flow limitations expressed as an index?

[sheepless]

I'm missing something. re the flow limit distribution graph: I'm not the sharpest tool in the shed but it appears that the combination of your lowest frequency & least %time (far right of the graph) is higher on resmed's scale of severity. I must be misreading the graph?

meanwhile, it's cool, and imo potentially very useful that you can separately display frequency & duration.

[2SleepBetta]

.. . . .  it appears that the combination of your lowest frequency & least %time (far right of the graph) is higher on resmed's scale of severity.

True. Only a few samples had the highest severity.

For the % of severity-duration extent (for the few FL of about  0.6 severity, [max= 0.63, as I recall]) the graph just shows an orange trace overlying the axis, a virtual  zero-% showing of the sleep time (considering, say, their negligible 10 seconds vs the  27k-seconds sleep session segment).

That smaller distribution graphic is close to being a representation of the tabular results for 20-second moving averages.

The upper FL chart's title erroneously shows 10 seconds instead of 20 seconds. That graph reflects 10 sample point (20 seconds) averaging.

[sheepless]

unfortunately, most of that is beyond me.

I'm (mis?) reading the graph to indicate that low count and low time = high severity on the resmed scale but since that's counterintuitive I don't think that's your intent. can you 's'plain it so even I can understand what that graph is telling us?

[2SleepBetta]

This attempts to clarify the percentages indicated in the graph.

Let's not think FL but of having an Rx or device to control pain and our logging the pain intensity whether of short or long duration.

To measure efficacy of therapy:

Count discrete pains or waves of pain (assume all are square waves here-sudden rise, constant level, sudden drop)
Log the duration of each pain
Log each pain intensity event on a scale 0-10.
Log how many pains of each intensity level 

Lets hope worst, most intense pain is far below 10. (the scale is at the bottom of  your graph of pain)
Lets hope worst, most intense pains are of short duration. (the percentage of time is at right of graph)
Lets hope worst, most intense pains are few (the count percentage is at left of graph)

As it turns out, all hopes are realized.

Accordingly there will be (in all probability we hope) the lowest count of most intense pain and the least percentage of total time suffering that pain intensity.

The illustration seems analogous, if not I'll give it another go. Or what is it I am missing ? My poor communication skills yet again.

Notes:

--My plot of the few points >=0.6 appears to lie at the zero count and percentage line that coincides with the severity axis; this is due to graph scale. For, say, a count of 5 vs a total count of, say, 25,000 the ratio of 1:5000 fraction looks like a zero value at the graph's scale. It is similar for a count of, say, 5 vs the 500 count grid line

--Time is not a factor in the severity level scale number, SFAIK.

However, I've almost come to think the algorithm may be designed to stack up some persistent continuous levels of severity (for uniformly deformed wave shapes) as a means of flagging possible seriousness of a persistent flow limitation. We sometimes see "marches" of visually the same wave form, wave after wave, all having a fixed specific severity individually; but we see their associated FL build upward stairways, one step higher for each additional "unvaried" wave form. For other marches of this kind the severity level changes very little; no stairway, just a "platform" of severity. Go figure. The mysteries of the algorithm. Different detection and flagging in different FL settings or "FL climates"?

[sheepless]

thanks.

it's not your communication skills. it's me being obtuse.

let's see if i have it now. it isn't correlating the extent of count & time with the resmed severity scale (where higher count & longer time should rank higher) as i first thought (hoped maybe). instead, it shows a specific example of the frequency of count & time at various levels of the resmed scale.

btw, I came away from an earlier exchange thinking you'd concluded time is a factor in resmed's severity scale. if it isn't, I'm back to suggesting reports of duration would be as important & useful a metric as frequency. idk how difficult it is, but that you are able to graph time tells us it's do-able.

[2SleepBetta]

thanks. 


it isn't correlating the extent of count & time with the resmed severity scale (where higher count & longer time should rank higher) as i first thought (hoped maybe). instead, it shows a specific example of the frequency of count & time at various levels of the Resmed FL scale.


btw, I came away from an earlier exchange thinking you'd concluded time is a factor in resmed's severity scale. if it isn't, I'm back to suggesting reports of duration would be as important & useful a metric as frequency. idk how difficult it is, but that you are able to graph time tells us it's do-able.

Correct. The graph is not exactly correlating count and time percentage with severity. But I think your hope was and is on the money and is partly fulfilled in the graph showing a moving average flow limitation (averaging and, in a sense, weighting the often ambiguous FL "severity" indicators). Graphs like mine, or better ones, can convey relative quantitative significance of moving averages of logged severities across the entire sleep session. Actual physiological severity of FL levels is another matter: how injurious is an O2 deprivation or CO2 excess level and how about numerous arousals? There are, I assume, normal ranges of tolerance for those metrics--both unknown to me. Scoring criteria for hypopnea may suggest an approach to setting a maximum TV drop range. 

The moving average (MAvg) approach factors in duration by including multiple FL events (5 events+/-? I don't know the optimal number to choose, but the graph illustrated 20 "seconds worth" of whole and fractional FL events), whether all were at one level of severity or at several levels. In that sense the graph does what you hoped. The  MAvg, as I see it, tames our often naturally erratic human biometric data. A sudden isolated high or low datum is blended in and has its limited effect in that way. Our respiration does not crash if a breath or two is skipped or huge.

All this must keep in mind that the Sao Paulo study strongly suggested that their control group (of "normals") sleep indicated that those who are free of recognized sleep troubles had flow limitations ("fL"?)---fL being related to but not the same as Resmed's FL---as much as 30% of their sleep time.

This thread of yours, with the issues just mentioned, together with high levels of FL of my Autoset days motivated my fussing around with one critical dimension of the multifactorial fL affliction that is variably and sometimes unreliably reflected in the FL flag: what are the impacts Resmed FL flags signify with respect to our vital Tidal Volume, TV, and more significantly, to Minute Volume (MV)--MV being a moving average of TV (by the way)? The intent of the thread "A call for VBA help . . . " was and is to tease breath by breath TV out of the FR curve and see how TV is reduced in FL locales. Work has shown significant agreement of FL with TV drops, but there can be episodes where FL--interpreting it as showing TV drop--vastly overstates degree of drop. The Resmed is responding to certain wave shape deformations and RR disturbances at those points it seems. Examination of curve shapes and inspiration times are often needed to make sense of continuous and/or high levels of FL.   

The big point in all as I would argue is that all sleep medicine, its devices and this forum are about maintaining sleepers' need of a certain sustained level of TVs per fractional minute, MV being a bit long for a therapy guideline.. One astute member has said--er, no, I think this is accurate paraphrase--TV is not important. I believe he meant to say that other channels of data presented by OSCAR are more important for diagnosing our symptoms and needs and for titrating our machines.

The concept and mechanics of doing that graph were easily done except for using a spreadsheet (Excel in my case) to clumsily deal with two huge data sets. The time and severity levels we see are just two columns of numbers in the PLD data file. Simple. But synchronizing the conceptually simple interleaving of the FR's 0.5 Hz data set with the separate FL 25 Hz data set is slow in Excel once you've figured out which and how to use unfamiliar math and selection functions as are available and most needed in Excel (and it will do the job).

Pure speculation here, as I go back to sheepless' time as severity and to real fL seriousness as well as Resmed FL scale factor:

 I 've wondered if the algorithm can sometimes get lost amid longer continuous low to high level fL or, on the other hand, does it simply begin to send a separate message overstating FL: "The length and/or level of this long, sustained FL can be serious and must be emphasized more by continually stacking up FL higher and higher", all that as if waves being evaluated by the algorithm were more and more misshapen though they are visually alike and would otherwise merit the same level of FL severity, not increases. There can seem to be reflection of a sensitive dependence on initial or preceding algorithm states, a kind of detection/evaluation hysteresis. This can partly explain why a train of like waves, slightly misshapen, will be flagged in one period or sleep session but visually similar waves will not be flagged at another time or other session.

Wow, I go on and on thinking I'm seeing and learning more but seeing clearly that the fraction I think I know is getting smaller faster.

[sheepless]

[cathyf]

Ok, this is a pretty basic question -- I don't believe any data taken during leaks. I see a lot of crazy data during even small leak, and don't believe it unless I can see the same thing without leaks.

Of course what really makes me nervous is the question of whether I believe that the sensors on the machine are always measuring leak properly...

[pholynyk]

It's important to remember that the sensors in the machine only measure pressure and total flow - in fact the total flow is calculated from the pressure differential across a known obstruction. Every thing else is calculated. I like to think of the breathing as ripple riding on the steady state venting of the mask. The venting of course varies with the pressure, and ResMed must assume some standard vent curve; also EPR complicates things. And transient leaks change things even more.
So suspecting the flow rate curve gets mangled by transient leaks is not unreasonable.

[cathyf]

Also what is interesting me more recently (because I'm trying to understand what vauto does for me that autoset does not) are the periods of wildly gyrating inhale curves

with EPR https://www.dropbox.com/s/2p9u2lwgpazdd9...0.png?dl=0
w/o EPR https://www.dropbox.com/s/vkttvomr0hwqv6...1.png?dl=0

If this is some sort of artifact, then I'm not willing to trust the data much at all!