Automated optimal glycaemic control using a physiology based pharmacokinetic, pharmacodynamic model

After decades of research, Automated Glucose Control (AGC) is still out of reach for everyday control of blood glucose. The inter- and intra-individual variability of glucose dynamics largely arising from variability in insulin absorption, distribution, and action, and related physiological lag-times remain a core problem in the development of suitable control algorithms. Over the years, model predictive control (MPC) has established itself as the gold standard in AGC systems in research. Models of glucose metabolism are a core element of MPC control schemes. The standard two- or three-compartmental models, i.e. the “Minimal-Model” [1], represent little biological detail, hampering the integration of multi-scale data, thus confining capabilities of model extrapolation. To overcome remaining challenges, a new approach to MPC AGC is developed here. The MPC uses, for the first time, an individualizable generic whole-body physiology-based pharmacokinetic/pharmacodynamic (PBPK/PD) model of the glucose-insulin-glucagon regulatory system. The model reflects detailed physiological properties of healthy populations and type 1 diabetes individuals expressed in the respective parameterizations. The model features a detailed representation of absorption models for oral glucose, subcutaneous insulin and glucagon, and an insulin receptor model relating pharmacokinetic properties to pharmacodynamic effects. Model development and validation is based on literature data. The quality of predictions is high and captures relevant observed inter- and intra-individual variability, thus improving model long-term predictions. This significantly strengthens the rationale for the use of MPC. To increase robustness vs. uncertainties (closed-loop stability), model kernels were updated with growing patient data and the MPC was integrated in a control cascade with a proportional, integrative, derivative (PID) based offset-control. Both, model and control algorithm, were validated and evaluated within an in-silico environment before testing the control approach within two 30-h clinical trials. The trials were each conducted in ten subjects with type 1 diabetes without endogenous insulin secretion. Blood glucose was controlled by subcutaneous delivery of insulin based on plasma glucose (PG, in trial #1) and continuous blood glucose monitors (CGMs, subcutaneous sampling, trial #2) measurements in 15 min intervals. Meal information, but no priming bolus (pre-meal insulin), was given to the controller at start of each meal. For the first clinical trial, the overall mean (n=10) PG was 156 mg/dL, with 74% time of PG values in the target range of 70–180 mg/dL. With 2 incidents during 240 h of closed-loop control, hypoglycemia (PG < 60 mg/dL) was rare. During nighttime control, prior to model adaptation (adaptation was slow if successful at all), mean PG was elevated (149 mg/dL, with 38% time in target 70–140 mg/dL). For the second clinical trial, control performance improved significantly due to an improved workflow and faster (earlier) model adaptation with an overall mean (n=10) PG of 127 mg/dL, with 76% time of PG values in the target range of 70–180 mg/dL. With 9 incidents during 240 h of closed-loop control, hypoglycemia (PG < 60 mg/dL) was slightly increased. Nighttime control improved the most with a mean PG exactly on target (110 mg/dL, with 78% time in target 70–140 mg/dL). Retrospective analysis of insulin and glucagon measurements collected during the trial, revealed significant glucagon surges, which were observed postprandial and coincided with severe morning insulin resistance for some patients. Whereas a consistent interpretation of the observed behavior is outstanding, the modeling framework allowed a structural mode-of-action evaluation to shed new light on the role of glucagon and nutrition (i.e. coffee) in the “dawn-effect” in Diabetes. This work shows that large-scale in-silico models of the glucose metabolism can provide a framework to improve diabetes research, the development of automatic control strategies for diabetes and ultimately every day diabetes management. The algorithm for the integrated closed-loop control system was benchmarked both, within in-silico clinical trials as well as within clinical feasibility studies. Once the relevance of (postprandial) glucagon in T1DM has been analyzed, fully understood and captured by PBPK/PD modeling, future trials testing the improved system seem very promising