Accelerometers are commonly used in clinical and epidemiological research for more detailed measures of physical activity and to target the limitations of self‐report methods. Sensors are attached at the hip, wrist and thigh, and the acceleration data are processed and calibrated in different ways to determine activity intensity, body position and/or activity type. Simple linear modelling can be used to assess activity intensity from hip and thigh data, whilst more advanced machine‐learning modelling is to prefer for the wrist. The thigh position is most optimal to assess body position and activity type using machine‐learning modelling. Frequency filtering and measurement resolution needs to be considered for correct assessment of activity intensity. Simple physical activity measures and statistical methods are mostly used to investigate relationship with health, but do not take advantage of all information provided by accelerometers and do not consider all components of the physical activity behaviour and their interrelationships. More advanced statistical methods are suggested that analyse patterns of multiple measures of physical activity to demonstrate stronger and more specific relationships with health. However, evaluations of accelerometer methods show considerable measurement errors, especially at individual level, which interferes with their use in clinical research and practice. Therefore, better objective methods are needed with improved data processing and calibration techniques, exploring both simple linear and machine‐learning alternatives. Development and implementation of accelerometer methods into clinical research and practice requires interdisciplinary collaboration to cover all aspects contributing to useful and accurate measures of physical activity behaviours related to health.

ActiGraph is the most common accelerometer in physical activity research, but it has measurement errors due to restrictive frequency filtering. This study investigated biomechanically how different frequency filtering of accelerometer data affects assessment of activity intensity and age-group differences when measuring physical activity. Data from accelerometer at the hip and motion capture system was recorded during treadmill walking and running from 30 subjects in three different age groups: 10, 15, and >20 years old. Acceleration data was processed to ActiGraph counts with original band-pass filter at 1.66 Hz, to counts with wider filter at either 4 or 10 Hz, and to unfiltered acceleration according to “Euclidian norm minus one” (ENMO). Internal and external power, step frequency, and vertical displacement of center of mass (VD) were estimated from the motion capture data. Widening the frequency filter improved the relationship between higher locomotion speed and counts. It also removed age-group differences and decreased within-group variation. While ActiGraph counts were almost exclusively explained by VD, the counts from the 10 Hz filter were explained by VD and step frequency to an equal degree. In conclusion, a wider frequency filter improves assessment of physical activity intensity by more accurately capturing individual gait patterns.

In objective physical activity (PA) measurements, applying wider frequency filters than the most commonly used ActiGraph (AG) filter may be beneficial when processing accelerometry data. However, the vulnerability of wider filters to noise has not been investigated previously. This study explored the effect of wider frequency filters on measurements of PA, sedentary behavior (SED), and capturing of noise. Apart from the standard AG band-pass filter (0.29–1.63 Hz), modified filters with low-pass component cutoffs at 4 Hz, 10 Hz, or removed were analyzed. Calibrations against energy expenditure were performed with lab data from children and adults to generate filter-specific intensity cut-points. Free-living accelerometer data from children and adults were processed using the different filters and intensity cut-points. There was a contribution of acceleration related to PA at frequencies up to 10 Hz. The contribution was more pronounced at moderate and vigorous PA levels, although additional acceleration also occurred at SED. The classification discrepancy between AG and the wider filters was small at SED (1–2%) but very large at the highest intensities (>90%). The present study suggests an optimal low-pass frequency filter with a cutoff at 10 Hz to include all acceleration relevant to PA with minimal effect of noise.

An improved method of physical activity accelerometer data processing, involving a wider frequency filter than the most commonly used ActiGraph filter, has been shown to better capture variations in physical activity intensity in a lab setting. The aim of the study was to investigate how this improved measure of physical activity affected the relationship with markers of cardiometabolic health. Accelerometer data and markers of cardiometabolic health from 725 adults from two samples, LIV 2013 and SCAPIS pilot, were analyzed. The accelerometer data was processed using both the original ActiGraph method with a low-pass cut-off at 1.6 Hz and the improved method with a low-pass cut-off at 10 Hz. The relationship between the physical activity intensity spectrum and a cardiometabolic health composite score was investigated using partial least squares regression. The strongest association between physical activity and cardiometabolic health was shifted towards higher intensities with the 10 Hz output compared to the ActiGraph method. In addition, the total explained variance was higher with the improved method. The 10 Hz output enables correctly measuring and interpreting high intensity physical activity and shows that physical activity at this intensity is stronger related to cardiometabolic health compared to the most commonly used ActiGraph method.

This review re‐examines the use of accelerometer and oxygen uptake data for the assessment of activity intensity. Accelerometers capture mechanical work, while oxygen uptake captures the energy cost of this work. Frequency filtering needs to be considered when processing acceleration data. A too restrictive filter attenuates the acceleration signal for walking and, to a higher degree, for running. This measurement error affects shorter (children) more than taller (adults) individuals due to their higher movement frequency. Less restrictive filtering includes more movement‐related signals and provides measures that better capture mechanical work, but may include more noise. An optimal filter cut‐point is determined where most relevant acceleration signals are included. Further, accelerometer placement affects what part of mechanical work being captured. While the waist placement captures total mechanical work and therefore contributes to measures of activity intensity equivalent by age and stature, the thigh and wrist placements capture more internal work and do not provide equivalent measures. Value calibration of accelerometer measures is usually performed using measured oxygen uptake with the metabolic equivalent of task (MET) as reference measure of activity intensity. However, the use of MET is not stringent and is not a measure of activity intensity equivalent by age and stature. A candidate measure is the mass‐specific net oxygen uptake, VO2net (VO2tot − VO2stand). To improve measurement of physical activity intensity using accelerometers, research developments are suggested concerning the processing of accelerometer data, use of energy expenditure as reference for activity intensity, and calibration procedure with absolute versus relative intensity.