The Pod is an intelligent sleep system that can be added to any bed. It integrates a network of sensors that continuously monitor vital sleep metrics, including heart rate, breathing rate, heart rate variability (HRV), sleep and wake times, and sleep stages. These data are analyzed by sophisticated algorithms to accurately track your sleep patterns.
Central to the Pod’s technology is Autopilot, the intelligence behind that Pod, which uses these algorithms to personalize your sleeping environment. One crucial component is our sleep staging algorithm, which assesses your sleep stage every minute. This enables Autopilot to adjust the Pod’s temperature dynamically throughout the night, ensuring each sleep stage is optimally tailored for maximum rest quality.
In 2023, Eight Sleep significantly enhanced our sleep algorithm, boosting its accuracy dramatically. Accurate sleep stage classification is pivotal not only for providing insightful user data but also for optimizing the Autopilot feature, which dynamically adjusts bed temperature to your sleep stages in real-time. Our algorithm uses real-time physiological data from Pod sensors to accurately predict your current sleep stage.
Understanding sleep architecture
Sleep architecture describes the structure of a typical night’s sleep, which consists of 4 to 6 sleep cycles. Each cycle lasts approximately 90-110 minutes, progressing through stages of light sleep, deep sleep, Rapid Eye Movement (REM) sleep, and brief periods of wakefulness. During a normal night, about 75% of your sleep is spent in the light and deep stages, with the remaining 25% in REM sleep (ref). Adults typically need 7-9 hours of sleep per night to function optimally. Insufficient sleep can impair concentration, learning abilities, and memory, reduce performance, negatively affect mood, weaken the immune system, and increase the risk of metabolic health issues (ref, ref, ref).
Exploring the four sleep stages and their importance
Throughout the night, your body cycles through different stages of sleep, each playing a unique role in your overall health and recovery.
- Wake: Throughout the night, it’s normal for healthy adults to wake briefly several times, though these awakenings are often so short that they are not remembered in the morning (ref).
- Light Sleep: This stage serves as a bridge from wakefulness to deeper sleep. Light sleep is marked by the beginning of physical relaxation and a reduction in heart rate and respiratory rate. Crucially, light sleep facilitates the consolidation of memories, particularly those involving new skills, making it fundamental for cognitive health and learning efficiency.
- Deep Sleep: During this stage, your body achieves its deepest relaxation, making it difficult to be awakened. It’s a critical phase for physical recovery, growth, and immune system strengthening. Since deep sleep is enhanced by a reduction in body temperature, the Autopilot feature of the Eight Sleep Pod proactively adjusts to cool your bed, thereby facilitating a more efficient transition into this restorative state (ref).
- REM Sleep: This phase features a highly active brain and rapid eye movements, with your body remaining essentially paralyzed to prevent acting out your dreams. REM sleep is vital for consolidating memories, processing emotions, and supporting overall brain health. During REM, your body cannot regulate temperature effectively; Autopilot steps in to help maintain temperature at an optimal setpoint (ref).
Accurate Sleep Stage Classification with Eight Sleep’s New Deep Learning Algorithm
To develop our algorithm, we utilized an initial dataset comprising over 5,700 subjects, each monitored with full polysomnography and professional sleep stage scoring. We excluded any subjects lacking complete sleep data, those with moderate to severe obstructive sleep apnea, and those who recorded less than four hours of sleep. The refined dataset was then segmented into training, validation, and test sets to ensure comprehensive learning and validation.
Our state-of-the-art machine learning (ML) algorithm marks a significant upgrade, from traditional ML feature engineering to advanced deep learning techniques. This transition allows for efficient extraction of critical data from raw inputs, resulting in robust and accurate sleep analysis. The new deep learning model has yielded a 50% increase in accuracy compared to our previous model, providing users with more accurate insights into their sleep patterns.
We benchmarked our algorithm’s performance against the gold standard of electroencephalogram (EEG) brain wave recordings. Impressively, our algorithm achieved an overall 78% accuracy in classifying the correct sleep stages. For context, agreement between trained sleep technicians using EEG typically ranges from 80% to 90%. This comparison underscores the accuracy of our algorithm in sleep stage classification (ref). An example of its outputs is illustrated in Figure 1, and the sensitivity and specificity for each sleep stage are reported in the Table.
Figure 1: Hypnogram illustrating algorithm accuracy, comparing our model’s predictions with the gold standard across sleep stages.
Table: Sensitivity & Specificity of Sleep Stages Compared to Ground Truth
Note: Sensitivity is the true positive rate and specificity is the true negative rate.
An algorithm that continues to improve
With a massive amount of real-world sleep data, we continue to refine the ML model to improve its accuracy even further. All Eight Sleep members who use Autopilot automatically benefit from the latest sleep algorithm every night.
