Specific validation analysis of stochastic ICING model based estimation of insulin sensitivity profile using clinical data

摘要

In this paper the performance of an insulin sensitivity profile estimation method is analysed from the the aspects of clinical applicability. In our previous studies it has been demonstrated that the grey box variant of the ICING (Intensive Control Insulin-Nutrition-Glucose) model may be successfully applied to improve the estimation accuracy of the insulin sensitivity profile of intensive care patients. This may enable the application of the model in the clinical treatment, especially in tight glycaemic control. The sensitivity estimation accuracy itself and the accuracy improvement compared to previous methods are highly variable, it strongly depends on the range of blood glucose concentration. In the present study the insulin sensitivity estimation is analysed from the clinically relevant aspects. Modelling error represented by the difference of measured and computed the blood glucose concentration was considered on the total glucose concentration range (measured in mmol/l) - 0 <; cG <; 20 - of the cohort data set as well as on its four subregions, namely hypoglycaemic (0 <; cG ≤ 4), normoglycaemic (4 <; cG ≤ 6.5), mild-hyperglycaemic (6.5 <; cG ≤ 10), and severe-hyperglycaemic (10 <; cG ≤ 20). The results of the computations indicate that the SDE model was significantly better in the normoglycaemic and mild-hyperglycaemic ranges, somewhat better in the hypoglycaemic range and nearly the same in the severe-hyperglycaemic range. The 97 % of all of the concentration values were in the normoglycaemic and the mild-hyperglycaemic range (5841 values), which amplifies our statement in these ranges, but further study is necessary the ensure the verdict for the hypoglycaemic and severe-hyperglycaemic ranges.