Five of fourteen consensus aging biomarkers predicted death in a 1,083-person cohort

Any drug proposed to slow human aging faces an arithmetic problem before it faces a scientific one. The outcome that matters — living longer, or living well for longer — takes decades to observe, which is longer than a trial, a patent, or usually a company.

The proposed workaround is a biomarker: a measurement that changes within a trial's lifespan and that stands in for the outcome nobody can wait for. Epigenetic clocks, which estimate biological age from DNA methylation patterns, are the leading candidates. Whether they can carry that weight is an open question, and 2026 produced several useful pieces of the answer.

What a surrogate endpoint has to do

To substitute for mortality, a biomarker needs two properties, and they are separate.

It must predict the outcome — people with worse readings must actually die sooner. And it must respond to intervention in a way that tracks the outcome — a treatment that improves the marker must also improve survival.

The first is testable in a cohort study. The second requires an intervention trial with long enough follow-up to observe the outcome — which is the problem the surrogate was introduced to avoid in the first place.

The ranking

A study published in March in Biomarker Research tested the first property directly, and did it against a defined slate rather than a favourite.

Researchers used the Berlin Aging Study II — 1,083 participants aged 60 to 80 at baseline, followed for an average of 7.4 ± 1.5 years, with a range of 3.9 to 10.4 years — to compare 14 biomarkers of aging that an expert panel had recently agreed on for use as outcome measures in intervention studies [s1].

The slate spanned four categories. The physiological markers were IGF-1 and DNA-methylation-derived GDF15; the inflammatory markers were high-sensitivity CRP and IL-6; the functional measures included muscle mass, muscle strength, hand grip strength, Timed-Up-and-Go, gait speed, standing balance, the frailty phenotype, cognitive health and blood pressure; and the epigenetic measures were an epigenetic clock and DunedinPACE [s1] — the latter a measure of the pace of aging rather than of biological age relative to chronological age [s2].

In models of all-cause mortality adjusted for age, sex, lifestyle factors and genetic ancestry, five predicted mortality significantly: hand grip strength, IL-6, standing balance, cognitive health, and DunedinPACE — with DunedinPACE the strongest [s1].

Eight did not. CRP, gait speed, IGF-1, blood pressure, muscle mass, DNAmGDF15, the frailty phenotype and Timed-Up-and-Go showed no association with mortality in this cohort [s1]. The results were corroborated in subgroup analyses stratified by cause of death [s1].

The result underneath the result

Feature selection identified a minimal set — muscle mass, standing balance and DunedinPACE — that predicted mortality with nearly the same discriminative accuracy as the full 14-biomarker model: a C-index of 0.63 against 0.65 [s1].

Two things about that deserve attention. First, muscle mass appears in the minimal set despite showing no independent association with mortality, which is a normal feature of feature selection and a normal source of confusion when such sets are quoted out of context.

Second, and more important: a C-index of 0.65 is modest. It means the full panel of consensus aging biomarkers, in a well-characterised cohort, correctly ranks which of two randomly chosen people dies first about 65% of the time. That is better than a coin flip and a long way from a precise instrument. Any trial powered on the assumption that these markers separate individuals cleanly is assuming more than this cohort supports.

What a clock does over four weeks

The second property — responsiveness — got a small, honest test in July.

A pilot feasibility study published in GeroScience enrolled 34 adults aged 48 to 81 with metabolic syndrome, who consumed one ounce of tree nuts and two tablespoons of extra virgin olive oil daily for four weeks [s2]. The primary objectives were feasibility, adherence and participant acceptability of epigenetic aging assessment, with DunedinPACE and AgeAccelGrim as exploratory outcomes [s2].

The feasibility findings were positive: adherence exceeded 95%, most participants said they would join a similar longer-term trial, and participants expressed strong interest in learning their biological age, indicating that evidence of slowed aging would motivate sustained dietary change [s2]. At baseline, all participants showed a faster-than-average pace of aging by DunedinPACE, supporting metabolic syndrome as a target population for this kind of research [s2].

No significant changes in epigenetic aging were observed over the four-week intervention [s2].

That null is the useful part. It is exactly what a well-behaved surrogate should do when a mild dietary change runs for a month — nothing. A clock that moved would raise the question of what it was actually responding to. The authors' framing is that biological aging measures may serve not only as surrogate outcomes but also as tools to support participant engagement [s2], which is a notably modest claim about the endpoint role.

The infrastructure question

The field is aware it is running ahead of its validation. A recommendations paper published in December 2025 in npj Aging argues that biomarkers of aging have the potential to transform geroscience clinical trials by stratifying participants, prioritising interventions and monitoring responses to geroprotectors — and then proposes standard practices for how longevity biotechnology companies collect that data [s3].

The rationale is that consistent collection would let the field support parallel and ongoing validation and benchmarking efforts, and would allow valuable clinical data to be reused through pre-competitive alignment on shared tools [s3]. In June, Nature Aging published a call for an open competition for biomarkers of aging [s4] — a further sign that head-to-head benchmarking is being treated as unfinished business rather than settled science.

How to read a biological age result

When a study, product or trial reports that an intervention reduced biological age, three questions apply.

Which clock, and has it been benchmarked against mortality in an independent cohort? The 14 candidates in BASE-II did not perform equivalently [s1].

Over what timescale, and is a change of that size distinguishable from measurement noise?

And has the clock been shown to respond to an intervention in a way that predicted that intervention's effect on survival? That is the question a surrogate endpoint ultimately has to answer, and it is the one the field's own benchmarking and validation efforts are still set up to address [s3][s4].