Wearable devices show how sleep patterns change with health conditions

In a recent study published in NPJ Digital Medicine, researchers used a large dataset consisting of five million nights of sleep monitoring data from wearable devices to examine changes in an individual’s sleep phenotype over time and determine if these changes in sleep patterns or phenotypes are informative about periods of acute illness such as fever, coronavirus disease 2019 (COVID-19), etc.

Study: Five million nights: temporal dynamics in human sleep phenotypes. Image Credit: New Africa/Shutterstock.com

Background

The rapid advancements in wearable device technology have made wearable health monitoring devices easily available and affordable. Apart from various other health parameters, these devices are widely used to monitor sleep patterns and quality.

However, despite the abundance of sleep monitoring data, converting the insights drawn from this data into actionable changes has been challenging due to the variability in sleep parameter combinations across individuals and within individuals across time.

The National Institutes of Health recommendations state that adults should get seven to nine hours of monophasic sleep every day. However, sleep studies have shown that sleep structures vary significantly in length and quality, and these variations are associated with lifestyle- and health-related factors.

Studies that have used clustering analyses for large-scale sleep data to quantify variations in sleep characteristics have been effective in characterizing sleep phenotypes but have only used cross-sectional data and have not considered the inferences that can be gained about illness and health from longitudinal sleep data.

About the study

In the present study, the researchers used a large dataset of sleep monitoring wearable data for over 33,000 individuals, adding up to over 5 million nights of data to determine changes in sleep phenotype over time. They also aimed to understand whether these changes were informative about health parameters or periods of acute illness.

The sleep periods in the large dataset were treated independently. Through clustering analyses, the researchers obtained a set of sleep phenotypes, including the insomnia-like phenotype, which consisted of segmented sleep of less than 6.5 hours a day, and the recommended sleep phenotype of 8 hours of monophasic sleep.

The relevance of these phenotypes and changes in sleep phenotypes in disease and health were tested by examining whether the transition probability patterns in a cohort of chronically ill individuals differed significantly from those in a healthy cohort.

The researchers also tested whether the transition probability patterns differed before and after illness in the same individuals.

The data was collected through self-reported survey responses and sleep-wake time series from 33,152 individuals over ten months. Sleep monitoring data from a wearable smart-ring device was also obtained from all the participants.

The data was divided into sleep periods, which were periods of non-overlapping three to six consecutive nights, which were then used to determine sleep phenotypes through clustering analyses.

The characteristics typical for each sleep period were used to identify the predefined sleep phenotype clusters. The patterns of transitions and distribution of sleep periods over time for each individual were used to determine changes in sleep phenotype.

Furthermore, the study also examined the distribution of shifts in sleep phenotypes among groups of individuals with sleep apnea, flu, diabetes, fever, and COVID-19.

Results

The results reported 13 sleep phenotypes linked to the quality and duration of sleep and found evidence of transitions between sleep phenotypes in an individual over time.

Furthermore, the patterns of sleep phenotype transitions showed significant differences between groups of individuals with and without chronic illnesses or health conditions and within an individual over time.

The study found that not only were sleep phenotypes of an individual dynamic, but the alterations in sleep phenotypes were informative about health conditions.

Furthermore, the assessment of temporal dynamics of sleep patterns revealed that current sleep patterns were indicative of potential changes in sleep phenotypes. For example, shorter periods of deep sleep indicated a shift to an insomnia-like sleep phenotype.

The temporal dynamics of sleep phenotype transitions were also found to indicate an individual’s chronic illness or health-related factors. The dynamic transition model was found to be more informative than the specific sleep phenotypes about an individual’s respiratory and cardiometabolic health factors.

Conclusions

Overall, the study identified 13 sleep phenotypes associated with the duration and quality of sleep and found that these phenotypes changed across individuals based on health conditions and within an individual over time.

The temporal transition patterns in sleep phenotypes also indicated chronic disease conditions such as respiratory and cardiometabolic illnesses.

These findings highlight the importance of longitudinal sleep analyses and temporal dynamics assessments in drawing actionable inferences from wearable sleep monitoring data.