Data Discrepancies

Since I gained access to my data ten days ago, I've been keeping track of it on an Excel spreadsheet, noting both the machine's readout and the figures from the software each day. I'm seeing some discrepancies in the numbers reported, as follows. (The numbers with a minus-sign indicate a lower figure reported by the software than by the machine, and vice-versa.)

Differences in Leak

2.63
-0.70
4.87
0.53
2.39
3.00
0.94
4.92
1.17
4.13

Differences in AHI

-1.70
0.20
1.00
-2.70
2.20
-0.30
0.10
0.20
-0.20
0.40

Anyone know what accounts for the discrepancies in the numbers? I assume the software is far more sophisticated in its reporting abilities, but some of the differences are marked (like the two nights where the leak rate was higher by nearly 5 points and the two nights where the AHI was off by more than 2). Makes me wonder just how accurate the machine's reporting capabilities are (and glad that I got the software and reader).

Just curious.

B.B.
The analysis of your numbers is still beyond me - but I am glad you are working so hard for success - you deserve it.

Jan

Well it is clear AGAIN (and again and keeps going like the energizer bunny) others have reported these problems.


- Respironics Math Sucks -


I've said it before and can say it again.

I also noticed significant differences in the averages (yes, I reset daily to zero to get just the daily average) when I compared software data to what I was given on the LED screen. I didn't take the time to document differences but after about 3 days I quit bothering to reset the machine averages to zero and don't even look at the machine's LED data. Do not know why the difference but does make a person question how much faith they can put in the machine's LED data. Better than nothing I suppose and I guess if feeling good and no problems and good data, it probably isn't critical but if not doing well, sure can muck up where to start fixing things.

Pugsy wrote:I also noticed significant differences in the averages (yes, I reset daily to zero to get just the daily average) when I compared software data to what I was given on the LED screen. I didn't take the time to document differences but after about 3 days I quit bothering to reset the machine averages to zero and don't even look at the machine's LED data.

I'm going to continue comparing the machine's readout to the software for a full month, just to see what's what. Then I'll stop resetting the machine every day and just rely on the software.

Do not know why the difference but does make a person question how much faith they can put in the machine's LED data. Better than nothing I suppose and I guess if feeling good and no problems and good data, it probably isn't critical but if not doing well, sure can muck up where to start fixing things.

Exactly.

I had my best night so far last night: Slept almost 7 hours, AHI was 0.6, and I feel good today! (Saturday night's AHI was 2.4, and I was yawning yesterday.)

BB (who's resisting the urge to kick myself in the a$$ for waiting so long to acquire a data-capable machine w/software & reader)

jules wrote:Well it is clear AGAIN (and again and keeps going like the energizer bunny) others have reported these problems.


- Respironics Math Sucks -


I've said it before and can say it again.

- So does DeVilbiss -
I'm trialing an Intellipap Autoset.

In SmarLink
15 minute "delay" is reported as 0.3 of an hour
20 minute "delay" is reported as.. you got it, 0.3 of an hour.

In other words, you add 33% to your time in "delay" (that's "ramp" or "settling") and the software doesn't care about the difference.

God knows how they make their other calculations -- the don't give you raw numbers -- maybe for good reason.

O.

Yes, they don't add up. Those of us that have noticed it call it Respirionics Math. Several of the numbers (Leak Rate) shows it to be carried to 2 decimal places. Pro analyzer and VIEWER rounds and always give XX.00 answer. That is why the average lines are useless. Pro analyzer does do a better job than EP in adding 2+2 and comming up with 5 as the answer.

Jerry

The software calculates a weighted average for Leak Rate while the hardware calculates a simple average. Based on my analysis of the difference in Leak Rate averages reported by my software and hardware, I reached the conclusion that the hardware (my M-Series Auto with Aflex) sums all of the Leak rates recorded during an interval and divides by the total number of leak rate changes - this is a referred to as a simple average. The software (I use Encore Pro) is provided, via the smartcard, all the leak rates and their durations as they occurred in the interval. It calculates a weighted average by summing the product of leak rate * (times) duration (for that leak rate occurence) for all the leak rate changes during the interval, and dividing by the total number of hours in the sleep interval.

I prefer the weighted average as it provides a more accurate view of the leakage during the full sleep interval. As an example, for a leak rate of 50 for 1 hour, a leak rate of 30 for 6 hours and a leak rate of 50 for 1 hour, the simple average is (50+30+50)/3 for an average of 43.3 LPM (reported by the hardware). The weighted average (software) for this 8-hr sleep interval is (50*1) + (30*6) + (50*1) or 280 divided by 8 or 35 LPM.

Previous posters have commented that the AHI 7-day and 30-day averages reported by some Respironics machines are averages of averages. I have not devoted any time to looking at the AHI averages reported by my machine and how they might differ from those reported by my software (Encore Pro).

6PtStar wrote:Pro analyzer does do a better job than EP in adding 2+2 and comming up with 5 as the answer.

Jerry

But this is exactly the math level Respironics uses. Their programmers never learned anything about sensitivity analysis, or round off errors propagating more round off errors, let alone basic arithmetic ----

I have posted before in viewtopic.php?p=338405#p338405

jules wrote:nah

not the way the programmers at respironics flunked 4th grade arithmetic and had an easy grader in 3rd grade

and if you do an advanced search for math sucks well, read the posts --

btw - in my previous incarnation I was a univ math prof who taught math modeling among other things

jules wrote:Well it is clear AGAIN (and again and keeps going like the energizer bunny) others have reported these problems.
- Math Sucks -
I've said it before and can say it again.

I Concur - Math Sucks
Really

cinco777 is right...I found the same thing. It all comes down to averages and averages for the "long run"! No different than the diabetes glucose readings on your meter. If you try to follow this stuff hour by hour, day by day...it'll drive you crazy. You want to have a good solid average per month and so on, instead of these averages day by day hour by hour. I used to be really bad at math, now I love it and I'm REALLY good at it...my artistic answer to math? 2+2 DOES equal 5!!

ozij wrote:

jules wrote:Well it is clear AGAIN (and again and keeps going like the energizer bunny) others have reported these problems.


- Respironics Math Sucks -


I've said it before and can say it again.

- So does DeVilbiss -
I'm trialing an Intellipap Autoset.

In SmarLink
15 minute "delay" is reported as 0.3 of an hour
20 minute "delay" is reported as.. you got it, 0.3 of an hour.

In other words, you add 33% to your time in "delay" (that's "ramp" or "settling") and the software doesn't care about the difference.

God knows how they make their other calculations -- the don't give you raw numbers -- maybe for good reason.

O.

Statistical overlap?
Since 1/10th of an hour ( .10 ) is 6 minutes, 2/10ths is 12 min., 3/10ths ( .30 ) is 18 min. and 4/10ths ( .40 ) would be 24 minutes.
"18" is halfway between 15 and 20. Not close enough to 12 (2/10ths) or 24 (4/10ths)


Den

Interesting.

Ramp (delay) is set in 5 minute increments, so my guess was:

• 15/60 =0.25 rounded comes to 0.3
  20/60 =0.33 rounded also comes to 0.3

O.