Resmed's Central Sleep Apnea Detection - Resmed publication

http://www.resmed.com/us/assets/documen ... -paper.pdf
Central Sleep Apnea Detection and the Enhanced AutoSet Algorithm
J P Armitstead, PhD; G N Richards, MB ChB; A Wimms, BSc; A V Benjafield, PhD
Applied Research and ResMed Science Center, ResMed Ltd, Sydney, Australia


Armitstead seem to be their new inventor - patents to his name.

Thank you very much for that link, ozij!

Extremely interesting.

THANKS, Ozij

Thanks, ozij.

The following is my attempt to be entertainingly flippant in responding to the paper, not any attempt by me to be scientific . . .

. . . the algorithm can determine whether the apnea is associated with a closed or open airway . . . This is not quite the same as determining whether the apnea is central or obstructive. . . .

Nice point. Too bad it wasn't called "open-airway-detection" then, instead of "central sleep apnea detection." Right?

. . . The 1 cm H2O amplitude pulses are able to be perceived by a patient if they are awake but are not large enough to arouse them from sleep. . . .

Ever? With any patient? Not even arousing some patients from one stage to another? Really? I wonder how they proved that? Well, they said it, so it must be true. They did say "arouse them from sleep," not 'cause arousals during sleep.'

. . . The CSAD [central sleep apnea detection] algorithm determines resistance of the upper airway. To do this accurately the impedance of the circuit needs to be known. CSAD works optimally when recommended circuit configurations are used . . . the accuracy falls rapidly with leaks greater than 30 L/min. . . .

So an unknown mask and hose can throw things off, much like leaks?

. . . unknown apneas do not cause an increase in delivered pressure. . . .

Pretty important that apneas be 'known' ones then, with that machine. Assuming they are centrals is an interesting choice, perhaps making the machine particularly suited to central patients with an obstructive component.

. . . Very high impedence combinations such as an AB filter and 15 mm diameter tubing should not be used. . . .

I hope CPAP.com takes note of that.

. . . No subjects reported disturbance of sleep due to the CSAD. . . .

And that proves what exactly? What about their EEG, etc?

. . . There were an additional 53 events classified as unknown (excluded from analysis). . . .

Even with lab conditions. Wow.

. . . a subject (treated with the previous AutoSet algorithm) experiencing obstructive apneas over a (approximately) one hour period. The treatment pressure in the first 8 minutes is limited to 10 cm H2O which is insufficient to prevent recurrent apneas, and the pressure rise between 8 and 10 minutes is due to snoring and flow limitation. . . .

So they were able to find a guy with apneas without any flow limitation or snoring for 8 minutes. Good thing they found one! Of course, if that patient was a tweaker, he would know to raise his minimum higher than 4 cm to make sure that didn't keep happening.

. . . Testing the accuracy of CSAD in clinical trials is problematic as the algorithm prevents most obstructive apneas during therapy. Therefore it is difficult to test multiple circuit configurations, and mask types at varying levels of pressure and inadvertent leak in a clinical trial. This was overcome by extensive bench testing using patient simulators . . .

It's the old 'if you can't prove it works on a sleeping human, prove it works on another machine and on people pretending to be asleep' trick! I like it! Let's hear it for bench tests and patient simulators!

. . . A further methodological problem is the difference between apnea classification (based on upper airway resistance) and the usual classification method using effort bands. In the clinical study the presence of cardiogenic flow was used as an indicator of airway state. . . . There is, however, limited published evidence of the use of cardiogenic flow to determine the state of the upper airway. . . .

So for a clinical study to validate your way of measuring the state of the airway, you used an unvalidated way to measure the state of the airway? I have to ask, Sir, do you have a drawer full of unpublished studies where the machine's scoring didn't do so well in comparison to effort belts, the" usual classification method"?

The problem that I see with this paper is that this is not independent research published in a peer-reviewed journal. This "paper" is proprietary material from Resmed and contains only a handful of referenced citations from other papers. The methodology and conclusions may very well be legitimate, but it would be nice to see an analysis from a peer-reviewed journal.

Laurie1041 wrote:The problem that I see with this paper is that this is not independent research published in a peer-reviewed journal. This "paper" is proprietary material from Resmed and contains only a handful of referenced citations from other papers. The methodology and conclusions may very well be legitimate, but it would be nice to see an analysis from a peer-reviewed journal.

I hear ya. But I'm glad they did it. Some info is better than none. I appreciate hearing them say what they are trying to do and how they are trying to do it. I learned a lot from that paper. No one else has any financial incentive to provide that info on ResMed's machines, and ResMed (and the other APAP manufacturers) would never let their secrets out to independents to study anyway.

Ozij, Thanks for the post! I think the run away pressure condition with APAP is exactly what happened to me until I went back to the plain old fixed pressure.

The paper implies that this algorithm is [already?] there in the S9 Autoset [enhanced?]. Is that made available on commercial units? Does anybody know if they have this enhanced algorithm on other models?

Curious: does Respironics have a similar algorithm too?

Laurle, such propitiatory papers are the earlier steps before peer review publications. Also needed is other independent researcher duplicating the results.

. . . The 1 cm H2O amplitude pulses are able to be perceived by a patient if they are awake but are not large enough to arouse them from sleep. . . .

Ever? With any patient? Not even arousing some patients from one stage to another? Really? I wonder how they proved that? Well, they said it, so it must be true. They did say "arouse them from sleep," not 'cause arousals during sleep.'


If you think 1 cmH2O oscillation during apneas is bad, try using Respironics. 1.5 cmH2O increase/decrease *EVERY* three minutes. (Apneas or not.)

[quote="feeling_better"]Ozij,

Laurle, such propitiatory papers are the earlier steps before peer review publications. Also needed is other independent researcher duplicating the results.[/quote]

I agree that the new technology is being incorporated in new S9s too soon. However, those S9s could run on plain CPAP mode also, and most of the clinical data would be available in this mode too (i think).

See a similar question about the S9 statistical accuracy by "zephyr" here:

http://www.apneasupport.org/viewtopic.php?t=26325

Laurie1041 wrote:The problem that I see with this paper is that this is not independent research published in a peer-reviewed journal. This "paper" is proprietary material from Resmed and contains only a handful of referenced citations from other papers. The methodology and conclusions may very well be legitimate, but it would be nice to see an analysis from a peer-reviewed journal.

And within that handful of referenced citations, the ones referring to the enhanced algorithm are nor really peer reviewed -- those were posters paid for by ResMed, at meetings.
However, when we're trying to learn about our equipment, proprietary papers are a beginning, and a description of how the algorithm has changed may not get into any peer reviewed journal.

jnk wrote:

. . . A further methodological problem is the difference between apnea classification (based on upper airway resistance) and the usual classification method using effort bands. In the clinical study the presence of cardiogenic flow was used as an indicator of airway state. . . . There is, however, limited published evidence of the use of cardiogenic flow to determine the state of the upper airway. . . .

So for a clinical study to validate your way of measuring the state of the airway, you used an unvalidated way to measure the state of the airway? I have to ask, Sir, do you have a drawer full of unpublished studies where the machine's scoring didn't do so well in comparison to effort belts, the" usual classification method"?

Despite the paper's title, containing the words "central sleep apnea detection" (a PR choice? a title meant for the bosses?) the paper actually makes it very clear that the aim of the enhanced algorithm was to detect whether the airway was open or not:
All emphasis mine:

Automatic algorithms respond to the presence of apneas (assumed to be obstructive) by raising delivered pressure because, although this does not treat the apnea that is detected, the pressure increase reduces the likelihood of further obstructive events occurring. If the apnea, however, is central an increase in pressure is inappropriate and may increase the chance of further central apneas.... By measuring upper airway resistance in this way the algorithm can determine whether the apnea is associated with a closed or open airway and can determine whether it is appropriate to increase the pressure or to make no response. This is not quite the same as determining whether the apnea is central or obstructive. By definition, all obstructive apneas are associated with a closed upper airway, but some central apneas can be associated with a closed airway if the pressure is below the airway closing pressure. If the airway is closed during a central apnea the appropriate algorithmic
response is to increase the therapeutic pressure.

It makes sense to check an "open airway detection" method against another open airway measure - and one such measure is the existence of cardiogenic oscillations on the flow. These have actually been shown - in the following, peer reviewed publication - to be a very valid indicator, since they only exist when effort belts show no effort:
http://chestjournal.chestpubs.org/conte ... 3/660.full
Cardiogenic Oscillations on the Airflow Signal During Continuous Positive Airway Pressure as a Marker of Central Apnea*
Indu Ayappa, PhD, Robert G. Norman, MS, RRT and David M. Rapoport
doi: 10.1378/chest.116.3.660
CHEST September 1999 vol. 116 no. 3 660-666

...a total of 648 obstructive and central apneas were identified (range, 1 to 102 events in each patient). These were subsequently classified using the following criteria: central apnea was present if no detectable chest or abdominal movement occurred during any part of the event; and obstructive apnea was present if there was any chest or abdominal movement indicating respiratory effort.

Sixty percent of the 351 central apneas and none of the 297 obstructive apneas showed cardiogenic oscillations (Table 1). In four obstructive apneas there was a suggestion of oscillation, but this did not meet our criterion of five consecutive oscillations at the cardiac frequency. Specificity of using cardiogenic oscillation on the flow tracing to identify central apnea was 100%, and sensitivity was 60%, yielding a positive predictive value of 100% and a negative predictive value of 68%.

http://chestjournal.chestpubs.org/conte ... nsion.html

So, as far as validating the positive predictive ability of the FOT - testing in against cardiogenic oscillations is a good idea.

However, David Rapoport sold his patent for an algorithm detecting cardiogenic oscillations to Puritan Bennett, and he himself is now working for Fisher & Paykel.

It seems to me that this Resmed proprietary paper must suffer from the "Not Invented Here" syndrom -- God forbid they should mention something done by a competitor -- they even avoid his term "cardiogenic oscillations".

In a sense, this paper is about "see, we can do what PB did, without mentioning their name - we no longer raise pressure in response to open ariway apanea below 10, Yay!".

Effort belts say nothing about the state of the airway, and in that sense, would not have helped with the problem of finding out whether the airway is open or closed.

Thanks, ozij. That helps. I just can't get past the idea that events should be defined by effort or lack thereof, not by airway status, from a PSG point of view. It's a mental glitch of mine. One of many. If you claim to discern centrals, compare it to effort belts, I say.

It was an entirely new thought to me in that paper that increased pressure is an appropriate response to a closed-airway central event. I don't completely understand why that is, if the point is to prevent further events. I guess closed-airway centrals often precede obstructive apneas--which makes sense, I guess, since if there wasn't enough pressure to keep the airway open during the central, more pressure must be needed to prevent obstruction. Still, I am not fully grasping that concept as well as I should.

Personally, I'm glad they are trying new things all the time. As long as the changes don't make things worse, and I'm satisfied the enhancement was not a step backward, in my mind. I do think it would be nice if all the inner workings were shared with the medical community to evaluate openly, though, for the sake of the patient over the patent.

jnk wrote:Thanks, ozij. That helps. I just can't get past the idea that events should be defined by effort or lack thereof, not by airway status, from a PSG point of view. It's a mental glitch of mine. One of many. If you claim to discern centrals, compare it to effort belts, I say.

Yes, you do keep saying that... however, an auto has to supply good therapy in any way it can -- can you imagine compliance rates (and sales rates, for that matter) if in addition a mask we had to strap on effort belts to get our therapy?

It was an entirely new thought to me in that paper that increased pressure is an appropriate response to a closed-airway central event. I don't completely understand why that is, if the point is to prevent further events. I guess closed-airway centrals often precede obstructive apneas--which makes sense, I guess, since if there wasn't enough pressure to keep the airway open during the central, more pressure must be needed to prevent obstruction. Still, I am not fully grasping that concept as well as I should.

I haven't quit figured out that part either....

Thanks, ozij!

ozij wrote:

jnk wrote: Yes, you do keep saying that...

If at PSG, it is the effort belts that tell you what is a central and what isn't, then any company claiming to make a machine that can differentiate and label centrals MUST have its machine compared to effort belts worn by sleeping patients to prove that claim.

There. I said it again!

Next thing you know, home-machine manufacturers will be claiming their machines can score hypopneas! Oh, wait. They already claim that, don't they? Never mind. Well at least they don't claim their hypopnea scores are comparable to PSG scoring.

jnk wrote:

ozij wrote:

jnk wrote: Yes, you do keep saying that...

If at PSG, it is the effort belts that tell you what is a central and what isn't, then any company claiming to make a machine that can differentiate and label centrals MUST have its machine compared to effort belts worn by sleeping patients to prove that claim. ...[/quote ]
They also use the EEG leads to determine a central apnea. If there is no breating for 10 seconds and no effort during that time (as determined by the belts) *AND* as determined by the EEG leads the patient is not in a transition to or from sleep, then it is determined to be a central sleep apnea.

We often see folks here, who are worried about "ca" being reported, but they cluster throughout the night. Usually, those clusters of "ca" are more an indication of when the person transitions to and from sleep than an indication of central apneas. In actuallity, they should be reported as a "clear airway" apnea and not as a "central apnea".

But I am glad that both Resmed and Respironics is making the effort to adjust their therapy to properly handle central apneas.