Ambulatory Glucose Profile

Some months ago Stuart - if I remember that correctly - published a link to the paper about the "Ambulatory Glucose Profile (AGP)".

The basic idea is to propose a standardized analysis and presentation of glucose data. Every application for data analysis (of CGM data for example) should at least integrate this AGP diagram. This standardization should help clinicians, researchers and patients in these aspects:

-learning one uniform interpretation of the AGP presentation (for safety)

-reducing the efford to handle software of different manufacturers (for time saving)

In my opinion this is an important development. Clinicians are confronted with a multitude of patients and their broad spectrum of equipment and devices. The more experience they have with one specific software the more reluctant they will get to work with the software choosen by their patients. As a result many conclusions and nuggets of wisdom will remain burried in the data. Here the standardization will help Clinicians by providing something they are familiar with. Thus they can utilize all of their experience to work on the interpretation of the data and their conclusions.

Because of these benefits we started to explore the possibility to create an Ambulatory Glucose Profile based on the data in our Glucosurfer Project. The following diagram is the first result of our effords that is based on the paper from Bergenstal et al:

What does it show?

It shows one day from 12 AM to 12 PM on the x axis. The five lines are showing 5 Quartiles of the blood glucose of the last 14 days. The main focus is the variability of the blood glucose.

What is a Quartile?

Let us assume we have the blood glucose at 8 AM for the last 14 days. Next you will find an example of 14 values at 8 AM:

109, 90, 180, 117, 78, 167, 112, 87, 176, 92, 148, 118, 136, 100

Now we order these 14 values in ascending order:

78, 87, 90, 92, 100, 109, 112, 117, 118, 136, 148, 167, 176, 180

Now we are interested at numbers at specific positions. These positions are 10%, 90%, 25%, 75% and 50% positions of the total number of values:

10% Quartile => Value at the position 0.1 * 14 = 1.4 = 1 Position => 78

90% Quartile => Value at the position 0.9 * 14 = 12.6 = 13 Position => 176

25% Quartile => Value at the position 0.25 * 14 = 3.5 = 4 Position => 92

75% Quartile => Value at the position 0.75 * 14 = 10.5 = 11 Position => 148

50% Quartile => Value at the position 0.5 * 14 = 7 Position => 112

The 50% Quartile has also the mathmatical name Median.

Quartiles to diagram

The previous calculation gives you the Quartile values for one specific time. By repeating the calculation for every minute of the day you will get a line of values for every Quartile.

Top dashed line: 90% Quartile for every minute of the last 14 days

Bottom dashed line: 10% Quartile for every minute of the last 14 days

Top yellow area: 75% Quartile for every minute of the last 14 days

Bottom yellow area: 25% Quartile for every minute of the last 14 days

Red line: 50% Quartile or Median for every minute of the last 14 days


The diagram represents the variability of the blood glucose for specific times of the day. By using Quartiles instead of averages the result is not influenced by sporadic outliers.

In general the yellow area or 25% and 75% Quartile has been mostly in the healthy range. So 50% of the numbers are showing good control. But at 8 AM you can see that the yellow area around the Median is growing. This represents the fact that my blood glucose in the morning is influenced by the dawn phenomenon. In the last 14 days I had a higher number of elevated numbers in the morning. This is clearly something that I need to address. The dashed lines also indicate that I need to invest more effords on reducing the outliers. So in general the reducion of deviation is the main goal.

By comparing this diagram with a diagram that will be created later in time I can evaluate my improvements. The general idea is to make the yellow area smaller. As a result the top and lower yellow line will be nearer to the red line. This would show that the deviation has been reduced. Furthermore the dashed lines should be as close to the yellow line as possible. This would show that the number of extreme outliers has been reduced. As a last aim the red line should be within the green area with a good safety margin. This would show that the Median of all glucose numbers for the last 14 days is well within the healthy range.

I am thankful for the input the working paper of Bergenstal et al. provides. There are many other great ideas in this paper that we would like to implement. The work on this new AGP analysis revealed that we need much more computing power to achieve this. For that we need new Hardware because our system from 2007 is clearly too limited to offer the AGP diagram to all of our users. If you would like to support our project please visit our donations page. Even the smallest donations can make a big difference. It just depends on the number of people to participate - which is also true for donations to