Lessons learned on the road to improve sleep data extracted from a Fitbit device

Extract

Dear Editor,

Wearable technologies developed for personal tracking of sleep and activity by consumers (consumer sleep technologies, CSTs) are becoming increasingly utilized as research tools. After the discontinuation of the Philips Respironics line of Actiwatches, a growing number of sleep researchers, including our team, adopted CST as a substitute tool to objectively estimate sleep.

Among the available CST, Fitbit is the most frequently used device in sleep research [1] and classifies sleep with proprietary algorithms applied to activity and heart rate data. Initial hesitation surrounding the use of Fitbit and other CST has been somewhat mitigated with increasing evidence that a variety of Fitbit models can classify polysomnography-defined sleep–wake, with superior performance as compared with research-grade actigraphy devices which use only activity to estimate sleep [2]. However, important differences exist between sleep tracking with CST and research-grade actigraphs, including the sensors used and the acquisition, processing, availability, and presentation of data, which has been detailed elsewhere [3]. Despite this, detailed and streamlined procedures for the use of CST in research, comparable to the SBSM Guide to Actigraphy Monitoring [4], are not available to our knowledge. Here, we have outlined our logistical and operational considerations for the use of CST, specifically the Fitbit Charges 5 and 6 (Table 1). This includes consideration of features of the Fitbit device itself, recommended accessory purchases, details of recommended settings on the Fitbit mobile app, syncing from the app to the cloud, use of Fitabase in data extraction, and recommendations for the recalculation of sleep parameters. We also provide code to automatically extract nightly summary sleep parameters (e.g. sleep onset time, final wake time, total sleep time, and sleep percentage) from the Fitabase epoch-by-epoch sleep–wake data export (Table 1). This guidance, based on our experiences and challenges utilizing Fitbit data, may benefit other study teams collecting sleep parameters with a Fitbit device; however, these solutions are specific to our work and may not generalize to other situations, highlighting the need for a diverse toolkit of vetted methods to process CST data. Additionally, we describe two key issues that we have encountered in analyzing the Fitbit data that may reduce the rigor and reproducibility of research with CST and outline our current approaches to them.