0. Petteri Väisänen November 4, 2015
An MMD and FDA compliant, intelligent self-care platform
Continuous Estimation of Glycosylated Hemoglobin (HbA1c) based on Self-monitoring Blood Glucose (SMBG) data and Laboratory HbA1c Measurements
Thank you Professor Herrara and good afternoon!
Hope You had a pleasant lunch and are full of energy to listen the last presentation of today.
Handouts and chewing gum. Unfortunately, it is not ERYTHRITOL but it is xylitol
TUT: Research project
QF: Past 1.5 years
Developing regulated medical device grade software solutions in the mHealth industry.
Petteri Väisänen
November 4, 2015

1. Agenda Introduction to established biomarker, HbA1c Objectives
Materials and Methods
Results
Discussion & Conclusions
(Fast trough) So lets get into it...

2. Introduction
Glycosylated hemoglobin (HbA1c, A1C, ‘long-term sugar level’) is the biomarker of glycemic control
Diabetes related risks are heavily linked to the glycosylation of proteins in all parts of the body
DCCT ( ) findings:
Intensive blood glucose control reduces risk of
eye disease by 76%
nerve disease by 60%
kidney diseas by 50%
EDIC findings ( ):
Intensive blood glucose control reduces risk of
any cardiovascular disease by 42%
nonfatal heart attack, stroke, or death from cardiovascular causes by 57%
HbA1c, glycosylated hemoglobin, long-term sugar level or just A1c is THE biomarker of glycemic control
Commonly known = Diabetes related risks are linked to the glycosylation of proteins
DCCT&EDIC: Type 1
Measurement of blood glucose control = HbA1c
National Institutes of Health; 2008;

3. Relative Risk of Complications versus HbA1c: DCCT
These are not all possible complications, because there is also wounds and ambutations,
but for the sake of time I think we already know WHY it is important to know ones HbA1c level
Then we can move on to measuring frequency of HbA1c…
Retinopathy = eye disease
Nephropathy = Kidney disease
Neuropathy = Nerve disese
Microalbuminuria = Describes a moderate increase in the level of urine albumin, valkuaisaine (symptom of renal disease)
Skyler, J.; 1996; Endocrinol Metab Clin North Am.; 25;

4. Other Diabetes Complications
Poor glycemic control is associated with increased risk of wounds and ambutations
HbA1c is predicting wound healing rate in diabetic patients
It has been shown that poor glycemic control, increased risk of wounds and ambutations.
And in addition to that, HbA1c is predicting the wound healing rate amongst diabetics
Diabetes related healthcare costs are reduced about 40 % when HbA1c gets 10 % lower, e.g, from 9 to 8.1.
Now we known why knowing one’s HbA1c level is important..
Gregg, E., et al.; 2004; Diabetes Care. Jul; 27:7;
Christman, A. et al.; 2011; J Invest Dermatol. Oct ;131;

5. Life time of erythrocytes is about 120 days HbA1c measured four times in a year?
-Functional life time of erythrocytes (red blood cell) is about days
Hemoglobin A1c changes slow
measured only every 3-4 months
BUT
Can change dramatically in a week
Because HbA1c is a "weighted" average of blood glucose levels during the preceding 120 days, meaning that glucose levels in the preceding 30 days contribute substantially more to the level of HbA1c
Yet, because of resources...

6. Common case in Finnish health care: HbA1c measured only once a year!
Although, this happens because of low resources in puplic health care sector, that is just not enough data to say wheather the diabetes management is going well or not.
This ultimately leaves unmonitored gaps
Then if we want more information, we could always measured it more often as in reasearch...

7. Reseach case: HbA1c measured about once a month
Tells a lot more about the diabetes management
Not only, expencive,
but also
impractical
Not pleasant for patient
One solution is already commonly used...

8. A1c-Derived Average Glucose (ADAG) study 2006-2008
Multicentered study included 507 subjects with Type 1 and Type 2 diabetes
Design to determine the mathematical relationship between HbA1c and average blood glucose (AG)
Subjects had 2 days of continuous glucose monitoring performed four times during the study period and seven-point self-monitoring of capillary glucose performed at least 3 days per week
Linear regression for estimated HbA1c (eA1c):
eA1c(%) = 0.63*AG(mmol/L)+1.63
ADAG study was designed to defined the mathematical relationship between HbA1c and average glucose
Works well over the population
Works if we know the actual average blood glucose
Measuring blood capillary blood glucose 7 times per day is quite a lot
If not done, measurements are not showing the real average glucose
I have actually an examble from our study…
Nathan, DM. et al; 2008; Diabetes Care; Aug.; 31:8;

9. Example from our study
HbA1c laboratory measurements do not match with average of self-monitored blood glucose (SMBG) values
Blood glucose measurement habits are biased
SMBG values are not resembling the average glucose, and thus, previous equation gives too low estimation for A1c
In this case, it is most likely because person measures only the low values
Therefore we sought define INDIVIDUALLY adaptive Continuous A1c...

10. Objectives Define individually adaptive Continuous A1c
 Fills unmonitored gaps
 Guide in day-to-day diabetes management
Our objective was to define INDIVIDUALLY adaptive Continuous A1c
Then more about materials and methods...

11. Materials and Methods 30 diabetics during 1 year period
Individual’s had their normal blood glucose measurement habits
No predifened measurement schedule
In average subjects reported
SMBG values per day
Laboratory HbA1c measurements per year
HbA1c change:
%-units/month
+0.95 %-units/month
Estimation algorithm utilizes a linear compination of multible statistical parameters such as mean, skewness, etc. in addition to laboratory HbA1c measurements
No predefined measurement schedule
Recommendation that subjects would take laboratory HbA1c as often as they could and all the measurements should be done in same laboratory
Averages
HbA1c change
Exclusion criteria
Linear combination of statistical parameters in addition to laboratory A1c
 Users have individual scaling factors defined
Then about the evaluation of accuracy and robustness...

12. Materials and Methods
Leave-one-out cross-validation was used to calculate mean absolute deviation (MAD) and mean absolute relative deviation (MARD)
Error grid analysis:
- Percentage of measurements within 10% error margin
- Percentage of measurements within 20% error margin
9 HbA1c
8 are used to train the algortihm and one which was left out, is to test it
Then we take another 8 HbA1c values to train the algorithm and evaluate how well it can predict the value which was left out
Then lets move on to Results...

13. Results: Example
Method can estimate A1c with biased measurement habits
An example shown previously…
Another example…

14. Results: Example 2
Method can estimate A1c with good accuracy and utilizes the available data efficiently
Good accuracy
Utilizes available data efficiently
Then the Error grid…

15. Results: Error grid 102 HbA1c measurements
Then more about the numerical results...

16. Results: Error Grid MAD and MARD were 0.41 and 5.38 %, respectively.
Population based eA1c
Continuous A1c
Dynamic eA1c*
Correlation (R)
0.46
0.89
0.76
Within
10 %
51.26 %
87.25 %
77.50 %
20 %
85.71 %
98.04 %
97.90 %
Then lets move on to discussion and conclusions...
* Different data
Kovatchev, B. et al; 2014; Diabetes Technology & Therapeutics; 16:5;

17. Discussion & Conclusions
HbA1c is the biomarker of glycemic control and it is measured 2-4 times in a year
Unmonitored gaps
Here, a mathematical model was developed to estimate individual’s HbA1c by using their routine SMBG and laboratory HbA1c measurements
Algorithm needs reference HbA1c for calibration
Sample size in the study was only 30 diabetics
Results show that the method works with biased and irregular measurement patterns
Although, HbA1c is the biomarker of glycemic control  measured every few months at best  Unmonitored gaps
ADAG defined simple way
We developed an individually adaptive Continuous A1c that routine SMBG and lab A1c, and can work with biased and irregular measurement patterns
Few issues: Algorithm needs reference HbA1c for calibration and sample size was only 30 diabetics, thus a larger study would give statistically more reliable results
Then to conclude...

18. Discussion & Conclusions
Continuous A1c…
utilizes routine monitoring data
fills unmonitored gaps
guides in day-to-day diabetes management
enables fast feed-back
increases reliability towards treatment
Continuous A1c…
utilizes routine monitoring data
fills unmonitored gaps
guides in day-to-day diabetes management
enables fast feed-back
increases reliability towards treatment
And the thing is:
We have the technology in our hands and we know that the information is there, the change in HbA1c is visible
And then if YOU put your self in a position of a diabetic: Would you want the feed-back in 3 months OR would you RATHER have it NOW?
Bonus: HbA1c -0.3units
Hypos reduced 30%

19. Thank You! For more information: petteri@quattrofolia.com