Diabetic controlling blood sugar, Spreadsheet to estimate HbA1c

Hello everyone:

I am relative newcomer to these groups, but I would like to share something
that I came up with to estimate my HbA1c levels. I have been following the
advice about the monitoring after meals in order to keep good control of
blood sugars, but could not find out anywhere in the net a place that would
let me estimate the impact that said control would have on my HbA1c levels.

As most of you are aware, a diabetic must try to keep this percentage under
6.5% in order to get better odds at not developing complications. Because I
could not find a tool to do this, I came up with one of my own by working a
formula to estimate the average blood sugar using HbA1c.

I created a small one page site for people to go download it. I would
appreciate it if people with data and an actual HbA1c test result would try
it out and tell me how accurate it is.

The spreadsheet can be found he

http://www.loanuniverse.com/Diabetes/

I await your feedback.

P.S: I am sorry for the cross-posting but I tried to keep the posts limited
to those groups that have a lot of diabetics as users.

R. loanguy wrote:
Hello everyone:

Hi Loanguy, I don't do spreadsheets but just want to say welcome

"R. loanguy" wrote in
:

Hello everyone:

I am relative newcomer to these groups, but I would like to share
something that I came up with to estimate my HbA1c levels. I have
been following the advice about the monitoring after meals in
order to keep good control of blood sugars, but could not find out
anywhere in the net a place that would let me estimate the impact
that said control would have on my HbA1c levels.

As most of you are aware, a diabetic must try to keep this
percentage under 6.5% in order to get better odds at not
developing complications. Because I could not find a tool to do
this, I came up with one of my own by working a formula to
estimate the average blood sugar using HbA1c.

I created a small one page site for people to go download it. I
would appreciate it if people with data and an actual HbA1c test
result would try it out and tell me how accurate it is.

The spreadsheet can be found he

http://www.loanuniverse.com/Diabetes/

I await your feedback.

P.S: I am sorry for the cross-posting but I tried to keep the
posts limited to those groups that have a lot of diabetics as
users.

FollowUp set to m.h.d.

You cannot make a sensible estimate of A1c from finger stick records.
There are several issues, I'll just give a short summary.

1. Sampling bias. If you take, for example, only fasting and pre-meal
glucose tests you will get a very different average than if you take
only 2 hour post prandial tests. This is called sampling bias. Worse
than that, different types of diabetes and different types of treatment
will vary the sampling bias differently. A type 2 treated with
exercise and diet will likely have higher postprandial bg levels than
preprandial. For a type 1 with a tight basal/bolus regimen with a fast
acting prandial insulin, 2 hour postprandials can be the lowest bg
levels of the day. Sampling bias is a difficult issue.

2. Sampling rate. If you assume three meals/day and the possible
occurrence of some events overnight (dawn phenomenon, Somoygi), you can
think of a time series of blood glucose as having 4 major cycles/day.
This is, of course, over simplified. Elementary sampling theory will
tell you that you need at least 8 samples/day to characterize the
signal. As with most theoretical constructs, elementary sampling
theory gives an answer which is insufficient for any real practical
use. The practical answer depends on a number of factors, not the
least of which is how big you want the error bars to be, but very few
data analysts would be happy with less than twice the theoretical
limit, i.e. 16 samples/day.

Worse than this, if you look at any MiniMed 3 day continuous blood
glucose record you will be able to see that there is much variation in
bg at much higher frequencies than the 4 cycles in the simple picture
above. This, of course, drive up the number of samples/day required to
get a sensible average.

3. Normal range. Every lab has its own normal range. There is a
movement to standardize on the DCCT normal range (4.0%-6.0%) but it is
progressing very slowly. Even if your spread sheet worked, it would
work for only labs with the same testing methodology as your lab and
the same normal range. Different methods for assaying A1c actually
measure slightly different things, so the comparison between labs is
difficult.

4. Many of the attempts to correlate daily bg testing rely on the data
published by the DCCT. They did a correlation between a 7 stick
regimen (3 preprandial, 3 postprandial, and 3 am) and A1c. Ignoring
sampling rate considerations, it was an attempt to correlate a standard
testing pattern and A1c in a specific population. It is difficult to
get even that limited objective accomplished. Several years after the
original publication, they found they had botched the original analysis
and published a correction. Most of the existing charts of bg vs. A1c
are based on the original analysis.

There are other more arcane technical issues, but the bottom line is
you just can't get there from here. Note, that for an individual with
stable blood glucose patterns it is quite possible to develop a
correlation between A1c and a stable pattern of bg measurements. If
the bg pattern changes, however, in a way masked by the sampling
interval or bias, then the resultant A1c changes will not be predicted
by the correlation which is no longer valid. This makes it, at best, a
futile exercise and, at worst, a misleading indicator.
The correlation will not, in general, be valid in any other diabetic.

--
-------------
Charly Coughran

Charly: I was aware of the sampling bias and sampling rate problems to being
with. In fact, as I was writing the spreadsheet I tried to address them in
the little FAQ, I created:

".........this model does not properly assign weights to the different
measurements. For example, if you measure your blood glucose one hour after
lunch you will have a high number, but this number might be short-lived. On
the other hand, if you measure before going to bed and get a low number and
this number lasts during the whole night the model gives each one of them
the same importance....."

and he

".......The more data you have, the better your average will be and
therefore the better the model should work in theory. In order to use more
data, some colums need to be added. I am using five colums because I try to
at least do those five tests if I have the time........"

I was not aware that there were problems with the available data that shows
a correlation between the average blood glucose and the HbA1c result is
flawed. I also did not consider the difference in the ranges as material
while I understand the basics of how the ranges are determined in different
populations.

Frankly, I am not sure if the result is going to be even close to the one
that might show on my next HbA1c test. Since I have only been diagnosed for
less than three months and I have to make a follow up appointment soon. The
reason why I posted here was to see if people out there had been measuring
themselves a lot, had access to a recent HbA1c result or were going to get
tested soon then it would help test how useful the spreadsheet is. The whole
thing might be a "futile exercise", but since some people are already
testing themselves a lot and the new meters have memory, it is a matter of
just going back and entering the information.

Not too much work in my humble opinion, at worst they can print the results
and take them as their log for their next doctor's visit.

Thanks for you comments

P.S: I agree with you that the whole thing is imperfect, it is a matter of
trying to come up with as good an estimate as one can with the available
data. Personally, I think that it will work best for people that test more
often than I do or that at least keep to the number of tests that they set
themselves to do everyday. I try to do at least five a day, but I have
missed many.

While researching the topic, I also read about how the process of binding to
hemoglobin accelerates a lot while having really high levels even if they
are short-lived so that is another monkey wrench on the whole thing.
Fortunately, I have only had one incidence of a reading over 200 in the last
couple of months, but for someone with worse control, using this might be
completely misleading instead of just a little misleading.

"Charly Coughran" wrote in message
1...
"R. loanguy" wrote in
:

Hello everyone:

I am relative newcomer to these groups, but I would like to share
something that I came up with to estimate my HbA1c levels. I have
been following the advice about the monitoring after meals in
order to keep good control of blood sugars, but could not find out
anywhere in the net a place that would let me estimate the impact
that said control would have on my HbA1c levels.

As most of you are aware, a diabetic must try to keep this
percentage under 6.5% in order to get better odds at not
developing complications. Because I could not find a tool to do
this, I came up with one of my own by working a formula to
estimate the average blood sugar using HbA1c.

I created a small one page site for people to go download it. I
would appreciate it if people with data and an actual HbA1c test
result would try it out and tell me how accurate it is.

The spreadsheet can be found he

http://www.loanuniverse.com/Diabetes/

I await your feedback.

P.S: I am sorry for the cross-posting but I tried to keep the
posts limited to those groups that have a lot of diabetics as
users.

FollowUp set to m.h.d.

You cannot make a sensible estimate of A1c from finger stick records.
There are several issues, I'll just give a short summary.

1. Sampling bias. If you take, for example, only fasting and pre-meal
glucose tests you will get a very different average than if you take
only 2 hour post prandial tests. This is called sampling bias. Worse
than that, different types of diabetes and different types of treatment
will vary the sampling bias differently. A type 2 treated with
exercise and diet will likely have higher postprandial bg levels than
preprandial. For a type 1 with a tight basal/bolus regimen with a fast
acting prandial insulin, 2 hour postprandials can be the lowest bg
levels of the day. Sampling bias is a difficult issue.

2. Sampling rate. If you assume three meals/day and the possible
occurrence of some events overnight (dawn phenomenon, Somoygi), you can
think of a time series of blood glucose as having 4 major cycles/day.
This is, of course, over simplified. Elementary sampling theory will
tell you that you need at least 8 samples/day to characterize the
signal. As with most theoretical constructs, elementary sampling
theory gives an answer which is insufficient for any real practical
use. The practical answer depends on a number of factors, not the
least of which is how big you want the error bars to be, but very few
data analysts would be happy with less than twice the theoretical
limit, i.e. 16 samples/day.

Worse than this, if you look at any MiniMed 3 day continuous blood
glucose record you will be able to see that there is much variation in
bg at much higher frequencies than the 4 cycles in the simple picture
above. This, of course, drive up the number of samples/day required to
get a sensible average.

3. Normal range. Every lab has its own normal range. There is a
movement to standardize on the DCCT normal range (4.0%-6.0%) but it is
progressing very slowly. Even if your spread sheet worked, it would
work for only labs with the same testing methodology as your lab and
the same normal range. Different methods for assaying A1c actually
measure slightly different things, so the comparison between labs is
difficult.

4. Many of the attempts to correlate daily bg testing rely on the data
published by the DCCT. They did a correlation between a 7 stick
regimen (3 preprandial, 3 postprandial, and 3 am) and A1c. Ignoring
sampling rate considerations, it was an attempt to correlate a standard
testing pattern and A1c in a specific population. It is difficult to
get even that limited objective accomplished. Several years after the
original publication, they found they had botched the original analysis
and published a correction. Most of the existing charts of bg vs. A1c
are based on the original analysis.

There are other more arcane technical issues, but the bottom line is
you just can't get there from here. Note, that for an individual with
stable blood glucose patterns it is quite possible to develop a
correlation between A1c and a stable pattern of bg measurements. If
the bg pattern changes, however, in a way masked by the sampling
interval or bias, then the resultant A1c changes will not be predicted
by the correlation which is no longer valid. This makes it, at best, a
futile exercise and, at worst, a misleading indicator.
The correlation will not, in general, be valid in any other diabetic.

--
-------------
Charly Coughran

Thank you Ozgirl


"Ozgirl" wrote in message
...
R. loanguy wrote:
Hello everyone:

Hi Loanguy, I don't do spreadsheets but just want to say welcome

The list of problems I gave you was not exhaustive. Spend the time
and effort learning as much as you can about your disease instead.
If you wish to understand the difficulties I didn't mention, I would
strongly suggest you read the FAQ section on A1c and look up the
references cited there.

The A1c reaction cannot be modeled as a simple reaction. It is a two
step reaction with differing forward and reverse reaction constants
for both steps. The average that A1c represents is quasi-
exponentially weighted in time and, if the models are to be believed,
not particularly sensitive to short term glucose fluctuations.

--
-------------
Charly Coughran




"R. loanguy" wrote in
:

Charly: I was aware of the sampling bias and sampling rate
problems to being with. In fact, as I was writing the spreadsheet
I tried to address them in the little FAQ, I created:

".........this model does not properly assign weights to the
different measurements. For example, if you measure your blood
glucose one hour after lunch you will have a high number, but this
number might be short-lived. On the other hand, if you measure
before going to bed and get a low number and this number lasts
during the whole night the model gives each one of them the same
importance....."

and he

".......The more data you have, the better your average will be
and therefore the better the model should work in theory. In order
to use more data, some colums need to be added. I am using five
colums because I try to at least do those five tests if I have the
time........"

I was not aware that there were problems with the available data
that shows a correlation between the average blood glucose and the
HbA1c result is flawed. I also did not consider the difference in
the ranges as material while I understand the basics of how the
ranges are determined in different populations.

Frankly, I am not sure if the result is going to be even close to
the one that might show on my next HbA1c test. Since I have only
been diagnosed for less than three months and I have to make a
follow up appointment soon. The reason why I posted here was to
see if people out there had been measuring themselves a lot, had
access to a recent HbA1c result or were going to get tested soon
then it would help test how useful the spreadsheet is. The whole
thing might be a "futile exercise", but since some people are
already testing themselves a lot and the new meters have memory,
it is a matter of just going back and entering the information.

Not too much work in my humble opinion, at worst they can print
the results and take them as their log for their next doctor's
visit.

Thanks for you comments

P.S: I agree with you that the whole thing is imperfect, it is a
matter of trying to come up with as good an estimate as one can
with the available data. Personally, I think that it will work
best for people that test more often than I do or that at least
keep to the number of tests that they set themselves to do
everyday. I try to do at least five a day, but I have missed many.

While researching the topic, I also read about how the process of
binding to hemoglobin accelerates a lot while having really high
levels even if they are short-lived so that is another monkey
wrench on the whole thing. Fortunately, I have only had one
incidence of a reading over 200 in the last couple of months, but
for someone with worse control, using this might be completely
misleading instead of just a little misleading.

"Charly Coughran" wrote in
message 1...
"R. loanguy" wrote in
:

Hello everyone:

I am relative newcomer to these groups, but I would like to
share something that I came up with to estimate my HbA1c
levels. I have been following the advice about the monitoring
after meals in order to keep good control of blood sugars, but
could not find out anywhere in the net a place that would let
me estimate the impact that said control would have on my HbA1c
levels.

As most of you are aware, a diabetic must try to keep this
percentage under 6.5% in order to get better odds at not
developing complications. Because I could not find a tool to do
this, I came up with one of my own by working a formula to
estimate the average blood sugar using HbA1c.

I created a small one page site for people to go download it. I
would appreciate it if people with data and an actual HbA1c
test result would try it out and tell me how accurate it is.

The spreadsheet can be found he

http://www.loanuniverse.com/Diabetes/

I await your feedback.

P.S: I am sorry for the cross-posting but I tried to keep the
posts limited to those groups that have a lot of diabetics as
users.

FollowUp set to m.h.d.

You cannot make a sensible estimate of A1c from finger stick
records. There are several issues, I'll just give a short
summary.

1. Sampling bias. If you take, for example, only fasting and
pre-meal glucose tests you will get a very different average than
if you take only 2 hour post prandial tests. This is called
sampling bias. Worse than that, different types of diabetes and
different types of treatment will vary the sampling bias
differently. A type 2 treated with exercise and diet will likely
have higher postprandial bg levels than preprandial. For a type
1 with a tight basal/bolus regimen with a fast acting prandial
insulin, 2 hour postprandials can be the lowest bg levels of the
day. Sampling bias is a difficult issue.

2. Sampling rate. If you assume three meals/day and the
possible occurrence of some events overnight (dawn phenomenon,
Somoygi), you can think of a time series of blood glucose as
having 4 major cycles/day. This is, of course, over simplified.
Elementary sampling theory will tell you that you need at least 8
samples/day to characterize the signal. As with most theoretical
constructs, elementary sampling theory gives an answer which is
insufficient for any real practical use. The practical answer
depends on a number of factors, not the least of which is how big
you want the error bars to be, but very few data analysts would
be happy with less than twice the theoretical limit, i.e. 16
samples/day.

Worse than this, if you look at any MiniMed 3 day continuous
blood glucose record you will be able to see that there is much
variation in bg at much higher frequencies than the 4 cycles in
the simple picture above. This, of course, drive up the number
of samples/day required to get a sensible average.

3. Normal range. Every lab has its own normal range. There is
a movement to standardize on the DCCT normal range (4.0%-6.0%)
but it is progressing very slowly. Even if your spread sheet
worked, it would work for only labs with the same testing
methodology as your lab and the same normal range. Different
methods for assaying A1c actually measure slightly different
things, so the comparison between labs is difficult.

4. Many of the attempts to correlate daily bg testing rely on
the data published by the DCCT. They did a correlation between a
7 stick regimen (3 preprandial, 3 postprandial, and 3 am) and
A1c. Ignoring sampling rate considerations, it was an attempt to
correlate a standard testing pattern and A1c in a specific
population. It is difficult to get even that limited objective
accomplished. Several years after the original publication, they
found they had botched the original analysis and published a
correction. Most of the existing charts of bg vs. A1c are based
on the original analysis.

There are other more arcane technical issues, but the bottom line
is you just can't get there from here. Note, that for an
individual with stable blood glucose patterns it is quite
possible to develop a correlation between A1c and a stable
pattern of bg measurements. If the bg pattern changes, however,
in a way masked by the sampling interval or bias, then the
resultant A1c changes will not be predicted by the correlation
which is no longer valid. This makes it, at best, a futile
exercise and, at worst, a misleading indicator. The correlation
will not, in general, be valid in any other diabetic.

--
-------------
Charly Coughran

R. loanguy wrote:

Hello everyone:

I am relative newcomer to these groups, but I would like to share
something that I came up with to estimate my HbA1c levels. I have been
following the advice about the monitoring after meals in order to keep
good control of blood sugars, but could not find out anywhere in the net a
place that would let me estimate the impact that said control would have
on my HbA1c levels.

As most of you are aware, a diabetic must try to keep this percentage
under 6.5% in order to get better odds at not developing complications.
Because I could not find a tool to do this, I came up with one of my own
by working a formula to estimate the average blood sugar using HbA1c.

I created a small one page site for people to go download it. I would
appreciate it if people with data and an actual HbA1c test result would
try it out and tell me how accurate it is.

The spreadsheet can be found he

http://www.loanuniverse.com/Diabetes/

Hi R,

HbA1c prediction was discussed in great detail about 2 months ago on MHD.
So I don't want to rehash the details again.

Basically, besides the sampling issue Charly mentioned, (i.e., Nyquist
sampling requirement for the AUC [integral] of the glucose profiles), your
model does not account for the red blood cell lifespan. 90 days of BG
averaging only accounts for 90 day old RBCs. But what about the new RBCs
made today? They have no glycosylation yet but are in the blood sample
drawn for the lab test of HbA1c. In short, the plasma compartment acts a
mixing vat for daily RBC cohorts and the mathematics must account for these
newer RBCs somehow. Also note that people without spleens will have longer
RBC lifespans and your model will have to find these values as well. (FYI,
the textbooks say RBCs live an average of 120 days in normals.)

Next, there are seasonal variations in HbA1c where 2 Swedish MDs (Diabetes
Care 12/96 if memory serves me) measured ~0.5% (converted to Bio-Rad HPLC
equivalent by me as it was about 0.8% for their methodology) yearly
variation in HbA1c in about 800-1000 patients over 10 years. Max was in
January and minimum was in July. I suspect the pineal gland in the brain
controls this hormonally and your model must account for this variation.

So your model is too simplistic to work universally for all DMs,
--
Jim Dumas
T1 4/86, background retinopathy, rarely hypoglycemic: 1/mo.
lispro+R+U+NPH daily, moderate exercise, typically 6% HbA1c

R. loanguy wrote:

snip

I await your feedback.

Hi R,

I decided to look for the seasonal variation of HbA1c DC reference. It is:

Seasonal variation of HbA1c in adult diabetic patients, Johan Asplund,
Diabetes Care, v20 #2, Feb 1997, p. 234.

500 to 800 patients beginning in 1987 for 8 years of data. 55% are type 1s.
HbA1c is measured via ion-exchange chromatography with reference 5.2% as
normal. So this is a different method than used in the DCCT with the
Bio-Rad HPLC instruments: ref 6.05% normal. Total HbA1c samples of
11,473. Average January HbA1c peak of 7.61% and July nadir of 7.23%
(student's t test P 0.0001!). So we have a difference of ~0.4% (round to
tenths) that must be converted to the US DCCT equivalent somehow. So I
guessed 0.5% for the US DCCT equivalent. The paper concludes patients have
better control in the summer and thus the lower HbA1c values.

But I disagree. I've kept BG data for myself since May, 1986. I don't make
any changes to my computer controlled insulin dosing and thus my BG control
is identical in January versus July. Exercise is likewise identical and
diet has little difference (more fresh veggies maybe). But I see an
average 0.7% summer-winter constant mean BG difference in HbA1c with the
Bio-Rad HPLC measured HbA1c over 1992 to 1999. I started using a computer
to dose in 1990 after a serious hypoglycemic event while driving my car.
This forced me to take action with tools available and I've not had another
ER visit since.

I also have taken great pains to use one BG meter assay methodology to keep
the BG data comparable over the years. This was all with the AccuChek III
double assay (two color Chemstrip bG) meter that is highly respected for
its accuracy. I have also forced my endos to use a lab that has the "Gold
Standard" Bio-Rad Diamat HPLC or equivalent HbA1c assay methodology to keep
the HbA1c assays comparable.

So my personal data hypothesis is the seasonal variation is caused by an
hormonal influence much like the artic fox's immune system slows down in
the winter months to conserve resources. This observation can be refined
easily now that OTC A1cNow kits are available. These assays are biased to
be equivalent to the DCCT methodology at a 7% HbA1c. They read slightly
higher when below 7% and slightly lower when above 7% per the professional
product literature in the Metrika 10 pack kits.

Thus my model (and yours by implication) must account for this variation.

The seasonal variation is real. What causes it is debatable,
--
Jim Dumas
T1 4/86, background retinopathy, rarely hypoglycemic: 1/mo.
lispro+R+U+NPH daily, moderate exercise, typically 6% HbA1c