The arithmetic behind one number.
BIOS Age is the sum of five terms, each a published effect converted into years. This page is the complete method as it runs on your iPhone — every constant, every cap, every study.
Three steps, no hidden weighting.
Every constant comes from a named study you can look up. A 10 % change in long-term risk is worth about one year of age, so a published risk ratio becomes a number of years directly — the effective-age method (Spiegelhalter 2016).
Steps, resting heart rate, cardio fitness, sleep duration and strength minutes — taken as medians, so one odd day can’t swing the result.
Each study reports how much a habit changes long-term risk. That ratio goes through the converter above and comes out as years — positive adds, negative removes.
The five terms are summed, clamped to ±10 years, tapered for age, and shown as the rows on your Age screen — which always add up to the headline.
Five terms. Five studies. Every constant shown.
For each term: what is measured, how it maps to years, where the mapping stops, and what the same input reads at a few reference points.
Daily movement
Saint-Maurice et al., JAMA 2020 ↗- 3,500
- +5.9
- 5,000
- +3.8
- 7,000
- +1.2
- 8,000
- 0
- 10,000+
- −2.0
28-day median of Watch-worn daily step totals, at least 14 valid days.
Ln-linear interpolation through the study’s step ladder — hazard 1.51 at 2,000 steps, 1.00 at 4,000, 0.68 at 6,000, 0.49 at 8,000, 0.40 at 10,000, 0.34 at 16,000 — with zero at 8,000 steps a day (7,000 from age 60).
- Penalty side attenuated ×0.7: people who are ill walk less, and the low end is inflated by reverse causation.
- Halved when a measured VO₂max is present — the two overlap.
- Capped at +6.0 and −2.0 years.
Resting heart rate
Zhang et al., CMAJ 2016 ↗- 45
- −1.0
- 55
- −0.1
- 60
- +0.4
- 70
- +1.3
- 96+
- +3.6
28-day resting heart rate from the wrist, recentred +4 bpm (wrist readings run low), clamped to 45–100.
Relative risk 1.09 per +10 bpm (CI 1.07–1.12), linear from 45 bpm. The conservative end of the published range, because fitness already carries part of this signal.
- No penalty at the low end — the study’s spline is flat there.
- Both directions: a low resting rate removes years.
Cardio fitness
Mandsager et al., JAMA Netw Open 2018 ↗- < 25th
- +5.0
- 25–49th
- +2.0
- 50–74th
- 0
- 75–97th
- −2.9
- ≥ 98th
- −4.4
Measured VO₂max only — the median of the latest six samples within 120 days, at least four required. Apple’s estimate never enters.
Your percentile band for age and sex, mapped through the study’s hazard by band, then shrunk by λ = 0.60 for device error (Watch MAE 6.9 ml/kg/min). Interpolated inside bands.
- Capped at ±5 years.
- No penalty for being fitter than the top band.
- Changes under 3.5 ml/kg/min are not shown — they are inside the device’s error.
Sleep duration
Windred et al., SLEEP 2024 ↗- 7 h 00
- 0
- 6 h 15
- +0.4
- 6 h 00
- +0.8
- ≤ 5 h 24
- +2.0
28-night mean of watch-recorded time asleep, at least 14 valid nights of 3 hours or more. iPhone-only nights are missing, not zero.
Flat at zero from 6.5 hours. A ramp to +0.76 years at 6.0 hours (hazard 1.075), then a steeper ramp that reaches the +2.0 cap at 5.4 hours.
- Penalty only. Long sleep costs nothing — the published long-sleep hazard came from self-report, not from watches.
- Four weeks or more above 9.5 hours raises an unscored note.
Strength training
Momma et al., Br J Sports Med 2022 ↗- 0 min
- 0
- 20 min
- −1.0
- 40+ min
- −1.9
Weekly minutes of logged strength-type workouts.
−1.9 years × min(1, minutes ÷ 40): a straight line to the study’s nadir (RR 0.83 at about 40 minutes a week), then flat.
- Never positive. The J-curve’s upper arm cannot be told apart from other training in Watch data, so lifting is never charged.
Rules of the reading.
The parts of Core 2 that are not a study — the guards that keep one number honest.
Behavioural terms use 28 days and recompute weekly, so consecutive readings share 21 of 28 days. Cardio fitness looks back 120 days — fitness moves slowly, and a shorter window made the term blink out when samples aged.
Nothing before 21 covered days and three active terms, one of them resting heart rate or a measured VO₂max. Until then the app says “Calibrating — N of M”.
A signal that is not measured is left out of the sum — never counted as zero. The rows on the Age screen say “not measured”.
The sum is clamped to ±10 years before it meets your calendar age. No profile, however extreme, reads more than ten years either way.
Most of the evidence comes from people over 40. From 40 the sum counts in full; at 30 it counts half, rising to full by 40; between 25 and 30 a quarter to a half. BIOS Age is not offered under 18.
Your watch’s own measurement error is propagated through the terms and printed beside the number — the ± years. It is why the headline is rounded to whole years, and why no band is quoted without a measured VO₂max.
When a term switches on or off between weeks — the first measured VO₂max, say — the history marks it, so a step in the line is never mistaken for a change in habit.
The converter’s constant, taken across a doubling time of 7 to 10 years, rescales every offset by ±18 %. The steps attenuation (0.7) against the wrist-cohort gradient is the other known lever. Both are stated, not hidden.
Specification.
- Inputs
- 5 terms — steps, resting heart rate, measured VO₂max, sleep duration, strength minutes
- Evidence
- 6 peer-reviewed papers — JAMA, CMAJ, JAMA Network Open, SLEEP, BJSM, BMC Med Inform
- Method
- Effective age (Spiegelhalter 2016): each hazard ratio becomes years via ln(HR) / ln(1.1)
- Window
- 28 days, median-based · recomputed weekly
- Resolution
- 0.5 years · sum clamped to ±10 · error band stated
- First reading
- After 21 covered days and 3 active terms — not before
- Runs on
- Your iPhone · network requests: 0
What Core 2 deliberately leaves out.
Each of these would make the number look more precise. None would make it more true.
Apple’s estimate is built from resting heart rate and activity — inputs already in the sum. Counting it would count them twice. Only a measured VO₂max enters.
Deep-sleep detection on a wrist runs at roughly 50 % sensitivity. Stages are shown, never scored.
The published hazard for long sleep comes from self-report. Device-measured long sleep is a different thing, so it costs nothing.
People who are ill walk less. The step penalty is attenuated ×0.7 so illness is not read as a habit.
Most of the evidence comes from people over 40. Younger users get a tapered reading rather than a confident wrong one.
Your first BIOS Age waits for 21 covered days and three active terms. A number built on less would be a guess wearing a decimal point.
Two worked profiles.
Computed by the shipping model and checked against an independent oracle to ±0.01 years — the same characterization test that guards every release.
| Profile | Steps | Resting HR | VO₂max | Sleep | Strength | Sum | Taper | BIOS Age |
|---|---|---|---|---|---|---|---|---|
Fit, 45 9,000 steps · 52 bpm · high band · 7 h 30 · 60 min a week | −0.5 | −0.4 | −2.9 | 0 | −1.9 | −5.7 | ×1.0 | 39(45) |
Sedentary, 55 3,500 steps · 74 bpm · low band · 6 h 05 · none | +3.0* | +1.6 | +5.0 | +0.6 | 0 | +10.0 | ×1.0 | 65(55) |
* halved: a measured VO₂max is present. The sum in the second profile is the clamp, not the raw total (+10.2).
Why “second generation”.
Core 1 blended estimated fitness, fabricated intermediate curves and two derived ages into one weighting. Core 2 retired all of it: only a measured VO₂max enters, every term names its study, every constant is in this page, and the history carries a labelled break where the method changed. On the same day, for the same person, Core 1 read 48; Core 2 read 34.
Sources.
The six papers behind the five terms and the converter. Each link resolves at the publisher.
- Daily movementSaint-Maurice PF, Troiano RP, Bassett DR Jr, et al. Association of Daily Step Count and Step Intensity With Mortality Among US Adults. JAMA 2020;323(12):1151–1160.doi.org/10.1001/jama.2020.1382 ↗
- Resting heart rateZhang D, Shen X, Qi X. Resting heart rate and all-cause and cardiovascular mortality in the general population: a meta-analysis. CMAJ 2016;188(3):E53–E63.doi.org/10.1503/cmaj.150535 ↗
- Cardio fitnessMandsager K, Harb S, Cremer P, et al. Association of Cardiorespiratory Fitness With Long-term Mortality Among Adults Undergoing Exercise Treadmill Testing. JAMA Network Open 2018;1(6):e183605.doi.org/10.1001/jamanetworkopen.2018.3605 ↗
- Sleep durationWindred DP, Burns AC, Lane JM, et al. Sleep regularity is a stronger predictor of mortality risk than sleep duration: A prospective cohort study. SLEEP 2024;47(1):zsad253.doi.org/10.1093/sleep/zsad253 ↗
- Strength trainingMomma H, Kawakami R, Honda T, Sawada SS. Muscle-strengthening activities are associated with lower risk and mortality in major non-communicable diseases: a systematic review and meta-analysis of cohort studies. British Journal of Sports Medicine 2022;56(13):755–763.doi.org/10.1136/bjsports-2021-105061 ↗
- The methodSpiegelhalter D. How old are you, really? Communicating chronic risk through ‘effective age’ of your body and organs. BMC Medical Informatics and Decision Making 2016;16:104.doi.org/10.1186/s12911-016-0342-z ↗
Open method, closed data.
Everything on this page runs on your iPhone. The app makes no network requests, and the arithmetic is the only thing that leaves it.