Flow Runs? (with 420e)

I'm puzzled about the data I'm seeing and don't know how to interpret. I've read what -SWS has to say about flow runs and their potential relation to the algorithm being used and other possible factors, but I still don't get what's happening in my case.

Briefly, I average about 25 flow runs per hour on a regular basis, and sometimes more.

What's puzzling is that this appears to be completely independent of all the other metrics (apneas, hypopneas, and acoustical vibrations). In fact, I'm looking at data for a few nights ago when I had 231 flow runs for an average of 33/hr while having 0 apneas, 1 CA, 21 hypopneas, 5 hypopneas (fl), and 2 acoustical vibrations. That is, all of the figures appear to be "good", but the flow runs number is HUGE. Is this bad? What's the story.

The settings at this point were a min pressure of 9 and a high of 15, but the pressure as a matter of fact never went above 11. IFL1 is off, IFL2 is on, and the max pressure for command on apnea is set at the default 10.

What's with this constantly high flow run figure?

Ghmerrill, bottom line is that an FL runs index of around 25 may or may not be problematic. If neither desaturations nor sleep arousals occur, then those flow limitations are of little or no consequence to you. Without equipment to measure both event types, your own subjective clinical assessment may or may not yield how well you are actually faring with a flow runs index of around 25. However, with IFL1 turned off, you are preventing the 420e from responding to flow limitation runs. You may get better results with IFL1 turned on.

If you haven't already tried it, put IFL1 back on to factory defaults and see if that helps or worsens your FL runs, AI and HI indices. Most people should run the 420e with IFL1 turned on, exactly as it was defaulted from the factory. Those who must turn IFL1 off do so because the 420e over-triggers in response to their flow limitations, thereby causing a pressure-runaway response from the 420e. And in a few of those cases the pressure itself may cause yet more flow limitations, a veritable albeit somewhat rare vicious cycle.

ghmerrill, my flow runs index from a 5 month period is 38.5. That's with IFL1 always turned off.

As -SWS said, most people using a 420E autopap should leave IFL1 "ON", just like it comes from the factory. I had to turn IFL1 off due to what he mentioned -- overtriggering to something about my natural breathing and running the pressure up unnecessarily. In fact, -SWS was the one who helped me. I'd have never known that IFL1 was the culprit in the pressure run-ups the 420E was doing with me. Most people do fine with IFL1 "ON".

My AHI is always quite low, and I feel fine every morning. When using the 420E, I pay no attention at all to my high flow runs index. Often my flow runs index will be in the 40's and 50's; even 60something a few times. I think it's trying to catch up with my age.

boy this sounds like a bad burrito

heheh, no kiddin'!

*note to self - no more Taco Bell*

Hi Guys
I have 420 E and have no idea what all of you are talking about .
But I would like to learn ,as I am getting the software next week and my heated hose I hope. Thats if my DME ever gets his act together.
Is there a more comprehensive manual that I can buy for the 420 e
If there is where might I get a copy.
thanks Jacko

_________________

CPAPopedia Keywords Contained In This Post (Click For Definition): hose, DME

hi jacko,

If you got your 420E from a local DME, you might go back to your doctor and ask him to send an order (Rx) over to your DME tellling them to give you the clinician's manual for your machine.

In the meantime, there are a lot of interesting posts by -SWS about how the 420E works, what the settings mean, etc:

LINKS to message board discussions with "-SWS"

Yep
Ok thanks again RG
regards Jacko

Thanks for the additional information.

I started out with IFL1 on, but it seemed as though the pressure was running up constantly. So I turned it off. I'll try turning it on again, now that I have a better idea of what the pressure range should be. I was titrated in the sleep study for 7, but have discovered that a low of 8 or 8.5, or even 9, seems best for me. And it looks as though for the most part I don't really gain anything from setting the high over 12, though on occasion it will go up to 14/15 if the high is set there.

Last night, for example, I set the range to 8-12, and 90% of the time it was under 10, with the graph from 8-10 (inclusive) looking very symmetric (23, 43, 28).

I remain befuddled by some of the details of the of the "Detailed record". There doesn't seem to be much correlation between pressure response and the number of events during a given period (apneas, hypopneas, etc.). I seem to be awakened at one or two points in the night by high pressure, but when I look at the graph, those points don't seem to be distinguished from others in terms of what's happening. It just appears that at times the pressure goes up in response to a sequence of events (over say an hour) and then stays up. I wake up, reinitialize the machine, the pressure stays down, and I go back to sleep. Is this to be expected?

I guess I need to look at the algorithm(s) -- if I can find a decent description -- in order to understand this better. I don't have a good grip on what criteria are used to lower the pressure once it has gone up. I would guess there's some sort of latency approach here, but I can't detect anything uniform from looking at the graphs. It obviously isn't "time since last event". But looking at the graphs I can't see anything that looks like a coherent predictive model either. I suppose data is being used in the model that isn't displayed by Silverlining. I'll look at some of the articles mentioned in earlier threads.

Ghmerrill, one document many people forget to read is the on-line document available within the "Help on 420e" section of Silverlining. I would recommend reading the "Help on 418P" section as well, since the 420e is an algorithmic carry-over of the 418p, and there are interesting and relevant tidbits found only in the 418p help section. There is much more algorithmic commonality between those two machines, by far, than there are subtle differences.

Also, one experiment that you might consider is a comparison of all your indices at your optimum fixed-pressure (perhaps still to be determined) with those same indices at your optimum APAP pressure settings. Don't forget to also factor in your own subjective assessment regarding how you feel by day, as well, when making that comparison. Good luck!

-SWS: Thanks for the additional advice. I keep meaning to look at the 420e online material, but just haven't managed to do it.

Are you aware of any good studies that analyze these things (e.g., genuine epi or clinical studies)?

I have now looked at the "Performance of Auto-adjust Nasal CPAP Devices ..." document by Eiken and McCoy. I guess I don't want to start a big debate about this work here, and consequently won't offer a scientific opinion of it. But I should say that I do not find it to be "helpful", if you take my meaning. I suppose that I could start with their references list and unravel things from there, though that material is now at least five years old. I suppose I should just contact our information management people and tell them to send recent articles on APAP studies .

I'm also assuming that the algorithms involved are proprietary and hence that finding descriptions of them may be difficult -- unless some of that is exposed in patent filings. Do you have any additional knowledge of this?

As far as I know, all modern autopaps are designed to bring pressure down slowwwwwly whenever they've had to go up. Otherwise, they'd be yo-yo'ing up and down all night in an abrupt fashion. That could disturb sleep.

All I look at on the 96 hour graph in the "Detailed" report are these things:

a. how does the leak line look -- staying fairly steady, well under the magenta pink line?

b. does the pressure being used stay for the most part on the minimum (I've got my minimum set for pretty close to what a "prescribed" pressure would be) even though it goes up occasionally during the night?

c. when the pressure did have to go up, was the highest point it reached at least two or three points below the max pressure I set?

d. are tickmarks for events pretty well scattered through the night -- no large blocks of marks?

If those things look as usual (there will be variances in pressures used from night to night) and my AHI was low and I feel good, I don't bother looking at other items in the data. I don't try to correlate "this with that" or try to figure out why it used more pressure on some nights, or more often, than other nights. I did seek help about the pegging out at the top pressure thing that resulted in advice to turn IFL1 off...but that was a rather drastic thing to be seeing on a graph.

Absence of tick marks in areas where you see pressure going up, staying up for awhile then settling back down slowly, are usually evidence that the machine was doing its job...sensing from our air flow that something was brewing...and using more pressure to prevent the event from happening.

Tick marks we see are events that sneaked through. It's normal for even people who don't have OSA to have those. That's why "AHI below 5" is widely regarded as "normal"...not requiring treatment. If we are getting an AHI below 5 WITH treatment, our machines are doing what they are supposed to do.

How the machines go about what they do is an interesting subject, but I don't think we can tell much from what we see on the software data the next morning WHY the machine did what it did. There are just too many sensing and deciding things going on in their little brains for us to see much from our overnight data about why they do what they do -- or why they decide to do it when they do it.

(if this is starting to happen, do this; but not if that happens next; but, yes if this happens too, or if that happens at almost the same time, or if this doesn't happen again, or only if that happens within this time frame...and hold what you've got for this long to be sure, etc.)

It would be especially difficult to make sense of from the data we see if the machine is doing its job well and tick marks don't appear much because the machine's actions prevented an impending event from occurring.

If a tree falls, and no one is in the forest....

ghmerrill wrote:I have now looked at the "Performance of Auto-adjust Nasal CPAP Devices ..." document by Eiken and McCoy.

I read that, too. Besides their turning off "IFL" in the 420E (surely they meant only IFL1, but it was not clear if they might have turned both IFL1 AND IFL2 off) I don't think that kind of study with machines hooked to an artificial breathing model is much help at all in understanding how any given autopap would actually treat individuals. It's sooooo far removed from what would happen if patient airflow response were in the loop.

The same problem that was in their first experiment is still there and always will be (imho) when using breathing machine models...there's no patient responding to the machine's first actions, which would greatly influence what any of the machines do next.

Interesting discussion about the first study done with the older autopaps and the same artificial models of sleep disordered breathing:

viewtopic.php?t=1715

-SWS wrote:

"The study only plays a non-responsive sleep-event "loop" to further demonstrate how each AutoPAP responds to a test dummy that doesn't respond at all to pressure. Or in other words: the study breaks the patient-to-machine feedback loop that would be absolutely crucial to any of those AutoPAP algorithms."

rested gal:

Sorry about the length of this. I got a little carried away.

Thanks for the additional information and how you approach things -- which seems emimently sensible. Part of my curiosity about how/why the machine does things is just that ... curiosity. Part of it is that if I understand this I may in fact be able to tune my set-up in a better manner -- or a somewhat easier manner than simply relying on what we refer to as "observational data".

a. how does the leak line look -- staying fairly steady, well under the magenta pink line?

For me, this is great. No problem in this regard.

b. does the pressure being used stay for the most part on the minimum (I've got my minimum set for pretty close to what a "prescribed" pressure would be) even though it goes up occasionally during the night?

If "for the most part" means 50% or more of the time, then for me the answer to this is "If the minimum is set to 9, then mostly for the most part the pressure stays on the minimum." What this means is that for most days, or most periods, the pressure is "mostly" on the minimum. But there are a number of days where the mode exceeds the minimum by as much as 3 cm. It's this variability that I'm trying to understand. Is it variability due to diet, state of tiredness, leaks, other factors, etc. So far I'm not satisfied in my ability to determine this.

But why is staying mostly on the minimum a goal to strive for? Why not set the minimum below the expected mode, thus taking more advantage of the automatic features and reducing the time spent on what would appear to be (for a significant amount of time) "excessive" pressure? Why not, for example, establish settings that result in the mode approximating the mean?

Now the answer to this question should seemingly lie in correlating the number events with each such setting. And I haven't sat down to do that systematically. Has anyone? Is there in fact reasonable empirical evidence for the guideline: Set your minimum to a value such that most of the time is spent at that pressure? What is gained by doing this? What is lost by doing it? What are the trade-offs as compared to establishing settings where the mode is the mean instead of the min?

If these relationships are well-known and established, I'd be happier to just go with it.

c. when the pressure did have to go up, was the highest point it reached at least two or three points below the max pressure I set?

No. I will have nights on which this is true, and I will have nights on which it is not true -- no matter what the max is set to. It's this last result (that no matter what the max is set to, I will have nights on which the max is reached) that puzzles me about how I should approach setting the max. I think I will have to simply make a decision here, based on experimentation, but I don't think there is a max that I would never reach (well maybe up around 18-20 -- haven't gone that far yet). That's one reason I'd like to know how the algorithm works.

d. are tickmarks for events pretty well scattered through the night -- no large blocks of marks?

Except for the runs (which I'm now prepared to ignore) this does seem to be to be the case.

Absence of tick marks in areas where you see pressure going up, staying up for awhile then settling back down slowly, are usually evidence that the machine was doing its job...sensing from our air flow that something was brewing...and using more pressure to prevent the event from happening.

Ah, silly me. I was assuming that all events were being reported on the graph. But if this is true, events are occurring, being responded to, but not being reported. That does make it difficult to infer much from the graph.

(if this is starting to happen, do this; but not if that happens next; but, yes if this happens too, or if that happens at almost the same time, or if this doesn't happen again, or only if that happens within this time frame...and hold what you've got for this long to be sure, etc.)

Gee, I hope the code doesn't look like this. If it does, that would certainly explain some of the slowness of response.

It would be especially difficult to make sense of from the data we see if the machine is doing its job well and tick marks don't appear much because the machine's actions prevented an impending event from occurring.

True. That's another reason I'd like to see the predictive model (even a reasonably coherent high-level description) that's being used.

If a tree falls, and no one is in the forest....

Ah, but the question isn't whether anyone will hear (have heard) the sound. The question is whether the hidden Markov model will have predicted the sound .

Seriously, I presume that some sorts of stochastic models are being employed. It would be nice to know (to some degree of specificity) what they are. One interesting question, for example, is whether the model is based on what you might call "population statistics" (i.e., statistics are arrived at across a population of APAP users, and then these "one size fits all" statistics are applied to each case by the algorithm) or whether the algorithm learns from the historical data in the individual user's case for a particular machine and uses an adaptive algorithm (so that how it responds for me is different from how it responds for you). I suspect the former, thought certainly the technology is available to support the latter. But more details of the model would be nice to know in any event.