What is Health Age?
Health Age sits within a new Longevity section of Apple's redesigned Health app. It compares your health metrics with what might be expected for your chronological age, using VO₂ max, resting heart rate, sleep and HRV collected through Apple Watch.
Lab data can also be incorporated, with A1c and LDL explicitly mentioned, alongside a broader panel available through Apple's partnership with Quest Diagnostics.
Apple describes the feature as being "informed by the best available science". What it hasn't published, at least yet, is the interesting bit: the weights, the formula, the underlying reference population, or validation numbers for the final consumer-facing Health Age itself.
What science might sit behind it?
The most relevant published work from Apple researchers is probably PpgAge, a wearable-based ageing clock developed using data from the Apple Heart and Movement Study.
It is important to be clear that PpgAge is not necessarily the algorithm behind Health Age. It is a substantially different model. PpgAge uses characteristics of the wrist photoplethysmography, or PPG, waveform itself to predict age, whereas Health Age appears to combine higher-level physiological measures such as VO₂ max, resting heart rate, sleep and HRV, with optional blood biomarkers.
There is an interesting extra layer here, though. Several of those higher-level measures ultimately depend on sensor-derived estimates themselves. VO₂ max, for example, is not being directly measured by the Watch in the way it would be in a laboratory metabolic test. It is inferred from a combination of signals including heart rate, which itself comes largely from PPG, alongside pace, movement and exercise intensity.
So although Health Age may operate on higher-level physiological variables rather than directly on the raw PPG waveform, some of those inputs are themselves products of earlier inference.
The study involved more than 200,000 participants. In a healthy validation group, the model predicted chronological age with a mean absolute error of roughly 2.4 years. More importantly, the difference between someone's predicted PPG age and chronological age was associated with cardiovascular and metabolic health.
That is real signal. It is not, however, the same as proving that a consumer-facing Health Age is an accurate measurement of an individual's "biological age". Those are quite different claims.
The population problem
This brings me to the part I find particularly interesting: populations.
At Terra, we've done some work thinking about generalized health-age and longevity-type scores, and philosophically I think most of these scores are fundamentally comparative. You take a collection of physiological measurements, work out where someone sits relative to some reference distribution, and translate that position into something intuitive. In this case, an age.
You therefore have to peg the score against some population. So the population you choose matters enormously.
Imagine two identical 45-year-olds. Compare one with the general population, including sedentary people and those with chronic disease, and they might look physiologically "young". Compare the same person with a carefully selected healthy population and suddenly the bar is much higher.
Interestingly, Apple's PpgAge paper demonstrates exactly this issue. The researchers deliberately selected a subgroup of participants who reported being healthy to construct the principal aging model. That methodological decision matters because your definition of "normal aging" determines what being older or younger than expected actually means.
We still don't know which reference population Apple uses for Health Age.
I actually like age as a communication tool because people immediately understand it. Telling somebody their cardiovascular fitness is 1.1 standard deviations above the age-adjusted mean is scientifically precise but not particularly motivating. Telling them that their health metrics resemble someone five years younger is much more intuitive.
But the trade-off is that the simplicity hides the assumptions underneath.
How I would think about building a Health Age score
If I were building a score like this, one relatively transparent starting point would be a z-score approach.
Take each metric, centre it around an age-adjusted population mean, divide by the standard deviation, then combine the resulting scores according to how strongly each variable predicts meaningful health outcomes.
This is my way of thinking about the problem, not a claim about how Apple has built Health Age.
In theory, I would expect VO₂ max to carry considerable weight because cardiorespiratory fitness is strongly associated with mortality and cardiovascular disease.
But there is another layer of inference here too. The VO₂ max reported by a wearable is itself an estimate, derived partly from PPG-based heart-rate data alongside pace, movement and other information rather than a direct laboratory measurement of oxygen consumption.
So you can end up with an estimated input that feeds into a second model, producing an estimated output.
That doesn't make it useless, but it matters when we think about precision and confidence.
This is another area I have been looking at recently, particularly how wearable-derived VO₂ max compares with direct measurement, so I will come back to it in a future blog.
Resting heart rate also contains useful signal. Sleep and HRV add further information, although both come with considerably more within-person and measurement variability. You could then combine those signals and map the result back onto an age distribution.
But there are lots of ways to build the model, and the choice of outcome matters. A model designed to predict chronological age is not necessarily the same model you would build to predict cardiovascular events, all-cause mortality, disability-free lifespan or metabolic disease.
That is another reason why "biological age" needs treating with some caution. There is no single biological clock inside us waiting to be read.
What do A1c and LDL add?
Apple also allows two particularly interesting lab values to contribute: A1c and LDL.
A1c measures the percentage of haemoglobin that has glucose attached to it. Because red blood cells circulate for roughly three months, A1c gives an integrated picture of blood glucose exposure over the previous two to three months, with more recent weeks contributing more heavily.
As a useful rule of thumb, each one-percentage-point increase in A1c corresponds to roughly 29 mg/dL higher estimated average glucose. Below 5.7% is generally considered normal, 5.7–6.4% is in the pre-diabetes range, and 6.5% or above is consistent with diabetes, usually requiring confirmation in the appropriate clinical context.
LDL, or low-density lipoprotein cholesterol, is different. LDL-containing particles play a causal role in atherosclerosis and cumulative exposure over decades matters enormously for cardiovascular risk.
Unlike A1c, however, a single LDL measurement is not really a three-month historical average. LDL can move substantially over weeks, and sometimes more quickly, in response to diet, medication and other changes.
What A1c and LDL both provide is a different layer of information from last night's sleep or today's resting heart rate. They describe important underlying metabolic and cardiovascular risk factors rather than acute recovery. This makes them particularly interesting additions to a longevity score.
What is Readiness?
Readiness is a zero-to-ten score that updates during the day. Apple says it draws from four broad areas: recent activity, training load, overnight vitals and sleep score.
Training load compares recent activity with a longer-term baseline, including the familiar seven-day versus 28-day relationship. Apple's sleep score incorporates sleep duration, bedtime consistency and interruptions.
The new Apple Watch Series 12 and Ultra 4 hardware also dramatically increases the density of the physiological data available. Heart rate can be recorded every five seconds throughout the day, with HRV measured as often as every five minutes.
Historically, one of the limitations of consumer wearables has been that apparently continuous metrics were often actually being inferred from relatively sparse measurements. Increasing sampling frequency should make it much easier to establish a reliable individual baseline and detect meaningful deviations from it.
I don't know much about hardware sensor engineering, but achieving that density of measurement while still maintaining practical all-day battery life seems like a significant step forward in itself.
Population comparison versus personal baseline
Apple's Recovery HRV is particularly interesting because it compares the user against their own baseline rather than against a population.
For something like Readiness, I think within-person comparison is generally far more valuable than population comparison.
An HRV of 40 milliseconds might be excellent for one person and unusually poor for another. What matters is whether your physiology has moved away from your own normal state.
Apple then translates the underlying data into four very simple labels: Recover, Pace Yourself, Ready and Go For It. Tap into the score and you can see which factors are pushing it up or down.
I really like how this is communicated to the user. I think many health and sports apps fail at exactly this point. They either give people a raw number with no context, or provide so much information that the useful explanation is buried three taps down and most people never find it.
Apple's approach is almost the opposite. The label tells you what you might want to do now. If you want to understand why, you can tap into the detail underneath.
From a behavior-change perspective, I think this might actually be more important than another marginal improvement in the algorithm.
What these scores don't actually measure
However, both Health Age and Readiness carry an important warning.
Neither directly measures the thing its name implies. There isn't a sensor inside the Watch measuring "readiness". And Health Age doesn't literally measure how old your biology is.
They are models built from proxies.
Readiness cannot directly see muscle glycogen, neuromuscular fatigue, motivation, immune status or whether you are actually going to perform well in a race that afternoon. It can measure physiological signals that tend to relate to recovery and infer something useful from them.
Similarly, a Health Age five years younger than your chronological age does not mean you have literally gained five additional years of life. This distinction matters because beautifully presented numbers can appear much more precise than the phenomenon they represent.
A Readiness score of 8 out of 10 doesn't mean that your body is objectively "80% ready". A Health Age of 37 doesn't mean there is a hidden physiological clock somewhere inside you reading exactly 37.
They are interfaces onto a collection of measurements and probabilities.
For me, the better question is whether the scores are directionally useful.
If I improve my aerobic fitness, sleep more consistently, lower my cardiovascular risk and become more physically active, does my Health Age move in the right direction?
If I accumulate excessive training load, sleep badly and show unusual overnight physiology, does Readiness pick that up?
And, crucially, does presenting those changes in a simple way cause me to do something useful?
That is ultimately much more important than whether the number itself can claim some philosophical status as my "true biological age".
Apple vs WHOOP, Oura and Garmin
Apple is now sitting squarely alongside WHOOP, Oura and Garmin in this space, but there are significant differences between the products.
WHOOP's Recovery is a 0–100 green-yellow-red score and is strongly influenced by HRV relative to your own baseline, alongside sleep and other physiological measures. Its conceptual strength is its relationship with Strain: yesterday's load and today's recovery sit side by side.
Oura's Readiness score is also 0–100, combining signals including resting heart rate, HRV balance, temperature, sleep and activity across both short- and longer-term baselines. I think of it as somewhat more of a holistic recovery and wellness signal than a direct training instruction.
Garmin probably has the closest mechanical cousin to Apple's approach. Training Readiness already combines sleep, recovery time, HRV status, acute load and stress, while its training-load system explicitly considers short-term load against a longer 28-day context.
Apple compresses this general concept into a score from zero to ten and four extremely clear labels.
Then, alongside it, Health Age becomes Apple's longevity counterpart to products such as WHOOP Age, with the additional step of incorporating clinical biomarkers such as A1c and LDL.
But the really big differentiator here is where all of this lives.
WHOOP is a dedicated recovery wearable. Oura is a ring heavily optimized around overnight physiology. Garmin is deeply embedded in training and performance.
Apple is putting these ideas into a mainstream smartwatch and health ecosystem that may already include your activity, sleep, ECGs, medications, medical records, and now bloodwork.
The Real Challenge is Changing Behavior
The important question isn't whether either score is perfect. They won't be.
It is whether these scores make enough people look at their health data, understand what is driving it and then actually change something.
Most of the physiological concepts Apple is using are not new. Biological-age models exist. Recovery algorithms exist. Training-load models exist. HRV-based readiness systems exist.
The genuinely interesting challenge is translating all of that complexity into something understandable enough to change behavior at enormous scale.
On that front, Apple's bet on simplicity might be the most important design choice of all.
References
- Apple. “Apple advances health and fitness capabilities using Apple Intelligence.” 9 September 2026. https://www.apple.com/newsroom/2026/09/apple-advances-health-and-fitness-capabilities-using-apple-intelligence/
- Apple. “Apple unveils Apple Watch Ultra 4.” 9 September 2026. https://www.apple.com/newsroom/2026/09/apple-unveils-apple-watch-ultra-4/
- Apple Support. “Track your sleep and view your sleep score on Apple Watch.” https://support.apple.com/en-gb/108906
- Nature Communications. “A wearable-based aging clock associates with disease and behavior.” 2025. https://www.nature.com/articles/s41467-025-64275-4
- Quest Diagnostics. “Apple Health app Users to be Able to Order Labs from Quest Diagnostics.” 9 September 2026. https://ir.questdiagnostics.com/press-releases/press-release-details/2026/Apple-Health-app-Users-to-be-Able-to-Order-Labs-from-Quest-Diagnostics/default.aspx
- American Diabetes Association. “Diagnosis.” https://diabetes.org/about-diabetes/diagnosis
- National Glycohemoglobin Standardization Program. “HbA1c and estimated average glucose.” https://ngsp.org/ifccngsp.asp
- American Heart Association / American College of Cardiology. “Updated Guideline for Managing Lipids.” 2026. https://newsroom.heart.org/news/accaha-issue-updated-guideline-for-managing-lipids-cholesterol
- WHOOP Support. “Healthspan: WHOOP Age & Pace of Aging Guide.” https://support.whoop.com/s/article/Healthspan-WHOOP-Age-Pace-of-Aging-Guide
- Oura. “Readiness Score.” https://ouraring.com/blog/readiness-score/
- Garmin Support. “Training Readiness.” https://support.garmin.com/en-GB/navionics/faq/SEkNpdGyhR917js0qQL3Q6/
