Understanding OSCAR statistics

[CPAPfriend]

hey dataq1,

do you know where I could find that equation by any chance?


@Gushauck

Imagine you took everyone in the world and asked them what their weight was. Then, you lined them all up from lowest to highest weight. The average weight would be the 50th percentile. The highest weight (in this case) would be the 99.5th percentile. The 95th percentile would be the weight where 95% of all the people in the world were at or below, so this would obviously be high, right? It's like asking, "how tall is the shortest person in the top 5% of people for height?", to put it differently.

Now, you might be asking, "well, why the 95th percentile?!? Why not the 87th percentile, or 61st percentile?". The reason we use the 95th percentile is because at about the 95th percentile, we are two standard deviations away from the average. Standard deviation is a way of signifying how spread out values are, so, in other words, when you get the average value given to you AND you get the 95th percentile, you get an idea for how spread out your values are. If the 95th percentile is very close to the average, then your data is clustered around less values, that is the range of values is crowded. If, on the contrary, your 95th percentile is very far from you average, then the values are spread out over a larger range. Most of this isn't going to mean much to the average user, but for someone with statistical intuition, it can be insightful and guide decision making.

But, like others have said, when it says the 95th percentile of your flow limit is at 0.16, it means that during 95% of the night, your flow limit was at or below that amount.

Flow limit is a unitless measurement of the flatness (and other things that we don't know) of the breath that you find in the flowrate graph. 1 = as bad as it can get. 0 = as good as it can get.

[BoxcarPete]

Excellent way of explaining it with the standard deviations. The 99.5 is close to a third standard deviation too, but it's not quite applicable here because OSCAR reports median instead of mean. Especially with flow limits, there are going to be many zeros throughout the night, so most people will see zero for the median because more than half their breaths are rated with a FL of 0. The mean would be a non-zero number as long as there were any FLs at any point in the night, so it isn't a perfect comparison here due to the skewed distribution.

[paulag1955]

Is there any way you can explain this without using standard deviation, which is meaningless to me?

[Gideon]

99.5% should be the Maximum value seen. OSCAR recognizes that most data sets have a totality unrealistic value as a max, usually see at startup or shutdown, so OSCAR uses a high statistical value, 99.5%, instead of the max. Assume it is the max and you will be fine. I have seen flow rate max numbers something like 3000 lpm. Keeping the max at a reasonable level keeps the flow rate chart from looking like a flat line with a single spike.

[hypopneac]

This is for dataq1, re your post in this thread on 11-29-23 at 9:35 p.m.

When you wrote about a "perfect inhalation curve" with a "nice rounded trace" as opposed to a "flattened top," specifically which graph in OSCAR are you referring to?  None of them bears the title "inhalation".   Thanks.

[Crimson Nape]

Flow Rate - Zoom in to a 3-minute window to see the waveform.
- Red

[hypopneac]

Thanks.  I did that and now I can see what was meant.  It's not like a nice sine wave, but it's not literally straight flat across the top either.  Not going to get crazy about it, lol.

[Crimson Nape]

I always described mine as resembling a cursive "r". I changed over to a bi-level (AirCurve 10 VAuto) and it corrected it in a heartbeat.
- Red

[hypopneac]

Red, that’s a good description for what I see when I expand the waveform to a 3 second window. I also experience some aerophagia, especially when starting therapy. I’ll take this up with the sleep doctor when I see him in two weeds.  Now I’ll probably bring my laptop to show him the OSCAR graphs. Thanks!