What is SRI?
The Sleep Regularity Index (SRI) was proposed by Phillips and colleagues (2017) and further developed and shared as open-source R package sleepreg by Windred and colleagues (2021). The SRI can have a value between -100 and 100, where 100 reflects perfect regularity (identical days), 0 reflects random pattern, and -100 reflects perfect reversed regularity in sleep.
Multiple implementations
R package ‘sleepreg’ was tailored to GGIR 2.0-0 as available at the
time and does not work with more recent GGIR versions. We now know that
GGIR 2.0-0 came with various bugs and is best not used. Therefore, an
effort has been made to address this. Once the ‘sleepreg’ maintainers
accept the proposed
changes, it will upgrade the sleepreg package code to work with both
historical and later GGIR versions. In the mean time you can use the
proposed upgrade to package ‘sleepreg’ by installing it via R command:
remotes::install_github("vincentvanhees/sleepreg", ref = "upgradeGGIRversion").
Complementary to this effort, GGIR offers two modified implementations of SRI, which we refer to as SRI1 and SRI2. In this Annex we discuss each aspect of their calculation and how they differ from the original SRI calculation in ‘sleepreg’.
However, first we would like to list all aspects that all these implementations have in common:
- SRI values are expressed on a scale from -100 to 100.
- Detected non-wear is treated as missing data and this is accounted for in the SRI calculation.
- Calculations are done based on 30 second epochs. Even if GGIR is configured to do its other analyses with a shorter epoch duration, e.g. 5 seconds, GGIR makes sure that values are aggregated to 30 seconds for the SRI calculation.
- All implementations derive regularity between days that are defined from noon to noon.
Regularity of what?
SRI as derived with github.com/dpwindred/sleepreg
In ‘sleepreg’, ‘Sleep’ is by default defined as the combination of:
- GGIR detected sleep period time window minus WASO episodes that last at least 30 minutes.
- Any 30 minute interval outside the GGIR_detect sleep period time window with 95% or more sustained inactivity bout. This intervals are assumed to equal naps.
The ‘sleepreg’ package comes various options to modify the configuration of the calculation as discussed in the package documentation and in the supplementary materials to Windred et al. 2021.
SRI1 as stored in GGIR part 4
In SRI1, ‘Sleep’ is defined as all sustained
inactivity bouts, which can optionally be smoothed with a rolling
average window. The rolling window size in hours is specified with
parameter SRI1_smoothing_wsize_hrs, and the fraction of
time that is needed to be sustained inactivity bout is set with
parameter SRI1_smoothing_frac.
The advantage of this calculation is that it does not rely on hard assumptions about accurate detection of sleep and naps by GGIR. However, SRI1 with default non-smoothing settings may need to be interpreted as a rest regularity index. Only when using the smoothing parameters SRI1 may better reflect a true indicator of sleep regularity.
SRI1 is calculated in GGIR part 3 and stored in the csv-output of GGIR part 4.
SRI2 as optionally stored in GGIR part 6
GGIR offers experimental functionality to classify daytime naps, which is subject to ongoing investigation in young population of pre-schoolers. We are still looking for community support to help improve and evaluate the nap detection for adult populations. Please reach out (v.vanhees at acceleting dot com) if you would be interested in applying for funding or have funding and want to sponsor this development. Once the nap detection functionality has matured we will document it more elaborately. However, if you have been helping to explore the nap detection and are confident that GGIR classifies both sleep and naps accurately, e.g. after inspecting the visual reports, then SRI2 would be superior over SRI1 as measure of sleep regularity.
In SRI2, ‘Sleep’ is defined as all sleep classified with GGIR part 4, minus WASO, plus daytime naps as classified in GGIR part 5.
SRI2 comes with parameter SRI2_WASOmin which allows us
to define the minimum WASO length to be considered, any WASO shorter
than this will be ignored and treated as sleep. In part 6, SRI is only
derived when parameter SRI2_WASOmin is specified. If no nap
parameters are set the resulting SRI estimate will only reflect the main
SPT window in the day.
At what resolution is SRI calculated?
SRI as derived with github.com/dpwindred/sleepreg
In the initial publication by Phillips 2017 it was proposed to only calculate SRI based on seven, or a multitude of seven, consecutive days of data without missing values. This to avoid a possible role of imbalanced data to the final estimate. However, this renders many datasets unsuitable for analysis and leads to a painful loss in sample size and statistical power. This was improved upon by Windred 2021, which allows for gaps in data. Next, SRI is calculated as one value across all valid days in the recording.
SRI1 and SRI2 in GGIR
GGIR calculates and stores the SRI per day-pair.
The benefit of the approach taken in GGIR is that it enables the user to study the day-pair to day-pair variation in SRI, and the role of day-pair inclusion criteria. Further, the access to SRI at day-pair level makes it possible to account for an imbalanced datasets via multilevel regression analysis applied to the output of GGIR, with day-pair as one of the model levels.
What level of nonwear is acceptable?
SRI as derived with github.com/dpwindred/sleepreg
Entire days are ignored when there are more than 6 hours of nonwear.
SRI1 and SRI2 in GGIR
Per day-pair GGIR calculates the fraction of the 30 second
epoch-pairs between both days that are valid, which can be found in the
GGIR part 4 output under the variable name
SriFractionValid.
By default, day-pairs are excluded if this fraction is below 0.66 or
if the fraction of valid data for the individual day is below 0.66. This
threshold is coupled with the 16-hour default value for parameter
includenightcrit (16/24 = 0.66). For example, if you set
parameter includenightcrit = 12, the fraction threshold
will be: 12 / 24 = 0.5.
The benefit of the approach taking by GGIR is that it looks at both the completeness of matching data points between day pairs and data completeness in the day itself.
Where to find SRI in the GGIR output?
SRI1 as stored in GGIR part 4
The day-pair level estimates are stored as variable
SleepRegularityIndex1 in the GGIR part 4 .csv-report on
sleep. Further, GGIR also stores the person-level aggregates such as:
the plain average over all valid days, the average of all valid weekend
days, and the average of all valid week days. No GGIR input arguments
are needed to invoke the SRI calculation in part 3. The calculation is
automatically performed after updating GGIR and processing your
data.