Evaluation of a model predictive control algorithm using time-variant sampling to establish tight glycaemic control in clinical practice

Introduction

Tight glycaemic control (TGC) in critically ill patients significantly improves clinical outcome. Even with increased workload for ICU nursing staff, targets for TGC are often not achieved. The aim of the present study was to evaluate in clinical practice a model predictive control algorithm (MPC) using time-variant sampling, which will be used in a fully automated insulin titration system (CLINICIP system).

Methods

This was an open randomized controlled clinical study. Fifty mechanically ventilated medical ICU patients were included for a study period of 72 hours. Patients were randomized either to a control group, treated by an insulin algorithm as routinely used in the ICU, or to the MPC group, using a laptop-based fully automated MPC algorithm. The target range for blood glucose (BG) was 4.4–6.1 mM for both groups. Efficacy was assessed by calculating the median BG, hyperglycaemic index (HGI) and BG sampling interval. Safety was assessed by the number of hypoglycaemic BG measurements < 2.2 mM.

Results

Patients were included for 72 (69–73) hours (median (IQR)) in the control group and 71 (70–73) hours in the MPC group. The median BG and HGI were significantly lower in MPC vs control patients (see Table 1). A single BG measurement < 2.2 mM was detected in the MPC group vs 0 in the control group. The sampling frequency was significantly higher in the MPC group.

Table 1 abstract P138)

Conclusion

The use of MPC improved BG and the HGI. This improvement was accompanied by an increased sampling frequency. MPC with time-variant sampling is a reliable tool for the implementation of TGC in patients in the medical ICU.