clocks, PhenoAge and GrimAge, were not consistently associated with measures of cognitive impairment. DunedinPACE, from the third generation, was, including for the clinical diagnosis of Alzheimer's disease.

So clocks cannot be discussed as one block. A measure validated on mortality is not automatically validated on the brain, and two tests sold under the same name of biological age can answer different questions. Asking on which criterion the algorithm was trained is the first question to put to a provider.

Two free measures already predict a great deal

A pooled analysis published in the JAMA in 2011 gathered nine cohorts and 34,485 people aged 65 and over, followed for six to twenty one years, with 17,528 deaths. Each 0.1 m/s step in walking speed was associated with a hazard ratio of 0.88 for mortality.

The magnitude is striking. At 75, predicted ten year survival ranged from 19% to 87% in men and from 35% to 91% in women on walking speed alone. Predicting from age, sex and walking speed did as well as predicting from age, sex, chronic disease, smoking, blood pressure, body mass index and hospitalizations.

Grip strength follows the same logic, with one useful limit to know. In a cohort of 861 people aged 61 to 74 followed for a median of 17.3 years, strength relative to weight was associated with mortality, hazard ratio 0.66 between extreme tertiles. But adding it to a classic cardiovascular risk model did not improve discrimination, with Harrell's C moving by only 0.0034. A marker can be associated with risk without adding anything to what is already known.

What a professional can reasonably do today

Announcing a biological age to a client commits the person who says it. As long as the algorithm, its training criterion and its reproducibility are unknown, the number remains a curiosity, not a basis for decision. It replaces neither a medical opinion nor a blood panel interpreted rigorously.

The defensible use is the opposite of what commercial offerings do with it. You do not start from the number to look for something to correct. You start from levers already identified, you choose reproducible and inexpensive indicators, walking speed, grip strength, cardiorespiratory capacity, and you measure again at a fixed interval with the same protocol.

A person who arrives with their test keeps the right to talk about it. The role of the professional is then to explain where the number comes from, what margin of error surrounds it, and to bring the conversation back to what can be measured without a laboratory and moves with training.

Sources

  1. 1.Klemera P, et al. A new approach to the concept and computation of biological age. Mech Ageing Dev. 2006.
  2. 2.Levine ME. Modeling the rate of senescence: can estimated biological age predict mortality more accurately than chronological age? J Gerontol A Biol Sci Med Sci. 2013.
  3. 3.Jia L, et al. Common methods of biological age estimation. Clin Interv Aging. 2017.
  4. 4.Levine ME, et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging (Albany NY). 2018.
  5. 5.Lu AT, et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging (Albany NY). 2019.
  6. 6.Belsky DW, et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. Elife. 2022.
  7. 7.Sugden K, et al. Association of Pace of Aging Measured by Blood-Based DNA Methylation With Age-Related Cognitive Impairment and Dementia. Neurology. 2022.
  8. 8.Higgins-Chen AT, et al. A computational solution for bolstering reliability of epigenetic clocks: Implications for clinical trials and longitudinal tracking. Nat Aging. 2022.
  9. 9.Studenski S, et al. Gait speed and survival in older adults. JAMA. 2011.
  10. 10.Laukkanen JA, et al. Handgrip strength is inversely associated with fatal cardiovascular and all-cause mortality events. Ann Med. 2020.

Reading studies is one thing, turning them into a plan is another

The Longevity Practitioner certification teaches the method that turns this kind of data into an accompaniment decision: assess, prioritise, plan over 90 days, re-measure, and know when to hand over.

See the certification

11 modules, €2,200. Enrol

Benjamin Sant

Co-founder of the School of Longevity

Cofounder of the School of Longevity with Tristan Maruyama. A former officer trained at Saint-Cyr from 2019 to 2022 and a coach since 2018, he has supported more than 4,000 people. He leads the School's editorial content with one constant principle: pass on only what can be situated, verified and applied seriously.

This article is published for information and education. It does not replace medical advice. For any personal health question, consult a healthcare professional.