06-09-2021, 09:44 AM
RE: Looking for some help - almost 5 years and not feeling a lot better
Almost there. Feeling a lot better. Decided to take a statistical approach to seeing what worked better. My apnea data is really noisy. It's hard to tell quantitatively if the numbers are going in the right direction. Sure, one can tell if it a setting is a trainwreck, but it's impossible by eye to note subtle improvements.
First thing I did was not to change things quickly. It takes me (my body's time constant?) at least two weeks for a setting to show it's true potential. My body's actual time constant is longer. This was a total surprise to me. I have plotted 6 months of data, and I can see there is some sort of coasting from the last setting, and then two weeks later there is a change in slope. These subtle changes are only visible if considerable smoothing is done to the data. I chose LOWESS smoothing. Locally weighted scatterplot smoothing is also known as locally weighted polynomial smoothing. wikipedia local regression article Through searching, I found someone had written a short python script to implement it. Also, when I changed machine settings they were relatively minor, rather than big changes. I want to see the effect of the setting, not of the step change in setting. In this graph, the important lines are black. Solid black is the smoothed AHI+UF2, and dotted black is my AHI. You can see breakpoints in the smoothed curves around 2 weeks after a setting change. The blue band is the standard deviation of the LOWESS curve. It gets wider at the endpoints because it is locally computed.
The second thing I did was to use the notes in OSCAR. In particular, I would note what foods I had the day before, including any spices, along with other notable things like strenuous activities, or whether I had been exposed to high levels of pollen, or grinding dust, or chemicals. This was enlightening to me, as it showed there were some spices that caused my AHI to peak. There was 100% correlation between that spice ingestion and higher apnea levels. (Doubling or tripling of mean scores.) As a result of eliminating this spice, my average AHI has decreased. Sure, I may consume the spice again, but I now know it comes with the cost of a poorer nights sleep. I found, as others have found out that proper hydration is a component of good sleep. Neither too little or too much help your sleep.
For every machine setting, I collected the data of each component of AHI and UF2. For me, I set the definition of UF2 to 50% flow reduction for 9 seconds or longer. Since my sleep issue seems to be micro-arousals more than apnea, I used UF2 as a proxy for sub-apnea events that seems to be slightly under the threshold of machine declared apnea. Yes, it is an imperfect proxy, but at least it is quantifiable. I computed the mean and standard deviation of CA, OA, H and UF2 for every setting. This way I can know how each setting affects each component. This gives me a clue for future changes. I appologize for the table formatting. I intended it to be fixed pitch, but the forum sw is changing it!
The goal of this effort was to minimize the sum of AHI+UF2. At this point, the sum is low enough that my life has greatly improved. There are up and down days, (remember the data is noisy) but overall things are so much better.
Posted this to give others some encouragement.
Don't know if this is possible, but, can we change the title now? The thread is more of a success story, now. I did get help here, and I do feel a whole lot better!
I want to thank everyone who offered suggestions and insight. 
Your encouragements helped more than you know.
First thing I did was not to change things quickly. It takes me (my body's time constant?) at least two weeks for a setting to show it's true potential. My body's actual time constant is longer. This was a total surprise to me. I have plotted 6 months of data, and I can see there is some sort of coasting from the last setting, and then two weeks later there is a change in slope. These subtle changes are only visible if considerable smoothing is done to the data. I chose LOWESS smoothing. Locally weighted scatterplot smoothing is also known as locally weighted polynomial smoothing. wikipedia local regression article Through searching, I found someone had written a short python script to implement it. Also, when I changed machine settings they were relatively minor, rather than big changes. I want to see the effect of the setting, not of the step change in setting. In this graph, the important lines are black. Solid black is the smoothed AHI+UF2, and dotted black is my AHI. You can see breakpoints in the smoothed curves around 2 weeks after a setting change. The blue band is the standard deviation of the LOWESS curve. It gets wider at the endpoints because it is locally computed.
The second thing I did was to use the notes in OSCAR. In particular, I would note what foods I had the day before, including any spices, along with other notable things like strenuous activities, or whether I had been exposed to high levels of pollen, or grinding dust, or chemicals. This was enlightening to me, as it showed there were some spices that caused my AHI to peak. There was 100% correlation between that spice ingestion and higher apnea levels. (Doubling or tripling of mean scores.) As a result of eliminating this spice, my average AHI has decreased. Sure, I may consume the spice again, but I now know it comes with the cost of a poorer nights sleep. I found, as others have found out that proper hydration is a component of good sleep. Neither too little or too much help your sleep.
For every machine setting, I collected the data of each component of AHI and UF2. For me, I set the definition of UF2 to 50% flow reduction for 9 seconds or longer. Since my sleep issue seems to be micro-arousals more than apnea, I used UF2 as a proxy for sub-apnea events that seems to be slightly under the threshold of machine declared apnea. Yes, it is an imperfect proxy, but at least it is quantifiable. I computed the mean and standard deviation of CA, OA, H and UF2 for every setting. This way I can know how each setting affects each component. This gives me a clue for future changes. I appologize for the table formatting. I intended it to be fixed pitch, but the forum sw is changing it!
Code:
Days PS EPAP IPAP avgAHI stdAHI avgCA stdCA avgOA stdOA avgH stdH avgUF2 stdUF2 avgHIF stdHIF
3 4.2 9.0 14.4 1.3 0.08 0.57 0.15 0.12 0.0 0.61 0.1 0.53 0.12 1.82 0.14
7 4.2 9.0 14.6 0.88 0.37 0.18 0.14 0.12 0.13 0.58 0.25 0.33 0.29 1.21 0.49
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 1.55 0.62
22 4.2 9.6 15.2 0.67 0.31 0.1 0.12 0.16 0.14 0.41 0.27 0.45 0.31 1.12 0.5
20 4.2 10.0 15.2 0.82 0.3 0.22 0.16 0.11 0.12 0.49 0.22 0.41 0.25 1.23 0.44
40 4.4 9.8 15.2 0.9 0.37 0.27 0.23 0.11 0.11 0.52 0.22 0.41 0.25 1.32 0.53
32 4.4 9.6 15.2 0.77 0.31 0.17 0.14 0.08 0.1 0.52 0.25 0.4 0.26 1.17 0.46The goal of this effort was to minimize the sum of AHI+UF2. At this point, the sum is low enough that my life has greatly improved. There are up and down days, (remember the data is noisy) but overall things are so much better.
Posted this to give others some encouragement.
Don't know if this is possible, but, can we change the title now? The thread is more of a success story, now. I did get help here, and I do feel a whole lot better!
I want to thank everyone who offered suggestions and insight. Your encouragements helped more than you know.
