In vivo comparison of different algorithms for the artificial beta-cell

Using an extracorporeal artificial beta-cell in chronically diabetic dogs, the effects of four different mathematical models of glucose-controlled insulin dosage were compared: the Biostator algorithm (quadratic equation), Toronto algorithm (hyperbolic tangent function), Karlsburg algorithm (modified first-order derivative controller), and Ilmenau algorithm (second-order linear difference equation). The constants of all formulas implemented for the artificial beta-cell were obtained by regression analysis of paired blood glucose and plasma insulin data from normal control animals. Thus, they were biologically equivalent for all formulas. The patterns of blood glucose, insulin doses, and plasma insulin before, during, and after an intravenous glucose infusion test performed during the glucose-controlled insulin infusion showed no significant differences between the experimental groups subjected to the different algorithms. However, in no case were really normal blood glucose response curves restored by the artificial beta-cell. This might be due, first, to the fact that the algorithm parameters were not adapted to the actual individual insulin responsiveness, second, to the unphysiological peripheral venous route of insulin administration, and, third, to the lack of appropriate adaptation of the animals to normoglycemia.