Automatic time in bed detection from hip-worn accelerometers for large epidemiological studies: The Tromsø Study.

Area of Science:

• Epidemiology
• Biomedical Engineering
• Data Science

Background:

• Accelerometers are vital for assessing physical activity in large epidemiological studies, typically worn on the wrist or hip.
• While wrist-worn devices are used for sleep analysis, hip-worn accelerometers have been underutilized for this purpose.
• Existing methods often neglect the potential of hip-worn accelerometers for detailed sleep and sedentary behavior analysis.

Purpose of the Study:

• To develop and validate a novel method for identifying time in bed using hip-worn accelerometer data.
• To enable advanced analyses of time in bed and sedentary behavior in large-scale epidemiological studies.
• To leverage existing datasets collected with hip-worn accelerometers for sleep-related research.

Main Methods:

• Developed accelerometer-specific data augmentation techniques (e.g., mimicking incorrect wear, adding noise, random cropping) to improve model training.
• Trained a neural network model on data from the Tromsø Study.
• Evaluated the model's performance on independent datasets, including data from a demographically different population.

Main Results:

• The algorithm achieved 94% accuracy on training data and 92% on unseen data from the same population.
• Comparable results were obtained when compared to consumer-wearable data from a different demographic group.
• While generalization was good, some instances of over/underestimation of time in bed were observed for specific days or participants.

Conclusions:

• The developed method offers a promising approach for identifying time in bed from hip-worn accelerometer signals.
• This technique can facilitate the re-analysis of existing data for longitudinal studies focusing on sleep and sedentary behavior.
• It serves as a foundation for developing more sophisticated algorithms to analyze sleep patterns from hip-mounted sensors.