How your sleep changes after one week on Orion

We collected continuous physiological measurements from users across seven nights before starting Orion, establishing a baseline for each user’s sleep architecture, sleep continuity, and overnight cardiovascular activity. We then collected the same measurements across their first seven nights on the Orion sleep system, to discover how those endpoints changed. Each participant’s nightly values were averaged within each window, producing one “before” and one “after” figure per person for each endpoint. The data was analyzed by an independent biostatistician with experience as a lead on FDA pivotal trials.

Every participant wore a wrist-worn physiological monitor to bed. The same class of device was worn across the whole cohort. The device carries a three-axis accelerometer, a high dynamic range gyroscope, an optical heart sensor that reads blood flow through the skin at the wrist, an electrical heart sensor, a blood oxygen sensor, and a wrist temperature sensor. Through the night it samples heart rate every five minutes, respiratory rate every ten minutes, blood oxygen every forty-five minutes, and heart rate variability every two hours. Sleep staging is not a sensor reading. It is a model prediction. All of those endpoints feed a state-of-the-art predictive model, which assigns every 30-second epoch of the night to one of four states: awake, light sleep, deep sleep, or REM. The classifier was developed and validated against overnight polysomnography at scale.

Orion showed its greatest benefit in what the wearables are best at measuring: cardiovascular activity. Minimum heart rate and average heart rate both fell across the full group, and heart rate variability rose. All three cardiac endpoints clear the statistical significance threshold by orders of magnitude. The improvement was close to universal, holding across everyone measured regardless of how they were sleeping before Orion arrived. It is also the measurement the device is most reliable at, read directly off the optical and electrical heart sensors. The significance and reliability of the measured benefit to cardiovascular activity are compounded by the strength of the wearable in measuring these endpoints.

The cardiovascular result above is a pooled one, and it held across the cohort. But not every benefit works that way. It is important to break out specific features of sleep in such a way that we are able to analyze the impact of the intervention on people who have a pronounced deficit on that feature. A benefit that shows up for a typical sleeper is not the same result as a benefit that reaches someone who has spent years short on deep sleep, and it is the second group for whom it matters most that Orion actually produces some form of improvement. Because these groups represent different classifications of sleepers, struggling with different kinds of problems, it is important that they are constructed and considered separately for each feature. For each feature, participants are ranked by their value before Orion and split into four equal groups, with Q1 the quarter performing worst on that feature and Q4 the best.

Across the sleep endpoints, the largest gains went to the people who needed improvement the most. The participants who struggled the most to get restorative deep sleep were the ones who saw the greatest increases in deep sleep when sleeping on Orion. The participants who struggled the most with wake-ups and interruption in the middle of the night were the ones who experienced the greatest increase in sleep continuity after sleeping on Orion. That pattern held as we moved through the other endpoints, showing that Orion made the biggest difference where it was needed the most.

Across the 158 participants with valid heart rate data, minimum heart rate fell for 118 of them and average heart rate for 113, roughly three in every four. Heart rate variability rose across the group, and all three cardiovascular p-values for the entire cohort round to 0.001 or lower.

Overnight heart rate is one of the most extensively studied markers in cardiovascular epidemiology. Every 10 beats per minute of higher resting heart rate carries a 17% higher risk of all-cause mortality, a 15% higher risk of cardiovascular disease, and an 18% higher risk of heart failure.¹ Heart rate variability is also of great importance to longevity and performance. Lower variability predicted higher mortality consistently across ages, sexes, continents, and populations.² It is also the number elite sport has converged on to represent recovery, used to track how well an athlete has bounced back, to detect overreaching, and to judge whether they are ready to train hard or need to back off.³

Total sleep time is a single number standing in for a lot of separate things going right. Time is lost at the front of the night, when sleep onset is delayed; at the end, when you wake early and cannot return to sleep; and in the middle, to awakenings that fragment the night. Those middle losses come in two forms. There are the awakenings you recall in the morning, and the smaller ones you do not, both more frequent when the body struggles to regulate core body temperature, when sleep is easily disturbed by movement, noise, or light, or when sleep cannot descend into slow-wave sleep and hold that depth. The smallest are micro wake events: brief arousals between sleep cycles and stage transitions that are never remembered. They are a normal part of sleep. But a greater number of them corresponds with more restless, less restorative sleep, and with less total sleep time. A rise in total sleep time can come from a reduction in any of those deficits, or several at once, and generally corresponds with a healthier, more restorative night. On Orion, total sleep time rose across the full group, from 410.3 to 417.6 minutes (p = 0.010).

Wake after sleep onset is the endpoint that captures the middle-of-the-night portion of that picture, counting the time spent awake between first falling asleep and finally getting up. It describes the most familiar experience of poor sleep there is: waking in the middle of the night and struggling to get back to sleep. The people living that experience are the ones who matter most for judging whether Orion works, so we looked at them on their own: the 43 participants who came in struggling most to stay asleep. In their first week on Orion they cut 16.3 minutes off the time they spent awake mid-night, a 26.0% reduction (p < 0.001). This is one of the strongest effects on any endpoint in the study, and it suggests Orion is an effective intervention for those who struggle to stay asleep.

More sleep is one of the few interventions that improves how you feel and how you perform at the same time. Athletes who extend their nightly sleep show measurable gains in physical and cognitive performance, including faster sprint times and quicker reaction times.⁴,⁵ The effect is not limited to athletes. Across a nationally representative sample of U.S. adults, people who met all five markers of healthy sleep were 30% less likely to die of any cause than those who met one or none, and at age 30 their life expectancy was nearly five years greater for men and over two years greater for women. Two of those five markers are simply not struggling to fall asleep and not struggling to stay asleep.⁶

Losing sleep works the other way, and the costs are steeper than most people expect. People sleeping under five hours a night are more than four times as likely to get sick after exposure to a common cold virus than those sleeping seven or more.⁷ Over the long term, those who consistently sleep six hours or less through midlife carry a 30% higher risk of dementia in later life.⁸

Those studies follow people over years, and what we measured is a single week, so the point here is not equivalence but stakes: total sleep time is worth moving, and the gains compound over a lifetime rather than a week. What our own data can show is who moved. Below, each of the 43 participants who began with the most wake time gets a single row, running from where they started to where they ended.

Deep and REM are not interchangeable minutes. Deep sleep is when the brain clears metabolic waste most efficiently, including the proteins whose accumulation is a hallmark of Alzheimer’s disease.⁹,¹⁰ REM does different work. It is where emotional experience is processed and emotional stability is gained.

Experiments that remove one stage and leave the other intact show how differently they fail. Selectively depriving twelve healthy middle-aged women of slow-wave sleep, with total sleep time unchanged, dropped their pain threshold 24% and produced the symptom profile of fibromyalgia, a diagnosis whose polysomnographic signature is reduced slow-wave sleep.¹¹ Suppressing REM instead produces something unrelated: more negative mood the following day and stronger neural responses to social rejection.¹²

Poor sleep is therefore not one condition. Deep and REM can be lost at independent rates, and every combination exists: people with a deep sleep deficit and adequate REM, people with the reverse, people short on both, people short on neither. Deep and REM minutes should be read separately because they describe independent classifications of sleep.

Reading them by quartile is what lets us see how Orion performs for each classification of deficit, and across the magnitudes of those deficits. Among the participants getting the least deep sleep, deep sleep rose 13.1% in the first week on Orion (p < 0.001). Among those getting the least REM, REM rose 19.4% (p < 0.001). These are two different groups of people, each carrying a different deficit, and each one closed the gap it came in with.

Nothing here is a diagnosis. We did not assess or screen for any sleep disorder in any participant, and none of the conditions referenced above is being attributed to anyone in this study. The research cited compares people over years or decades; what we measured is one week, and it speaks to the direction sleep architecture moved, not to the outcomes those studies describe.