Spreadsheet Analysis of SpO2: Median of +/- one minute looks helpful

[DavidEsp]

Not yet the "all singing" metric I seek, but a step in the right direction - definitely helpful in removing the clutter (would this be a helpful OSCAR feature?)...

While analysing my oximeter SpO2 data,
which is at "one sample every 4 seconds" resolution,
when exported as CSV from oximeter app O2 Insight,
and subsequently imported to a spreadsheet (LibreOffice Calc)...

I found that plotting a graph of the Median (not Average) over each (sliding) period of 2 minutes gave a much clearer picture of the main trends, e.g. the SpO2 troughs, and retained the original integer value-type, convenient for filtering

In practice I took the median 1 minute before to 1 minute after each timestamp, so as to avoid time/phase-shifting.

[chemmkl]

A sliding window of the median is a well-known method for removing outliers in time-series data, so I think that is what you would be effectively doing with the presented approach. In you are looking for a trend then it seems to me like a good method, however I am not sure that we want to disregard severe desaturations at outliers. Check out the MAD statistic (Median Absolute Deviation, a robust statistics because is not sensible to outliers) for outlier removal that adapts to slow changes in the trend.

This article is interesting as it presents the problem of determining when to alert when desaturations happen: https://www.aami.org/docs/default-source..._based.pdf

I like the idea that is not about an absolute threshold like "below 88%" but also it has to be sustained maybe for 15 seconds before sounding the alarm.

Finally, keep in mind that the data coming from the device is often not raw data. For example, the O2 Ring reads the SpO2 every second but the data we get is the average of the last 4 seconds in 4-second interval.

Since the clinicians seem to be worried with desaturations per hour (drops higher than 3%) maybe that could be a good magnitude to track the trend of?