useless FL column in the Machine Settings log on the Statistics page

[cathyf]

On the OSCAR "Statistics" page there is the log of changes to machine settings, and there is a column for FL. Mine is a whole column of 6-1/2 years of zeros! It appears that it's the average minimum FL, not the median or the even-more-useful 95%.
[attachment=31877]
I changed my settings 5 days ago, so it's pretty simple to look at those 5 data points:
Date:    Min  / Med / 95% / Max
25-Apr: 0.00 / 0.01 / 0.22 / 0.66
26-Apr: 0.00 / 0.03 / 0.29 / 0.45
27-Apr: 0.00 / 0.02 / 0.25 / 0.52
28-Apr: 0.00 / 0.01 / 0.16 / 0.40
29-Apr: 0.00 / 0.00 / 0.19 / 0.42
----------------------------------------
Avg:     0.00 / 0.01 / 0.22 / 0.49

I've got a 2-day segment

Date:    Min  / Med / 95% / Max
17-Mar: 0.00 / 0.02 / 0.36 / 0.60
18-Mar: 0.00 / 0.00 / 0.19 / 0.41
----------------------------------------
Avg:     0.00 / 0.01 / 0.27 / 0.50



(Sleepyhead does this, too -- first thing I noticed 6-1/2 years ago!)

reporting the 95%FL would seem to be way more useful, since that's the number that tells us if the FLs are low or high or ridiculously high...

Also on the topic of that list -- average of total time in apnea might be useful, too.

[GuyScharf]

That column currently contains the flow limits index -- an index of events. But ResMed doesn't report flow limits as events; it reports them in the flow graph. Thus this column is always zero for ResMed machines. It does have values for some other brands of machines.

We are considering changing what is in that column for ResMed machines.

[cathyf]

Ah, makes sense -- for ResMed machines, the 95%FL would seem to be the number we are most paying attention to.

However those flow limit min/med/95%/max numbers are calculated by OSCAR, right? That would be different from the other columns in that report, where it is simply the value reported by the machine.

And we've been discussing elsewhere that we really don't know which data regions the reported FL numbers are applying to. The particular sort of statistic that we are reporting there only apply if you are sure that the data at each and every time slice is real, and the data collection/calculation is uniform over the time slices. Because of how my particular data looks, I've been complaining for my entire 6-1/2 years as a Sleepyhead/OSCAR user that some very significant fraction of my FL data points that are being reported by the machine as zero are obviously ND or NC (No Data or Not Calculated). And the work that 2SleepBetta has been doing suggests that the calculation that ResMed is doing is working in the time direction is a very rich and complicated way. So simply counting the milliseconds that the machine reported each particular FL value (which is what you are counting when you are calculating percentiles of time-series data) is not a particularly exact thing to count anyway.

[Sleeprider]

Cathy, you read my mind. I had discussed this with the development team just a few weeks ago. Hopefully this statistic is revised in a future edition.

[GuyScharf]

All of the numbers are calculated by OSCAR, with the intent of replicating the numbers computed by the CPAP machine. We do that because not all CPAP machines report the same numbers. This can result in slight differences between what the machine reports and what OSCAR reports. For ResMed machines, there is a difference in the way ResMed and OSCAR compute AHI (and other indices). The ResMed machine rounds all session elapsed times to the nearest minute, then adds and divides into the number of events. Using the high-resolution flow and other data from the machine, OSCAR knows elapsed times to the millisecond. That rounding error in the machine can thus create a different (and sloppier) calculation.

[cathyf]

All of the numbers are calculated by OSCAR, with the intent of replicating the numbers computed by the CPAP machine. We do that because not all CPAP machines report the same numbers. This can result in slight differences between what the machine reports and what OSCAR reports. For ResMed machines, there is a difference in the way ResMed and OSCAR compute AHI (and other indices). The ResMed machine rounds all session elapsed times to the nearest minute, then adds and divides into the number of events. Using the high-resolution flow and other data from the machine, OSCAR knows elapsed times to the millisecond. That rounding error in the machine can thus create a different (and sloppier) calculation.

Yeah, I'm was a math major and I have this OCD thing with doing arithmetic in my head. Many a morning I've woken up and looked at the time and the calculated AHI and solved for the number of events and said "Huh?" Recently, I woke up to exactly 7 hours of run time, and an AHI of 1.2.

Rounded to one decimal place, 8/7=1.1 while 9/7=1.3.

So I had 8.5 events?!? hmmmmmmmm....