Paper 026: When Clocks Disagree — Interpreting Biological Age Measurements in the Presence of Model Variability

Position paper. Hypothesis, not conclusion. This paper does not present product behavior as scientific validation, nor does it present hypotheses as established biological mechanisms.

Biological aging clocks have rapidly evolved from research tools into technologies increasingly available to clinicians, longevity companies, and individuals. They offer an appealing promise: a numerical estimate of biological aging that may provide insight beyond chronological age.

At the same time, the field recognizes an important characteristic of these models. Different biological aging clocks can produce substantially different estimates for the same individual.

This is not evidence that biological clocks are fundamentally flawed. Rather, it reflects differences in their design, objectives, training data, and biological assumptions.

As biological age measurements become more accessible outside research settings, a practical question emerges:

How should an individual interpret a biological age measurement when different validated models may disagree?

This paper explores that question. We propose that continuity—the preservation of personal context across measurements over time—may improve interpretation without attempting to resolve disagreements between biological clocks themselves.

In July 2026, researchers at Tally Health published an analysis illustrating a phenomenon that has long been recognized within the aging clock community but is less widely appreciated outside it.

Applying eight blood-trained epigenetic clocks to the same publicly available whole blood dataset, they found that biological age estimates frequently differed substantially for identical samples.

Across the dataset:

• the average difference between the youngest and oldest prediction for the same sample was 17 years;
• the smallest observed difference was 4 years;
• the largest reached 45 years.

The biological sample remained unchanged.

Only the analytical model changed.

The authors concluded that clock variability has important implications for interpretation, intervention studies, personalized monitoring, and the broader concept of biological age itself.

Rather than viewing this finding as a limitation of the field, we view it as an invitation to improve how biological age measurements are interpreted.

Biological clocks are designed to answer different scientific questions.

Some estimate chronological age from DNA methylation patterns.

Others are optimized to predict mortality risk, disease burden, healthspan, functional decline, or the pace of aging.

They differ in:

• training populations;
• statistical methodology;
• molecular features;
• biological endpoints;
• tissue specificity;
• intended clinical application.

Consequently, clocks that are each scientifically valid may legitimately produce different estimates for the same individual because they are not measuring precisely the same biological construct.

A Horvath clock, a GrimAge model, a PhenoAge model, and DunedinPACE each describe aging from different perspectives.

Agreement should therefore not be assumed.

Variability is an expected property of models trained for different purposes.

Modern biological clocks can generate highly reproducible numerical outputs.

However, numerical precision and biological interpretation are not identical concepts.

A model may consistently estimate a biological age of 56 years.

Another equally validated model may estimate 63 years.

Both estimates may be internally consistent.

The remaining question is not whether the algorithms function correctly.

The question is:

What does either number actually mean for the individual sitting in front of the result?

This distinction becomes increasingly important as biological age testing moves beyond research laboratories into routine personal use.

A measurement can be technically precise while still requiring careful interpretation.

Most discussions surrounding biological clocks focus on improving measurement.

Better biomarkers.

Better algorithms.

Better prediction.

We propose that an additional challenge exists.

Interpretation.

Biological age should not be viewed solely as an isolated numerical output.

It should also be understood as a measurement occurring within an individual’s biological timeline.

Like many laboratory values in medicine, biological age gains meaning from its surrounding context.

Questions naturally arise:

• Was the individual recovering from illness?
• Had medication recently changed?
• Were multiple measurements obtained over time?
• Was the sample collected under similar conditions?
• Which biological clock generated the estimate?
• How does today’s measurement compare with previous measurements from the same individual?

The biological clock does not answer these questions.

Yet they may influence how its output is interpreted.

We propose continuity as an interpretive framework rather than a measurement framework.

Continuity does not attempt to replace biological clocks.

It does not determine which clock is correct.

Instead, it preserves the contextual information surrounding measurements so they can be interpreted within a broader longitudinal record.

This includes preserving:

• measurement history;
• sampling conditions;
• biological baseline;
• timing;
• relevant health events;
• medications;
• travel or environmental changes where appropriate;
• the specific clock used for analysis.

Viewed individually, a biological age measurement represents a snapshot.

Viewed longitudinally, it becomes part of a trajectory.

Trajectory may provide information that isolated measurements cannot.

A persistent change observed across multiple measurements may deserve different interpretation than a single unexpected result.

Continuity therefore complements measurement rather than competing with it.

Medicine rarely relies on isolated measurements alone.

Blood pressure is interpreted differently depending on repeated observations, stress, medication use, illness, and measurement conditions.

Inflammatory markers are interpreted alongside symptoms.

Laboratory values are routinely evaluated relative to previous results.

Clinical interpretation depends not only on numbers but also on their context.

We suggest that biological age measurements may ultimately benefit from similar longitudinal interpretation.

Whether this proves clinically valuable remains an open research question.

It is equally important to define what this framework does not propose.

Continuity does not:

• identify the most accurate biological clock;
• resolve disagreement between existing models;
• establish biological truth;
• replace clinical judgement;
• validate biological age as a definitive measure of health.

Our proposal is considerably more modest.

We suggest that preserving contextual information surrounding biological measurements may improve how those measurements are interpreted over time.

This remains a hypothesis requiring future study.

• Which contextual variables most influence interpretation of biological age measurements?
• Can longitudinal tracking distinguish model variability from genuine biological change?
• How should disagreement between clocks be communicated to patients?
• Should biological age reports routinely include methodological context alongside numerical estimates?
• What role should clinicians play in interpreting biological age outside research settings?
• Can continuity-based interpretation improve clinical usefulness without altering the underlying biological clocks?

We believe these questions deserve collaborative investigation across aging biology, clinical medicine, digital health, and computational modeling.

Recent work by Tally Health provides a clear public demonstration of a property long recognized within the biological aging field: different validated clocks can produce substantially different biological age estimates for the same individual.

Rather than interpreting this variability as a weakness, we believe it highlights a different challenge.

As biological age measurements become increasingly available, improving interpretation may become as important as improving measurement.

We propose that continuity—the preservation of personal context across measurements and over time—deserves consideration as part of that interpretive process.

Continuity does not determine which biological clock is correct.

It does not eliminate uncertainty.

It does not replace biological measurement.

Instead, it may provide the missing information required to understand biological age as part of an individual’s evolving health history rather than as an isolated numerical output.

This is a hypothesis, not a conclusion.

We hope it encourages further discussion among researchers working at the intersection of biological aging, clinical medicine, and digital health.

Tally Health. (2026). Epigenetic clock variability analysis. July 2026.

Steve Horvath. (2013). DNA methylation age of human tissues and cell types. Genome Biology, 14, R115.

Eric Topol & Tony Wyss-Coray. (2026). Biological Aging Clocks in Health and Disease. Nature Medicine.

ZKOMI Research Team. (2026). Paper 022: From Measurement to Context.