Development and application of an automated algorithm to identify a window of consecutive days of accelerometer wear for large-scale studies

Abstract

Background: Some accelerometer studies ask participants to document in a daily log when the device was worn. These logs are used to inform the window of consecutive days to extract from the accelerometer for analysis. Logs can be missing or inaccurate, which can introduce bias in the data. To mitigate this bias, we developed a simple computer algorithm that used data within the accelerometer to identify the window of consecutive wear days. To evaluate the algorithm's performance, we compared how well it agreed to the window of days identified by visual inspection and participant logs.

Findings: Participants were older women (mean age 79 years) in a cohort study that aimed to examine the relationship of objective physical activity on cardiovascular health. The study protocol requested that participants wear an accelerometer 24 h per day over nine calendar days (to capture seven consecutive wear days) and to complete daily logs. A stratified sample with (n = 75) and without (n = 100) participant logs were selected. The Objective Physical Activity and Cardiovascular Health (OPACH) algorithm was applied to the accelerometer data to identify a window of up to seven consecutive wear days. Participant logs documented dates the device was first put on, worn, and removed. Using pre-established guidelines, two independent raters visually reviewed the accelerometer data and characterized the dates representing up to seven consecutive days of 24-h wear. Average agreement level between the two raters was 90%. The percent agreement was compared between the three methods. The OPACH algorithm and visual inspection had 83% agreement in identifying a window with the same total number of days, if one or more shifts in calendar dates were allowed. For visual inspection vs. logs and algorithm vs. logs, this agreement was 81 and 74%, respectively.

Conclusion: The OPACH algorithm can be efficiently and readily applied in large-scale accelerometer studies for the identification of a window of consecutive days of accelerometer wear. This algorithm was comparable to visual inspection and participant logs and might provide a quicker and more cost-effective alternative to selecting which data to extract from the accelerometer for analysis.

Trial registration: clinicaltrials.gov identifier: NCT00000611.

Figures

Figure 1

Example of an annotated accelerometer signal used during visual inspection. Each box represents a day, with time on the x-axis and total counts on the y-axis. Day 5 would be considered the first day and day 11 would be considered the last day of the wear window, for a maximum 7 consecutive days of wear. Day 6 would best represent the first day of wear documented in the log.