A Preliminary Study for Development of A New Identification Algorithm for Objectively Measured Sedentary Behavior

Introduction Sedentary behavior (SB) has been defined as any purposefully engaged activity that is mainly dominated by prolonged sitting with low energy expenditure (METs<1.5). Objective measures (e.g., accelerometer) has widely been used for monitoring SB in free-living settings; however, relatively little effort has been made to develop an appropriate algorithm to identify true SB bout in accelerometry data.

Purpose The primary aim of this pilot study is to develop SB identification algorithm for accelerometry data collected in free-living settings.

Methods Two of the primary investigators wore an accelerometer (ActiGraph GT1M) on the waist and engaged in 6 consecutive activities for 30 to 60 minute durations in a free-living setting that include 3 SBs (i.e., watching TV, reading, and computer use) with typical sedentary break (SBK) activities (i.e., standing) and 3 light to moderate intensity physical activities (i.e., house work and slow walking). The starting and ending times for each activity including the times for SBK were also recorded. SB patterns in accelerometry data were visually scrutinized and SAS v9.3 was used to create an automated macro algorithm to identify true SB bouts using a threshold of <100 counts per minute (cpm) as a potential sedentary minute.

Results A new algorithm developed to identify a true SB bout was featured by the following conditions; 1) start the first screening window at the minute when activity counts <100 cpm; 2) within the first screening window allow up to 2 consecutive minutes with >99 cpm (condition 1); 3) start a second screening window up to 10 minutes after the last minute of allowed consecutive minutes with >99 cpm (condition 2) [50% of the second screening window has to be sedentary in order to continue the first screening window]; and 4) start a third screening window across entire bouts observed from first screening window to check the existence of 5 consecutive sedentary minutes (condition 3), when the first screening window is closed (i.e., failed to meet condition 1 and 2). A duration that meets all of conditions was considered a SB bout. Using the new algorithm, an average of 97.8% of true SB times across 3 SBs for two accelerometry data were correctly identified (1.1% of misidentification of non-SB as SB).

Conclusions The present study is a preliminary study to develop the new SB algorithm for accelerometry data. A relatively high accuracy for identification of true SB bouts could be achieved by applying sequential screening processes in the algorithm. A future study is warranted to investigate the patterns of SB and to develop more elaborate screening processes based on a large sample size.