Trusting the Right Tool for the Job
Why measuring HRV accurately starts with the right device (and why the right device is not a wearable)
This blog was prompted by recent papers showing issues in HRV measurements for the Apple Watch and Fitbit as well as preliminary data showing similar issues for Garmin devices.
In recent years, we’ve seen an explosion in the popularity of wearable devices that claim to measure everything from sleep to stress, recovery, and more. Among these metrics, heart rate variability (HRV) has emerged as a key parameter for understanding stress and recovery. As readers here will know already, HRV is one of the very few things that wearables can actually measure - even though this is the case only at rest (and we should call it PRV, or pulse rate variability, which can differ from HRV, or the electrical activity of the heart - but that’s another story). All other parameters are either estimated (e.g. sleep time and stages) or completely made up (recovery, readiness, stress, etc.).
Yet, not all HRV measurements are created equal, and recent research highlights the pitfalls of relying on hardware or algorithms not specifically optimized for the job. The prime example of this issue is certainly the Apple Watch (oh dear), but the situation seems rather similar for Fitbit, Garmin, Whoop and others.
The Apple Watch remains a prime example because not only does it not comply with standard protocols to communicate with third-party apps, but Apple Watch's automatically collected data is also subject to changes in a way that makes it unreliable. See for example in the image below the differences in HRV when exporting twice the same data. If your HRV goes from 50ms to 100ms when exporting it a second time, something is really wrong. We cannot trust the processing provided.
On top of these issues, the recent paper mentioned above shows poor accuracy for the Apple Watch when used to compute HRV, despite good results for heart rate data (at rest). While I’ve used the Apple Watch as an example here, mainly given its popularity and poor performance at the task (no comment on the apps using it as the only data source - which seem to be rather popular), this is a much larger problem when using wearables and automatic measurements. For example, years ago, after showing how the only reliable way to use HRV data collected during the night was to average it (learn more here), Whoop quietly changed their algorithms, potentially leading to inconsistencies in users’ data - data that remains more noisy than what you get from other wearables when the analysis is automated. Similarly, Reddit is full of users whose HRV has changed when upgrading to a new Oura ring, something I discuss further below.
Another paper recently published showed similar issues for Fitbit, with large errors, and particularly incorrect measurements for HRV values above 50 ms.
Where do we go from here?
The first step to avoid the issues above and take control of your measurements and data is to measure intentionally (i.e. first thing in the morning), using a device that either 1) shows you the raw data, e.g. the actual PPG signal, 2) at least provides the RR intervals, or beat to beat data, which can then also be compared with a gold standard (e.g. electrocardiography).
HRV is not the easiest signal to measure and is prone to artifacts as well as systematic issues depending on the type of filtering used (which I believe might be the issue we see in the Fitbit data above, given the underestimation for high HRV, they are probably overcorrecting with an artifact threshold that is too strict - you are welcome). Without an accurate and consistent device, you are collecting noise. This is what you are doing in many cases when using a wearable.
The Problem with "All-in-One" Devices
Many wearables today, such as smartwatches and rings, aim to do it all. They track your steps, sleep, workouts, HRV, etc. However, as highlighted in the study linked above about the Apple Watch, the tradeoffs in hardware and algorithm design often result in inconsistent or inaccurate HRV measurements. It is common for users to report quite different HRV values when e.g. upgrading to a newer version of their wearable (or just updating software). Why does this happen?
It boils down to hardware constraints and various tradeoffs. A smartwatch for example must be relatively small, lightweight, and have a long battery life. These tradeoffs mean compromises in sensor design, sampling frequency, and artifact correction algorithms. For most of these devices - the Apple Watch in particular, but also Fitbits or Garmins or the latest iterations of other wearables that are now providing lots of “features” - HRV often ends up being an afterthought, not the core functionality of the device. While in the early days we might have had a wearable with the goal of providing accurate HRV, now I feel like the goal is to “try not to break the HRV measurement too much” while adding lots of unnecessary things. Get what I mean? The outcome is then the one we see, with inaccurate data, values changing based on whatever fix is implemented later on, new hardware, etc.
In contrast, tools designed to do one job, i.e. measuring HRV, like a chest strap or a dedicated HRV app - even using the camera as in HRV4Training - are designed specifically for this purpose and this purpose only. These tools are optimized for accurate data collection over a short period without being burdened by competing priorities and will provide you with more accurate data.
The same considerations apply to software aggregators - you can get all data in one place, but none of the meaning. In contrast, using different tools, carefully designed with the specific goal of helping you interpret one type of data, can provide more value.
A single, morning measurement, taken according to best practices (a simple protocol consisting of waking up, going to the bathroom, and afterward measuring your HRV while sitting), is all that is required to capture individual responses to stressors in a way that is meaningful and actionable.
Personally, I use a chest strap (Polar H10 paired with the HRV Logger app) every time I want to:
Check HRV before and after exercise to assess exercise intensity (learn more here).
Do a deep breathing exercise using the HRV4Biofeedback app.
Check ectopic beats to keep track of premature ventricular contractions or else (also with the HRV Logger).
On the other hand, I use the phone camera in HRV4Training to assess my morning physiology.
I spent the last 11 years of my life building this app with the one goal of measuring HRV accurately, and as such, I trust it (and I have validated it against electrocardiography).
If you do not trust this method, just use a chest strap for your morning measurements with HRV4Training as well. You will be one of the few people that actually knows their HRV, and not just their PRV.
I currently do not use any wearable, despite developing this technology 10-15 years ago, way before it came to the consumer. Sleep metrics are inaccurate or not useful, resting physiology is low priority and lower accuracy with each new iteration (as discussed in this blog), made-up scores are front and center, intentional measurements are typically not possible, pairing with third-party apps for manual measurements is not possible, etc - this is pretty far from what I had envisioned when building the early prototypes.
A morning measurement instead will do pretty well (see example above).
Wrap up
HRV is a sensitive metric. Small changes in data collection methodology—sampling frequency, sensor placement, or artifact correction algorithms—can lead to significantly different outcomes. For instance, when wearable manufacturers update their algorithms or hardware, your HRV trend might suddenly shift, not because your physiology has changed, but because the device’s interpretation of your data has changed.
It’s tempting to rely on a single wearable or platform to track everything. The convenience of having all your data in one place is hard to resist. But the reality is that this convenience often comes at the expense of accuracy. When HRV is just one of many features, it’s unlikely to receive the dedicated attention required to ensure high-quality measurements.
Using a wearable and automated data collection, you never have control over firmware or software updates that might change your data in ways that mess up your analysis and interpretation.
Imagine using HRV to guide your recovery, only to find out that your "improvement" was actually due to a software update or a change in the device's artifact correction method: this is a lot more common than you’d think.
While wearables started as promising devices, at this point I do not feel comfortable recommending any athlete to use these devices to track their own physiology in response to training. The accuracy of wearables is insufficient and constantly prone to changes due to hardware and software updates, and the protocol itself (measuring during sleep) is suboptimal to track training-related responses. Let alone all the noise you have to work hard to keep your mind free of, due to an ever-expanding number of features with no utility for athletes (readiness and recovery scores, sleep scores, stress scores, etc - all things that influence you psychologically without actually reflecting your body’s response).
If you care about measuring your HRV accurately, don’t settle for a general-purpose device where HRV is an afterthought. Instead, consider using tools designed specifically for HRV. Whether it’s a chest strap or a camera-based app, these tools provide the reliability needed to take control, measure intentionally and make meaningful decisions based on your data.
For those seeking the most accurate HRV measurements, a chest strap remains the gold standard. Chest straps are specifically designed to capture precise beat-to-beat heart rate data (RR intervals), with high sampling rates and minimal noise. These devices don’t have to worry about conserving battery life for days or fitting a dozen sensors into a sleek form factor, or powering a screen.
They do one job—and they do it exceptionally well.
Remember, it’s not all about convenience; it’s about trust. Trust the right tool for the job, and you’ll be on the path to meaningful insights and better health and performance.
Marco holds a PhD cum laude in applied machine learning, a M.Sc. cum laude in computer science engineering, and a M.Sc. cum laude in human movement sciences and high-performance coaching.
He has published more than 50 papers and patents at the intersection between physiology, health, technology, and human performance.
He is co-founder of HRV4Training, advisor at Oura, guest lecturer at VU Amsterdam, and editor for IEEE Pervasive Computing Magazine. He loves running.
Social:
Such a good article! I am citing you in my next YouTube video so other cyclists like me can up there game and make HRV useful!
It's interesting these devices seem to be getting worse when you'd expect them to get better. I guess they're all trying to outdo each other with features to keep the sales flowing, assuming that most people don't know (or care) that accuracy is being sacrificed. For many people, I think it's probably not that much of an issue; but for those, like you, who rely on monitoring for high-level training, or those of us using devices to try to help us navigate and manage chronic illnesses, it *does* matter. Ah well, I guess 'jack of all trades, master of none' comes to mind here ... 😏