mrueschman

Thanks for clarifying. You are correct - the original collection dates were all scrubbed from the EDFs as part of our normal de-identification process. We set all the calendar dates to 01/01/1985 - the first allowable date per the EDF specification.

My question to the STAGES team focused on the time mismatch. I opened 1-2 files from all the other sites and did not see similar mismatches. Looking back through old emails, I did see a couple mentions of the MAYO files being particularly difficult to process/harmonize, so perhaps some sort of inadvertent time shift occurred.

Thanks for raising the issue. I see what you mean in those (and other) MAYO files. I'll inquire with the STAGES data contributor.

nuMoM did not have EEG signals, so traditional sleep staging was not possible. The Stage 2 indicators you see in the annotation files represent the estimated sleep period. There is some description in the Flacco 2015 paper:

Level 3 sleep monitoring devices, as used in this study, do not record sleep directly (i.e. with EEG and electromyography signals). Therefore, sleep onset and offset and periods of prolonged wakefulness are identified by certified polysomnologists using information from both participant completed questionnaires on bed and wake times and the visualized patterns of heart rate, breathing pattern, movement and artifact. The sleep period is defined as the period between sleep onset and offset. Sleep onset is identified based on self-reported bedtime plus visualization of reduced signal artifact, decreased heart rate, regularization of breathing pattern, and stability of the position sensor (indicating a supine or lateral position). Conversely, sleep offset is identified by evidence of sustained movement artifact, changing position, increase in heart rate, and participant reported wake time. Following the initiation of sleep, segments of the study were excluded from analysis if significant movement or artifact occurred for a period of ≥ 20 minutes.

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4410772/

Thanks for using the site!

Thanks for using the site. Your choice for SHHS2 makes sense.

I also see this in MrOS (older adults): https://sleepdata.org/datasets/mros/variables/poxusual

Maybe also CFS (mostly adults; some adolescents): https://sleepdata.org/datasets/cfs/variables/desslp

There are also a couple blog posts about NSRR's harmonization efforts:

• https://sleepdata.org/blog/2022/10/harmonizing-sleep-and-circadian-data-why-is-it-important
• https://sleepdata.org/blog/2022/11/a-sneak-peek-into-the-nsrr-standardized-metadata-for-validated-survey-instruments

The STAGES contributor got back to me about the first question.

This question came up from another user as well which I tried to answer here: https://github.com/Stanford-STAGES/stanford-stages/issues/31

Essentially the cohorts_deid Excel file lists the studies used for the narcolepsy classification portion and not the sleep staging, however we still wanted to provide these additional EDF studies despite them not being used as controls or narcoleptics.

Unfortunately, the MESA (Exam 5) sleep journals were never entered electronically. ☹️

They were filled out on paper and scanned to PDF at the field sites. The actigraphy scorer kept the PDF open for review while scoring.

I received confirmation of the following from the STAGES data contributor:

1. All the PSG annotations (staging, respiratory events) came from human scoring (sleep labs at each clinical site)
2. The "SRBDVariables" spreadsheet was derived from the PSG annotation files.

I think the references in that supplemental documentation were discussing the possibility of using automated algorithms to further analyze the STAGES PSG data.

Hey CC - thanks for checking out the site. I agree this could be made clearer in the STAGES documentation.

Another user asked about this on the Forum last year; I provided the details I could find about the manual sleep scoring: https://sleepdata.org/forum/stages-ground-truth-hypnogram/

In short: Yes, the STAGES PSG data were scored by sleep experts.

Thanks, you're right, I should have clarified further. The "immobile time" counting is done on a continuous basis by the Actiware software, so if there were 5 minutes of immobile time before the beginning of a REST interval this can result in a value of 0 minutes of sleep latency.

I think the point I was trying to get at is that sleep latency in actigraphy was most commonly in the 0-10 minute range because the scoring rules called for REST intervals to be set approximately around the time activity counts started to decrease (in other words, around the time the subject started becoming more immobile than mobile). People are typically lying still when they start to try and fall asleep, yet (of course) the actigraphy device can't tell when the person truly transitions from wake to sleep.

The scorers did reference self-reported sleep diaries in the setting of REST intervals, however this was just one input used (others being: activity count levels; light levels; event markers). Perhaps if the sleep diary alone was used in the setting of REST intervals we might get "truer" estimates of sleep onset latency. For instance, a subject reports they got in bed and tried to sleep at 10 p.m., yet the actigraphy device still shows sporadic movement until 10:30 p.m., when the subject triggered the "5 minutes of immobile time" for sleep onset, thus giving ~30 minutes of sleep onset latency for that night.

Bottom line: sleep latency and actigraphy is very tricky!