I had just finished looking at my pump BG/insulin charts. And there is a "standard deviation" number in the data. I am wondering if anyone knows what that number means, and what might be a good goal for it? Is a high number good, or a low number?
It's how much your BG varies from the average, I think that a number in the 30s is decent, 20s is good? I am not sure there's a medical recommendation on it but that's what I shoot for?
Standard deviation is a method for looking at a data set (in our case Bg numbers) and determining the variation from the mathmatical average. Looking at 90 days worth of numbers can be daunting as we can easily see the high, the low and the average, but what does this really tell us about our control. Statistical mathmatics can dig out additional information. BTW, lower is a better number, for example:
Given a meter average of 120 mg/dl and a deviation of 30 gives us a range of 90-150 mg/dl (120-30 and 120+30) tells us that 68% of all readings fall within this range. This is known as "one" standard deviation. If we double this standard devation to 60 (which is two standard deviations) then the range becomes 60-180 mg/dl which tells us that 96% of all readings fell between 60-180. Additional statiscal formulas are avaible to further break down a large data set of Bg number.
Thanks for the detailed explanation, I now understand what this number means..
Monitoring standard deviation and striving to keep it low (in the 30s or less) will help keep you off of the "roller coaster." I think that BG variability is a more important measure of your overall health than the A1c number.
A diabetic with a high standard deviation could have a good A1c and receive a "good report" from the doctor. I think a combination of the two numbers is a better overall indicator of glucose control.
There are studies that show lower variability (low SD) is associated with a lower incidence of complications, correcting
for A1C.
See, for example
http://www.ncbi.nlm.nih.gov/pubmed/18824382
The same study showed that high variability (high SD) is associated with a higher incidence of hypoglycemic unawareness (not surprising, really, as high variability likely means you are hitting lows more often).
Another paper is here:
http://www.mendosa.com/minimal_variability.pdf
That is assuming that the meter reading is a random variable that follows a Gaussian distribution. Other distribution will have other percentages falling under a certain multiple of standard deviations.
It is worth to note that the deviation of blood glucose for healthy people is very small. For them even the range from 70 to 140 is unlikely and most of the time they will have numbers from 90 to 110. The fluctuation itself has been recognized as an additional factor for complications. Groups of people with the same A1c but different glucose deviation will have a different likelyhood for complications. The higher the deviation the higher the risk of complications. This can go so far that a higher A1c combined with a lower deviation can lead to less complications. But the true nature is still under question. To me it seems that we have not much scientific data for diabetics below 6%. We do not know if this more healthier A1c is capable to compensate for some of the negative effects of deviation. All we can do is to shoot for both: good A1c und small deviation.
It is utterly unfair that with the loss of insulin we have also lost the capability to produce amylin - also produced by the beta cells. In healthy people this hormone will slow the digestion so insulin can catch up in the blood stream with the consumed glucose. With our fast digestion we will often face the problem of high glucose spikes after meals. The compensation for this can be a higher waiting time after injection or higher fat intake or the build-up of muscle mass to buffer the glucose spike. Even a mild gastroparesis might have the positive effect of moderating the spikes - but I do not want to paint that in positive colors.
For many insulin dependent diabetics that strive to "normalize" their A1c, the SD becomes a major issue. You can't safely bring your average blood sugar down into the 80s without limiting your blood sugar swings. It just places you at too much risk of hypos. And make no mistake, you cannot just limit lows, swings high invariably result in higher rates of swings low. And the efforts that you need to make to reduce your SD are along many dimensions, it is not that easy. A key element is of course postprandial variance, which can most readily be reduced by reducing your carb load and with accurate dosing, timing and profiling of insulin. Once you have brought that in line reducing variation in morning and fasting blood sugars becomes much more complicated. And life is full of disruptions, like gastroparesis, but also exercise, sleep variations, stress and who knows what. In the end, even with a perfect CGMS, it is impossible for most people to hope to get their SD below 10. Not only do our bodies not properly produce amylin, but we also don't properly produce glucagon (which raises blood sugar) and the insulin that slowly absorbs from our injections cannot hope to achieve the fine tuning that normal bodies can achieve with insulin released directly into the portal vein.
+1
As general rule, a smaller standard deviation means a smaller variation around the mean, but using it to compare your data to another individuals data is not how it's meant to be used.
Standard deviation is simply an [i]estimate[/i] of the variation within a group or population and assumes that those data are normally distributed. Assuming a normal distribution of numbers around the average, you can then compare the variation for those two or more groups.
You can't do that with your own BG numbers because they are not normally distributed. One way to look at it is to assume that a normal or average BG will be 80. Assuming your BG goes up to a high of 120, you're also assuming that your BG will drop to a low of 40 and that there will be an equal distribution of BGs between 80 and 120 on the high side, aND 80 and 40 on the low side. Our BG numbers are going to be skewed which violates the use of standard deviation as a way to measure total daily variation.
We shouldn;t have to estimate variation because, if we are measuring our BGs concientiously, we alrady know what our average, high, and low BG's are. In a population of individuals without a complete data set, variation HAS to be estimated which is what standard deviation is for.
A better way for us to use standard deviation would be to treat our fasting and post meal BGs as two seperate groups of data. You would need multiple days worth of those data to make the comparison. You would expect those to be tightly controlled if you are under good control. You would then compare those two means, generate a standard deviation for each, then compare those two standard deviations.
Unless everybody you are comparing yourself to does pretty much the same thing, you can't compare your data to someine elses. I can generate misleading low SDs by simply measuring my BG right before a meal, then two hours later when hope my BG has come back to normal. If I'm not measuing peak BG, I'm not generating an accurate measurement of my daily variation.
Yes, these percentages do assume a Gaussian distribution (bell curve). My discourse above was meant only to explain how SD works and what information it can divulge. Bg readings are not going to be evenly distributed across the bell curve.
As to Bg readings the only ones I can attest to are my own. For myself the actual percentages are only 2-3 points off from the percentages above. Keep in mind my sugars are fairly consistent, such that my fasting, pre and post meal numbers don't deviate more than 30-40 points from one day to the next. Sure . . . there are occasional boo-boo's and what the hell's but overall I end up fairly close.
Also for me, the actual Bg results contained within one standard deviation doesn't necessarily imply that the numbers within that group are evenly distributed, or for that matter acutally contain a result at the min and max of the SD. only that they represent roughly 68% of all the Bg results being examined including those readings that are contained within 2 SD's.
I haven't done any additional study of this as I only have my data to work with, but I suspect stability contributes to SD working fairly closely to the "text books" on the subject and that even a tighter stability, maybe even that of a non diabetic would even be closer yet.
I wonder if I could get the wife or one of my children to collect 90 days worth of readings 6-7 times a day? Regardless, lower SD is better and I know for a fact my Doc always looks at this information.
Hi all,
Sorry, long time lurker who hasn't checked in in a long time.
I actually run a little diabetes blog and have two articles on this very question. I'm a bit of a math geek so I go into detail, but with examples. It's best to read these two in order...
http://www.diydiabetic.com/blood-sugars/blood-sugar-by-the-numbers-averages/
http://www.diydiabetic.com/blood-sugars/blood-sugar-by-the-numbers-standard-deviation/
I'd be interested in what people think of the rest of the site! Plus, we're looking for contributors. :)
All the best,
Steve
Thanks for your input, Steve. I did read the article you referenced and got a lot out of it. I respect Gary Scheider very much. Your site is very appealing and well thought out. I can see it becoming popular. I hope you will lurk less and contribute more to this site!
Thanks for the kind words, Cindy!