OSCAR preferences for flagging HR & SpO2 events?

[cathyf]

I've been tinkering with the preferences for flagging oximetry data, and I can't seem to find a happy medium between flagging everything and flagging nothing, especially with the SpO2 data.

Can somebody explain what I should set the % and seconds to in various scenarios?

[Crimson Nape]

Knowing that the Viatom/Wellue uses a 4-second sampling, I think I would use something that would be divisible by 4. Something like 20 to 32 seconds. This would allow for repeatability of readings and throw out any artifacts. As far a percent; Since the percent value is based on your average, I would try to come up with an acceptable average for yourself. Then, determine what is a value you want flagged. Determine the percentage difference between the two, and you will have your percent value.

. . . Or did you want something that involves Gauss-Jordan least-squares regression analysis?

[Ratchick]

This seems like as good a place as any to mention what I discovered from talking to Viatom.

They calculate their baseline by taking the MOD (edit: MODE) of the values greater than or equal to the mean (by mean, I assume they mean rounded to the nearest integer, based on my own testing). They then calculate their drops based on a fall of more than 3 or 4 percentage points for ANY length of time, and the drop ends when the value rises to within 2 percentage points of the baseline.

So if your baseline is 95%, to trigger a 3% drop, the value has to be lower than 92% (which, in practice, is 91% or less, because it measures in whole integers) and the drop will end when the level rises to 93% again. And a 4% drop is calculated from the moment the value falls to 90% and ends the moment it gets to 93% again.

In the movement graph, every integer is 1/32nd of a G (the acceleration felt due to the force of gravity, or 9.8m/s²). So a movement value of 32 means the device was accelerating at 1G and 64 would be accelerating at 2G, etc. It doesn't seem to differentiate what direction that acceleration is in (and I'm not sure if it's additive, average or what).

Hope that's at least interesting to someone!

As to what settings to choose, that depends on your purpose. If you want to see a more accurate reading of how much time a night you spend under a certain level, then go with shorter periods. If you're only interested in prolonged drops, then set it longer. Though most doctors use 3 or 4% over either 8 or 10 seconds, you can also limit it to the really big drops too, if you really want to.

Just bear in mind that (from my experience at least) you need to reimport the viatom data to have the stats recalculated. It doesn't recalculate it on the fly.

[kappa]

They calculate their baseline by taking the MOD of the values greater than or equal to the mean (by mean, I assume they mean rounded to the nearest integer, based on my own testing).

In the movement graph, every integer is 1/32nd of a G (the acceleration felt due to the force of gravity, or 9.8m/s²). So a movement value of 32 means the device was accelerating at 1G and 64 would be accelerating at 2G, etc. It doesn't seem to differentiate what direction that acceleration is in (and I'm not sure if it's additive, average or what).

That is interesting. I'm guessing they might mean the "MODE" of the values greater than or equal to the mean ("MOD" is usually short for "modulo" which doesn't make sense here). i.e. the most common reading greater than or equal to the average.
OSCAR uses the median SpO2 value of the first hour of a session to determine the baseline, which is likely to be similar. A drop is reported when value is <= baseline-change% for >= minimum_duration. 
The acceleration calculation would likely be something like abs(sqrt(ax^2+ay^2+ax^2)-32) - I assume they have a 3 axis accelerometer with 1g = 32.

[kappa]

Correction: OSCAR uses the 90% percentile of the first hour, not the median as I previous stated.

[Ratchick]

That's good to know, thank you. Now I can definitely understand the differences in figures more clearly. I hadn't yet gone through the code to figure out exactly how the baseline was calculated in OSCAR.

You're right, I did mean mode, sorry, it was late and I was tired and caffeine-deficient.

You're also correct that they have a three-axis accelerometer setup. I honestly never bothered to ask the exact formula as I was just interested to know what the numbers represented rather than exactly how they were calculated for movement.