Effect of Carbohydrate Counting Errors on Glycemic Control - A Hybrid in Silico Study

Many T1DM patients have difficulties to correctly estimate meal carbohydrates. These estimates are used for calculating meal boluses, but there is a lack of data on how estimation errors affect glycemic control. 6 days of data of a recent clinical trial with 37 patients are used for identifying carbohydrate-to-insulin-ratios (CIRs) and insulin-sensitivity-factors (ISFs) based on a previously published method, either using the correct carbohydrate amount or an estimated amount, affected by some random estimation error. On the 7th day the identified CIRs and ISFs are used to calculate the insulin doses based on estimated carbohydrate amounts, including estimation errors. A deviation based method is used to simulate the effect of carbohydrate counting errors on glycemic control. In the simulations a whole range of values is tested regarding estimation bias (between -20% and +10%) and uncertainty of estimation (between 0% and 60%). The figures show the time in hypoglycemia (thypo)and hyperglycemia (thyper) and the value of a combined cost function (V = 0.05* hypo+0.01*thyper). The left ones correspond to the case of CIRs and ISFs identified using the correct carbohydrate amounts, whereas for the right ones CIRs and ISFs identified based on the estimated carb amount are used, affected by the same bias and uncertainty that is afterwards used in the simulations. Uncertainty in carb estimates always leads to inferior glycemic control, whereas an estimation bias hardly affects the results since it is usually implicitly included in the CIRs and ISFs used by the patients.