GGIR does not have a build-in step detection algorithm, but does facilitate the embedding of an external function for event detection such as steps. Such algorithms typically count steps per epoch, where an epoch is usually a 5 second or larger time interval. Steps and cadence per epoch alone is not of direct value for research. Therefore, GGIR facilitates the extraction of summary statistics, which will be discussed below.
The only external algorithm for step detection currently available for GGIR is the Verisense algorithm, which was design for data collected on the wrist. The algorithm has been described and evaluated in studies by Maylor 2022 and Rowlands 2022.
The original Verisense algortihm code is not actively maintained at the time of writing this documentation. An improved copy of the Verisense algortihms code with minor bug fixes is part of the GGIR GitHub repository.
To use the algorithm, copy-paste the following code to the top of your R script and update the file path in the first line to the Verisense function file on your computer. A discussion of this code will follow further down.
source("C:/path_to_file/verisense_count_steps.R")
myfun = list(FUN = verisense_count_steps,
parameters = c(4, 4, 20, -1.0, 4, 4, 0.01, 1.25),
expected_sample_rate = 15,
expected_unit = "g",
colnames = "step_count",
outputres = 1,
minlength = 1,
outputtype = "numeric",
aggfunction = sum, # aggregate step by taking sum
timestamp = FALSE,
reporttype = "event",
ilevels = c(0, 100, 250), # acceleration levels to be used
clevels = c(0, 80, 100, 120), # cadence levels to be used
qlevels = c(0.5, 0.9), # quantiles to be used
ebout.dur = c(1, 5, 10), # event bout duration to extract
ebout.th.cad = 30, # event bout threshold for cadence
ebout.th.acc = 50, # event bout threshold for acceleration
ebout.criter = 0.8, # event bout criteria (same as boutcriter)
ebout.condition = "AND") # event bout logic (see below)
Note the that
parameters = c(4, 4, 20, -1.0, 4, 4, 0.01, 1.25) is based
on Rowlands et al. “Stepping up with GGIR” from 2022.
Next, use GGIR as you normally do, but include
myfun = myfun as input argument. If we would rely on all
GGIR default parameter values and segments days in 6 equal parts then
this would look as follows:
GGIR(myfun = myfun,
outputdir = "C:/myresults",
datadir = "C:/mydata",
qwindow = c(0, 4, 8, 12, 16, 20, 24))All step and cadence statistics are derived as summary per recording day and per recording as a whole, and stored in results/part2_dayeventsummary.csv and results/part2_eventsummary.csv, respectively. The column name extensions AD_, WD_, WE_, WWD_, and WWE_ inside the file part2_eventsummary.csv have the same meaning as elsewhere in the GGIR part2 output, e.g. AD means aggregation across all recording days.
When you use GGIR’s day segment analysis functionality the statistics are also derived per time segment within a day. The time segment for which each statistic is derived is clarified in the ending of a column name as _startHour-endHourhr. For example, _0-14hr indicates that it was extracted from data between midnight and 14:00.
Total step count is stored with tot_step_count in the column name. Cadence is derived in the unit ‘steps per minute’, even if the epoch size for step detection is not 1 minute. Cadence is abbreviated as cad in the column name. Mean cadence is indicated as mn_cad.
Note that Chapter 6 discusses how acceleration metrics are imputed based on a mean of valid time points on other days of the recording. For steps and cadence this would be problematic as the average between walking and non-walking is not informative. Therefore, imputation of steps is done by the median instead and by that represents typical stepping behaviour at that time of the day.
As discussed in Chapter 7 accelerometer data can be described in terms of acceleration levels. As such, we can also describe step count and cadence per acceleration level.
To control this aspect, include in your myfun object the
item ilevels as a numeric vector with the acceleration
values to define the acceleration levels,
e.g. ilevels= c(0, 50).
Output:
Mean cadence for the acceleration range 0-50 m_g_ when working
with acceleration metric ENMO will be stored with column
name mn_cad_acc0-50mg_ENMO. Note: The step bout
detection functionality automatically uses the acceleration metrics
specified by the user. So, if you want to try it out with both ENMO,
ENMOa, and MAD metric then that should work.
Total step count for the acceleration range 0-50m_g_ when working
with acceleration metric ENMO will be stored with column
name tot_step_count_acc0-50mg_ENMO. Here
step_count is the value as provided via
myfun parameter colnames.
Similarly, we can also describe the data based on cadence level and use absolute cadence thresholds.
To control this aspect, include in your myfun object
item clevels as a numeric vector with the cadence values
that define the cadence levels,
e.g. clevels = c(0, 30, 120).
Output:
Instead of setting absolute thresholds to define cadence levels we can use percentiles.
To control this aspect, include in your myfun object the
item qlevels as a numeric vector with fractions of 1 to
define the quantiles, e.g. qlevels = c(0.5, 0.9).
Output:
cad_p50 for qlevels = 0.5 and
cad_p90 for qlevels = 0.9.The same statistics can also be derived from the consecutive X hours with the least and most cadence per (segment of a) day.
To control this aspect, include in your myfun object the
item winhr, e.g. winhr = c(5, 10) which works
the same as winhr
in GGIR itself.
Output:
L10_cad_.M5_cad_.This is then followed by:
Similar to the detection of physical activity bouts as discussed in Chapter 11 we can detect bouts of stepping behaviour. In GGIR we define these based on duration, minimum cadence AND/OR minimum acceleration and the fraction of the bout for which these criteria need to be met.
The step bout detection functionality automatically uses all acceleration metrics specified by the user. So, if you want to try it out with both ENMO, ENMOa, and MAD metric then that should work.
To control the criteria for bout detection, include in your
myfun object the items:
ebout.dur a numeric vector of length 1 or larger with
the minimum bout duration(s) of interest,
e.g. ebout.dur = c(1, 5, 10).ebout.th.cad a single number being the minimum cadence
value in steps per minute, e.g. ebout.th.cad = 30.ebout.th.acc a single number being the minimum
acceleration value in steps per minute,
e.g. ebout.th.acc = 50.ebout.criter a single number as a fraction of 1 being
the faction of a bout for which the inclusion criteria need to be met
identical to boutcriter in the context of physical activity
bouts, e.g. ebout.criter = 0.8.ebout.condition whether cadence and acceleration
condition are both required (ebout.condition = "AND") or
either of the conditions is required
(ebout.condition = "OR"). If both cadence and acc need to
meet a thresholds then fill in "AND". If it also acceptable
if only one of the thresholds is met then fill in "OR". If
you do not want cadence or acceleration to be used in the equation then
simply set ebout.th.cad = 0 or
ebout.th.acc = 0.Output:
Columns names related to detected bouts start with the word Bout_, e.g. “Bout_meandur_E5S_B10M80%_cadT30_AND_accT50_ENMO”, “Bout_number_E5S_B5M80%_cadT30_AND_accT50_ENMO” and “Bout_totdur_E5S_B5M80%_cadT30_AND_accT50_ENMO”.