Objective sleep measures collected

Actiwatches are small data loggers that are worn on the non-dominant wrist and record digitally integrated measures of gross motor activity. These devices contain accelerometers and light sensors to objectively assess sleep and physical activity of the individual wearing them. Prior studies have demonstrated actigraphy’s high sensitivity with moderate accuracy for assessing sleep parameters in populations with normal and disturbed sleep when compared to polysomnography^34–36. Our participants were instructed to begin wearing the Actiwatch Spectrum Plus (Philips Respironics) continuously from the time of enrollment into the sleep study (beginning within seven days prior to their planned discharge date) until the fourth week after discharge, which typically coincided with the final visit for follow-up assessments.

After downloading the Actiwatch data from each participants’ watch (using Philips Actiware 6.0.9 software, https://www.usa.philips.com/healthcare/sites/actigraphy/solutions/actiware), the study team screened the files for any malfunctioning watches, corrupt data or required adjustments using daily self-reported sleep diaries for cross-validation. Major rest intervals were marked using programmed functions of the software and were adjusted (inserted a new sleep interval OR extended, shortened, or split an existing one) only if any discrepancy was noticed based on sleep diary data. A total of 22 of the 77 participants did not require any modifications in sleep intervals marked by Actiware. adjustments of up to 9 days of actigraphy recorded sleep intervals were made by the group as part of the cleaning process for 53 participants. There were 5 participants who required modifications in sleep intervals for 10 or more days with no more than a maximum of 23 days. Minor rest intervals (naps) were manually entered based on activity patterns during the day only if the participant mentioned taking a nap on that day in their sleep diary. The recording of a nap was not based on a time of day but was based on a short interval of rest independent of the major rest interval (typically nighttime sleep). For all instances where the two group members did not agree with a marked interval, the questions were brought to the whole team for a consensus meeting. Weekly averages of the following variables were calculated from cleaned actigraphy data: sleep duration, wake after sleep onset time (WASO), sleep onset latency (SOL), sleep efficiency, and nap duration for each 24-h period. Actigraphy based sleep efficiency is calculated as the percentage of sleep during a given rest interval which is the time in bed (i.e., total sleep time divided by interval duration minus total invalid time (sleep/wake)).

The SRI python script (https://github.com/mengelhard/sri) was adapted into the JMP (version 14, SAS Headquarters, Cary, NC, www.jmp.com) Scripting Language (JSL) with additional customization considerations specific to this project. Much of the workflow was followed as described in Brooks et al.¹³ with a few other edits described herein. Actigraphy data were recorded starting at 5:00:00 PM on the day of discharge and collected for the next 28 days (24-h intervals with 1440 epochs for each day where an epoch is 1 min). One matrix of interval status columns and epochs for each participant was saved and used to calculate the weekly SRI and total SRI scores (across 28 days, if available). For this analysis, each participant was required to have at least seven days of actigraphy data. The customized JSL script calculates a total average SRI over 28 days (when available) and a seven-day average SRI for weeks 1,2, 3 and 4 (i.e., days 1–7, days 8–14, days 9–21 and days 22–28, respectively). The formula for the SRI calculated was first introduced by Lundsford-Avery et.al and was used in a previous publication from our group^12,13. The formula to calculate the SRI based off of the publication from Lundsford-Avery et. al¹² is shown below:

where δ(si,j,si+1,j) =1 if si,j=si+1,j and 0 otherwise.

The SRI can be interpreted as higher values indicate more regular sleep/wake patterns and lower values indicate irregular sleep/wake patterns. For example, if a person goes to sleep and wakes up at the same time every day during an assessment period, this individual will have a SRI close to (or exactly) 100. The total SRI and weekly SRI was calculated by averaging all SRIs for the specific time interval (i.e., total for 28 days denoted total SRI and weekly for week 1, week 2, or week 3 and week 4). To assess the completeness of the actigraphy data, the script will calculate the total number of ‘EXCLUDED’ minutes recorded by the watch for each day.

An individual was included in the analysis if they had at least seven days of actigraphy data captured. In addition to this requirement for inclusion, there could be no more than 1 day with 6 or more excluded hours over the 28-day assessment period. The justification behind this was based on the allowance of removing the watch for activities such as showering, bathing, and/or swimming. This filter removed 43 participants, leaving a subset of 77 participants to be included in this analysis.

Daily light exposure was investigated in this patient cohort to determine if an association could be observed between the SRI and daytime and nighttime light exposure as was previously reported in Philips, et al¹¹. Daily white light exposure was collected for each patient per 24-h period (comprised of 1440-min epochs). Light exposure data capture for each patient began at 10:00 PM on the day of discharge. A binary variable of 0 and 1 indicating no light exposure or light exposure for each epoch was created. If the watch detected ≥ 250 lm of white light, the variable was coded as 1 and if the watch detected < 250 lm of white light, the variable was coded as 0. Three white light exposure variables were created as follows by summing over the binary (0,1) variable for three different time intervals: total 24-h white light exposure, total 12 h of white light exposure during the interval of 10:00 AM to 9:59 PM (for clock day) and a total 12 h of white light exposure during the interval of 10:00 PM to 9:59 AM (for clock night). Clock day and clock night time intervals were chosen based on previous literature^11,12. The result of this calculation gives a total number of minutes of white light exposure for each time interval of interest. The three light exposure variables were averaged over the entire 28 days (total) or were averaged weekly.