8. Sleep Fundamentals: Sustained Inactivity Bout Detection (SIB)

Sleep analysis in GGIR comes at three stages:

1. The discrimination of sustained inactivity and wakefulness periods, discussed in this chapter.

2. Identification of time windows that guide the eventual sleep detection, as discussed in the next chapter.

3. Assess overlap between the windows identified in step 1 and 2, which we use to define the Sleep Period Time window (SPT) or the time in bed window (TimeInBed) as discussed in chapter 10.

The term “sleep” is somewhat controversial in the context of accelerometry, because accelerometers can only capture lack of movement, while sleep is not formally defined by a lack of movement. To acknowledge this, GGIR refers to these classified ‘sleep’ periods as sustained inactivity bouts abbreviated as SIB. However, we use the term sleep in GGIR to denote the broader topic of describing the main resting period in the day.

Most SIB detection algorithms are designed for wrist-worn data. No evidence exists that they are informative for other wear locations. However, if your main interest is in daytime behaviour and not in sleep, you can still use them. In that case set parameter relyonguider = TRUE in combination with any of the SIB algorithms described below. GGIR will then rely on the guider (discussed in next chapter) to define the Sleep Period Time window (this is, the main episode of sleep in the day) and by that the waking hours of the day. The sleep analysis based on the SIB detection may not be informative though and best ignored.

GGIR offers the user the choice to identify SIB period with the following algorithms:

SIB: vanHees2015

This algorithm looks for the periods of time where the z-angle does not change by more than 5 degrees for at least 5 minutes. This algorithm was proposed in a 2015 article. The idea behind the algorithm is that it is a more interpretable heuristic compared with the conventional approaches that use the magnitude of acceleration to distinguish sustained inactivity bouts. There is no reason to assume that vanHees2015 is a better or worse reflection of sleep, the advancement is purely intended to be in terms of interpretability. The vanHees2015 algorithm is the default. The values 5 and 5 in the algorithm can be modified with parameters anglethreshold and timethreshold, but we currently see no basis to recommend doing this and advise sticking to default values.

SIB: NotWorn (EXPERIMENTAL)

Disclaimer: The status of this SIB algorithm is experimental because it has not been described and evaluated in a peer-reviewed publication yet. This means revisions to the algorithm can be expected as the algorithm matures.

The algorithms named “NotWorn” for both sib and guider (next chapter) are designed for studies where the instruction is to not wear the accelerometer during the night. It should be obvious that this does not facilitate any meaningful sleep analysis. Nonetheless we need a crude estimate of night time versus day time in order for GGIR part 5 to characterise day time behaviours.

If this is the case for your dataset then use guider setting HASPT.algo = "NotWorn" as discussed in the next chapter. Further, we recommend combining the using of “NotWorn” with: do.imp = FALSE, HASPT.ignore.invalid = NA, and ignorenonwear = FALSE.

The detection of sib periods is based on the acceleration metric as defined with parameter acc.metric.

Combined with count acceleration metrics

If you are using a count accelereration metric then set HASIB.algo = "NotWorn". In part 4 the sib will be set equal to the detected guider window. So, effectively guider and sib algorithm are identical in this case.

Combined with gravitational unit acceleration metrics

If you are using an acceleration metrics the expresses acceleration in gravitational units then set HASIB.algo as you would do if the accelerometer was expected to be worn and specify a second guider for parameter HASPT.algo as discussed int he next chapter, e.g. HASPT.algo = c("NotWorn", "HDCZA)". In this way GGIR will first search for long non-wear periods as indicator of sleep and use those to define the sleep window but if those are not found it will fall back on the sib-algortihm as specified with HASIB.algo, e.g. "vanHees2015".

SIB: Count based algorithms (EXPERIMENTAL)

Disclaimer: The status of this SIB algorithm is experimental because it has not been described and evaluated in a peer-reviewed publication yet. This means revisions to the algorithm can be expected as the algorithm matures.

Accelerometers have been used for sleep research since the 1990s. However, the initial accelerometers did not store their data in gravitational units at sub-second level as we use nowadays but stored their data in 30 or 60 second epoch aggregates. Although these aggregates are referred to as counts by many manufacturers the calculation of counts differs by manufacturer.

We have attempted to facilitate several sleep detection algorithms from the literature in that period such as “Sadeh1994”, “ColeKripke1992”, and “Galland2012”. A problem with all these algorithms is that the preprocessing done to generate the counts is insufficiently described in the literature. For the zero-crossing count as used by Sadeh1994 and ColeKripke1992 an attempt has been made to collect as much information as could be found and made an educated guess on the missing information. The zero-crossing counts are discussed in the chapter 4 about acceleration metrics.

When counts are calculated we can use the following SIB algorithms. For all of these it is uncertain whether the Y-axis direction on modern accelerometers matches the direction of the Y-axis in literature because for old studies the direction of Y-axis was to our knowledge never clarified.

Sadeh1994

The algorithm proposed by Sadeh et al. link. To use this set parameter HASIB.algo = "Sadeh1994" and argument Sadeh_axis = "Y" to indicate that the algorithm should use the Y-axis of the sensor.

Galland2012

The algorithm proposed by Galland et al. link. To use our implementation of the Galland2012 algorithm specify parameter HASIB.algo = "Galland2012". Further, set Sadeh_axis = "Y" to specify that the algorithm should use the Y-axis.

ColeKripke1992

The algorithm proposed by Cole et al. link, more specifically GGIR uses the algortihm proposed in the paper for 10-second non-overlapping epochs with counts expressed average per minute. We skip the re-scoring steps as the paper showed marginal added value of this added complexity. To use the GGIR implementation of this algortihm, specify parameters HASIB.algo = "ColeKripke1992" and Sadeh_axis = "Y" to indicate that the algorithm should use the Y-axis of the sensor.

Dealing with expected or detected nonwear time segments

Depending on study protocol we may want to interpret invalid data (typically non-wear) differently. When you set parameter ignorenonwear=TRUE (default) it will ignore any non-wear period for the SIB detection. This is useful to prevent nonwear episodes before going to bed or after waking up from contributing to sleep.