Better SpO2 Metrics? Integrative/Cumulative?
Shortcomings of Current Metrics
From my very limited perspective as an oximeter (O2Ring) user, including its apps (phone, computer) and OSCAR, a huge shortcoming of current SpO2 metrics sticks out. They all seem based on near-instantaneous situations. Consequently they are jittery and don't indicate (only ambiguously correspond to) the more consequential kinds of disruption in sleep breathing patterns.
For example, in OSCAR, an event/alarm/reminder can be triggered on SpO2 dropping below a specific threshold value. But this happens even when it only occurs for an instant (single sample). Or events can be flagged when SpO2 drops ("rapidly") by some proportion/percentage (eg 3%). Such metrics are way too jittery - inducing a user temptation to set thresholds lower than one would like e.g. in order to avoid being woken up unneccessarily (thereby causing sleep-disruption health-problems) or to avoid clutter in event-flags.
Even the more sophisticated O2 Score seems too blunt a metric. I find it too ambiguous to be useful. An SpO2 trace with lots of downward spikes spread over the whole night can give the same score as one with a sustained period of reduced oxygen. Possibly reflecting this, sometimes it tallies with my feeling/headaches, sometimes not.
Specific Example
A couple or so days ago I woke up feeling heavy. Looking at my SpO2 trace, at one point (in time) I noted a sustained period of about half an hour of significantly lower than normal SpO2.
Reasoning
Surely it is sustained patterns, i.e. cumulative situations which matter, not sporadic separated transients (over short periods let alone single samples).
As a non-expert, I'd regard my own (at least) "ignorable short periods" of low SpO2 as being anything under say 20 seconds. Provided any such instances were separated by say another 5 minutes from any other ones, I'd regard them as being totally ignorable - just part of "my normal". I will have to test this as an SpO2 pattern, but at least I am comfortable with this as an occasional daytime (non) breathing pattern (e.g. while watching TV).
Proposed Solution
(Slightly inspired by Loudness (LUFS etc.) metrics in audio measurement).
What's really needed is a metric corresponding to how much distruption is likely being caused to the body, primarily to to the brain. The "disruption" concept here including a range of possible minor as well as major effects. Then place alarm-generating threasholds on that metric.
As a first layer, I imagine a graph where, starting from a point at which SpO2 drops, brain-cell "happiness" only begins to drop, continuing slowly over time (numerous seconds). Could be inferred via a time-constant type model (of the general human brain oxygen and metabolism etc. biology). Then when breathing starts again, I imagine that curve recovering more quickly (than it dropped). As (presumably) happens when resuscitating people.
As a second layer, I imagine what really matters is whether (and by how much and for how long) this curve has passed beyond/below (as you prefer) some "threshold of significance". Perhaps the area between that curve and that threshold (implying a further level of integration/accumulation - representing the consequence of sustained "brain cell unhappiness").
Questions
* Anyone heard of anything like this? Relevant jargon/search-terms?
* Is anything like this already being done in practice or under research?
* Would it be a useful addition to SpO2 devices (e.g. as basis for vibrate-alarms) and their apps and/or to OSCAR?
* What would be the best way to organise collaborative research into this? At the very least as an amateur project.
Immediate Idea
I guess one could immediately play with such ideas based on spreadsheet/CSV data (exported from any of the oximeter apps or OSCAR) or in code (e.g. R, Python). But it would be great to first firm up on the principles and numbers - and especially to avoid "reinventing the wheel" at all (if that applies).
From my very limited perspective as an oximeter (O2Ring) user, including its apps (phone, computer) and OSCAR, a huge shortcoming of current SpO2 metrics sticks out. They all seem based on near-instantaneous situations. Consequently they are jittery and don't indicate (only ambiguously correspond to) the more consequential kinds of disruption in sleep breathing patterns.
For example, in OSCAR, an event/alarm/reminder can be triggered on SpO2 dropping below a specific threshold value. But this happens even when it only occurs for an instant (single sample). Or events can be flagged when SpO2 drops ("rapidly") by some proportion/percentage (eg 3%). Such metrics are way too jittery - inducing a user temptation to set thresholds lower than one would like e.g. in order to avoid being woken up unneccessarily (thereby causing sleep-disruption health-problems) or to avoid clutter in event-flags.
Even the more sophisticated O2 Score seems too blunt a metric. I find it too ambiguous to be useful. An SpO2 trace with lots of downward spikes spread over the whole night can give the same score as one with a sustained period of reduced oxygen. Possibly reflecting this, sometimes it tallies with my feeling/headaches, sometimes not.
Specific Example
A couple or so days ago I woke up feeling heavy. Looking at my SpO2 trace, at one point (in time) I noted a sustained period of about half an hour of significantly lower than normal SpO2.
Reasoning
Surely it is sustained patterns, i.e. cumulative situations which matter, not sporadic separated transients (over short periods let alone single samples).
As a non-expert, I'd regard my own (at least) "ignorable short periods" of low SpO2 as being anything under say 20 seconds. Provided any such instances were separated by say another 5 minutes from any other ones, I'd regard them as being totally ignorable - just part of "my normal". I will have to test this as an SpO2 pattern, but at least I am comfortable with this as an occasional daytime (non) breathing pattern (e.g. while watching TV).
Proposed Solution
(Slightly inspired by Loudness (LUFS etc.) metrics in audio measurement).
What's really needed is a metric corresponding to how much distruption is likely being caused to the body, primarily to to the brain. The "disruption" concept here including a range of possible minor as well as major effects. Then place alarm-generating threasholds on that metric.
As a first layer, I imagine a graph where, starting from a point at which SpO2 drops, brain-cell "happiness" only begins to drop, continuing slowly over time (numerous seconds). Could be inferred via a time-constant type model (of the general human brain oxygen and metabolism etc. biology). Then when breathing starts again, I imagine that curve recovering more quickly (than it dropped). As (presumably) happens when resuscitating people.
As a second layer, I imagine what really matters is whether (and by how much and for how long) this curve has passed beyond/below (as you prefer) some "threshold of significance". Perhaps the area between that curve and that threshold (implying a further level of integration/accumulation - representing the consequence of sustained "brain cell unhappiness").
Questions
* Anyone heard of anything like this? Relevant jargon/search-terms?
* Is anything like this already being done in practice or under research?
* Would it be a useful addition to SpO2 devices (e.g. as basis for vibrate-alarms) and their apps and/or to OSCAR?
* What would be the best way to organise collaborative research into this? At the very least as an amateur project.
Immediate Idea
I guess one could immediately play with such ideas based on spreadsheet/CSV data (exported from any of the oximeter apps or OSCAR) or in code (e.g. R, Python). But it would be great to first firm up on the principles and numbers - and especially to avoid "reinventing the wheel" at all (if that applies).
