Imputing missing sleep data from wearables with neural networks in real-world settings.

Sleep is a critical component of health and well-being but collecting and analyzing accurate longitudinal sleep data can be challenging, especially outside of laboratory settings. We propose a simple neural network model titled SOMNI (Sleep data restOration using Machine learning and Non-negative matrix factorIzation [NMF]) for imputing missing rest-activity data from actigraphy, which can enable clinicians to better handle missing data and monitor sleep-wake cycles of individuals with highly irregular sleep-wake patterns. The model consists of two hidden layers and uses NMF to capture hidden longitudinal sleep-wake patterns of individuals with disturbed sleep-wake cycles. Based on this, we develop two approaches: the individual approach imputes missing data based on the data from only one participant, while the global approach imputes missing data based on the data across multiple participants. Our models are tested with shift and non-shift workers' data from three independent hospitals. Both approaches can accurately impute missing data up to 24 hours of long dataset (>50 days) even for shift workers with extremely irregular sleep-wake patterns (AUC > 0.86). On the other hand, for short dataset (~15 days), only the global model is accurate (AUC > 0.77). Our approach can be used to help clinicians monitor sleep-wake cycles of patients with sleep disorders outside of laboratory settings without relying on sleep diaries, ultimately improving sleep health outcomes.

Efficient assessment of real-world dynamics of circadian rhythms in heart rate and body temperature from wearable data.

Laboratory studies have made unprecedented progress in understanding circadian physiology. Quantifying circadian rhythms outside of laboratory settings is necessary to translate these findings into real-world clinical practice. Wearables have been considered promising way to measure these rhythms. However, their limited validation remains an open problem. One major barrier to implementing large-scale validation studies is the lack of reliable and efficient methods for circadian assessment from wearable data. Here, we propose an approximation-based least-squares method to extract underlying circadian rhythms from wearable measurements. Its computational cost is ∼ 300-fold lower than that of previous work, enabling its implementation in smartphones with low computing power. We test it on two large-scale real-world wearable datasets: [Formula: see text] of body temperature data from cancer patients and ∼ 184 000 days of heart rate and activity data collected from the 'Social Rhythms' mobile application. This shows successful extraction of real-world dynamics of circadian rhythms. We also identify a reasonable harmonic model to analyse wearable data. Lastly, we show our method has broad applicability in circadian studies by embedding it into a Kalman filter that infers the state space of the molecular clocks in tissues. Our approach facilitates the translation of scientific advances in circadian fields into actual improvements in health.