The Promise and Pitfalls of At-Home Sleep Tracking: A Deep Dive into the Withings Sleep Analyzer

Introduction

Sleep is an essential pillar of health and well-being. The clinical gold standard for sleep assessment, polysomnography or PSG, provides a detailed analysis of sleep architecture but is impractical for routine or long-term monitoring. Its reliance on complex equipment, high cost, and typically in-lab application make it an intrusive process. The proliferation of consumer wearable and nearable devices offers more accessible alternatives, yet their accuracy often lacks rigorous scientific validation, particularly in home environments.

A recent study sought to address this gap by evaluating the accuracy and reliability of the Withings Sleep Analyzer (WSA). This contactless sleep mat, placed under the mattress, was compared directly against simultaneous PSG recordings in a large and diverse group of individuals in their own homes. This research investigates the sensor’s performance in real-world conditions, offering critical insights into the current state of consumer sleep-tracking technology.

Methods

The study involved 117 healthy participants, with 69 women, and a mean age of approximately 40 years. Each participant slept in their own bed for one night with both the PSG equipment and the under-mattress device active. This setup allowed for a direct, epoch-by-epoch comparison of the data recorded by the consumer device against the clinical reference standard. The analysis focused on two primary objectives: the accuracy of distinguishing sleep from wakefulness and the precision of classifying distinct sleep stages, including light, deep, and REM sleep. Performance was assessed using standard classification metrics to ensure a robust evaluation.

Results

The investigation found that the contactless device performs effectively in identifying sleep and wake states. It achieved an overall accuracy of 87% in this core task, demonstrating a high sensitivity of 93% for detecting sleep and a moderate sensitivity of 73% for detecting wakefulness. A key strength observed was the sleep mat’s consistent performance across various subgroups. The accuracy of sleep-wake detection remained stable regardless of participant age, BMI, sex, mattress type, mattress thickness, sleep quality or the presence of a bed partner.

Challenges emerged in the classification of specific sleep stages. The sensor’s mean accuracy for staging sleep was 63%, with a Cohen’s Kappa of 0.49. The primary difficulty was in distinguishing between light and deep sleep. This led to systematic biases in sleep duration estimates; the device tended to slightly overestimate total sleep time by an average of 20 minutes but substantially overestimated light sleep by 1 hour and 21 minutes. Conversely, it moderately underestimated REM sleep by 15 minutes and deep sleep by a more significant 46 minutes.

Notably, a notable proportion of misclassifications made by the sensor mirrored disagreements found between the expert human reviewers who scored the PSG data, especially concerning the boundary between light and deep sleep. Furthermore, participants reported that their perceived sleep quality was significantly altered for the worse on the night they used the PSG equipment, highlighting the intrusive nature of the gold standard itself.

In a comparative context, the Withings Sleep Analyzer exhibits highly competitive performance in sleep-wake discrimination relative to other devices on the market. For the more nuanced task of sleep stage classification, its accuracy is comparable to that of similar products. This level of performance is particularly noteworthy given the systemic challenges in sleep staging.

Conclusion

For individuals seeking to understand their sleep over weeks and months, the primary benefit of a device like the Withings Sleep Analyzer lies in its practicality. Its contactless, ‘set-and-forget’ nature eliminates the nightly burden of wearing a device and avoids the discomfort that can disrupt sleep, a notable issue even with the clinical gold standard. While the sensor’s accuracy in distinguishing specific sleep stages requires further refinement, its strong performance in tracking overall sleep and wake times provides reliable insights into sleep duration and consistency. This capability for accessible, unobtrusive, and longitudinal monitoring is where at-home sensors currently provide the most value, empowering users with meaningful data on their long-term sleep trends.

Poster Session: Time and Location

“Evaluation of a Contactless Sleep Monitoring Device for Sleep Stage Detection against Home Polysomnography in a Healthy Population”

Session Title: Poster abstract group 2

Session Date: Monday, September 8, 2025

Presentation Time: 6:00pm to 7:00pm (Presenting authors will be present near their assigned poster board throughout the scheduled one-hour presentation window.)

Poster Board Number: 531

Location: Posters will be displayed in the exhibit hall on Level 4 and accessible during regular congress hours.

About Marie-Ange Stefanos

Marie-Ange Stefanos is a Machine Learning Research Scientist and a PhD candidate pursuing a joint doctorate in Computer Science and Neuroscience from Université Paris Cité (France) and Reykjavik University (Iceland). Building on her background with an Engineering degree in Signal Processing from Grenoble INP – Phelma and an M.Sc. in Machine Learning from KTH Royal Institute of Technology, her path into health research was driven by a central question: how can my technical background be best applied to solve meaningful challenges in human health?

Her doctoral research focuses on insomnia, where she develops algorithms using data from wearables and self-reports to identify predictive biomarkers and differentiate subtypes of the disorder. This work depends entirely on data integrity, which is why she believes the rigorous validation of consumer devices, as discussed in this article, is the essential first step in translating complex signals into reliable, actionable insights for users.

Interested in partnering with us?

Contact Us