Fast estimation of meal/insulin bolus effects in T1DM for in silico testing using hybrid approximation of physiological meal/insulin model
Sprache des Vortragstitels:
American Diabetes Association's 70th Scientific Sessions (ADA)
Sprache des Tagungstitel:
In silico testing is becoming the longer the more a sensible alternative instead of tests on animals for many applications, and in the case of T1DM a meal model (MM) has been accepted by the FDA for this use. This physiologically oriented model requires a set of parameters specific to the patient.
Two different treatments of T1DM are common, continuously insulin administration and single insulin injections. In the first case, the focus of this paper, it is possible to use a behavioural model to simplify the MM without a loss of quality of the simulated blood glucose behaviour. To this end, this paper proposes the use of a hybrid model (HM), which uses distribution functions instead of a compartment structure like the MM, to approximate the original MM. In the case of an insulin pump, the proposed method can be used to simulate the insulin peaks.
To test the result, 1 identification and 15 validation datasets have been created with the MM using a realistic and physiologically correct set of parameters for the MM, each of the length of 2 days. The input values are 0-120 g for CHO and 0-40 units for insulin. The inputs have been combined randomly, the time windows of the meals are 7:00-10:30 for breakfast, 11:00-16:00 for lunch and 17:00-21:00 for dinner. The mean fit value for the 15 validation dataset is 81.34±10.59%. Figure 1 shows the result of a validation dataset compared with the output of the MM: the loss of precision is only limited and very local, ands concerns for instance the second peak after the lunch, the important features for the use in diabetes therapy like the increases, the descents and the duration of the blood glucose after a meal are well reproduced.
While the MM requires 35 parameters, the HM requires only 8. This is especially important if the model is used inside an optimization loop.