Standard deviation

I was told that basically if you take your hba1c level.. Or your 90day overall figure on your combo pump.. Thenlook at you SD, then that is just the variance in levels that you are getting with your blood tests.

Ie...
If you have an hba1c or 90day reading of 5.5 and you SD is 5.5 then your average readings from your blood meter will be showing you in the ranges of nithing to 11.0. (this will deoend on your qty of tests.... And other factors).

However, if you are at 5.5 and you have a SD of 2.0... Your variances would be more eithin the 3.5 to 7.5 range.

However... Again... This all depends on the qty of nlood tests that you take, otherwise its a useless bit of info.
Ie if you were taking 2 bloodtests a day they may always be within the target ranges and you could be thinking yippee, whilst in reality you blood during the rest of the 24 hours could be ****, and so the SDwould be nonsense.

I would only recommend looking at your SD if you do 8 or more bloodtests a day at regular times. I do, so I tryto keep my SD's to less than 3.0.

I've only had one consultant that ever looked at my SD range. I dont think they bother with it, but I personally think my retinopathy disappeared because my SD was really good.

Standard deviation is a measure of variance of a normal distribution (often referred to as a bell curve). If you add up all the values of a range of measures and divide by the number of values you obtain an arithmetic mean. Let's say that mean is 5.5 - the SD is calculated by a well know formula; it is easily worked out on a calculator or on an Excel spread sheet. If the SD is 2 it means that about 68% of the values will lie between 3.5 and 7.5. About 95% of the values will lie within 2 SD i.e 1.5 to 9.5.

I have no idea if the values of glucose blood levels will form a normal distribution. However it would seem common sense to try and keep variations to a limited range if possible.

The big advocate of using SD is Irl Hirsch an American diabetologist. He says that ideally the standard deviation should be less than a third of the average, this is though too difficult for most people with T1 so he suggests a target of less than half.

The big problem is though there is no actual evidence that it makes any difference to the development of complications. According to the DCCT/EDIC trials it was the HbA1c and the average glucose level that determined risk and whether that level was achieved with large or small variations didn't make any difference. Longer term variability was found to be a risk factor (ie HbA1cs being low for a while then high, then low) Of course it may be that they simply haven't found the evidence. That trial is the best long term trial we have and lasted over 20 years but the glucose measurements were only taken from a day every 3 months. What is needed is a long term trial using continuous monitoring and that would be hugely expensive.

You might like to read these: The first is a presentation by Dr Hirsch. The second a couple of more general blog posts about using SD.
http://www.slidefinder.net/u/updownhand ... dout/80869
http://www.diabetesmine.com/2009/07/way ... -buzz.html
http://www.diabetesmine.com/2010/12/the ... -care.html

Personally, I think that looking at a graph or indeed just being aware if you are having lots of highs/lows is as useful as any quantitative statistic . I think it's more sensible to consider the reason that individual levels are too high/low/out of target at the time they happen rather than looking back over them . The aim is to try to avoid those out of target readings in the future

If you do use SD, as dd says it is meaningless if you don't have enough data. If I only tested before lunch and before dinner I would have a miniscule SD but even if I tested 8+ times a day I would might miss highs ( less likely to miss lows and this could skew things)

There is an interesting (long) discussion on the validity of SD on another forum. (the point made by the original poster in the discussion is that glucose levels are not normally distributed).
Google Tu diabetes. Why Standard Deviation of Blood Glucose can be a meaningless number for diabetics

Do all pumps give SD Readings or just Accuchek combo pump?

Thanks Ferrodene... Never knew just what percentage lay where... I just took it as smaller SD the better providing I try to keep my hba1c around 6.5 to 7.0.

Great answer from Ferrocene. Standard Deviation is a benchmark of variation that we can use to compare variation between data sets. To calculate is you first need to calculate the mean of your BGs. Then for each data point subtract the individual BG from the mean and square it. Add all these up, divide by the number of readings and take the square root.

For example, take these numbers: 5, 6, 8, 11. The mean is 7.5. To get the standard deviation, do the following:

(5-7.5)^2 + (6-7.5)^2 + (8-7.5)^2 + (11-7.5)^2 = 21

Standard deviation = square root of 21/4 = 2.29

If you can get all the readings into Excel, there's a functions called STDEV() that calculates the SD for a column on data.

phoenix said:

The big advocate of using SD is Irl Hirsch an American diabetologist. He says that ideally the standard deviation should be less than a third of the average, this is though too difficult for most people with T1 so he suggests a target of less than half.

Click to expand...

Gary Scheiner says the lower the SD the better, if I remember right he says if the SD is more than half of your meter average then this would signify a lot of high's and low's, if it's around a third less then this would be considered good control.

It's good that weezer has brought up a question about SD as it's an important aspect of diabetes management

There are a few averages: mean, mode and median. When people speak of an average, they normally mean the mean! Mode is used for most common value in a data set and median is used for the middle value in a data set. Sorry, I've got a background in stats and science so it's natural for me to say "mean" instead of "average".

Crokey... More andmore data... Honest.. I couldnt do iwith putting all my data in on an excel sheet.. But if ever I lose my remote or theariva expert blood tester,, then I know where I could get the data.

Why is it that from my 5 consultants/dsn's in past 3 years that inky one seemed to have an interest in SD? Isthis usual? I understand it with non pump people, but whenits there in the readings of the pump, I would have thought that more consultants would look at this.

Thanks guys, this is a really useful discussion.

If I were to go back over a year of readings and calculate the SD on a month by month basis and plot a graph, I guess I could see any trend in SD. Do you think this would be a valid indication of the tightness of control? My HbA1cs are always around 5.9, but I have a feeling my range of readings has been getting worse over the last 6 months.

If this is valid, do you think I should take out the subsequent readings where I've had mild hypos and tested a few times over the course of an hour to make sure the BG was recovering, or should I leave those in? I'm just wondering how much it would skew results if, for example my BG had been at 3.1 and I corrected with carb and then tested again after 15 minutes and again half hour later. I guess I mean that I tend to test more when out of range low than when out of range high - or does this not matter?

I generally test at least 7 times a day and upload all data to my PC, so i have a good set of data to work from.

Thanks

Smidge

Good question, Smidge. I reckon that using all of the hypo, 10-minute-post-hypo, 30-minute-post-hypo readings would add a bias to the data so would be tempted to only include the hypo event and your usual pre- and post-meal readings.

SamJB said:

Good question, Smidge. I reckon that using all of the hypo, 10-minute-post-hypo, 30-minute-post-hypo readings would add a bias to the data so would be tempted to only include the hypo event and your usual pre- and post-meal readings.

Click to expand...

But it shows exactly why I dont put much store in statistics, 90% of statistics are wrong and 85% of statistics are made up on the spot including the two I just quoted.

So statistically more statistics are inaccurate than accurate!!

Love it!

(Although I still stand by my comment!)

smidge said:

If this is valid, do you think I should take out the subsequent readings where I've had mild hypos and tested a few times over the course of an hour to make sure the BG was recovering, or should I leave those in? I'm just wondering how much it would skew results if, for example my BG had been at 3.1 and I corrected with carb and then tested again after 15 minutes and again half hour later. I guess I mean that I tend to test more when out of range low than when out of range high - or does this not matter?

Click to expand...

No you shouldn't exclude the low's Smidge, if your SD is more than half your average because of the low's then you need to address why you are having so many hypo's, likewise could be said of someone experiencing a lot of hyper's, ideally the SD should a be around a third as has been discussed on the thread.

From a statistical point of view it means that you have uneven sampling. It would add a higher statistical weighting to the hypo, decrease your average and increase your and SD.

I would only include the first initial hypo reading... As it is within the same time frame,-10, 25 mins later. That will just distort readings, as I guess normally you would have readings for pre meals, after meals, during night etc.

If you're on a combo pump, it breaks these down into your timeslots anyway. Quite easy to see which ones are deviating most...

Thanks for the advice, guys

I ran the Excel SD function through several months of readings last night and it was very interesting (I know, I should get out more! :lol: )

First thing I noticed is that last year, my SD was 1.4 to 1.8. It has now risen to 1.8 to 2.3 for the last 5 or 6 months. The rise seems to coincide with my change from Insuman basal to Levemir in the middle of last September. My meter averages have remained pretty constant between 5.6 and 6.2, and my HbA1c is generally about 5.9. So, this seems like something to spend more time looking at. So far, I haven't taken out the post-hypo readings so the results will be heavily weighted to that, but when I get some time I'll do that and see if it makes a big difference.

Anyway, I'll post the results once I've got them if they seem to show anything interesting once I've looked at them properly.

Smidge

I should have said all the readings are per month.

I've been quietly reading this to get my head around things (I love a good statistic)

I've run it through excel for the last month and got

Average BG: 7.5
Standard deviation 2.28

That is just under a third of the average, and so can be considered relatively good control right?

My priority now would be to reduce the average BG, without causing lots of hypos, and hence try to reduce the standard deviation also?