A new study has highlighted the effectiveness of using wearable technology, such as Fitbits, to predict long-term sleep deprivation. A group of researchers from Florida International University and the University of South Florida conducted a study to explore the potential of wearable technology and health surveys to predict sleep patterns over extended periods of time.
The research, published in Computers in Biology and Medicine, focuses on predicting an individual’s average sleep duration across 30 days using physical activity data collected by Fitbit devices. The study utilised machine learning techniques, particularly Recursive Feature Elimination with Random Forest (RFE-RF), to identify which features were most predictive of sleep patterns. These features included not only the physical activity data but also previous sleep patterns and psychological factors such as depression and exercise motivation.
The significance of this study lies in its ability to accurately predict whether an individual will suffer from long-term sleep deprivation, a condition linked to numerous adverse health outcomes, including obesity, diabetes, and cardiovascular disease. The researchers found that previous sleep behaviour and physical exercise were the most relevant predictors. Interestingly, the study also revealed that personality traits and depression levels played a crucial role in determining sleep patterns, with distinct differences observed between male and female participants.
For the analysis, the research team used data from the NetHealth project, which tracked 698 university students wearing Fitbit Charge HR devices over several months. The students also completed various health surveys, which provided additional insights into their mental and physical health. By combining this data, the researchers developed predictive models that performed well in classifying individuals as long-term sleep-deprived or not, achieving an impressive Area Under the Curve (AUC) score of 0.9762 in the best-performing model.
One of the key findings was that previous sleep patterns were more predictive of future sleep for females than males, indicating possible gender differences in sleep behaviour and its determinants. The study also underscored the importance of physical activity, with metrics such as daily steps and time spent in different heart rate zones being consistently selected as important features by the machine learning models.
Beyond physical activity, the study highlighted the relevance of psychological factors. For example, higher levels of depression were found to correlate with poorer sleep outcomes. Moreover, motivation for exercise, measured through a specific questionnaire, emerged as a significant predictor, suggesting that individuals’ reasons for engaging in physical activity might influence their sleep quality and duration.
This research not only advances the understanding of sleep prediction using wearable technology but also opens up new avenues for personalised health interventions. The potential to integrate sleep prediction functionality into mobile health (mHealth) applications could lead to more effective management of conditions like hypertension and mental health disorders, which are closely linked to sleep patterns.
