What Do Biological-Age Apps Actually Measure?

Biological-age apps measure a model-defined pattern, then express that pattern on an age-like scale. Depending on the product, the inputs may be DNA methylation, blood chemistry, organ-function tests, body composition, fitness, movement, sleep, or heart signals. Because the input set and target outcome differ, two “biological ages” can be mathematically correct for their own models and still disagree for the same person.

Four common kinds of biological-age model

1. Chronological-age clocks

These models learn which biological measurements best predict calendar age. Horvath’s multi-tissue DNA-methylation clock is a well-known example. A gap between predicted and calendar age may be studied as age acceleration, but the original prediction task remains chronological age.

2. Risk- or outcome-oriented ages

These models use biomarkers related to mortality, disease, or functional outcomes and then translate estimated risk back to an age scale. Phenotypic Age and related clinical-biomarker methods are examples. They are not measuring the same construct as a pure calendar-age clock.

3. Pace-of-aging measures

A pace measure tries to represent how quickly multi-system function is changing, often around a reference value of one biological year per calendar year. That is conceptually different from asking, “What age does my current profile resemble?”

4. Wearable or functional estimates

These models use signals such as activity patterns, cardiorespiratory fitness, heart-rate features, sleep, gait, or body composition. A 2021 study, for example, trained a model on wearable activity patterns and evaluated its biological-age-acceleration output in large observational datasets. These estimates can update frequently but inherit the limits of both the sensors and the population used to build the model.

What the number is—and is not

• It is a weighted summary of selected inputs under a particular algorithm.
• It may be trained to resemble age, risk, function, or longitudinal decline.
• It is not a direct count of cellular damage or years of life remaining.
• It does not automatically describe every organ system.
• It cannot be compared fairly with another model until you know the input types, target, scale, and population.

What the evidence says

Klemera and Doubal began from the problem that biological age has no exact universal definition. Later work compared multibiomarker algorithms by their association with mortality, while DNA clocks modeled methylation patterns and wearable models learned activity-related patterns. Together, this research supports biological age as a family of estimates, not one interchangeable measurement.

How Metric estimates biological age

Metric uses authorized Apple Health data rather than a blood or saliva test. It averages eligible values over rolling windows, converts each available contributor into an age-equivalent estimate, and combines the deviations from chronological age. Current model weights are: VO2 max 18%, resting heart rate 14%, HRV 10%, body-fat percentage 10%, sleep duration 10%, daily steps 9%, lean-mass ratio 9% when body fat is unavailable, BMI 5%, active-energy context 4%, sleep regularity 4%, systolic blood pressure 4%, and diastolic blood pressure 3%.

Missing or stale contributors are not replaced with a made-up measurement; their contribution stays neutral relative to chronological age. Metric also shows data freshness and coverage because a broad estimate with recent inputs should be interpreted differently from one built from sparse data.

Questions to ask any biological-age app

• Which exact inputs does the model use, and which are missing for me?
• What was the model trained or calibrated to predict?
• How are chronological age and biological sex used?
• How much history is averaged, and how fresh must each input be?
• What happens when a metric is absent or implausible?
• Can I see contributors, weights, uncertainty, and algorithm-update dates?
• Is the output an age, an age gap, an age ratio, or a pace?

Limitations

• A model can omit important domains and still produce a precise-looking number.
• Population averages may not fit an individual’s physiology, training status, medication use, disability, or device context.
• Frequent wearable updates can reflect sensor variation or behavior change rather than a durable change in aging biology.
• Associations with future outcomes do not prove that moving the score itself changes those outcomes.
• Different model families should not be ranked by whose output is youngest.

Sources and further reading

• Mechanisms of Ageing and Development (2006): A new approach to the concept and computation of biological ageExplains why biological age lacks a single exact definition and proposes a multibiomarker estimation method.
• Genome Biology (2013): DNA methylation age of human tissues and cell typesIntroduced a multi-tissue DNA-methylation age estimator, illustrating a molecular-clock approach to biological age.
• Aging (2018): An epigenetic biomarker of aging for lifespan and healthspanDeveloped DNAm PhenoAge from a mortality-related clinical phenotype and evaluated associations with aging outcomes.
• The Journals of Gerontology: Series A (2013): Modeling the rate of senescence: can estimated biological age predict mortality more accurately than chronological age?Compared five biological-age algorithms in NHANES III and their associations with mortality.
• Aging (2021): Deep longitudinal phenotyping of wearable sensor data reveals independent markers of longevity, stress, and resilienceDeveloped a wearable-activity-based biological-age-acceleration model and evaluated it in large observational datasets.
• Apple Developer Documentation (2026): HealthKit data typesDefines HealthKit records for heart rate, resting heart rate, HRV (SDNN), sleep, workouts, nutrition, oxygen saturation, respiratory rate, and other categories.

Continue exploring

• Biological age app
• Biological age versus chronological age
• How accurate are biological-age estimates?
• Metric methodology