Sampling frequency affects ActiGraph activity counts

Introduction: The ActiLife Data Analysis Software processes ActiGraph accelerometer data and has a bandpass filter attenuating accelerations falling outside the normal human frequency passband considered to be 0.25-2.5 Hz. This frequency passband disfavour vigorous physical activity that is normally performed at frequencies higher than 2.5 Hz. With the ActiGraph model GT3X one has the option to select sample frequency from 30 to 100 Hz. This study investigated the effect of the sampling frequency on the ouput of the bandpass filter.Methods: A synthetic frequency sweep of 0-15 Hz was generated in Matlab and sampled at frequencies of 30-100 Hz. Also, acceleration signals during indoor walking and running were sampled at 30 Hz using the ActiGraph GT3X and resampled in Matlab to frequencies of 40-100 Hz. All data was processed with the ActiLife software.Results: Acceleration frequencies between 5-15 Hz escaped the bandpass filter when sampled at 40, 50, 70, 80 and 100 Hz, while this was not the case when sampled at 30, 60 and 90 Hz. During the ambulatory activities this artifact resultet in different activity count output from the ActiLife software with different sampling frequency. The difference increased with increasing activity intensity, with up to 1000 counts per minute at fast running.Discussion & conclusions: Activity counts from vigorous physical activity is highly attenuated with the ActiLife software. High frequency movement and noise information escape the bandpass filter depending on the sampling frequency of data collection, adding unexplained variation in activity counts. Therefore, the choice of sampling frequency may be an additional source of error with large impact on phsyical activty research as many investigators are using the ActiGraph with the ActiLife software. Accordingly, the outcome from and comparability between studies can be questioned.