mrueschman

Aaron,

I don't have any meaningful updates on MESA availability; I know discussions are still ongoing. I am not aware of other public sources of concurrent PSG/actigraphy.

I met with another colleague recently who showed interest in sharing some of his historical data on NSRR (including joint PSG/actigraphy), but I think that will be another slow moving process.

On the upside, we should have HCHS/Sueno actigraphy data shared sometime this summer, although those do not have concurrent PSG.

Fengzhen,

I am not that familiar with the EDF and the sleep recordings myself, but I am going to ask a couple other Core Members to review your question and comment if they have thoughts.

Have you tried EDFbrowser? http://www.teuniz.net/edfbrowser/

The site there mentions Nihon Kohden's .EEG files. Unfortunately I don't have any personal experience with the type of stuff you are attempting.

Winda,

Sorry, this is a bit confusing with things labeled similarly but with different underlying data. slpprdp is the most commonly used variable for "total time spent asleep between sleep onset and lights on" and it is the denominator for the AHI variables. It is available for SHHS1 and SHHS2.

At a glance, I think slptime is something that was recorded by the sleep scorer on a grading/quality sheet. It probably isn't the most useful variable. If you wanted to have a "total time in bed" variable (includes sleep latency, sleep time, and wake time from lights off to lights on), I would suggest timebedp.

Winda,

rdi0p (with that formula) represents the AHI included events at all oxygen desaturations. From what I understand, at the time these data were originally scored/created (late 1990s), the AHI/RDI terms were essentially interchangeable and did not (yet) include the RERA distinction. "rdi*" variables have carried through into many of our other datasets, though these more closely match an AHI because we don't typically score RERAs.

Thanks,

Mike

Zhiz,

I will ping a couple other NSRR Core Members to see if they have any thoughts/ideas about using the Spectral tool with non-NSRR data.

Mike

Zhiz,

Are you trying to use this tool with an EDF that you have created? I believe the tool was designed with the NSRR data in mind, for which we have EDF (signal data) and XML (annotation data) pairs. Please let us know more about your particular use case and I can ask someone more familiar with the tool to comment.

Thanks,

Mike

I don't see anything about retirement/career status in our SHHS datasets. I suspect this is known within the parent cohorts (e.g. Framingham, ARIC, CHS), yet wasn't pulled over by the dataset owners into the official SHHS datasets. To get at this information you would probably have to access the parent cohort datasets and SHHS dataset on BioLINCC, which has its own set of access rules.

As for dates, these were purposefully removed upon posting to BioLINCC (which we use as the source for NSRR) for de-identification purposes. The "index date" referenced throughout indicates the "date of the baseline polysomnogram", which essentially represents "date of the SHHS1 visit". I think we have sent other users the "year of index date" (e.g. 1995 or 1996 in most cases for SHHS1) in a secondary dataset before, so I could get that for you if it would be of any help. Unfortunately we aren't permitted to share anything more specific.

Winda,

It sounds as if the interim follow-up medications data were never fully processed (i.e. translating medication names into the classification types we see in SHHS1 and SHHS2). Here's what I got back from the dataset owners:

You are correct, there are no medication variables included in the FUP dataset. I looked through my notes and emails and can’t find any documentation as to why we decided to exclude these data from the final analytic dataset. I believe that the medication data collected at the FUP time point were neve categorized as they were at the SHHS 1 and SHHS 2 time point and this may be why we excluded the medication data from the final dataset.

PJ,

Thanks for inquiring and checking out the SHHS data. Good questions!

• My initial thought is that subjective sleep efficiency would be difficult to assess in SHHS. On other sleep diaries we often have questions about awakenings during the night, yet the "Morning Survey" you cited is limited in that it doesn't really ask any questions about the time between going to sleep and waking up (besides Q4 for a subjective quality rating).

• An accurate sleep efficiency measurement in SHHS1 was difficult to ascertain due to some of the limitations of the home sleep testing devices. As such, the slp_eff variable filters out many cases where it was noted during scoring that the entire in bed period may not have been captured on the recording. Some of the Signal Quality variables were used to censor values from slp_eff. Variables like latreliable, restan2, and restan3 were probably used. I think the exact filtering code may have been lost over time, though it has been on my back-back-burner to try and reverse engineer the filters that were used and make these more apparent in the variable metadata.

To the question about if it's "all right" to use slpeffp in SHHS1, it's a decision that each researcher can make to filter on certain criteria or not. Some of the efficiency values are probably not all that reliable. Unfortunately, this is one of the prevailing problems in analyzing SHHS1 data and the issue comes up fairly often. Thankfully most of the other studies do not share these sleep latency/efficiency challenges!