The rumors had it there would be sleep functionality announced at the Apple Special Event on September 10. Why was there no word about this? And what does this say about the health capabilities in the future for Apple?

Navigating the moving line between consumer and medical

The expectation

There were numerous posts about the rumored sleep features, such as Apple Watch sleep tracking revealed: sleep quality, battery management, more: "While asleep, the Apple Watch will track the user’s quality of sleep using its multiple sensors and inputs, including the person’s movement, heart rate, and noises. Data about the user’s quality of sleep will be made available in the Health app and a new Sleep app for the Apple Watch."

Bloomberg predicted Apple Plans Sleep Tracking Feature for Future Watch: "The company has been using the sleep-tracking feature for several months with testers at secret sites around its Cupertino, California, headquarters, according to people familiar with the work."

So what happened?

What was announced at the event were new cases and a built-in compass. And lots of attention to health and medical implications: two new health-related capabilities, and three new health research studies Apple will be involved with, involving top medical centers. A video was shown of people whose lives were saved by the Watch. Clearly Apple is notching up its ability to play in the medical world. One article suggests this was exactly the problem; they're working to demonstrate their excellence, but Sleep trackers suck, that's why Apple Watch still doesn't have one: "Half-useful data sets are the antithesis of everything this platform is doing right now... Here's a big secret - something we don't talk about enough - sleep tracking in almost every fitness tracker is broken and terrible. And I couldn't be happier Apple isn't shipping a similarly poor feature just to say the Watch had something similar."

Does Apple feel they can't afford the hit of a health-related feature that is unproven, or underperforms? Was a fitness-related functionality so far from the medical message on which they were focusing that they left it on the side?

Can a wrist-worn device accurately determine sleep stages? The National Sleep Foundation cautions Are Sleep Trackers Accurate? "Activity trackers, such as FitBit, might help you see your general sleep patterns, but they are limited in their usefulness. Since they may not be as helpful as they promise, interpret what they tell you with a grain of salt - at least for now."

Good point. But what does this leave out? I agree that 'staging sleep' (the technical term for determining when a person is awake or in light, deep, or REM sleep) cannot be done accurately using only the motion sensors (accelerometers) in wearable devices. But as pointed out above, the Series 5 was going to use additional sensors: "including the person’s movement, heart rate, and noises". Does that additional information make it possible to accurately stage sleep? The evidence is yes, here is at least one example of a device, that does use additional sensors and advance machine learning, the WHOOP. BenGreenfieldFitness reports "the WHOOP has 90% sleep/wake accuracy compared to gold standard sleep labs". The WHOOP is a device for high-end athletes to train better, and part of what it provides is full sleep tracking. Full disclosure: WHOOP.com is a client.

Advanced Machine Learning

Separating out the different sleep stages is a challenging ML problem, which calls for a customized ML (machine learning) approach.

To train an algorithm like this, one needs a reference to train against, a gold standard. The gold standard for sleep is 'polysomnography' - you go into a sleep lab, they put lots of electrodes on your head, and sleep experts 'stage' your sleep. This is then used as the reference to train the algorithms. Making and interpreting these recordings is cumbersome and expensive. It is usually considered desirable to train ML algorithm on hundreds of thousands of data sets. This would not be practical with sleep.

We have developed approaches that allow us to synthesize data from a limited number of data sets, such as in using multiple Generative Adversarial Networks (GANs), as we demonstrated in Vincent. This permits getting effective results in ML development with limited data sets.

Conclusions

Apple made clear its commitment to playing a role in the advancement of medicine, in announcing the three upcoming clinical studies involving top institutions: Harvard School of Public Health, the NIH, the AHA, and Brigham and Women's Hospital. They are in a position to be a game-changer, and they are capitalizing on that. They will continue to play in the fitness arena, but I predict they will maintain some distance between their clinical vs. fitness initiatives.

It takes a long time to validate a health-related feature well. As we previously reported in Apple's Watch Series 4 - a demonstration of their new relationship with the FDA? At last year's Apple event, they announced the AF detection feature of the Watch Series 4 on September 12. The FDA had announced clearance of the device the previous day. It wasn't until December 6 that Apple released a new version of iPhone and Watch software to enable the new heart rhythm capabilities. So even if they do decide to release this functionality, we might have to wait. As Bloomberg also reported, "Before Apple launches new health and fitness-related features for the Watch, it puts the functionality through rigorous testing at labs around its campus. The company also conducts in-house testing for new sensors on exercise equipment such as treadmills and bikes." What isn't mentioned here is a sleep lab, which is what one needs to verify that a device performs well in tracking sleep. Verifying sleep tracking performance is much more challenging than verifying other fitness features.

Will the next Watch software release have sleep capabilities? We'll know soon! The scheduled release date is September 19 2019, with the Watch 5 available for purchase the next day. If sleep functionality is included, it should work on current watches, since there are no new sensor changes reported on the Watch Series 5.

There is much talk of the current 'sleep epidemic' we're facing. Could an Apple sleep tracker help with this? Would the information be accurate and actionable? How will it integrate with the overall fitness and medical aspects of the Watch and Apple's long-range vision? Check back here - we will continue to monitor the new capabilities and implications.

Author
David Ritscher
Connected System Architect
With a focus on connected devices, consumer products, wearables, and implantable medical  devices; design of sensor systems, algorithms, DSP, machine learning; experienced in bringing new concepts from ideation to research and development through successful product launch