Blog 1 - Verisense Data Processing
In the first of our blog series our lead data scientist Dr. Matt Patterson gives an overview on how data from Verisense is processed and the benefits of this.
The Verisense sensor collects raw acceleration data from the wrist and transmits it to the Verisense cloud platform via the Verisense base station. The raw data is processed in the cloud to obtain activity and sleep metrics for the time in which the patient wore the sensor. Site managers have access to both the raw data and the processed data. This is very beneficial for data transparency, future algorithm development and re-processing raw data with updated algorithms in years to come.
Verisense uses proven algorithms to obtain activity and sleep data from the raw acceleration signal that comes from the wrist sensor. Sedentary time and activity levels are calculated by using a cut point analysis on the combined acceleration signal. Sleep time is obtained by looking for sustained periods of nocturnal inactivity based on the angle of the forearm.
There are a significant amount of metrics provided in the Verisense output. Understanding them all can be tricky, but at the highest level, Verisense provides day-time metrics and night-time metrics. Day-time metrics include sedentary time as well as various levels of physical activity. These are also provided in continuous periods of time, so they conform more closely to governmental health agency guidelines on exercise. Night-time metrics include bed time, wake up time, amount of time in bed as well as sleep efficiency. Many more metrics are provided, but these metrics are a good place to start.
The Verisense Base Station is a smart phone device that receives data from the Verisense Sensor, and using mobile data or WiFi networks automatically uploads this data to the Verisense Cloud platform.
Ideally, at least seven full days of patient wear should be obtained. This way, week-day and week-end behaviour can be analysed. The absolute minimum amount of data required for one night and one days worth of data is 36 hours. Non-wear time is automatically calculated by searching for 15 minute chunks of no movement in the data. Non-wear time periods have activity imputed from the average of other days at the same time.
Example output from the Verisense system
Have some questions? Our second blog post answers questions we frequently receive from users on data processing and metrics. Read Here.
