Hello Guest, Welcome to Apnea Board !
As a guest, you are limited to certain areas of the board and there are some features you can't use.
To post a message, you must create a free account using a valid email address.

or Create an Account


New Posts   Today's Posts

flow limitations expressed as an index?
#31
RE: flow limitations expressed as an index?
looking back at pholynyk's post at #7, it's clear the flow limitation graph has an x as well as a y dimension, y being severity and x being duration. what's not clear (to me) is whether the severity index of 0-1 incorporates duration. it looks to me like it doesn't. if duration is factored in, my proposal is moot. if not, an important bit of information is overlooked. in that case, 2SleepBetta's suggestion at the top of post #25 would capture both dimensions. simply sum the x axis times at each y axis increment (in whatever incremental units the machine reports), divide by usage hours. this next step is beyond my mathematical capabilities but if the result is a huge number, divide it by some number to reduce it.
Post Reply Post Reply






Donate to Apnea Board  
#32
RE: flow limitations expressed as an index?
(04-13-2021, 08:33 PM)sheepless Wrote: I'd like this deeper dive into flow limitations to continue but hope my much less complicated proposal doesn't get lost.

"in some cases bigger FL number means more flow-limited than smaller FL number, where in other cases a smaller FL number means more flow-limited than a bigger FL number."

probably just me being obtuse but I'm not sure I follow. are your references to the flow rate graph, the flow limitations graph under the Daily tab, the flow limitations graph under the Overview tab or the flow limitations stats under the Daily tab? like, when you refer to the FL curve, it sounds like you might mean the flow rate curve?

"Without a cervical collar, I have dramatically more flow limits than when I wear one"

IF I understand your post (emphasis on the IF), I still think it makes sense as a definitional issue. I'd say the collar is keeping the airway open enough to reduce the restrictions from 90 or 100% (whatever the definition of oa is) to 40 or 50% (whatever the definition of flow limitation is). partial opening of the airway producing fewer apnea, more flow limitations.

To give an example of what I mean... Here's a 3-minute 45-second sample of me without a cervical collar on.
   

Look at the first four breaths on the left vs the last 7 breaths on the right. All those are scored as FL of zero. Look at 2:20:44 and 2:21 -- more ludicrously non-zero zeros! Those last 7 breaths on the right there are truly not flow-limited (or not much, anyway). But the preceding breaths are a mess! And how the machine scores that mess is pretty haphazard, I think. Here's the actual data, with the FLG's marked in blue and green:
   

When you look at the data, there is a time stamp associated with an amount of flow limitation. When the machine drops an FLG at a particular second, it's very unclear as to what interval that flow limitation is supposed to be associated with. In the OSCAR graph, the value begins with the time stamp and is drawn as persisting until the next FLG that gets logged to the card. Looking at my data, that's obviously not true in any general sense. It looks to me like the machine is calculating flow limitations based upon some data in the neighborhood of the timestamp, but I'm pretty sure that the time interval that the machine uses to produce the value is usually not all the way until the next FLG drop. It looks to me like sometimes the machine is paying attention to the FLs and doing a calculation and logging it. But whether the machine is paying attention to the level of FLs all the way to the next FLG data that gets logged -- uh-uh, I don't think so!
Post Reply Post Reply
#33
RE: flow limitations expressed as an index?
@cathyf I don't really understand where you are going with that.
If you are interested in the "why": there are crucial graphs missing.
Flow rate, the high res (mask) pressure, your Flowlimits and finally the leaks would be the minimum.

What is quite clear so far: quite heavy artefacts from heart (I would estimate a heart rate around 90). From the looks of the flow rate I would suspect EPR is working overtime and obviously triggered early or mis-timed ... which than of course results in those flowrates. ... we'll see about that in the high-res pressure curve^^
Where is your data coming from? Right from the SD-Card / EDF-file or from SleepyHead's own data-storage? (most likely the latter as ResMed reports that data at 0.5 Hz continuously and only SH with the time-delta-format is compressing that a bit)
Post Reply Post Reply
#34
RE: flow limitations expressed as an index?
(04-14-2021, 01:48 PM)TBMx Wrote: @cathyf I don't really understand where you are going with that.
If you are interested in the "why": there are crucial graphs missing.
Flow rate, the high res (mask) pressure, your Flowlimits and finally the leaks would be the minimum.

What is quite clear so far: quite heavy artefacts from heart (I would estimate a heart rate around 90). From the looks of the flow rate I would suspect EPR is working overtime and obviously triggered early or mis-timed ... which than of course results in those flowrates. ... we'll see about that in the high-res pressure curve^^
Where is your data coming from? Right from the SD-Card / EDF-file or from SleepyHead's own data-storage? (most likely the latter as ResMed reports that data at 0.5 Hz continuously and only SH with the time-delta-format is compressing that a bit)

The place I'm coming from is the suggestion "let's multiply the reported flow limit numbers by their durations and add it all up to get the area under the curve" -- because it looks to me like we don't know the durations. Sure, the duration is bounded by the timestamp of the next data point, but I'm maintaining that it looks significantly shorter than that.

As my physicist husband talks about data, the way that the data jumps around "is not physical". Reading the patent applications that makes total sense -- the algorithm is doing a lot of "if this look at that" and using numbers to mean big/little rather than more exact measurements, and not calculating at all when it doesn't make sense. And we look at the data and eyeball "this looks pretty good, that looks ugly" and I don't think you can get there by doing something as basic as multiply-and-add. Because there's way too much going on you are doing too much apples-to-oranges.

(And, yeah, I'm dumping the data out of OSCAR with the export data command.)

(And I would put more data up but I'm out of room for allocated attachments and need to spend some time cleaning up files.)

(And, annoyingly enough, my fitbit seems to be missing my sleeping heart rate data for just that night, although it's there in the whole day's data, and the statistics are calculated. I seem to have stumbled upon a fitbit bug!)
Post Reply Post Reply






Donate to Apnea Board  
#35
RE: flow limitations expressed as an index?
"The place I'm coming from is the suggestion "let's multiply the reported flow limit numbers by their durations and add it all up to get the area under the curve" -- because it looks to me like we don't know the durations. Sure, the duration is bounded by the timestamp of the next data point, but I'm maintaining that it looks significantly shorter than that.""

clearly most of us don't understand how fl are reported (maybe the oscar folks do?). I hope all you smart folks figure out how it can be better understood and improved.

while you work on that, why not take advantage of what we have. it gives us more information than currently reported (I haven't been told differently yet anyway; nor has anyone said it's a bad idea and why) to help determine the significance of fl in individuals and to compare in relative terms the results of settings changes. ideal or not, completely accurate or not, it moves us forward.

I've had my say (ad nauseum maybe). I don't sense much enthusiasm but it'll be up to the membership, moderators, advisory group and especially the oscar team to decide whether to do anything with it.

I hope you-all continue the discussion and I hope you figure it out!
Post Reply Post Reply
#36
RE: flow limitations expressed as an index?
(04-14-2021, 07:00 PM)cathyf Wrote:  I don't think you can get there by doing something as basic as multiply-and-add. Because there's way too much going on you are doing too much apples-to-oranges.
Well, mainly because the "Flow Limitation" reported by the machine is just a "Flow Grade" and as such measured on an ordinal scale.  Data of that kind must not be added, subtracted or divided by themselves or by other numbers.  Like school grades: It's pretty senseless to add an A to a C or multiply a D by 3 ...
Post Reply Post Reply
#37
RE: flow limitations expressed as an index?
@cathyf: "How" 'FLs' are reported is quite simple: ResMed logs every 2 seconds a number for that. This number refers to the last 2 seconds, which in the end means: the last breath (to be precise: the last completed inhalation as only the inhalation is taken into account) ... this means there can be 2 datapoints (= 4 seconds in total) for 1 single breath or 1 for 2 - depending on the respiratory rate at that moment.
The question is: what do these "FL" mean (exactly) and is there some value or deeper meaning to them?
ResMed says in their "Clinical Manual" for ResScan regarding this: 'Flowlimitations: from flat to round'

Well, Flowlimitations are 'a bit' more than just flattenings:
Sleeprider worded it quite nicely:
(04-09-2021, 04:17 PM)Sleeprider Wrote: but I have seen considerable differences between individual results at the same 95% flow limit. I generally like to see less than 0.1 in that statistic, and it is still disruptive for some people and impossible to get for others.

I wouldn't be that polite and frankly say: the real correlation is only slightly above a random number generator. This might or does change for those with non-hypoxic breathing disorders - BUT: the devices are not meant for diagnostic purposes - those are devices for treatment.
If you have a problem with flowlimitations and only take the flattenings this might have nevertheless 'some' meaning but that would be comparable to just taking OAs (or CAs or ... - just pick one) and call THAT "AHI".
AHI is a well established surrogate for sleep apnea syndrome (meaning a HYPOXIC SRBD) but that consists of apneas AND hypopneas (both obstructive and non-obstructive and even 'mixed') - if one takes only 1 single kind of those events and then says that this number is the same or even correlates to the AHI that is simply false or the result would be overall meaningless - for some that would show some kind of correlation but overall it would be hit or miss and that is NOT a correlation.

If you take an incomplete value and compress that even further you always end up with nothing (or: a random number would not be much worse).
If you nevertheless do that and put a shiny label on it, people will start to put even more meaning in it - but in this very case it never had that to begin with!

However: as OSCAR seems to be able to export that data, one could do that (e.g. in Excel) for oneself. As it seems to be in the time-delta-format that is no magic:
supposing the FL-values are all ordered by time (and already filtered to be just the FLs), time beeing column A and in C the value for that timestamp. In the original data from the machine the FL-value that is in row 'x' would repeat itself (every 2 seconds) until the timestamp that is in row x+1.
Sou you could sum them up over: Cx * (A(x+1)-A(x)) [that would be in seconds - divide it by 2 if you want the sum of the original datafrequency]
The Index would than be that sum divided by the total duration (well - more in hours to get a reading different from 0^^).

Expectation: meaningless and completely obscure value. (and I mean even more obscure than before)

If we stick to the school-grades (where at least for me just talking about the average or index is quite common): if someone applies to be a rockt-scientiest how much value is in the index of the scool-grades in sports, arts and history??
Well ... as "Turnvater Jahn" said: 'inside a healthy body is a healthy mind', for arts you have to be creative and it is always good to know where you are coming from ... all true - but the main focus for a rocket-scientist would be the STEM fields.


What might yield some information would be the index AND the average distance of prolongued episodes with those flattenings. (but this information is far easier and directly seen if one simply looks at that graph).
Keep in mind: the sao paulo cohort established up to 30% of the night with flowlimitations is healthy and normal. That would mean the 70-percentile needs to be different from 0.00 - If that actually would be the case this could also be seen at the first view: the pressure would be equal to CPAP and the FL-graph would be (nearly completely) black (or colored or whatever).
Flattenings are not all there is to flowlimitations - 95% is way to high ... something around 80 to 90 together with the amount of time where the resp. rate doubles should give a starting point for the actual amount of the residual flowlimitations. (personally I would consider the latter as worse ... but I know that I am sensitive to intra-breath pressure changes - so it is quite safe to assume that those would strike me more than just flattenings ... but I have 'just' OSAS and thus flowlimitations do not play such an important role for me anyway^^)

But to introduce an overall new index with real value it would be necessary to establish the importance or at least a clear (and preferably strong) correlation with the underlying data in the first place! And this - at least from my point - not given for ResMeds "FLs" - so far it is hit or miss: many have no trouble with that or at least the change in that value does NOT correlate to the subjective feeling (or as presented by sheepless: make it even worse), some have trouble with flowlimitations but the reading from the machine says there are none.
In most (if not all) cases other values or just the overall "look" from the respiratory flow are much better and have a much stronger correlation. (and that still does not mean that there is any causality in it)
FLs I would at most attribute a relation but definitely no correlation.
Post Reply Post Reply






Donate to Apnea Board  
#38
RE: flow limitations expressed as an index?
@TBMx: "If you take an incomplete value and compress that even further you always end up with nothing (or: a random number would not be much worse). If you nevertheless do that and put a shiny label on it, people will start to put even more meaning in it - but in this very case it never had that to begin with!"

Obviously, if GI, then GO, as in GIGO. However, as one who also read the Sao Paulo study months ago, here are a few comments about the quotation above and its context: 

(1) Limited in knowledge, I believe ResMed's FL is a valuable indicator of short-period flow limited breaths, if only for those with a high level of inspiratory flow limited breathing--the FL thing is more than marketing pandering. 

(2) We see and rely on high levels of FL "coaching us" to use EPR and pressure support, (PS), to fight it. Neither TBMx's quote of SR nor SR's apt observation supports dismissal or downplay of FL. The 95% number is patient relative, is relative to a particular sleeper's individual level of FL, his particular set of overall losses of air volume as a result of brief drops (neonate hypopneas?) in airflow delivery after and relative to a short period of higher delivery of air volume to the lungs. For patients A and B, 95% for A means nothing with respect to B's 95%.  Resmed, right or wrong, is credited by many knowledgeable PAPers as being more responsive to FL than other makers. Some (or all?) other makers do not even provide explicit FL markers. 

(3) sheepless's OP called for more revealing presentation of the overall effect of the FL we see. The simple summation of areas of the given indicators, as I suggested (and am not invested in), would simply consolidate the areas of the scattered FL. If the FL are garbage, of course the sum is.

Thanks to cathyf and her probing we are digging into and learning more about the matter of inspiratory flow limited breathing. I appreciate the OP question and all the solid contributions, TBMx's and cathyf's  particularly, which have helped me to learn more, to plow new ground as I hope to be converging on a helpful understanding of my arousals and micro arousals. I'd fiddled with the EDF and the large data file a year or two back, but in revisiting those last night, I am (and was) "all thumbs" with them.

In an effort to understand the airflow volume changes the FL indicates, my intent was to use EDF (its Excel CSV) to numerically integrate flow data and determine the areas (air volumes) of a span of inspiratory airflows--a span which had a variety of FL presences. I would begin that effort with singlets. That done, then take on multiples, some of those FL with near and far neighbors, including those situations where clustered FLs' indicators stack upwards. With a sub-goal integration done, then compare and contrast, try to reconcile areas and cumulative times found to the FL indications smart ResMeds provide. (In passing, my mostly locally-lonesome "M-tipped" inspiratory peaks with their own lonesome FL are most often flagged with a 10-second FL marker of 0.02 to .04. An "M-tip" peak looks like a good candidate to analyze first as there is a definite one to one correspondence when, as often, the preceding and following few  inspiratory wave are relatively uniform (i.e., are generally fairly rounded). Hopefully, the 25 Hz signal is sufficient for the short span of time. . . . But, whew, so many things for us to balance in our "look at" and "must do" prioritizations of time.)

Thanks to TBMx I now know or have some idea of what time frames Resmed use for registering a FL. But, as cathyf observed in a post, regularities we observe are often affected by overriding if-then controls, so anomalies and compromises (and probably omissions) confound our discovery process.

Harking back to a far-back post (maybe by sheepless?), I see my and our intent in trying to unravel the exact meaning of ResMed FL reporting is to get a better understanding of how to improve our sleep. Most of the relevant FL information is publicly available in ResMed patent applications. We are not trying to reverse engineer their algorithm for economic gain, but are trying to understand what FL means in our quest for our most restful sleep. To whom, for what, and how is the FL mark helpful or indicative? ResMed could have been more helpful to our understanding (or I haven't looked in its right places re FL). I would not be surprised to find that there are Resmed advances in handling of FL in its new 11 series of machines.
Post Reply Post Reply
#39
RE: flow limitations expressed as an index?
(04-15-2021, 07:26 AM)TBMx Wrote: @cathyf: "How" 'FLs' are reported is quite simple: ResMed logs every 2 seconds a number for that. This number refers to the last 2 seconds, which in the end means: the last breath (to be precise: the last completed inhalation as only the inhalation is taken into account) ... this means there can be 2 datapoints (= 4 seconds in total) for 1 single breath or 1 for 2 - depending on the respiratory rate at that moment.

First off, TBMx, I think we are at pretty much the same place as far as the numeric usefulness of the flow limit numbers and durations, it's just that I think we got there from different directions. What I am seeing from my data is that while ResMed may log a flow limit value every 2 seconds it is not calculating that number every two seconds. I do have long stretches where each breath looks pretty similar to the breaths before it and the breaths after it. But I also have lots and lots of time every night where every block of 1-5 breaths looks totally different from the little groups of breaths before and after -- a wild, jaggy, ride. When I see long stretches where my breathing looks pretty consistent -- whether good, bad, or really ugly -- it looks like the flow limit graph is "tracking" the flow limitation in the neighborhood. But where the pattern of breathing is jumping up and down all over the place, the flow limit graph looks like it's unconnected from the waveforms in the same time slices. I think that a lot of those data points are "leftover" from calculations done shortly before over breaths before that, and it also looks like the calculations are not always looking at breaths over the same length of timeslice, too. Given what the patent applications say, that's really not surprising. As multicast says, it's more a "grade" than a "score".




(04-15-2021, 07:26 AM)TBMx Wrote: However: as OSCAR seems to be able to export that data, one could do that (e.g. in Excel) for oneself. As it seems to be in the time-delta-format that is no magic:
supposing the FL-values are all ordered by time (and already filtered to be just the FLs), time beeing column A and in C the value for that timestamp. In the original data from the machine the FL-value that is in row 'x' would repeat itself (every 2 seconds) until the timestamp that is in row x+1.
Sou you could sum them up over: Cx * (A(x+1)-A(x)) [that would be in seconds - divide it by 2 if you want the sum of the original datafrequency]
The Index would than be that sum divided by the total duration (well - more in hours to get a reading different from 0^^).

Expectation: meaningless and completely obscure value. (and I mean even more obscure than before)

I actually did that calculation. (but you knew that, LOL)

This was the night I was working with:
http://www.apneaboard.com/forums/attachm...?aid=31089

As you can see, the summary statistic median flow limit is worthless. (An aside -- in the statistics screen there is a column for flow limit which is the average median over the period -- it's worthless -- maybe replace it with either the average 95% or average max?) The 95% is 0.28 and the "max" is 0.79.

The calculation I did...
Took all of the FLG data points out of the file.
If a data point is in the same session as the data point that follows, I calculated a duration by subtracting this time stamp from the following time stamp. (if it's the last data point of a session, duration is zero.)

Total seconds: 30148
Total in-session seconds: 29580
Total seconds where FLG value > 0.00: 11544

Then, in the same spirit as we measure power in KilowattHours, I created the concept FlowLimitSeconds. I multiplied each FLG value by its duration, and then summed them all.

FlowLimitSeconds: 1432.76

1432.76 / 11544 = 0.12
1432.76 / 29580 = 0.05

So, are either of those numbers less useless than 0.00 / 0.28 / 0.79 ?

yeah, "meaningless and completely obscure value" -- I think that about sums it up!


(04-15-2021, 07:26 AM)TBMx Wrote: Keep in mind: the sao paulo cohort established up to 30% of the night with flowlimitations is healthy and normal. That would mean the 70-percentile needs to be different from 0.00 - If that actually would be the case this could also be seen at the first view: the pressure would be equal to CPAP and the FL-graph would be (nearly completely) black (or colored or whatever).

When was the Sao Paulo study done? Back at the beginning of my adventure I did an ApneaLink home study -- May 2014. None of my 3 in-lab studies have mentioned anything like flow limits, but the ApneaLink did.
breaths: 6324
Flow lim. Br. w/o Sn (FL): 3824
Flow lim. Br. w/ Sn (FS): 164
% Flow lim. Br. w/o Sn (FL): 60%
%Flow lim. Br. w/ Sn (FS): 3%
But the kicker is that under the "Normal" column it says that "normal" %FL is "< about 60%", and "normal" %FS is "< about 40%". So if ALL your breaths are flow-limited, with 60% w/o snore and 40% w/ snore, that's still "normal"!!! (AND -- the ApneaLink measures from power-on to power-off, so includes all of the awake time, and that's 60%/40% of the awake time, too!)

Also, as to your "nearly completely solid" -- no, that's not what they do. The solid color comes from the FL number whipsawing back and forth between zero and large numbers, which in my case is indication of arousals. because when I'm awake my FLs are zero. The more interesting cases are where the FL goes off zero for long enough to be seen as a "hump" on the zoomed-out whole-night view -- which indicates long periods of even, flow-limited breathing that I'm not fighting back against.

I'm back at seeing that the machine is fighting back against this thing that it is reporting, but I don't think that it's doing a very good job of reporting what it is calculating. And I'm not convinced that it's doing a very good job of adjusting pressure the way it should be. Because I know from long experience that if I constrain the pressures such that the machine can't react to the FLs, then that means the machine measures more FLs, but does NOT in general mean that I get apneas or hypopneas. The rationale for reacting to the flow limitations is the theory that 1) raising the pressure will eliminate or reduce the FLs, and 2) if the FLs are allowed to continue without a pressure change then apneas and hypopneas will follow, while if the FLs are reduced, the apneas/hypopneas will be prevented.

In my case, I think 1) is true, but 2) is not -- FLs are their own thing, and NOT early indicators of impending apneas/hypopneas.

Which I think is what you are saying, too, TBMx
Post Reply Post Reply


Possibly Related Threads...
Thread Author Replies Views Last Post
  Blank flow rate in Oscar pappingit 7 58 05-02-2021, 12:58 PM
Last Post: sawinglogz
  Flow Limitation Units in Statistics cathyf 15 410 04-01-2021, 02:10 AM
Last Post: cathyf
  Oscar: Weird spikes in flow rate and other feedback readings MikeBob 5 228 03-05-2021, 07:52 PM
Last Post: MikeBob
  Sleepyhead using Icon auto / Flow chart John McGowan 10 818 02-22-2021, 02:46 AM
Last Post: GuyScharf
  PS BIPAP ASV not showing Flow Rate in Oscar 19MaleMabyeUARS? 1 164 01-06-2021, 10:59 PM
Last Post: Sleeprider
  Flow rate graph shirsh98 9 329 12-29-2020, 02:32 PM
Last Post: Sleeprider
  reading the flow chart staceyburke 7 443 11-19-2020, 09:21 AM
Last Post: multicast


New Posts   Today's Posts






About Apnea Board

Apnea Board is an educational web site designed to empower Sleep Apnea patients.