Table 5.
Fitbit derived behavioral activity features.
| Steps activity features | Sleep activity features | Heart rate features | ||
|---|---|---|---|---|
| Percent minutes with steps | Maximum steps in 30-min period | Time of day of last step | Main sleep minutes | Resting heart rate (beats per minute) |
| Percent minutes with 1–30 steps | Maximum steps in 60-minute period | Time of day of 6 min max active | Total minutes spent in bed | Skewness of heart rate |
| Percent minutes with 30–100 steps | Average streak duration (minutes) | Time of day of 30 min max active | Total minutes asleep in day | Kurtosis of heart rate |
| Percent minutes with >100 steps | Maximum streak duration (minutes) | Time of day of 60 min max active | Sleep efficiency score (Fitbit) | Max rolling 5-min average |
| Steps taken per day | Average streak (steps) | Skewness of step distribution (all values) | Percent main sleep interval spent really awake | 95th percentile of heart rate |
| Length of longest inactivity period (<1 step) | Maximum streak (steps) | Skewness of nonzero step distribution | Minutes during main sleep spent awake | |
| Length of longest inactivity period (<5 step) | Number of nonzero step streaks | Kurtosis of step distribution (all values) | Number of naps in a day | |
| Maximum steps in 6-min period | Time of day of first step | Kurtosis of nonzero step distribution | Number of contiguous wakeful states | |
| Time went to sleep |
Data streams from the Fitbit API were downloaded at minute-level granularity, and then aggregated into day-level features with input from a medical domain expert.