• Guest - w'd love to know what you think about the forum! Take the 2025 Survey »

Standard deviation

This sounds like a really good handle on self-management. I'd like to start doing it. But how do you get the SD figures?

I use the Diabetes Co Uk tracker app. Perhaps I shd switch over to sth else. Or else use Andrew Colvin's program. For metering, I just use the Libre's strips.

Recommendations please?

Non-numerate Lucy (I'm very literate, honest guv)
 
If you haven't got it built in to your app or other software then you can work SD out for a set of numbers from here - assuming they fall in a normal distribution (i.e they are not skewed in any way).

http://www.mathsisfun.com/data/standard-deviation-formulas.html (basically what @FergusCrawford posted earlier as the raw formula).

If you have your data on a spreadsheet it will be easy to automate this.
 
Lucy, if you are using the Libre, you can get the same information from the data you export to your PC.

There are Excel functions that provide Mean (average) and Standard Deviation for a data series and you can use these to give you the same check.
 
tim2000s said:
There are Excel functions that provide Mean (average) and Standard Deviation for a data series and you can use these to give you the same check.
Click to expand...

Of course, doh! :banghead:
 
Lucy - I use the Excel function like Tim. I used to use it on all my blood tests but I have changed to using it on my Libre scans. The erroneous scans obviously do skew things a little, but there is so much data that I think it won't be very significant. The BG scans were always skewed a little by the fact I test more frequently when hypo or close to it (so low BG) rather than when at normal levels. During the period I was SD'ing both BG tests and Libre tests, they were remarkably similar - as was my average!

It was actually a thread on the forum a year or more ago that got me started keeping track of my SDfromMean and as I had all my old data in a spreadsheet it was very quick and easy to see how it had changed over the course of a couple of years. It actually led eventually to my recent change of insulin because although my HbA1c had dropped from 5.9% to 5.4% since using Levemir, my SDfromMean had risen from 1.7 to 2.9 over the same period and was on a continual increase. Without the SDfromMean data, I would not have been able to convince my consultant that my control was slipping and that it needed further investigation by way of a sensor.

Smidge
 
Thank you again, Smidge. And do you do and compare just weekly SD figures? I'm thinking days too micro to be useful.
 
LucySW said:
Thank you again, Smidge. And do you do and compare just weekly SD figures? I'm thinking days too micro to be useful.
Click to expand...
You'll get a better result if you do it across the largest data set that you have.
 
In excel from Libre exported background data (i.e.: not the scan or blood stick data) 7130 records
Average BG from 22 Oct to date is 5.742692847
Standard deviation on same data is 1.316767376
Pretty much what I got from mysugr
 
tim2000s said:
You'll get a better result if you do it across the largest data set that you have.
Click to expand...
And then also for each period where you want to compare something, no? - For me, eg, basal only cf. basal bolus.
 
LucySW said:
And then also for each period where you want to compare something, no? - For me, eg, basal only cf. basal bolus.
Click to expand...
You can do this, yes, but make sure you are clear about the parameters you are using and your dataset is clean.
 
Today I learned what SD stands for on my Codefree chart. Apart from false readings which we may or may not discover by repeat testing, and Smidge's view that testing more frequently when hypo or close to it gives low BG, what will skew an SD calculation?
 
SD is a measure of average variance from the mean. Having a high range of readings from low to high will widen the SD. If your values don't fit a normal distribution then the SD will not make sense either.
 
LucySW said:
Thank you again, Smidge. And do you do and compare just weekly SD figures? I'm thinking days too micro to be useful.
Click to expand...

I let Excel keep a running total across a complete data set, so I do actually see daily changes, but as you say, daily changes are not really of any value. I break it down on a monthly basis for comparison. Obviously working in months is pretty arbitrary, but I need some basis on which to make a comparison and it seems about the right level - any less than a month would be a bit too much I think. Being able to look at it on a month by month basis and compare back to previous years gives me a trend and also lets me spot any seasonal patterns. As Tim says, the bigger the data set the more significance the SD has.

Smidge
 
The Libre data proves fascinating. Christmas has a significant impact on my overall average and SD.

If I take Christmas out, my average blood sugar is 6.95 mmol/l with an SD of 2.98.

If I include Christmas, I get 7.27 mmol/l and 3.34. As an indicator, this is absolutely in line with my Hba1C, which was 44, and equates to 7.3 mmol/l.

So what I take from this is that in the run up to Christmas, my BS variance was a little on the high side.
 
Last edited by a moderator:
One problem about using SD is that, mathematically,a condition for it's use is that the data is normally distributed and blood glucose data isn't. There are lots of papers which ignore this and others in which they use other methods to get around it and yet others that argue that it's a good tool nonetheless..
There are lots of different methods used though it's hard to actually find how the calculations are done
The MAGE. is quite often used for CGM data This is the mean amplitude of glycemic excursions) is a calculation that often seems to be used to compute glucose variability.
I've found out how to do that one ( I'm going to have a go at this later on in the month ;or at least I'm going to see if OH will help me do it in excel)

http://www.ehow.com/how_5616113_compute-mage-diabetes.html

But then of course there is the argument about whether glucose variability actually matters (as opposed to HBA1c and average levels) And whether there is a difference here between T1 and T2
http://www.ncbi.nlm.nih.gov/pubmed/25023992
 
Last edited by a moderator:
That MAGE value is an interesting one. Following the instructions that are given in the ehow, I initially ended up with a MAGE that is 4.1

Taking a further look at the dataset, I can see that some of these are due to gaps in the data, most likely where my sensor has been replaced so I discount these. This leaves me with a result of 3.8.

Digging into these occurrences of notable change, there are 13 of them and I can see that in every case they are either the result of compensating for a hypo, or the follow up to a correction dose of insulin.

I'm therefore not entirely sure what MAGE is usefully indicating!
 
tim2000s said:
That MAGE value is an interesting one. Following the instructions that are given in the ehow, I initially ended up with a MAGE that is 4.1

Taking a further look at the dataset, I can see that some of these are due to gaps in the data, most likely where my sensor has been replaced so I discount these. This leaves me with a result of 3.8.

Digging into these occurrences of notable change, there are 13 of them and I can see that in every case they are either the result of compensating for a hypo, or the follow up to a correction dose of insulin.

I'm therefore not entirely sure what MAGE is usefully indicating!
Click to expand...
The argument that it should be use comes very much from the 'side' that considers intra daily variability important.
Here we have Monnier arguing the case (he is responding to Kilpatricks analysis of the DCCT data which showed that the daily ups and downs weren't a factor above HbAc ) Kilpatrick used standard deviation. Monnier says this misses the peaks and troughs. http://care.diabetesjournals.org/content/31/Supplement_2/S150.full
.
Consider two patients with type 2 diabetes who have similar A1C and SD of glucose fluctuations around the mean. Assume that one subject has many minor glucose fluctuations and one or two major swings per day, whereas the other patient exhibits moderate glucose fluctuations over 24 h. Despite similar SD of glucose around the mean, these two patients should exhibit very different MAGE values, and thus Kilpatrick's use of SD as a definitive measure of glucose variability is questionable. Even though the MAGE determination requires continuous glucose monitoring, our opinion is that this index should be the gold standard for assessing glucose fluctuations in all prospective interventional trials designed to estimate glucose variability.
Click to expand...
Kilpatrick did do some further analysis of the same data using MAGE and said this didn't make any difference either.(but the DCCTdidn't use CGM)
 
Back
Top