Resources required
In addition to the monitors, sufficient computers or laptops are required to initialise and download the accelerometers. Additional docking stations may be needed to recharge the accelerometers. In the field, attention must be paid to the logistics of distributing and collecting and re-distributing the accelerometers.
Administration of accelerometers
If there is going to be a time lag between initialising the monitor and participant wearing the device make sure that the data analysis program can handle this delay. Think about how accelerometers are going to be distributed and returned e.g. face-to-face, by mail? Be aware that late or non-returns impact on the number of instruments available. Are participants going to be asked to keep concurrent activity logs or detail wearing and non-wearing times? And how will this information be linked to the accelerometer data?
Compliance
Compliance may be low in some populations, e.g. adolescents. The following points may enhance compliance:
- Show participants graphical output data which will includes non-wearing time
- Provide incentives for wearing the accelerometer for the desired period
- Provide encouragement and support for participants through phone calls, SMS or email messages
- Logs of wearing may help self-monitoring
- Provide clear instructions
- Enlist the support of others such as teachers, parents, family members
- Investigate and overcome any barriers to wearing, e.g. a waist belt may not feel comfortable in obese subjects.
Decision rules
Ward et al (2005) comprehensively described a series of decision rules for accelerometry data reduction, which provides an excellent starting point for any researcher and these are summarised below.
1. Defining a day – what constitutes a day? Waking time, a specific period of time, or percentage of time for an individual’s given day? The 70/80 rule has been suggested; a day is defined as the period which at least 70% of the population have recorded data and 80% of that period constitutes a minimal day for inclusion in data analysis (Catellier et al, 2005).
2. Identification of wearing period, i.e. distinguishing between removing accelerometer during water based activity, genuine inactivity and omission is difficult. Different algorithms may be used for different periods of continuous zero movement to account for inactivity, e.g. 20 mins and 60 mins.
3. Handling missing data is a difficult issue given the irregular pattern and variability of physical activity. Using days where measurements are missing may underestimate physical activity, while leaving out such days may introduce selection bias.
4. There is debate in the literature about what constitutes a bout of activity, e.g. various health authorities have recommended 10 continuous minutes of physical activity as a bout. When analysing bout data, algorithms that allow for 1-2 minute interruptions in the bout is recommended.
5. Data cleaning should identify spurious data, which may have resulted from the accelerometer malfunctioning or the participant interfering with the instrument.
6. It is good research practice to report wearing time and the number of interruptions observed and any decisions that were made to handle assumed non-wearing time.
The impact of using different ‘decision rules’ has been investigated when four data reduction algorithms were used to analyse the same data set and compared (Masse et al, 2005).
- More stringent inclusion criteria of ‘valid data’ had a significant negative influence on sample size.
- Average activity counts per minute and average activity counts per day were most significantly affected by varying the periods of accelerometer activity included in the dataset.
- Shorter interval of continuous bouts of zeros, increasing wearing time, and increasing minimum number of valid days required were consistently associated with an increase in light activity and moderate physical activity.
- Outcome variables were affected differently by the use of different algorithms.
- Varying other decision rules such as cut-offs for sedentary, light, moderate and vigorous activity may further exacerbate the differences observed (Masse et al, 2005).
Data handling
The data produced from accelerometry are complex and lengthy but freely available programmes exist for data reduction and analysis (see examples and links section for downloadable programs). One of the most challenging issues when working with accelerometry output data is managing ‘missing data’ i.e. periods of time quantified as zero activity which is difficult to interpret. The first decision is whether it is actually missing data rather than a sustained period of inactivity. Concurrent completion of an activity log can help identify reasons for non-wearing of the accelerometer and help in ascribing appropriate values to these times. There are two basic methods of data imputation: interpolation and average replacement (Catellier et al, 2005). Imputation may work better on weekdays than weekend days as they tend to have lower levels of missing data. Imputation is essentially using existing data to better estimate what is missing. How close the estimate of the missing value is to its true value depends on how many predictors are used and the correlation with the missing variable (Catellier et al, 2005).
More sophisticated approaches to data processing have been suggested to detect a multidimensional movement signature and assign a degree of membership to a (limited) set of activity types. Traditionally, activity levels have identified activities of similar total acceleration but which may have different energy costs. This resulting estimation error may be overcome by first making an inference of activity type (Prober et al, 2006).