Simulation and initial proof-of-concept validation of a glycaemic regulation algorithm in critical care

Glycaemic control can reduce mortality in intensive care by 45%. A model-based control algorithm utilising insulin and nutritional glucose inputs is presented. Simulated long-term gluco-regulatory trials with the virtual-patient method using retrospective ICU patient data (n=19) validate the approach. Two short-term proof-of-concept clinical trials test glycaemic predictive capability and the ability to adapt to patient condition. In simulation, a 312% increase in time spent in the 4-6 mmol/l euglycaemic band compared to retrospective patient data is recorded while feeding 39% greater nutrition. In the clinical trials, mean target error was 8.7% with hourly prediction horizon. About 61% of targets were achieved within ±5%, and only two targets had errors >20%, occurring during rapid deterioration in patient condition. Overall, the protocol demonstrated effective glycaemic control across the selected cohort in simulation and in highly dynamic patient conditions observed in initial proof-of-concept clinical trials.