(344c) Model-Based Control of Blood Glucose in Intensive Care Unit (ICU) Patients

Individuals who are critically ill may suffer from hyperglycemia (high blood glucose) and insulin resistance, even if they do not have diabetes. A healthy individual has a basal (steady-state) glucose concentration of approximately 80 mg/dl, but patients suffering from stress hyperglycemia can easily have blood glucose values greater than 200 mg/dl. A landmark study by Van den Berghe et al. (2001) showed that maintaining blood glucose below 110 mg/dl reduced overall in-hospital mortality by 34%, bloodstream infections by 46%, and acute renal failure by 41%. In addition, patients receiving intensive insulin therapy were less likely to require prolonged mechanical ventilation. It is thus becoming standard to monitor blood glucose in the ICU by taking blood samples every hour or two, depending on the stability of the blood glucose values. Table-lookup protocols are often used by the critical care nursing staff to determine insulin infusion rates, but a high-error rate naturally motivates a more automated procedure. Davidson et al. (2005) developed the Glucommander algorithm, which is a proportional-only control algorithm with a time-varying proportional gain (Bequette, 2007). There are several challenges to the development of a fully closed-loop glucose control for the ICU. The current generation of continuous glucose monitors (CGM) that measure interstitial fluid glucose and are becoming more widely used in ambulatory care of individuals with type 1 diabetes, are not reliable for use in the ICU; there are a number of continuous glucose sensors under development for ICU applications, however. One of the major challenges that we have tackled in this research effort is the wide variability in insulin sensitivity from individual-to-individual, as well as for specific individuals during the course of their hospital stay. We have extended a multiple model predictive control (Rao et al., 2003; Kuure-Kinsey and Bequette, 2009) approach to adapt to this time-varying insulin sensitivity.

A number of different models are available for studies of insulin-glucose dynamics in critical care (Parker and Doyle, 2001; Florian and Parker, 2005). In this work a three-state physiological model developed by Chase et al. (2006) is used for simulation studies based on 15 sets of model parameters (that is, a population of 15 distinct in silico subjects). In the first part of this presentation, insulin infusion rate is the only manipulated input to control the glucose concentration. While the focus is often on regulating blood glucose by infusing insulin, it should be recognized that ICU patients also have nutritional needs that are satisfied by both enteral (feeding tubes) and parenteral (IV) delivery (Wong et al., 2006). In the second part of our presentation, glucose feed rate is used as a manipulated input, in addition to insulin infusion. Our control algorithms assure that a specified nutrition rate (grams of glucose per day) is achieved over a long time scale (several hours to a day), yet vary the glucose infusion over the short time scale to improve glucose control. These ideas can be considered extensions of the habituating control strategy proposed by Henson et al. (1995), which uses an additional degree of freedom to improve disturbance rejection. The proposed multiple model predictive control strategy, using both insulin and glucose infusion, leads to better performance than fixed-model and single input control strategies.

Literature Cited

Bequette, B.W. ?Analysis of Algorithms for Intensive Care Unit Blood Glucose Control,? Diabetes Science and Technology, 1(6), 813-824 (2007). Chase JG, Shaw GM, Wong XW, Lotz T, Lin J, Hann CE. Model-based glycemic control in critical care-A review of the state of the possible. Biomed Signal Process Control. 1(1):3-21 (2006).

Davidson, P.C., R.D. Steed and B.W. Bode ?Glucommander: a computer-directed intravenuous insulin system shown to be safe, simple, and effective in 120,618 hours of operation,? Diabetes Care, 28(10), 2418-2423 (2005).

Florian Jr, J.A. and R.S. Parker. ?Empirical Modeling for Glucose Control in Diabetes and Critical Care.? European Journal of Control, 11(6), 616-616, 2005.

Henson, M.A., B.A. Ogunnaike and J.S. Schwaber ?Habituating control strategies for process control,? AIChE Journal, 41(3), 604-618 (1995).

Kuure-Kinsey, M. and B.W. Bequette ?Multiple Model Predictive Control of Nonlinear Systems,? in Nonlinear Model Predictive Control, Lecture Notes in Control and Information Sciences Vol. 384, pp. 153-165, L. Magni, D.M. Raimondo, F. Allgower (eds), Springer (2009).

Parker, R.S. and F.J. Doyle III. ?Control?relevant Modeling in Drug Delivery.? Advances in Drug Delivery Reviews, vol. 48, pp. 211-248, 2001.

Rao, R., C.C. Palerm, B. Aufderheide and B.W. Bequette ?Experimental Studies on Automated Regulation of Hemodynamic Variables,? IEEE Engineering in Medicine and Biology Magazine, 20(1), 24-38 (Jan/Feb, 2001).

Van den Berghe, G., P. Wouters, F. Weekers, C. Verwaest, F. Bruyninckx, M. Schetz, D. Vlasselaers, P. Ferdinande, P. Lauwers, and R. Gouillon, ?Intensive insulin therapy in critically ill patients,? New Eng. J. Med., 345(19), 1359-1367 (2001).

Wong, X.W., I. Singh-Levett, L.J. Hollingsworth, G.M. Shaw, C.E. Hann, T. Lotz, J. Lin, O.S.W. Wong, and J.G. Chase ?A Novel, model-based insulin and nutrition delivery controller for glycemic regulation in critically ill patients,? Diabetes Technology and Therapeutics, 8(2), 174-190 (2006).