Philippe Wagner, presenterar sitt examensarbete  

On blood glucose based prediction of glycated haemoglobin A (HbA1c) in type II
diabetes subjects.
	
Abstract.  Type II diabetes is, at least in western society, growing
to be a substantial health issue. In the U.S. alone estimates are that
6,3% of entire the population carry some form of the disease. In this
context it is essential that treatments are developed to be effective
in a medical as well as an economical sense.  It has been shown that
glycemic control is fundamental in decreasing risks for diabetic
complications. In order to assess long-term glycemic control the
amount of glycated haemoglobin A (HbA1c) contained in the blood is
measured. The ratio between the HbA1c level and the total amount of
haemoglobin is believed to be a good indicator of the mean blood
glucose levels of the four months preceding HbA1c determination.
Since measuring HbA1c levels is a costly procedure, this thesis is
concerned with trying to predict HbA1c levels by using the less costly
blood glucose measurements. As a consequence, it would enable more
rapid drug evaluation in addition to reducing pharmaceutical
development costs.    In many cases modelling of the blood glucose -
HbA1c relation is attempted by application of univariate linear
models. Most commonly using the measured or approximate area under the
curve corresponding to the blood glucose profile as explanatory
variable. It is shown in this thesis, however, that modelling results
can be significantly improved by using a multivariate approach. Both
approaches are put in context of the process bio-kinetics in order to
yield a possible explanation for the advantage of the multivariate
approach.  Furthermore, the process bio-kinetics suggest that the
blood glucose - HbA1c relation is in fact non-linear. This is
confirmed by use of polynomial models, leaving room for the
possibility that modelling results may be further enhanced if a
suitable non-linear model was to be determined. Nevertheless, in spite
of the improvements made to existing models, prediction and modelling
results remain poor. This may have several possible
explanations. However, in order for prediction of HbA1c to be possible
a necessary condition is that the blood glucose profile is
approximately periodic, with a period of roughly 24 hours. Examination
of data variation lead us to believe that the major factor in imposing
variation on data is that population blood glucose profiles do not
fulfil this condition. Subsequently, it is in the thesis concluded
that the limitation of this type of prediction lay primarily with the
treatment of population subjects and not in the inadequate description
of the blood glucose - HbA1c relation.