Thanks for checking out the site and data, Richard.
Yes, the XML files as-is will not work in Polyman. The "NSRR" versions of the XML annotations are designed to work with EDF Viewer. The "Profusion" XML are straight from Compumedics Profusion where the sleep studies were originally scored.
To make these "openable" in Polyman you will have to write some sort of script that translates the existing XML into the format readable by Polyman. You can explore the XML structure by opening one of the files in a text editor.
That's correct. There are individual "time until X" variables like MI, CABG, PTCA, etc. I think the CVD death variable is a composite that the dataset owners created from the other "time until fatal X event" variables, so you could probably create your own "time until first instance of CVD" from some of the other CVD Outcomes variables.
Some information here from the SHHS authors: https://sleepdata.org/datasets/shhs/files/documentation?f=SHHS_CVD_Outcomes_Protocol.pdf
Sorry for the confusion. The "time until death" is the correct definition. We will update this an upcoming release for these variables:
Thanks for the question. All of the "Start" times are in "seconds from recording start time". You are correct about the Durations, i.e. the SpO2 artifact (#2) lasted 3 seconds and the Arousal (#3) lasted 25.4 seconds.
Thanks for raising this on the forum. Yes, at my own glance, bmiz_change appears to be wrong. It looks like the data owners had another variable, chg_bmiz, which is more correct but was removed for redundancy with bmiz_change. I will swap the variables in a future release.
For the time being, I suggest using a calculation of your own for the BMI Z-score change between baseline and follow-up.
Baseline = readings from the baseline visit
Followup = readings from the follow-up visit (approximately 3 months after baseline)
Subjects were randomized at baseline, followed for 3 months, and then given another sleep study. The numbers in the filenames are the Subject Codes, so many subjects have both a "baseline" and "followup" reading. (E.g. heartbeat-baseline-700001.edf is a reading from the same person as heartbeat-followup-700001.edf).
There were 3 treatment arms (described here) and the subjects are all mixed together (i.e. not separated out by apnea or control status).
An entry criteria for HeartBEAT was an AHI >= 15, so all the subjects in the study had some level of sleep apnea to begin with.
The EDF files contain EKG and Pulse signals, which will give second-by-second tracings from the heart-related signals. You will find some EDF visualizing/processing tools on our directory here.
Some information on the signals here: https://sleepdata.org/datasets/heartbeat/pages/equipment
The EDFs are here: https://sleepdata.org/datasets/heartbeat/files/polysomnography/edfs
Most of our datasets contain subjects with sleep apnea and heart rate data. Please read through the dataset documentation to see which ones might suit your research questions best.
Yes, sleep and wake are determined using an automatic algorithm. Here are a couple links with more description:
Yes, REST intervals were manually annotated. Anything that is not REST ends up as ACTIVE, assuming the data are not EXCLUDED. The MESA actigraphy scoring manual describes the process used by the actigraphy reviewer.
Hope that helps!
The XML annotation files contain information and timestamps for respiratory events and oxygen desaturations. Sleep staging for each epoch is also available in the XML files.
If you wish to analyze the airflow signal then you will need to look into the raw signal EDF files. There should be packages for statistical software to read in the EDF for further analysis. Here's a tool for reading EDFs into MATLAB: https://sleepdata.org/tools/dennisdean-block-edf-loader