Common mistakes when using ActiGraph and how to avoid them

Written by human movement scientist Carmen Brouwers

The use of accelerometers such as those from ActiGraph has increased significantly in both research and applied settings over the past years. This is not surprising: they provide an objective way to measure physical activity, whereas traditional questionnaires can introduce bias.

They are also particularly useful in populations where questionnaires are not feasible or reliable, such as children, animals, or patients with conditions like Parkinson’s disease, dementia, or ALS/MS.

At the same time, this comes with a challenge. While the threshold to start measuring is low, the risk of misinterpretation is high. Small decisions in setup and analysis can have a major impact on your results. In this blog, I discuss several common mistakes and how to avoid them.

Sensor placement

A first important aspect is sensor placement. Whether a device is worn on the hip, wrist, or ankle makes a significant difference in the data you collect. These positions are not interchangeable within a study. The advantage of these sensors is that they can be worn at multiple locations, but it is essential to carefully consider why a specific placement is chosen. For physical activity assessment, the hip is generally recommended, as it best represents whole-body movement. Wrist placement may overestimate activity and intensity, since arm movements during sedentary behavior can still generate signals. For sleep measurement, however, the wrist is often preferred, as it better captures restlessness during the night.

In practice, it is sometimes not possible for participants to wear the device at a specific location. This is not necessarily a problem, as long as the placement is consistent across participants and within a study. Consistency is key for comparability.

Epoch length

Another commonly underestimated factor is epoch length. This refers to the time window over which raw acceleration data is summarised. Longer epochs (e.g. 60 seconds) provide a more general overview but may miss short bursts of intense activity. Short epochs (e.g. 1 second) provide more detail but also introduce more noise and increase data complexity. The optimal choice depends on the research question and population. Most commonly used algorithms are also based on 60-second epochs. For short monitoring periods, such as a few hours, shorter epochs may be preferable. For long-term monitoring over weeks or months, longer epochs are often sufficient.

Cut-points

A classic issue in accelerometer research is the use of cut-points. These are threshold values used to classify activity into intensity levels such as light, moderate, or vigorous. While this is useful for interpreting intensity and duration of activity, cut-points are highly population- and context-specific. Values developed for healthy adults are not necessarily valid for children or clinical populations such as Parkinson’s disease or MS. Despite this, they are still often applied without critical consideration, which can lead to misleading conclusions. It is important to carefully justify the chosen cut-points and apply them consistently.

Non-wear time

Another common source of error is non-wear time detection: periods in which the device is not worn. This step is sometimes overlooked during analysis, which can bias sedentary behaviour estimates. Participants cannot be expected to wear the device 24/7 over long periods, and they may remove or forget it occasionally. Fortunately, non-wear detection can be done using software algorithms that identify periods of prolonged inactivity, as well as through device-based indicators such as skin contact detection. Correctly defining and excluding non-wear time is essential to avoid misclassifying it as sedentary behaviour.

Measurement duration

Measurement duration is another important consideration. One or two days of data rarely provide a representative picture of habitual activity. There is substantial variation between weekdays and weekends, and factors such as holidays or vacation periods should also be considered. Guidelines often recommend a minimum of four to seven valid days of measurement, although this is not always achieved in practice.

Initialization settings

Finally, device initialization settings are sometimes not carefully checked. Parameters such as sample rate, subject IDs, and device configurations in software like ActiLife must be set correctly. These may seem like minor details, but they cannot be corrected afterwards and can significantly affect your dataset. Using standardized templates in ActiLife can help ensure consistency across studies and participants.

In summary, using an accelerometer is not simply a matter of attaching a device and downloading data. It requires careful methodological choices tailored to the research question and population. While it may seem like a lot to consider, once properly set up, it becomes a robust and reliable way of measuring physical activity.

If you have any questions about this or would like to discuss it, feel free to send me an email.