S9 data skew - my latest take on it (long discussion)

[idamtnboy]

At the risk of being accused of beating a dead horse to a pulp I'd like to share what my latest conclusions are.

First, some of you will recall some previous posts where I placed blame on the Resscan software for the S9 data skew problem that others discovered last spring. Then in one of my posts after that I tell how I finally got the picture that what I was seeing, and discussing, was not the original problem but a different Resscan problem that turns out to be unique to my PC, as best as we know now. I then had begun to also see the data skew problem in my own data. So I set on quest last weekend to try to decipher the S9 data skew problem for myself.

Here, in a nut shell, are my observations. Data gets recorded correctly on a newly formatted SD card. This has been previously determined by others. The data skew gets progressively greater with time. If this has to do with time only, or with how much data has been written to the card, which of course increases with time, I have not determined yet. I can tell you a data card with 14 days and 20 mb of data has data skew of only a few seconds, whereas a card with 95 days and 28.7 mb of data has some skew greater than 1 minute. The skew problem is a result of the S9 doing a poor job of defining and recording the data recording start time for each file. Fragmenting of the files on the card doesn't seem to a cause. I took a card that Defrag said was badly fragmented, defragged it, and put it back into the S9. The next day data files had the same problems as before.

There are four different files written that have data that is shown in the details graphs. The BRP file is the high resolution flow data, the PLD file is the detail data such as leak and mask pressure, the EVE file has info about the time and duration of each apnea, and the SAD file readings for O2 and pulse. The BRP data is what you see in the Resscan flow graph. The EVE data is what is shown in the event graph. All of the files have a date & time stamp in the filename. All four files always have the same time, within a second or two, in the filename. Some of the components of each of the files are recording start date, recording start time, recording duration, descriptions of the data, and the data. I looked in detail at data in 4 days of files on the card with 95 days data, and 4 days on the card with 5 days data. I looked at times on several files on the card with 14 days data. I used two basic tools, EDF Browser, and edf2ascii.exe, EDF Browser to view the BRP file flow data, and edf2ascii to convert the files to plain text files which I then looked at in Notepad and Lotus 1-2-3. The time stamp for each line of data in an edf is given as number of seconds after the recording start times.

In the 95 day card the recording start times vary significantly, by as much as 54 seconds between the BRP and EVE files and as much as 79 seconds after the filename time. The time of the first apnea, as given in the BRP file and in the EVE file varied from 20 to 38 seconds. But in every case I looked at, the time span from the first apnea to all the subsequent apneas was exactly the same, within a second, in the BRP and EVE files. Also, the time between the start of recording and first apnea is different between the BRP and EVE files. My conclusion is that once data skew begins to be seen in Resscan the start times of the EVE files and BRP files both are suspect. Subsequently, the exact time of the apneas is questionable. I put this info into a Lotus graph but it turned out to be more confusing than helpful.

The start times in the BRP files vary from the start times in the PLD files. What this results in is the beginning of the flow graph does not match the beginning of the leak graph, or any of the other graphs. But, the recording duration is almost always the same, which means that if the flow graph begins 10 seconds before the leak graph it also ends 10 seconds before the leak graph. What little I looked at the SAD files there appears to be the same situation there. The question that remains unanswerable at this time is what is the correct start time for the flow graph, and all the other graphs?

None of this variation exists in the files on a newly formatted SD card. In fact, the filename time and the start times in all the files agree within one second. This leads me to believe that if you want to know the exact time of an apnea look at the BRP file in EDF Browser and add to the time shown the difference between the filename time and the data start time. In Resscan you can look at the flow graph and determine the approximate time of the apnea by looking at the time ruler. If the SD card is recently formatted that time will be within a very few seconds.

I think this has been mentioned by others already, but the time S9 uses as the time of the apnea event is the very end of the non breathing span of time. Interestingly it is labeled in the data as Onset time.

In short, the S9 has a problem writing the proper start times when there are many days of data on a card. It looks to me like the S9 also fails to record data at the start of a mask session when there are many days of data on the card. The question is, why? Does the data recording routine in the S9 spend time reviewing all data in the summary file before it starts writing data to the card? The presence of 7 days of high resolution data seems to have little effect, and neither does 14 days of detail data.

Unless you really need, for some reason, to know to the exact second when an apnea occurred, there is no reason to be concerned about the data skew problem. As long as the AHI chart can help you find the corresponding segment of the flow graph that should be enough. That's all it's really good for anyway, plus giving the length of the apnea.

I don't know if any of this will help anyone else, but at least I'm comfortable at having defined the problem for myself and deciding it's nothing to get wrapped around the axle about.

[jlk]

Nice description of your subject line! I can't find a smart-a$$ smilie for my post. I have been missing 5 minutes to 5.5 hours of data, Is that also a skew? (serious question) john

[idamtnboy]

Nice description of your subject line! I can't find a smart-a$$ smilie for my post. I have been missing 5 minutes to 5.5 hours of data, Is that also a skew? (serious question) john

I would say no, that is not skew, but it may be related. I can't see any reason why you would be losing data unless your S9 has some sort of serious recording problem. I'd like to see a file listing from your card and a few files. My Resscan installation has an issue caused by s/w conflicts I'm surmising. Maybe yours does too, in a different fashion. PM me with an offline email address so we can discuss how to best get them to me, if you're willing. Fortunately, HIPAA doesn't apply to patient to patient conversations and data sharing.

[jlk]

[/img]

This is the worst of my missing data. Unit was on the whole night. I took a potty break at 5:30 am, don't know what happened from 10:15 till 4:45 ish? john

[idamtnboy]

This is the worst of my missing data. Unit was on the whole night. I took a potty break at 5:30 am, don't know what happened from 10:15 till 4:45 ish? john

Do you by chance have Smartstart turned on?

[jlk]

Yes and I love that feature, must be lazy. john