Blog 2 - FAQs on Data Processing
In the second of our blog posts, we build on the topic of data processing by answering common queries that you want us to address
1. What is a defined as a valid day?
A valid day is defined by the existence of more than 16 hours of wear time in a midnight to midnight range. Ideally, the participant would have closer to 24 hours of wear time in a day. Such low hours of wear time are still accepted because otherwise, the 16 or more hours of wear time would be lost. The missing data on a given day (due to non-wear) is imputed from the average of the value at the same time on other days from the recording.
2. Are the Verisense algorithms proven?
Verisense algorithms have been used extensively by academic and clinical researchers since 2014. They have been used to process data from nearly 200,000 patients across many different patient cohorts including dementia, obesity, stroke, cystic fibrosis and coronary artery disease, just to name a few. Verisense raw data processing algorithms have been validated for sleep period window detection, sleep efficiency and sedentary and physical activity level classification.
3. How much data is needed – no. of days – for a valid study?
At the very minimum 36 hours of recording is required to get one night and one day worth of data. However, this may not be fully reflective of the patients behaviour in the home and community setting. Most studies collect data for at least a full week to get weekday data and weekend data, as these often have different characteristics.
4. How is non-wear determined?
Non-wear periods of time are found by looking for 60 minute periods of very little movement in the Verisense accelerometer readings. If the 60 minute period meets the criteria, then the middle 15 minutes of that period is classified as non-wear. This was found to be the best way to avoid erroneously classifying sleep and sedentary periods as non-wear periods. Very short periods of non-wear (ie, 5min) cannot be detected.
5. How is sleep calculated?
Sleep onset time and wake time are calculated by using an adaptive threshold on the fore-arm angle, as measured from the Verisense sensor while the arm is relatively still. Since movement is the only factor considered in the measurement, it can sometimes happen that sedentary behaviour before or after sleep is erroneously classified as sleep.
6. How is sleep efficiency calculated?
Once sleep onset and wake time are obtained, the time between these two points is known as the sleep period window. Five minute chunks within the sleep period window are classified as either nocturnal sleep or nocturnal wake. This is done based on the fore-arm angle deviation over each 5 minute period of time. When there is a lot of movement, the period is classified as nocturnal wake and when there is little movement, the period is classified as sleep. Sleep efficiency is the percentage of nocturnal sleep time within the sleep period window.
7. How do you calculate between different activity intensities?
A combined signal (ENMO) is calculated from the raw tri-axial acceleration data that comes off the Verisense sensor. A cut-point approach is then applied to this signal to classify periods into sedentary, light, moderate or vigorous physical activity. These cut points on the ENMO signal have been validated in several studies (Hildebrand et al, 2014 & Hildebrand et al, 2016).
Activity metrics can be accessed as total time in an activity level over the course of a day. However, as health guidelines generally recommend continual exercise over a period time, it is often convenient to look at activity that take place over a set period of time. Verisense provides light, moderate, vigorous and moderate-to-vigorous physical activity in the following time periods: 1-10min, 10-30min and over 30 min. This is also provided for MVPA (moderate-to-vigorous physical activity), which is all activity over the moderate activity threshold.
8. Where can I find more information on how data is processed?
This paper provides an overview of how the Verisense algorithms works. For a detailed description of the algorithms used to obtain activity and sleep metrics from Verisense please refer to this page and this pdf document.
9. How is data presented?
Processed results for each participant in a trial are available in .csv format. Different .csv files are created for different aspects of analysis. Numerical results from the .csv files should be used in studies. Visualizations of each participants information is also generated and saved into .pdf format. Samples of raw data and participant reports can be found here.
Example output from the Verisense system
