In Silico Monte Carlo Virtual Trials of a Model-Based Adaptive Type 1 Diabetes Mellitus Control Protocol

Objective: To test an in silico Type 1 diabetes control protocol while accounting for realistic physiological variability, and measurement and delivery error Methods: A Monte Carlo (MC) analysis uses clinically reported variations in physiological parameters, subcutaneous insulin absorption and delivery, nutritional carbohydrate counting intake, and SMBG error to test robustness. The model-based protocol is repeatedly tested on a 40 patient virtual cohort over 1.4M patient hours. The analysis is repeated for SMBG frequency of 2-10/day. Long term HbA1c is estimated from clinically reported formulas to assess performance. Results: The protocol controlled 100% of the cohort to ADA recommended HbA1c with SMBG frequency of 6/day. Peak control is achieved at SMBG frequency of 8/day. A small but significant decrease in time in the 72-144mg/dL band and consequent increase in mild and severe hypoglycaemia occurs at SMBG frequency of 10/day. Time spent in the 72-108mg/dL band is not significantly different to a no error and no variability simulation. Cohort HbA1c is reduced for all SMBG frequencies. Hypoglycaemia increases over the no error simulation, as expected. The difference in 95% confidence band for time in severe (=54mg/dL) and mild (=71mg/dL) hypoglycaemia spans an acceptable [1-6%] or 0.24-1.44hours/day versus the no error simulation for 6/day SMBG frequency. Conclusions: A MC simulation tool predicts long-term glycaemic control outcomes to test an adaptive control protocol in conditions of realistic variability and error. The protocol is shown to be robust, remaining effective and safe from hypoglycaemia compared to perfect no error or variability simulations, and clinical cohort control data. The protocol utilises the most commonly used forms of intervention (SMBG and MDI) and is thus applicable for most T1DM individuals