How Sleep and Fitness Trackers Work, and Whether They’re Worth It

If you’ve read articles about fitness trackers, they were probably written by compulsive workout junkies who compare them for how well they can track those zillion mile bike rides or marathon training runs. Well, I’m not one of them. But the tech in sleep and fitness trackers is pretty amazing and well worth writing about. And yes, they can also provide health benefits for the rest of us who get exercise as time permits.

Trackers, as with much of the digital health movement, have come a long way in the last few years. From the simple and not-very-accurate step counters of a few years ago, they have evolved into devices that can monitor your heart rate, sleep, and other vital signs. However, they’re far from perfect, so they can also provide an undeserved impression of accuracy.

How Step-and-Stair Tracking Works

The simplest form of counting steps is to use the data from the device’s accelerometer and Inertial Measurement Unit (IMU) to detect rhythmic motions that are consistent with the back-and-forth movement that typically goes along with walking or running. By using the data from both sensors, the device tries to filter out false positives.

Once the device has a step count, then it multiplies that by an estimate of your stride to calculate how far you’ve walked or run. Worst case it uses a generic guess at your stride, but typically you’re able to enter your height to give it a more accurate starting point, or even enter your stride length directly. Some devices go a bit further and will calibrate your stride by comparing GPS results with its estimates. Because consumer GPS has limited accuracy, this process usually requires several minutes of traveling at a consistent speed. Some also calculate separate stride lengths for walking and running. Until recently, that meant remembering to tell the tracker when you started a hike or run. But many newer devices do a good job of auto-detecting when you start some type of exercise and classifying it appropriately.

From having owned various fitness trackers over the years, it’s clear that counting steps and stairs is as much of an art as a science at this point. Even when using several current state-of-the-art trackers at the same time, their step counts can differ by as much as 15 percent. Typical wrist trackers and watches don’t have the processing power to run a lot of sophisticated AI-based analytics to help clean up the data, either.

Devices with altimeters often also let you count how many flights of stairs (or equivalent) you’ve climbed. Here, too, sensor fusion is required, so that altitude gained while driving or flying doesn’t get credited to your fitness (a shame for tech journalists who spend a lot of time on airplanes).

Tracking climbing can be even more of a crapshoot. For example, my Fitbit Versa regularly reports dozens of floors climbed while I’m playing tennis — even though each floor is supposed to represent 10 feet of altitude gained while walking or running. In contrast, my Huawei Band 3 Pro isn’t fooled. However, the Versa does a better job keeping up with my running up and down stairs during the day.

Fitness Tracking: Another Field Turned on Its Head by AI

As with so many areas of technology, digital health has been vastly improved through the use of AI. For example, instead of writing long sequences of complicated code based on physical models to count steps, modern trackers rely on neural networks that use machine learning to determine strides. Similarly, instead of relying on human analysis of sleep data for each patient, trackers have systems that are trained on huge amounts of human-labeled sample data. As a result, they can categorize the sensor information from users into not just sleeping or awake, but even the specific type of sleep.

How Heart Rate Monitoring Works

If you’ve ever had a heart issue, you may have been hooked up to a machine with a variety of electrodes to monitor your heart (an ECG or EKG). Those electrodes measure the small electrical currents emitted by the “pacemaker cells” in your heart. The best consumer-grade heart monitors use a simplified version of the same technique. A chest strap with electrodes on the inside is used. With that approach, it is possible to get both a very accurate measurement of heart rate, and also calculate Heart Rate Variability (HRV), an increasingly popular metric of fitness.

As you can imagine, that’s something of a hassle, so most trackers rely on a less accurate but lower-hassle optical system. Optical heart rate monitors use a process called photoplethysmography (PPG) to calculate your heart rate by shining light into your skin and measuring the reflectance. The light is emitted from LEDs (usually at least two) on the inside of the tracking device. Multiple LEDs at different frequencies help provide better results across the wide range of possible skin colors and thickness.

Unfortunately, the readings from an optical tracker placed on your wrist, or in a ring, are susceptible to fluctuations as you move. In particular, if you are running or jogging at a similar pace to your heart rate, then it is possible for a tracker to pick up on that cadence and think it is your heartbeat. This is often referred to as the “crossover problem.” Since only about .1 percent of the light reflected from your skin is related to the heart rate signal, there are plenty of opportunities for error to creep in.

To help with this, many trackers also incorporate an accelerometer to help them disregard incorrect data. The amount of light reflected also varies with ambient light level, as unless you are in a dark room or have your hand and wrist completely covered, some pollution of the light from the LEDs will occur. Higher-end devices include ambient light monitoring to minimize this problem.

Because of these issues, the most accurate of the optical heart rate devices appear to be armbands and clips that go on your finger. Of course, neither is quite as easy to use as a wrist-based tracker or even a ring, so a lot of work has gone into making more accurate heart rate tracking for popular devices that can be worn all day (and night). Manufacturers of brand-name models from Garmin, Fitbit, and others claim accuracy within 5 percent of a medical-grade device for their wrist-worn trackers. That’s pretty reasonable if you just want a general measure of your health, and an estimate of how much “cardio” time you’re getting from exercise each day, but certainly not good enough for training elite athletes.

As an experiment, I outfitted myself with five different heart-rate-capable tracking devices. For starters, we have a Sleeptracker from FullPower under our mattress (which uses pressure and vibration to measure heart rate while asleep). Then I tried a ZeTime watch, a Fitbit Versa, a Huawei Band 3 Pro, and an inexpensive fingertip pulse-oximeter. While the data from the ZeTime nearly gave me a coronary (it showed some massive spikes while sleeping that certainly didn’t look healthy), the other four trackers were generally consistent in pattern, and fairly close in actual values. I’m sure some of the differences were caused by having to wear several at once, so none of them were really in an ideal location. None of these devices are accurate enough to calculate HRV, though. Leading HRV app maker EliteHRV only fully supports chest strap devices for that purpose.

Using an ECG to Detect A-fib

While the Apple Watch 4 isn’t the first wearable to be able to provide users with an electrocardiogram (ECG), it is by far the most popular. Specifically, on demand, the latest Apple Watch can provide an ECG trace and detect whether the user may be suffering from an irregular heartbeat — in this case atrial fibrillation or a-fib. It does that by measuring the electric pulses sent out by the heart as they reach the watch. To get a reading, the user lays their finger alongside the watch for 30 seconds to close the circuit. By itself, diagnosing an irregular heartbeat may not mean much, but it is enough reason to consider further evaluation by a medical professional. Apple helps the process along by providing a PDF of the ECG that the user can forward to their physician.

To validate the effectiveness of this capability, Apple has funded an extensive study showing that wearers of its Watch 4 using this feature receive similar benefits to those wearing a medical device in a more typical week-long evaluation. There are clearly benefits to early detection of symptoms of possible heart disease. However, the medical community is divided over the value of diagnosing a-fib in otherwise healthy people with no specific propensity for heart disease. In any case, this capability is certainly a taste of what are likely to be further developments in tracking heart health through popular wearables.

Sleep Tracking Compared With Sleep Studies

If you have a sleep disorder or have ever suspected that you have an issue with sleep apnea, you were probably referred to a clinic that could load you up with electrodes and charge you a ton to let you know how, and how well, you sleep. But if you simply wanted an idea of how well you’re sleeping, and what you might be able to do to improve it, wearing a dozen electrodes every night certainly isn’t practical. Enter sleep trackers. Using one or more sensors, they rely on science and machine learning to estimate when you are sleeping, what phase of sleep you’re in, and suggest various health tips and tidbits.

My experience with five different trackers that report on sleep indicates that consumer products can do a reasonable job of creating a rough outline of your sleep and waking states, and perhaps of roughly the total time spent in each of the labeled sleep states. These are commonly called Light, Deep, and REM, although a sleep researcher I spoke with said that medically REM is important enough that they start by classifying sleep into REM and non-REM. In any case, no two of the trackers matched on a consistent basis.

The sleep tracker I’ve been using the longest is the Sleeptracker. The sensor pod goes under your mattress so it is totally hassle-free. Fullpower has also done an excellent job of building health statistics based on your demographic profile compared with its community of users. That lets them provide some interesting and potentially useful coaching tips. Placement of the sensor also helps them measure breathing rate — something the typical fitness trackers I’ve used couldn’t estimate. From speaking with Fullpower, the company credits its in-house design sensors that detect motion from under your mattress coupled with an AI model based on 250 million nights of sleep with giving the device 90%+ accuracy and putting it ahead of the competition. I also asked about the large variety of mattresses and learned that the device auto-calibrates, again based on the company’s cloud-based AI models. By using dual sensors, Sleeptracker can also monitor two sleepers in the same bed, which is pretty impressive.

Most other sleep trackers in use are simply fitness trackers that can do continuous heart-rate monitoring. They analyze data including how much you are moving and your heart rate to estimate whether you are asleep or awake, and which stage of sleep you’re in. Currently, none of the standard fitness trackers are certified as medical grade devices or for use in diagnosing sleep apnea. However, startup Beddr has a device you can attach to your forehead that also includes a pulse-oximeter and can be used to detect apnea events. Fitbit markets that its Charge 3 and Versa have SpO2 (pulse-ox) sensors, but they don’t actually do anything currently.

The Quantified Self Is on the Way

While consumer fitness and sleep trackers clearly have a long way to go before they are on a par with medical-grade procedures, progress has been and is likely to continue to be rapid. Sensors are getting smaller, less expensive, and more accurate at the same time that increased processing power and improved analysis tools are becoming available. What took a large watch a couple of years ago can now be done with a ring. As a next step, look for increased integration of personal tracking devices with the professional health care system. It is already starting to happen on a limited basis but is likely to become commonplace.

[Image Credit: PPG]

Now read from PCMag: