Apnea Board Forum - CPAP | Sleep Apnea
Apnea statistics question, how tightly controlled is your apnea? - Printable Version

+- Apnea Board Forum - CPAP | Sleep Apnea (https://www.apneaboard.com/forums)
+-- Forum: Public Area (https://www.apneaboard.com/forums/Forum-Public-Area)
+--- Forum: Main Apnea Board Forum (https://www.apneaboard.com/forums/Forum-Main-Apnea-Board-Forum)
+--- Thread: Apnea statistics question, how tightly controlled is your apnea? (/Thread-Apnea-statistics-question-how-tightly-controlled-is-your-apnea)

Pages: 1 2 3


Apnea statistics question, how tightly controlled is your apnea? - happydreams - 03-05-2021

This question isn't about my therapy, but more about how well does PAP treatment work from a statistical sense.  If your eyes glaze over even thinking about statistics, this isn't the thread for you...  Moderators, my apologies if this belongs under my treatment thread.  I think it may be of more general interest.

Specifically, has anyone run statistics on (your) apnea stats to see how well the therapy is working for a specific apnea index?  If one considers the standard deviation as a figure of merit (and obviously the mean) how small is your standard deviation?  When does one consider the therapy to be truly effective?

For my machine, a VAuto, it doesn't control CA, so I disregard the stats on CA, but I have been looking at statistics for UA, OA, H & UF2 to see if my settings are actually effective.  This means one discounts single awesome (or horrible days) to find out what works.  I think this would be called a data driven approach.  Since the beginning of the year (actually started in October) I have logged the apnea occurrences from OSCAR and the sleep time and determined the stats as well as histograms for these parameters.  Actually, I disregard UA's since I've only had 1 this year.

For myself, I find CA's are effectively uncontrolled (as expected for my machine), OA's are tightly controlled in a statistical sense, and Hypopneas and UF2's are relatively weakly controlled at my current (and also cumulative) settings.  Is this typical?  Or do the standard deviations get tighter with a better therapeutic match to the patient?

As an example: (apologies for formating, forum converts code to variable font and handles tabs oddly!)
Code:
Days PS   EPAP  IPAP    avgAHI  stdAHI  avgCA   stdCA    avgOA   stdOA   avgH   stdH   avgUF2   stdUF2
3    4.2  9     14.4    1.3     0.08    0.57    0.15     0.12    0.0     0.61   0.1    0.53     0.12
7    4.2  9     14.6    0.88    0.37    0.18    0.14     0.12    0.13      0.58   0.25   0.33     0.29
35   4.2  9.4   15.0    1.05    0.48    0.29    0.26     0.2     0.14    0.56   0.35   0.5      0.28
18   4.2  9.6   15.2    0.69    0.29    0.1     0.13     0.16    0.15    0.43   0.27   0.49     0.32
Statistically, obviously one would not consider the 3 day session (the setting is actually 7 days, but I only collected information for 3).  The 7 day session is also weak from a statistical sense, but it showed little change in the control over hypopneas.  (EPAP is actually EPAP_min, and IPAP is actually IPAP_max, the table headings were quick and dirty, to get a sense of what is going one.)

Sometimes it is useful to see graphically how the data is distributed.  To me this can be a more effective tool to understanding.  As an example here is aggregate data since January 1st.  The histograms were plotted on a common binning, so that it is easy to compare the relative process control.  The x-axis is the bin index.  So using OA as an example, since 1/1/2021, there has only been 1 night with and OAI=0.5 or greater.
[attachment=30584]
It's pretty obvious from the histograms that the OA is tightly controlled.  CA isn't bad, which is fortunate since there's no process control available on my machine.  It's clear both the Hypopneas and UF2's are relatively weakly controlled by the VAuto, at least under these settings.  This tells me there may be some possible improvement to be made with enough persistence.  It's the higher standard deviation of UF2 & H that seem to be responsible for my roller coaster evenings.

For me, as an apnea sufferer, what has been difficult is dealing with the wild and random nature of my sleep quality.  Having a awesome day followed by a near horrible one is really hard to manage.  Reducing the standard deviation could ameliorate this somewhat.  For me, I feel significantly worse at higher AHI+UF2 indices.  Higher values are typically accompanied by headaches and lethargy, so I've been methodically attempting to reduce the standard deviations to achieve a more normalized day.

Have any of you looked statistical or similar methods to effectively manage and control your apnea?  Please discuss.


RE: Apnea statistics question, how tightly controlled is your apnea? - srlevine1 - 03-05-2021

Do you have room temperature control, similar sleep times, a standardized diet, or do anything else to help remove additional variables?
Aren't your sample sizes too small to draw meaningful statistical inferences?
In looking at my data -- logged on ResScan since 2008 -- I sort of believe that all of the results are suggestive but not conclusive. The fact that I can use statistical functions does not mean they are helpful over the longer time frames because my data seems to regress to the mean.


RE: Apnea statistics question, how tightly controlled is your apnea? - happydreams - 03-05-2021

Thank you for the discussion.

Ahh, there's the rub.  To be honest, nothing really is under that level of control.  I don't know whose life really is.  It doesn't mean that the data is useless.  I too have data for 5 years on a different machine.  It told me that no setting ever controlled my apnea effectively.  I had 3 good days a year.  And more than 250 pretty bad days a year, over 5 years.  It was a painful experience.  That's why there's a VAuto on my night stand now.  Overall, my stats are far superior.  Now I have about 4 good days a month, and 9 bad days a month.  That's an order of magnitude better.  It's not due to me changing diet, or keeping my room at 72F.  (Can't control room temperature tightly on a one zone steam heating system.)  My sleep times are about the same as before.  Things are better now, but there is room for more improvement.  

The numbers that I have may not be truly statistically significant, I grant you, since the sample set is small.  On the other hand, how is one to get a sense if one's therapy is effective?  By determining how you feel, to be sure, and some of your numbers...  I consider my settings to be not optimal (yet).  How would someone scientifically improve their therapy?  One cannot wait an infinite time to tell if something is working, at least not for us mere mortals.  So we need to use sufficiently long times, say a month to get an idea of the trend.  Is there a more sound way to do this?  Do you know of a better way?  As the patient and titrator, one doesn't want to suffer any longer at a setting than necessary, but want to know when it's better to collect more data or to institute more changes.  Have any guidance on this?

Most people can't control their diet to the extent of a "perfect test", never mind the efficacy or potency of their medication.  Most medications are only tested for overall efficacy, not their physical actual daily dose (as delivered).  Some medications such as powdered inhalers have been documented (in the medical literature) as being difficult to use effectively.  I have a lot of personal experience with powdered inhalers and know they are extremely variable in their daily dosage.  "Better nights" are strongly correlated with effective dosages being delivered.  (It's incredibly easy to get your tongue and roof of your mouth coated with medicine rather than into your lungs, where the medication will actually be effective.)  I've tried to take daily notes and have identified some spices that give me a poor or restless night's sleep.  I try to avoid those spices now.  But I'm not disciplined enough to eat just PB&J sandwiches for a year to have a more controlled experiment.  

So how does one know when to hang in, or to change something?  What's an effective and practical guideline, given that (at least for me) apnea scores appear to be statistical?


RE: Apnea statistics question, how tightly controlled is your apnea? - Sleeprider - 03-05-2021

Statistics can express the distribution of something like AHI, a trend, frequency, probability and confidence that it will fall within a particular range, and even a correlation to pressure and other variables. There is a statistic that can do pretty much anything you want to prove or disprove, which is why they are both useful and useless depending on how they are constructed, particularly as we get into more complex multivariate testing. Your qualitative sense of therapy being much improved, is probably as valuable as any attempt at a quantitative analysis.


RE: Apnea statistics question, how tightly controlled is your apnea? - sheepless - 03-05-2021

I have little to offer beyond support. I had a basic stats class in college and a discipline-specific applied stats class in graduate school. I barely scraped by but found it fascinating. when I first started papping, I tried to use sleepyhead's exportable data in a spreadsheet but as I recall not all the data I was interested in was exportable. in addition, I was deep in the throes of sleep deprivation and couldn't handle the learning curve to figure out what to do with the data that was available. it would be cool if someone like yourself figured out a way to more easily gather the data and for the math challenged like me to analyze and interpret them. like an oscar-specific add-on statistics program, maybe even someday built into oscar.

meanwhile, I do think the machines and oscar already give us a broad range of basic stats that are quite useful. maybe only for lack of more precise analytic capabilities, at present there's probably as much art as science in titrating effectively. no matter how systematic the analysis, the subjective how-we-feel will likely always be a useful guide.

thanks for sharing this, happydreams.


RE: Apnea statistics question, how tightly controlled is your apnea? - Gideon - 03-05-2021

My gut sees some flaws in your analysis, but I'm not a statistics expert.

Event totals are only valid when they are over a constant time period. The "Index", the Rate per hour seems to be a much more valid number. 17 events in 1 hour, or 17 events all night is a big difference. Granted your nights are all x hrs +/- some amount.

You are using a variable pressure, Recording the settings is good, but realize when we evaluate we note that a person is doing better at 11, than he does at 6 and make recommendations accordingly.

The other 'parameter' you should track is subjective, How did you feel that night, how well did you sleep, figure a scale 1-5 or 1-10, 0 being not recorded.

Once you are fairly well dialed in, what really matters is how you feel the next morning/day. Comfort and sleeping well is what it is all about.


RE: Apnea statistics question, how tightly controlled is your apnea? - happydreams - 03-05-2021

I've generally found lower scores lead to feeling better, but how you feel is better than any statistic.  Last night was a relatively great night's sleep.  But I have a pretty bad headache today.  By the numbers I'm terrific.  My pounding head begs to differ with that numerical assessment at the moment.

The numbers I generated were an attempt to guide me in some sort of sensible fashion.  Longer trials are more meaningful, but one can't wait an infinite time to move on.  Short trials don't mean too much, but at least tell us if something is colossally going wrong.  The short trial results may be misleading, however, so they are worth revisiting should things not go as expected.  

All in all, titration is empirical, so I thought it an interesting topic to discuss to enable or guide users to get to where they need to be in a shorter time span.  If stats aren't the right guideline, so be it.  As a person who has worked in the engineering field, I thought some numbers might be useful.  From a process control perspective, large standard deviations relative to the mean are indicative of poorly controlled process.  Narrow distributions tell us that whatever we are doing is controlled.  

Plotting feelings is a little harder, but I suppose it's doable, given the discipline to accurately assess them.  At the moment, I don't know how to write down exactly how I feel every day.  Not even sure I have the vocabulary or ability to assess them accurately.  Can we think of some guideline or factors that could be useful to assess?  The slider bar in OSCAR is something, but I often find even that is too subjective to me to assess objectively every time.  Maybe a couple of categories could help assess that setting?  Just throwing out an idea.

I'd like to add something like what I've done to OSCAR, but it may be misused.  We (I) wouldn't want someone to draw a false conclusion on two days data!  I know enough to not do this myself, but others may not.  Perhaps a quality factor (color continuum from red to green?) assessing how much stock to take in the numbers?  Or even greying out the numbers until a larger set is collected?  

The idea is somehow to enable our users to evaluate (fairly and unbiased, given the data) how they are doing and how it compared to what settings (or bed times!) they were doing before.  With the plethora of numbers that OSCAR has, it's hard to evaluate if setting X or situation Y is all that different than another, without some sort of analytic tools.  To enable the analysis of "soft variables" one would need a way to score daily each of them.   It's hard to tell if people would have the discipline to do this.  I'm motivated to do this, and still forget to enter stuff.  I try, but there's quite a few days where the notes are blank.  I get busy and forget to enter stuff.  The longer the delay from the night in question to adding data (how I felt) the more inadvertent bias creeps in.  Same day recollections are usually better than a week later...  I see no issue to score "soft variables", as long as they were entered daily.  If the entries were older than that (weekly or longer), I'd be hesitant to trust them at all.  How that would be done 

Personally, after a while, one gets a feeling of what to do next.  The hard part is to both be patient and bold.  Patient enough to give a setting some time to prove it's worth, and bold enough to change when it really is necessary.  If there was a way to expedite others to learn this it would be a good thing for the community.

Good discussion, thanks for continuing this.


RE: Apnea statistics question, how tightly controlled is your apnea? - happydreams - 03-05-2021

@Gideon, my stats are based on index.  They really are AHI, CAI, OAI, HI, UF2I.  The histograms are how many days that index occured.

My monthly files have apnea events and sleeptime, (for every day) from that I compute index, just like OSCAR.  I did it this way because it was far easier for me to eliminate SWJ events.  The events I have logged are non-SWJ.  No point of me scoring apnea when I'm not asleep.  I only can estimate this by my sleep onset, which for me is rather easy to identify.  Other SWJ gets counted as apnea, because I have no obvious way to eliminate it without an EEG.  

Have to admit, at the moment, I'm grasping for a way to do this reasonably and effectively.  I've only captured EPAP_min and IPAP_max, not their stats or time at pressure.  As you say, this can matter.  I'm assessing manually if the events are occuring at say low EPAP and going from there.  I'm just looking for a way to construct a method to assess numerical progress from a statistical sense.  A single day at 0.0, while lovely to boast about, isn't indicative of proper therapeutic settings.  Now, if that 0.0 happens often & one feels pretty good, well that's sweet, or I think it would be.  (Wouldn't know, it's never happened to me!)

As mentioned earlier, if it could be done, (and I'm not sure how reliably ensure unbiased entry,) one could just as well plot or assess how well one feels.  I can tell if I feel good a single day, but it's hard to pull up any stats on how often that is, (or compare 'goodness') without crawling through lots of notes.  For myself, and probably others, it would be good to know how often one feels good at a PAP setting.  In the end, that's what we want to titrate to, isn't it?


RE: Apnea statistics question, how tightly controlled is your apnea? - sheepless - 03-05-2021

the general discussion is a necessary starting place but it strikes me as too vague to know how to proceed. seems to me much depends on precisely what one wants to accomplish and methods and/or inputs will vary accordingly. if feeling better is what we're after, comparing ahi or it's components (or whatever) to a numerical representation of restfulness seems simple enough. if avoiding morning headaches is the goal, plot the presence/absence or a graded scale of severity against suspected variables, like oa or flow limitations or O2 levels, probably one by one. I guess multivariate analysis is possible too. maybe it's as simple as stating multiple hypotheses and testing them to better understand the relationships between variables, then hope something new and useful in the numbers & relationships declares itself.

as I recall, one reason I wanted to analyze the detailed exportable data was to help find optimal pressure settings that were eluding me. I failed to pursue this very far, but a while back member mper6794 helped me titrate my vauto with scatterplots and if I recall correctly, some basic statistical analyses. a site search may reveal others that applied or attempted to apply statistical analyses.


RE: Apnea statistics question, how tightly controlled is your apnea? - happydreams - 03-05-2021

@sheepless, good comments.  If I knew what to do, and how to do it, I'd have done it by now Smile  This was intended as an exploratory discussion, not a proposal, nor a fait accomplis.  It's revealing that you mention pursuit of optimal pressure.  Now what is optimal pressure, exactly?  Optimal for what?

It's a minimization problem of sorts, find a group of variable settings x,y, z, & P such that something is minimized.  A useful thing to minimize might be the sum of "feels good", headache index, and AHI?  Or a weighted sum, depending what one wants to prioritize.  To do this one would need numerical metrics for the three categories.  Not saying this is a trivial effort.  Just sort of thinking out loud for the moment.

Thanks for the lead on mper6794 and scatter plots.  It gives me something to ponder.  I'll trawl the site looking for it and similar terms.